System.Data.SQLite
Artifact Content
Not logged in

Artifact 7ddfde6395ade2914485d2e73160b346e08a9e3e:


/******************************************************************************
** This file is an amalgamation of many separate C source files from SQLite
** version 3.4.1.  By combining all the individual C code files into this 
** single large file, the entire code can be compiled as a one translation
** unit.  This allows many compilers to do optimizations that would not be
** possible if the files were compiled separately.  Performance improvements
** of 5% are more are commonly seen when SQLite is compiled as a single
** translation unit.
**
** This file is all you need to compile SQLite.  To use SQLite in other
** programs, you need this file and the "sqlite3.h" header file that defines
** the programming interface to the SQLite library.  (If you do not have 
** the "sqlite3.h" header file at hand, you will find a copy in the first
** 2702 lines past this header comment.)  Additional code files may be
** needed if you want a wrapper to interface SQLite with your choice of
** programming language.  The code for the "sqlite3" command-line shell
** is also in a separate file.  This file contains only code for the core
** SQLite library.
**
** This amalgamation was generated on 2007-07-20 11:05:39 UTC.
*/
#define SQLITE_AMALGAMATION 1
#ifndef SQLITE_PRIVATE
# define SQLITE_PRIVATE static
#endif
#ifndef SQLITE_API
# define SQLITE_API
#endif
/************** Begin file sqlite3.h *****************************************/
/*
** 2001 September 15
**
** The author disclaims copyright to this source code.  In place of
** a legal notice, here is a blessing:
**
**    May you do good and not evil.
**    May you find forgiveness for yourself and forgive others.
**    May you share freely, never taking more than you give.
**
*************************************************************************
** This header file defines the interface that the SQLite library
** presents to client programs.  If a C-function, structure, datatype,
** or constant definition does not appear in this file, then it is
** not a published API of SQLite, is subject to change without
** notice, and should not be referenced by programs that use SQLite.
**
** Some of the definitions that are in this file are marked as
** "experimental".  Experimental interfaces are normally new
** features recently added to SQLite.  We do not anticipate changes 
** to experimental interfaces but reserve to make minor changes if
** experience from use "in the wild" suggest such changes are prudent.
**
** The official C-language API documentation for SQLite is derived
** from comments in this file.  This file is the authoritative source
** on how SQLite interfaces are suppose to operate.
**
** The name of this file under configuration management is "sqlite.h.in".
** The makefile makes some minor changes to this file (such as inserting
** the version number) and changes its name to "sqlite3.h" as
** part of the build process.
**
** @(#) $Id: sqlite3.c,v 1.4 2007/07/21 08:51:43 rmsimpson Exp $
*/
#ifndef _SQLITE3_H_
#define _SQLITE3_H_
#include <stdarg.h>     /* Needed for the definition of va_list */

/*
** Make sure we can call this stuff from C++.
*/
#if 0
extern "C" {
#endif

/*
** Make sure these symbols where not defined by some previous header
** file.
*/
#ifdef SQLITE_VERSION
# undef SQLITE_VERSION
#endif
#ifdef SQLITE_VERSION_NUMBER
# undef SQLITE_VERSION_NUMBER
#endif

/*
** CAPI3REF: Compile-Time Library Version Numbers
**
** The version of the SQLite library is contained in the sqlite3.h
** header file in a #define named SQLITE_VERSION.  The SQLITE_VERSION
** macro resolves to a string constant.
**
** The format of the version string is "X.Y.Z", where
** X is the major version number, Y is the minor version number and Z
** is the release number.  The X.Y.Z might be followed by "alpha" or "beta".
** For example "3.1.1beta".
**
** The X value is always 3 in SQLite.  The X value only changes when
** backwards compatibility is broken and we intend to never break
** backwards compatibility.  The Y value only changes when
** there are major feature enhancements that are forwards compatible
** but not backwards compatible.  The Z value is incremented with
** each release but resets back to 0 when Y is incremented.
**
** The SQLITE_VERSION_NUMBER is an integer with the value 
** (X*1000000 + Y*1000 + Z). For example, for version "3.1.1beta", 
** SQLITE_VERSION_NUMBER is set to 3001001. To detect if they are using 
** version 3.1.1 or greater at compile time, programs may use the test 
** (SQLITE_VERSION_NUMBER>=3001001).
**
** See also: [sqlite3_libversion()] and [sqlite3_libversion_number()].
*/
#define SQLITE_VERSION         "3.4.1"
#define SQLITE_VERSION_NUMBER 3004001

/*
** CAPI3REF: Run-Time Library Version Numbers
**
** These routines return values equivalent to the header constants
** [SQLITE_VERSION] and [SQLITE_VERSION_NUMBER].  The values returned
** by this routines should only be different from the header values
** if you compile your program using an sqlite3.h header from a
** different version of SQLite that the version of the library you
** link against.
**
** The sqlite3_version[] string constant contains the text of the
** [SQLITE_VERSION] string.  The sqlite3_libversion() function returns
** a poiner to the sqlite3_version[] string constant.  The function
** is provided for DLL users who can only access functions and not
** constants within the DLL.
*/
extern const char sqlite3_version[];
SQLITE_API const char *sqlite3_libversion(void);
int sqlite3_libversion_number(void);

/*
** CAPI3REF: Database Connection Handle
**
** Each open SQLite database is represented by pointer to an instance of the
** opaque structure named "sqlite3".  It is useful to think of an sqlite3
** pointer as an object.  The [sqlite3_open] interface is its constructor
** and [sqlite3_close] is its destructor.  There are many other interfaces
** (such as [sqlite3_prepare_v2], [sqlite3_create_function], and
** [sqlite3_busy_timeout] to name but three) that are methods on this
** object.
*/
typedef struct sqlite3 sqlite3;


/*
** CAPI3REF: 64-Bit Integer Types
**
** Some compilers do not support the "long long" datatype.  So we have
** to do compiler-specific typedefs for 64-bit signed and unsigned integers.
**
** Many SQLite interface functions require a 64-bit integer arguments.
** Those interfaces are declared using this typedef.
*/
#ifdef SQLITE_INT64_TYPE
  typedef SQLITE_INT64_TYPE sqlite_int64;
  typedef unsigned SQLITE_INT64_TYPE sqlite_uint64;
#elif defined(_MSC_VER) || defined(__BORLANDC__)
  typedef __int64 sqlite_int64;
  typedef unsigned __int64 sqlite_uint64;
#else
  typedef long long int sqlite_int64;
  typedef unsigned long long int sqlite_uint64;
#endif

/*
** If compiling for a processor that lacks floating point support,
** substitute integer for floating-point
*/
#ifdef SQLITE_OMIT_FLOATING_POINT
# define double sqlite_int64
#endif

/*
** CAPI3REF: Closing A Database Connection
**
** Call this function with a pointer to a structure that was previously
** returned from [sqlite3_open()] and the corresponding database will by
** closed.
**
** All SQL statements prepared using [sqlite3_prepare_v2()] or
** [sqlite3_prepare16_v2()] must be destroyed using [sqlite3_finalize()]
** before this routine is called. Otherwise, SQLITE_BUSY is returned and the
** database connection remains open.
*/
SQLITE_API int sqlite3_close(sqlite3 *);

/*
** The type for a callback function.
** This is legacy and deprecated.  It is included for historical
** compatibility and is not documented.
*/
typedef int (*sqlite3_callback)(void*,int,char**, char**);

/*
** CAPI3REF: One-Step Query Execution Interface
**
** This interface is used to do a one-time evaluatation of zero
** or more SQL statements.  UTF-8 text of the SQL statements to
** be evaluted is passed in as the second parameter.  The statements
** are prepared one by one using [sqlite3_prepare()], evaluated
** using [sqlite3_step()], then destroyed using [sqlite3_finalize()].
**
** If one or more of the SQL statements are queries, then
** the callback function specified by the 3rd parameter is
** invoked once for each row of the query result.  This callback
** should normally return 0.  If the callback returns a non-zero
** value then the query is aborted, all subsequent SQL statements
** are skipped and the sqlite3_exec() function returns the SQLITE_ABORT.
**
** The 4th parameter to this interface is an arbitrary pointer that is
** passed through to the callback function as its first parameter.
**
** The 2nd parameter to the callback function is the number of
** columns in the query result.  The 3rd parameter to the callback
** is an array of strings holding the values for each column
** as extracted using [sqlite3_column_text()].
** The 4th parameter to the callback is an array of strings
** obtained using [sqlite3_column_name()] and holding
** the names of each column.
**
** The callback function may be NULL, even for queries.  A NULL
** callback is not an error.  It just means that no callback
** will be invoked.
**
** If an error occurs while parsing or evaluating the SQL (but
** not while executing the callback) then an appropriate error
** message is written into memory obtained from [sqlite3_malloc()] and
** *errmsg is made to point to that message.  The calling function
** is responsible for freeing the memory that holds the error
** message.   Use [sqlite3_free()] for this.  If errmsg==NULL,
** then no error message is ever written.
**
** The return value is is SQLITE_OK if there are no errors and
** some other [SQLITE_OK | return code] if there is an error.  
** The particular return value depends on the type of error. 
**
*/
SQLITE_API int sqlite3_exec(
  sqlite3*,                                  /* An open database */
  const char *sql,                           /* SQL to be evaluted */
  int (*callback)(void*,int,char**,char**),  /* Callback function */
  void *,                                    /* 1st argument to callback */
  char **errmsg                              /* Error msg written here */
);

/*
** CAPI3REF: Result Codes
** KEYWORDS: SQLITE_OK
**
** Many SQLite functions return an integer result code from the set shown
** above in order to indicates success or failure.
**
** The result codes above are the only ones returned by SQLite in its
** default configuration.  However, the [sqlite3_extended_result_codes()]
** API can be used to set a database connectoin to return more detailed
** result codes.
**
** See also: [SQLITE_IOERR_READ | extended result codes]
**
*/
#define SQLITE_OK           0   /* Successful result */
/* beginning-of-error-codes */
#define SQLITE_ERROR        1   /* SQL error or missing database */
#define SQLITE_INTERNAL     2   /* NOT USED. Internal logic error in SQLite */
#define SQLITE_PERM         3   /* Access permission denied */
#define SQLITE_ABORT        4   /* Callback routine requested an abort */
#define SQLITE_BUSY         5   /* The database file is locked */
#define SQLITE_LOCKED       6   /* A table in the database is locked */
#define SQLITE_NOMEM        7   /* A malloc() failed */
#define SQLITE_READONLY     8   /* Attempt to write a readonly database */
#define SQLITE_INTERRUPT    9   /* Operation terminated by sqlite3_interrupt()*/
#define SQLITE_IOERR       10   /* Some kind of disk I/O error occurred */
#define SQLITE_CORRUPT     11   /* The database disk image is malformed */
#define SQLITE_NOTFOUND    12   /* NOT USED. Table or record not found */
#define SQLITE_FULL        13   /* Insertion failed because database is full */
#define SQLITE_CANTOPEN    14   /* Unable to open the database file */
#define SQLITE_PROTOCOL    15   /* NOT USED. Database lock protocol error */
#define SQLITE_EMPTY       16   /* Database is empty */
#define SQLITE_SCHEMA      17   /* The database schema changed */
#define SQLITE_TOOBIG      18   /* String or BLOB exceeds size limit */
#define SQLITE_CONSTRAINT  19   /* Abort due to contraint violation */
#define SQLITE_MISMATCH    20   /* Data type mismatch */
#define SQLITE_MISUSE      21   /* Library used incorrectly */
#define SQLITE_NOLFS       22   /* Uses OS features not supported on host */
#define SQLITE_AUTH        23   /* Authorization denied */
#define SQLITE_FORMAT      24   /* Auxiliary database format error */
#define SQLITE_RANGE       25   /* 2nd parameter to sqlite3_bind out of range */
#define SQLITE_NOTADB      26   /* File opened that is not a database file */
#define SQLITE_ROW         100  /* sqlite3_step() has another row ready */
#define SQLITE_DONE        101  /* sqlite3_step() has finished executing */
/* end-of-error-codes */

/*
** CAPI3REF: Extended Result Codes
**
** In its default configuration, SQLite API routines return one of 26 integer
** result codes described at result-codes.  However, experience has shown that
** many of these result codes are too course-grained.  They do not provide as
** much information about problems as users might like.  In an effort to
** address this, newer versions of SQLite (version 3.3.8 and later) include
** support for additional result codes that provide more detailed information
** about errors.  The extended result codes are enabled (or disabled) for 
** each database
** connection using the [sqlite3_extended_result_codes()] API.
** 
** Some of the available extended result codes are listed above.
** We expect the number of extended result codes will be expand
** over time.  Software that uses extended result codes should expect
** to see new result codes in future releases of SQLite.
** 
** The symbolic name for an extended result code always contains a related
** primary result code as a prefix.  Primary result codes contain a single
** "_" character.  Extended result codes contain two or more "_" characters.
** The numeric value of an extended result code can be converted to its
** corresponding primary result code by masking off the lower 8 bytes.
**
** The SQLITE_OK result code will never be extended.  It will always
** be exactly zero.
*/
#define SQLITE_IOERR_READ          (SQLITE_IOERR | (1<<8))
#define SQLITE_IOERR_SHORT_READ    (SQLITE_IOERR | (2<<8))
#define SQLITE_IOERR_WRITE         (SQLITE_IOERR | (3<<8))
#define SQLITE_IOERR_FSYNC         (SQLITE_IOERR | (4<<8))
#define SQLITE_IOERR_DIR_FSYNC     (SQLITE_IOERR | (5<<8))
#define SQLITE_IOERR_TRUNCATE      (SQLITE_IOERR | (6<<8))
#define SQLITE_IOERR_FSTAT         (SQLITE_IOERR | (7<<8))
#define SQLITE_IOERR_UNLOCK        (SQLITE_IOERR | (8<<8))
#define SQLITE_IOERR_RDLOCK        (SQLITE_IOERR | (9<<8))
#define SQLITE_IOERR_DELETE        (SQLITE_IOERR | (10<<8))
#define SQLITE_IOERR_BLOCKED       (SQLITE_IOERR | (11<<8))

/*
** CAPI3REF: Enable Or Disable Extended Result Codes
**
** This routine enables or disables the
** [SQLITE_IOERR_READ | extended result codes] feature.
** By default, SQLite API routines return one of only 26 integer
** [SQLITE_OK | result codes].  When extended result codes
** are enabled by this routine, the repetoire of result codes can be
** much larger and can (hopefully) provide more detailed information
** about the cause of an error.
**
** The second argument is a boolean value that turns extended result
** codes on and off.  Extended result codes are off by default for
** backwards compatibility with older versions of SQLite.
*/
int sqlite3_extended_result_codes(sqlite3*, int onoff);

/*
** CAPI3REF: Last Insert Rowid
**
** Each entry in an SQLite table has a unique 64-bit signed integer key
** called the "rowid". The rowid is always available as an undeclared
** column named ROWID, OID, or _ROWID_.  If the table has a column of
** type INTEGER PRIMARY KEY then that column is another an alias for the
** rowid.
**
** This routine returns the rowid of the most recent INSERT into
** the database from the database connection given in the first 
** argument.  If no inserts have ever occurred on this database
** connection, zero is returned.
**
** If an INSERT occurs within a trigger, then the rowid of the
** inserted row is returned by this routine as long as the trigger
** is running.  But once the trigger terminates, the value returned
** by this routine reverts to the last value inserted before the
** trigger fired.
*/
sqlite_int64 sqlite3_last_insert_rowid(sqlite3*);

/*
** CAPI3REF: Count The Number Of Rows Modified
**
** This function returns the number of database rows that were changed
** (or inserted or deleted) by the most recent SQL statement.  Only
** changes that are directly specified by the INSERT, UPDATE, or
** DELETE statement are counted.  Auxiliary changes caused by
** triggers are not counted.  Use the [sqlite3_total_changes()] function
** to find the total number of changes including changes caused by triggers.
**
** Within the body of a trigger, the sqlite3_changes() interface can be
** called to find the number of
** changes in the most recently completed INSERT, UPDATE, or DELETE
** statement within the body of the trigger.
**
** All changes are counted, even if they were later undone by a
** ROLLBACK or ABORT.  Except, changes associated with creating and
** dropping tables are not counted.
**
** If a callback invokes [sqlite3_exec()] or [sqlite3_step()] recursively,
** then the changes in the inner, recursive call are counted together
** with the changes in the outer call.
**
** SQLite implements the command "DELETE FROM table" without a WHERE clause
** by dropping and recreating the table.  (This is much faster than going
** through and deleting individual elements from the table.)  Because of
** this optimization, the change count for "DELETE FROM table" will be
** zero regardless of the number of elements that were originally in the
** table. To get an accurate count of the number of rows deleted, use
** "DELETE FROM table WHERE 1" instead.
*/
SQLITE_API int sqlite3_changes(sqlite3*);

/*
** CAPI3REF: Total Number Of Rows Modified
***
** This function returns the number of database rows that have been
** modified by INSERT, UPDATE or DELETE statements since the database handle
** was opened. This includes UPDATE, INSERT and DELETE statements executed
** as part of trigger programs. All changes are counted as soon as the
** statement that makes them is completed (when the statement handle is
** passed to [sqlite3_reset()] or [sqlite_finalise()]).
**
** See also the [sqlite3_change()] interface.
**
** SQLite implements the command "DELETE FROM table" without a WHERE clause
** by dropping and recreating the table.  (This is much faster than going
** through and deleting individual elements form the table.)  Because of
** this optimization, the change count for "DELETE FROM table" will be
** zero regardless of the number of elements that were originally in the
** table. To get an accurate count of the number of rows deleted, use
** "DELETE FROM table WHERE 1" instead.
*/
int sqlite3_total_changes(sqlite3*);

/*
** CAPI3REF: Interrupt A Long-Running Query
**
** This function causes any pending database operation to abort and
** return at its earliest opportunity.  This routine is typically
** called in response to a user action such as pressing "Cancel"
** or Ctrl-C where the user wants a long query operation to halt
** immediately.
**
** It is safe to call this routine from a thread different from the
** thread that is currently running the database operation.
**
** The SQL operation that is interrupted will return [SQLITE_INTERRUPT].
** If an interrupted operation was an update that is inside an
** explicit transaction, then the entire transaction will be rolled
** back automatically.
*/
SQLITE_API void sqlite3_interrupt(sqlite3*);

/*
** CAPI3REF: Determine If An SQL Statement Is Complete
**
** These functions return true if the given input string comprises
** one or more complete SQL statements. For the sqlite3_complete() call,
** the parameter must be a nul-terminated UTF-8 string. For
** sqlite3_complete16(), a nul-terminated machine byte order UTF-16 string
** is required.
**
** These routines are useful for command-line input to determine if the
** currently entered text forms one or more complete SQL statements or
** if additional input is needed before sending the statements into
** SQLite for parsing. The algorithm is simple.  If the 
** last token other than spaces and comments is a semicolon, then return 
** true.  Actually, the algorithm is a little more complicated than that
** in order to deal with triggers, but the basic idea is the same:  the
** statement is not complete unless it ends in a semicolon.
*/
SQLITE_API int sqlite3_complete(const char *sql);
SQLITE_API int sqlite3_complete16(const void *sql);

/*
** CAPI3REF: Register A Callback To Handle SQLITE_BUSY Errors
**
** This routine identifies a callback function that might be invoked
** whenever an attempt is made to open a database table 
** that another thread or process has locked.
** If the busy callback is NULL, then [SQLITE_BUSY]
** (or sometimes [SQLITE_IOERR_BLOCKED])
** is returned immediately upon encountering the lock.
** If the busy callback is not NULL, then the
** callback will be invoked with two arguments.  The
** first argument to the handler is a copy of the void* pointer which
** is the third argument to this routine.  The second argument to
** the handler is the number of times that the busy handler has
** been invoked for this locking event. If the
** busy callback returns 0, then no additional attempts are made to
** access the database and [SQLITE_BUSY] or [SQLITE_IOERR_BLOCKED] is returned.
** If the callback returns non-zero, then another attempt is made to open the
** database for reading and the cycle repeats.
**
** The presence of a busy handler does not guarantee that
** it will be invoked when there is lock contention.
** If SQLite determines that invoking the busy handler could result in
** a deadlock, it will return [SQLITE_BUSY] instead.
** Consider a scenario where one process is holding a read lock that
** it is trying to promote to a reserved lock and
** a second process is holding a reserved lock that it is trying
** to promote to an exclusive lock.  The first process cannot proceed
** because it is blocked by the second and the second process cannot
** proceed because it is blocked by the first.  If both processes
** invoke the busy handlers, neither will make any progress.  Therefore,
** SQLite returns [SQLITE_BUSY] for the first process, hoping that this
** will induce the first process to release its read lock and allow
** the second process to proceed.
**
** The default busy callback is NULL.
**
** The [SQLITE_BUSY] error is converted to [SQLITE_IOERR_BLOCKED] when
** SQLite is in the middle of a large transaction where all the
** changes will not fit into the in-memory cache.  SQLite will
** already hold a RESERVED lock on the database file, but it needs
** to promote this lock to EXCLUSIVE so that it can spill cache
** pages into the database file without harm to concurrent
** readers.  If it is unable to promote the lock, then the in-memory
** cache will be left in an inconsistent state and so the error
** code is promoted from the relatively benign [SQLITE_BUSY] to
** the more severe [SQLITE_IOERR_BLOCKED].  This error code promotion
** forces an automatic rollback of the changes. See the
** <a href="http://www.sqlite.org/cvstrac/wiki?p=CorruptionFollowingBusyError">
** CorruptionFollowingBusyError</a> wiki page for a discussion of why
** this is important.
**	
** Sqlite is re-entrant, so the busy handler may start a new query. 
** (It is not clear why anyone would every want to do this, but it
** is allowed, in theory.)  But the busy handler may not close the
** database.  Closing the database from a busy handler will delete 
** data structures out from under the executing query and will 
** probably result in a segmentation fault or other runtime error.
**
** There can only be a single busy handler defined for each database
** connection.  Setting a new busy handler clears any previous one.
** Note that calling [sqlite3_busy_timeout()] will also set or clear
** the busy handler.
*/
int sqlite3_busy_handler(sqlite3*, int(*)(void*,int), void*);

/*
** CAPI3REF: Set A Busy Timeout
**
** This routine sets a busy handler that sleeps for a while when a
** table is locked.  The handler will sleep multiple times until 
** at least "ms" milliseconds of sleeping have been done.  After
** "ms" milliseconds of sleeping, the handler returns 0 which
** causes [sqlite3_step()] to return [SQLITE_BUSY] or [SQLITE_IOERR_BLOCKED].
**
** Calling this routine with an argument less than or equal to zero
** turns off all busy handlers.
**
** There can only be a single busy handler for a particular database
** connection.  If another busy handler was defined  
** (using [sqlite3_busy_handler()]) prior to calling
** this routine, that other busy handler is cleared.
*/
int sqlite3_busy_timeout(sqlite3*, int ms);

/*
** CAPI3REF: Convenience Routines For Running Queries
**
** This next routine is a convenience wrapper around [sqlite3_exec()].
** Instead of invoking a user-supplied callback for each row of the
** result, this routine remembers each row of the result in memory
** obtained from [sqlite3_malloc()], then returns all of the result after the
** query has finished. 
**
** As an example, suppose the query result where this table:
**
** <pre>
**        Name        | Age
**        -----------------------
**        Alice       | 43
**        Bob         | 28
**        Cindy       | 21
** </pre>
**
** If the 3rd argument were &azResult then after the function returns
** azResult will contain the following data:
**
** <pre>
**        azResult[0] = "Name";
**        azResult[1] = "Age";
**        azResult[2] = "Alice";
**        azResult[3] = "43";
**        azResult[4] = "Bob";
**        azResult[5] = "28";
**        azResult[6] = "Cindy";
**        azResult[7] = "21";
** </pre>
**
** Notice that there is an extra row of data containing the column
** headers.  But the *nrow return value is still 3.  *ncolumn is
** set to 2.  In general, the number of values inserted into azResult
** will be ((*nrow) + 1)*(*ncolumn).
**
** After the calling function has finished using the result, it should 
** pass the result data pointer to sqlite3_free_table() in order to 
** release the memory that was malloc-ed.  Because of the way the 
** [sqlite3_malloc()] happens, the calling function must not try to call 
** [sqlite3_free()] directly.  Only [sqlite3_free_table()] is able to release 
** the memory properly and safely.
**
** The return value of this routine is the same as from [sqlite3_exec()].
*/
int sqlite3_get_table(
  sqlite3*,              /* An open database */
  const char *sql,       /* SQL to be executed */
  char ***resultp,       /* Result written to a char *[]  that this points to */
  int *nrow,             /* Number of result rows written here */
  int *ncolumn,          /* Number of result columns written here */
  char **errmsg          /* Error msg written here */
);
void sqlite3_free_table(char **result);

/*
** CAPI3REF: Formatted String Printing Functions
**
** These routines are workalikes of the "printf()" family of functions
** from the standard C library.
**
** The sqlite3_mprintf() and sqlite3_vmprintf() routines write their
** results into memory obtained from [sqlite_malloc()].
** The strings returned by these two routines should be
** released by [sqlite3_free()].  Both routines return a
** NULL pointer if [sqlite3_malloc()] is unable to allocate enough
** memory to hold the resulting string.
**
** In sqlite3_snprintf() routine is similar to "snprintf()" from
** the standard C library.  The result is written into the
** buffer supplied as the second parameter whose size is given by
** the first parameter.  Note that the order of the
** first two parameters is reversed from snprintf().  This is an
** historical accident that cannot be fixed without breaking
** backwards compatibility.  Note also that sqlite3_snprintf()
** returns a pointer to its buffer instead of the number of
** characters actually written into the buffer.  We admit that
** the number of characters written would be a more useful return
** value but we cannot change the implementation of sqlite3_snprintf()
** now without breaking compatibility.
**
** As long as the buffer size is greater than zero, sqlite3_snprintf()
** guarantees that the buffer is always zero-terminated.  The first
** parameter "n" is the total size of the buffer, including space for
** the zero terminator.  So the longest string that can be completely
** written will be n-1 characters.
**
** These routines all implement some additional formatting
** options that are useful for constructing SQL statements.
** All of the usual printf formatting options apply.  In addition, there
** is are "%q" and "%Q" options.
**
** The %q option works like %s in that it substitutes a null-terminated
** string from the argument list.  But %q also doubles every '\'' character.
** %q is designed for use inside a string literal.  By doubling each '\''
** character it escapes that character and allows it to be inserted into
** the string.
**
** For example, so some string variable contains text as follows:
**
** <blockquote><pre>
**  char *zText = "It's a happy day!";
** </pre></blockquote>
**
** One can use this text in an SQL statement as follows:
**
** <blockquote><pre>
**  char *zSQL = sqlite3_mprintf("INSERT INTO table VALUES('%q')", zText);
**  sqlite3_exec(db, zSQL, 0, 0, 0);
**  sqlite3_free(zSQL);
** </pre></blockquote>
**
** Because the %q format string is used, the '\'' character in zText
** is escaped and the SQL generated is as follows:
**
** <blockquote><pre>
**  INSERT INTO table1 VALUES('It''s a happy day!')
** </pre></blockquote>
**
** This is correct.  Had we used %s instead of %q, the generated SQL
** would have looked like this:
**
** <blockquote><pre>
**  INSERT INTO table1 VALUES('It's a happy day!');
** </pre></blockquote>
**
** This second example is an SQL syntax error.  As a general rule you
** should always use %q instead of %s when inserting text into a string 
** literal.
**
** The %Q option works like %q except it also adds single quotes around
** the outside of the total string.  Or if the parameter in the argument
** list is a NULL pointer, %Q substitutes the text "NULL" (without single
** quotes) in place of the %Q option.  So, for example, one could say:
**
** <blockquote><pre>
**  char *zSQL = sqlite3_mprintf("INSERT INTO table VALUES(%Q)", zText);
**  sqlite3_exec(db, zSQL, 0, 0, 0);
**  sqlite3_free(zSQL);
** </pre></blockquote>
**
** The code above will render a correct SQL statement in the zSQL
** variable even if the zText variable is a NULL pointer.
*/
SQLITE_API char *sqlite3_mprintf(const char*,...);
SQLITE_API char *sqlite3_vmprintf(const char*, va_list);
SQLITE_API char *sqlite3_snprintf(int,char*,const char*, ...);

/*
** CAPI3REF: Memory Allocation Functions
**
** SQLite uses its own memory allocator.  On some installations, this
** memory allocator is identical to the standard malloc()/realloc()/free()
** and can be used interchangable.  On others, the implementations are
** different.  For maximum portability, it is best not to mix calls
** to the standard malloc/realloc/free with the sqlite versions.
*/
SQLITE_API void *sqlite3_malloc(int);
SQLITE_API void *sqlite3_realloc(void*, int);
SQLITE_API void sqlite3_free(void*);

/*
** CAPI3REF: Compile-Time Authorization Callbacks
***
** This routine registers a authorizer callback with the SQLite library.  
** The authorizer callback is invoked as SQL statements are being compiled
** by [sqlite3_prepare()] or its variants [sqlite3_prepare_v2()],
** [sqlite3_prepare16()] and [sqlite3_prepare16_v2()].  At various
** points during the compilation process, as logic is being created
** to perform various actions, the authorizer callback is invoked to
** see if those actions are allowed.  The authorizer callback should
** return SQLITE_OK to allow the action, [SQLITE_IGNORE] to disallow the
** specific action but allow the SQL statement to continue to be
** compiled, or [SQLITE_DENY] to cause the entire SQL statement to be
** rejected with an error.  
**
** Depending on the action, the [SQLITE_IGNORE] and [SQLITE_DENY] return
** codes might mean something different or they might mean the same
** thing.  If the action is, for example, to perform a delete opertion,
** then [SQLITE_IGNORE] and [SQLITE_DENY] both cause the statement compilation
** to fail with an error.  But if the action is to read a specific column
** from a specific table, then [SQLITE_DENY] will cause the entire
** statement to fail but [SQLITE_IGNORE] will cause a NULL value to be
** read instead of the actual column value.
**
** The first parameter to the authorizer callback is a copy of
** the third parameter to the sqlite3_set_authorizer() interface.
** The second parameter to the callback is an integer 
** [SQLITE_COPY | action code] that specifies the particular action
** to be authorized.  The available action codes are
** [SQLITE_COPY | documented separately].  The third through sixth
** parameters to the callback are strings that contain additional
** details about the action to be authorized.
**
** An authorizer is used when preparing SQL statements from an untrusted
** source, to ensure that the SQL statements do not try to access data
** that they are not allowed to see, or that they do not try to
** execute malicious statements that damage the database.  For
** example, an application may allow a user to enter arbitrary
** SQL queries for evaluation by a database.  But the application does
** not want the user to be able to make arbitrary changes to the
** database.  An authorizer could then be put in place while the
** user-entered SQL is being prepared that disallows everything
** except SELECT statements.  
**
** Only a single authorizer can be in place on a database connection
** at a time.  Each call to sqlite3_set_authorizer overrides the
** previous call.  A NULL authorizer means that no authorization
** callback is invoked.  The default authorizer is NULL.
**
** Note that the authorizer callback is invoked only during 
** [sqlite3_prepare()] or its variants.  Authorization is not
** performed during statement evaluation in [sqlite3_step()].
*/
int sqlite3_set_authorizer(
  sqlite3*,
  int (*xAuth)(void*,int,const char*,const char*,const char*,const char*),
  void *pUserData
);

/*
** CAPI3REF: Authorizer Return Codes
**
** The [sqlite3_set_authorizer | authorizer callback function] must
** return either [SQLITE_OK] or one of these two constants in order
** to signal SQLite whether or not the action is permitted.  See the
** [sqlite3_set_authorizer | authorizer documentation] for additional
** information.
*/
#define SQLITE_DENY   1   /* Abort the SQL statement with an error */
#define SQLITE_IGNORE 2   /* Don't allow access, but don't generate an error */

/*
** CAPI3REF: Authorizer Action Codes
**
** The [sqlite3_set_authorizer()] interface registers a callback function
** that is invoked to authorizer certain SQL statement actions.  The
** second parameter to the callback is an integer code that specifies
** what action is being authorized.  These are the integer action codes that
** the authorizer callback may be passed.
**
** These action code values signify what kind of operation is to be 
** authorized.  The 3rd and 4th parameters to the authorization callback
** function will be parameters or NULL depending on which of these
** codes is used as the second parameter.  The 5th parameter to the
** authorizer callback is the name of the database ("main", "temp", 
** etc.) if applicable.  The 6th parameter to the authorizer callback
** is the name of the inner-most trigger or view that is responsible for
** the access attempt or NULL if this access attempt is directly from 
** top-level SQL code.
*/
/******************************************* 3rd ************ 4th ***********/
#define SQLITE_CREATE_INDEX          1   /* Index Name      Table Name      */
#define SQLITE_CREATE_TABLE          2   /* Table Name      NULL            */
#define SQLITE_CREATE_TEMP_INDEX     3   /* Index Name      Table Name      */
#define SQLITE_CREATE_TEMP_TABLE     4   /* Table Name      NULL            */
#define SQLITE_CREATE_TEMP_TRIGGER   5   /* Trigger Name    Table Name      */
#define SQLITE_CREATE_TEMP_VIEW      6   /* View Name       NULL            */
#define SQLITE_CREATE_TRIGGER        7   /* Trigger Name    Table Name      */
#define SQLITE_CREATE_VIEW           8   /* View Name       NULL            */
#define SQLITE_DELETE                9   /* Table Name      NULL            */
#define SQLITE_DROP_INDEX           10   /* Index Name      Table Name      */
#define SQLITE_DROP_TABLE           11   /* Table Name      NULL            */
#define SQLITE_DROP_TEMP_INDEX      12   /* Index Name      Table Name      */
#define SQLITE_DROP_TEMP_TABLE      13   /* Table Name      NULL            */
#define SQLITE_DROP_TEMP_TRIGGER    14   /* Trigger Name    Table Name      */
#define SQLITE_DROP_TEMP_VIEW       15   /* View Name       NULL            */
#define SQLITE_DROP_TRIGGER         16   /* Trigger Name    Table Name      */
#define SQLITE_DROP_VIEW            17   /* View Name       NULL            */
#define SQLITE_INSERT               18   /* Table Name      NULL            */
#define SQLITE_PRAGMA               19   /* Pragma Name     1st arg or NULL */
#define SQLITE_READ                 20   /* Table Name      Column Name     */
#define SQLITE_SELECT               21   /* NULL            NULL            */
#define SQLITE_TRANSACTION          22   /* NULL            NULL            */
#define SQLITE_UPDATE               23   /* Table Name      Column Name     */
#define SQLITE_ATTACH               24   /* Filename        NULL            */
#define SQLITE_DETACH               25   /* Database Name   NULL            */
#define SQLITE_ALTER_TABLE          26   /* Database Name   Table Name      */
#define SQLITE_REINDEX              27   /* Index Name      NULL            */
#define SQLITE_ANALYZE              28   /* Table Name      NULL            */
#define SQLITE_CREATE_VTABLE        29   /* Table Name      Module Name     */
#define SQLITE_DROP_VTABLE          30   /* Table Name      Module Name     */
#define SQLITE_FUNCTION             31   /* Function Name   NULL            */
#define SQLITE_COPY                  0   /* No longer used */

/*
** CAPI3REF: Tracing And Profiling Functions
**
** These routines register callback functions that can be used for
** tracing and profiling the execution of SQL statements.
** The callback function registered by sqlite3_trace() is invoked
** at the first [sqlite3_step()] for the evaluation of an SQL statement.
** The callback function registered by sqlite3_profile() is invoked
** as each SQL statement finishes and includes
** information on how long that statement ran.
**
** The sqlite3_profile() API is currently considered experimental and
** is subject to change.
*/
SQLITE_API void *sqlite3_trace(sqlite3*, void(*xTrace)(void*,const char*), void*);
SQLITE_API void *sqlite3_profile(sqlite3*,
   void(*xProfile)(void*,const char*,sqlite_uint64), void*);

/*
** CAPI3REF: Query Progress Callbacks
**
** This routine configures a callback function - the progress callback - that
** is invoked periodically during long running calls to [sqlite3_exec()],
** [sqlite3_step()] and [sqlite3_get_table()].  An example use for this 
** interface is to keep a GUI updated during a large query.
**
** The progress callback is invoked once for every N virtual machine opcodes,
** where N is the second argument to this function. The progress callback
** itself is identified by the third argument to this function. The fourth
** argument to this function is a void pointer passed to the progress callback
** function each time it is invoked.
**
** If a call to [sqlite3_exec()], [sqlite3_step()], or [sqlite3_get_table()]
** results in fewer than N opcodes being executed, then the progress 
** callback is never invoked.
** 
** Only a single progress callback function may be registered for each
** open database connection.  Every call to sqlite3_progress_handler()
** overwrites the results of the previous call.
** To remove the progress callback altogether, pass NULL as the third
** argument to this function.
**
** If the progress callback returns a result other than 0, then the current 
** query is immediately terminated and any database changes rolled back.
** The containing [sqlite3_exec()], [sqlite3_step()], or
** [sqlite3_get_table()] call returns SQLITE_INTERRUPT.   This feature
** can be used, for example, to implement the "Cancel" button on a
** progress dialog box in a GUI.
*/
void sqlite3_progress_handler(sqlite3*, int, int(*)(void*), void*);

/*
** CAPI3REF: Opening A New Database Connection
**
** Open the sqlite database file "filename".  The "filename" is UTF-8
** encoded for sqlite3_open() and UTF-16 encoded in the native byte order
** for sqlite3_open16().  An [sqlite3*] handle is returned in *ppDb, even
** if an error occurs. If the database is opened (or created) successfully,
** then SQLITE_OK is returned. Otherwise an error code is returned. The
** sqlite3_errmsg() or sqlite3_errmsg16()  routines can be used to obtain
** an English language description of the error.
**
** If the database file does not exist, then a new database will be created
** as needed.  The default encoding for the database will be UTF-8 if
** sqlite3_open() is called and UTF-16 if sqlite3_open16 is used.
**
** Whether or not an error occurs when it is opened, resources associated
** with the [sqlite3*] handle should be released by passing it to
** sqlite3_close() when it is no longer required.
**
** Note to windows users:  The encoding used for the filename argument
** of sqlite3_open() must be UTF-8, not whatever codepage is currently
** defined.  Filenames containing international characters must be converted
** to UTF-8 prior to passing them into sqlite3_open().
*/
SQLITE_API int sqlite3_open(
  const char *filename,   /* Database filename (UTF-8) */
  sqlite3 **ppDb          /* OUT: SQLite db handle */
);
SQLITE_API int sqlite3_open16(
  const void *filename,   /* Database filename (UTF-16) */
  sqlite3 **ppDb          /* OUT: SQLite db handle */
);

/*
** CAPI3REF: Error Codes And Messages
**
** The sqlite3_errcode() interface returns the numeric
** [SQLITE_OK | result code] or [SQLITE_IOERR_READ | extended result code]
** for the most recent failed sqlite3_* API call associated
** with [sqlite3] handle 'db'.  If a prior API call failed but the
** most recent API call succeeded, the return value from sqlite3_errcode()
** is undefined. 
**
** The sqlite3_errmsg() and sqlite3_errmsg16() return English-langauge
** text that describes the error, as either UTF8 or UTF16 respectively.
** Memory to hold the error message string is managed internally.  The 
** string may be overwritten or deallocated by subsequent calls to SQLite
** interface functions.
**
** Calls to many sqlite3_* functions set the error code and string returned
** by [sqlite3_errcode()], [sqlite3_errmsg()], and [sqlite3_errmsg16()]
** (overwriting the previous values). Note that calls to [sqlite3_errcode()],
** [sqlite3_errmsg()], and [sqlite3_errmsg16()] themselves do not affect the
** results of future invocations.  Calls to API routines that do not return
** an error code (examples: [sqlite3_data_count()] or [sqlite3_mprintf()]) do
** not change the error code returned by this routine.
**
** Assuming no other intervening sqlite3_* API calls are made, the error
** code returned by this function is associated with the same error as
** the strings returned by [sqlite3_errmsg()] and [sqlite3_errmsg16()].
*/
SQLITE_API int sqlite3_errcode(sqlite3 *db);
SQLITE_API const char *sqlite3_errmsg(sqlite3*);
SQLITE_API const void *sqlite3_errmsg16(sqlite3*);

/*
** CAPI3REF: SQL Statement Object
**
** Instance of this object represent single SQL statements.  This
** is variously known as a "prepared statement" or a 
** "compiled SQL statement" or simply as a "statement".
** 
** The life of a statement object goes something like this:
**
** <ol>
** <li> Create the object using [sqlite3_prepare_v2()] or a related
**      function.
** <li> Bind values to host parameters using
**      [sqlite3_bind_blob | sqlite3_bind_* interfaces].
** <li> Run the SQL by calling [sqlite3_step()] one or more times.
** <li> Reset the statement using [sqlite3_reset()] then go back
**      to step 2.  Do this zero or more times.
** <li> Destroy the object using [sqlite3_finalize()].
** </ol>
**
** Refer to documentation on individual methods above for additional
** information.
*/
typedef struct sqlite3_stmt sqlite3_stmt;

/*
** CAPI3REF: Compiling An SQL Statement
**
** To execute an SQL query, it must first be compiled into a byte-code
** program using one of these routines. 
**
** The first argument "db" is an [sqlite3 | SQLite database handle] 
** obtained from a prior call to [sqlite3_open()] or [sqlite3_open16()].
** The second argument "zSql" is the statement to be compiled, encoded
** as either UTF-8 or UTF-16.  The sqlite3_prepare() and sqlite3_prepare_v2()
** interfaces uses UTF-8 and sqlite3_prepare16() and sqlite3_prepare16_v2()
** use UTF-16.
**
** If the nByte argument is less
** than zero, then zSql is read up to the first zero terminator.  If
** nByte is non-negative, then it is the maximum number of 
** bytes read from zSql.  When nByte is non-negative, the
** zSql string ends at either the first '\000' character or 
** until the nByte-th byte, whichever comes first.
**
** *pzTail is made to point to the first byte past the end of the first
** SQL statement in zSql.  This routine only compiles the first statement
** in zSql, so *pzTail is left pointing to what remains uncompiled.
**
** *ppStmt is left pointing to a compiled 
** [sqlite3_stmt | SQL statement structure] that can be
** executed using [sqlite3_step()].  Or if there is an error, *ppStmt may be
** set to NULL.  If the input text contained no SQL (if the input is and
** empty string or a comment) then *ppStmt is set to NULL.  The calling
** procedure is responsible for deleting the compiled SQL statement
** using [sqlite3_finalize()] after it has finished with it.
**
** On success, [SQLITE_OK] is returned.  Otherwise an 
** [SQLITE_ERROR | error code] is returned.
**
** The sqlite3_prepare_v2() and sqlite3_prepare16_v2() interfaces are
** recommended for all new programs. The two older interfaces are retained
** for backwards compatibility, but their use is discouraged.
** In the "v2" interfaces, the prepared statement
** that is returned (the [sqlite3_stmt] object) contains a copy of the 
** original SQL text. This causes the [sqlite3_step()] interface to
** behave a differently in two ways:
**
** <ol>
** <li>
** If the database schema changes, instead of returning [SQLITE_SCHEMA] as it
** always used to do, [sqlite3_step()] will automatically recompile the SQL
** statement and try to run it again.  If the schema has changed in a way
** that makes the statement no longer valid, [sqlite3_step()] will still
** return [SQLITE_SCHEMA].  But unlike the legacy behavior, [SQLITE_SCHEMA] is
** now a fatal error.  Calling [sqlite3_prepare_v2()] again will not make the
** error go away.  Note: use [sqlite3_errmsg()] to find the text of the parsing
** error that results in an [SQLITE_SCHEMA] return.
** </li>
**
** <li>
** When an error occurs, 
** [sqlite3_step()] will return one of the detailed 
** [SQLITE_ERROR | result codes] or
** [SQLITE_IOERR_READ | extended result codes] such as directly.
** The legacy behavior was that [sqlite3_step()] would only return a generic
** [SQLITE_ERROR] result code and you would have to make a second call to
** [sqlite3_reset()] in order to find the underlying cause of the problem.
** With the "v2" prepare interfaces, the underlying reason for the error is
** returned immediately.
** </li>
** </ol>
*/
SQLITE_API int sqlite3_prepare(
  sqlite3 *db,            /* Database handle */
  const char *zSql,       /* SQL statement, UTF-8 encoded */
  int nByte,              /* Maximum length of zSql in bytes. */
  sqlite3_stmt **ppStmt,  /* OUT: Statement handle */
  const char **pzTail     /* OUT: Pointer to unused portion of zSql */
);
int sqlite3_prepare_v2(
  sqlite3 *db,            /* Database handle */
  const char *zSql,       /* SQL statement, UTF-8 encoded */
  int nByte,              /* Maximum length of zSql in bytes. */
  sqlite3_stmt **ppStmt,  /* OUT: Statement handle */
  const char **pzTail     /* OUT: Pointer to unused portion of zSql */
);
SQLITE_API int sqlite3_prepare16(
  sqlite3 *db,            /* Database handle */
  const void *zSql,       /* SQL statement, UTF-16 encoded */
  int nByte,              /* Maximum length of zSql in bytes. */
  sqlite3_stmt **ppStmt,  /* OUT: Statement handle */
  const void **pzTail     /* OUT: Pointer to unused portion of zSql */
);
int sqlite3_prepare16_v2(
  sqlite3 *db,            /* Database handle */
  const void *zSql,       /* SQL statement, UTF-16 encoded */
  int nByte,              /* Maximum length of zSql in bytes. */
  sqlite3_stmt **ppStmt,  /* OUT: Statement handle */
  const void **pzTail     /* OUT: Pointer to unused portion of zSql */
);

/*
** CAPI3REF:  Dynamically Typed Value Object
**
** SQLite uses dynamic typing for the values it stores.  Values can 
** be integers, floating point values, strings, BLOBs, or NULL.  When
** passing around values internally, each value is represented as
** an instance of the sqlite3_value object.
*/
typedef struct Mem sqlite3_value;

/*
** CAPI3REF:  SQL Function Context Object
**
** The context in which an SQL function executes is stored in an
** sqlite3_context object.  A pointer to such an object is the
** first parameter to user-defined SQL functions.
*/
typedef struct sqlite3_context sqlite3_context;

/*
** CAPI3REF:  Binding Values To Prepared Statements
**
** In the SQL strings input to [sqlite3_prepare_v2()] and its variants,
** one or more literals can be replace by a parameter in one of these
** forms:
**
** <ul>
** <li>  ?
** <li>  ?NNN
** <li>  :AAA
** <li>  @AAA
** <li>  $VVV
** </ul>
**
** In the parameter forms shown above NNN is an integer literal,
** AAA is an alphanumeric identifier and VVV is a variable name according
** to the syntax rules of the TCL programming language.
** The values of these parameters (also called "host parameter names")
** can be set using the sqlite3_bind_*() routines defined here.
**
** The first argument to the sqlite3_bind_*() routines always is a pointer
** to the [sqlite3_stmt] object returned from [sqlite3_prepare_v2()] or
** its variants.  The second
** argument is the index of the parameter to be set.  The first parameter has
** an index of 1. When the same named parameter is used more than once, second
** and subsequent
** occurrences have the same index as the first occurrence.  The index for
** named parameters can be looked up using the
** [sqlite3_bind_parameter_name()] API if desired.  The index for "?NNN"
** parametes is the value of NNN.
** The NNN value must be between 1 and the compile-time
** parameter SQLITE_MAX_VARIABLE_NUMBER (default value: 999).
** See <a href="limits.html">limits.html</a> for additional information.
**
** The third argument is the value to bind to the parameter.
**
** In those
** routines that have a fourth argument, its value is the number of bytes
** in the parameter.  To be clear: the value is the number of bytes in the
** string, not the number of characters.  The number
** of bytes does not include the zero-terminator at the end of strings.
** If the fourth parameter is negative, the length of the string is
** number of bytes up to the first zero terminator.
**
** The fifth argument to sqlite3_bind_blob(), sqlite3_bind_text(), and
** sqlite3_bind_text16() is a destructor used to dispose of the BLOB or
** text after SQLite has finished with it.  If the fifth argument is the
** special value [SQLITE_STATIC], then the library assumes that the information
** is in static, unmanaged space and does not need to be freed.  If the
** fifth argument has the value [SQLITE_TRANSIENT], then SQLite makes its
** own private copy of the data immediately, before the sqlite3_bind_*()
** routine returns.
**
** The sqlite3_bind_zeroblob() routine binds a BLOB of length n that
** is filled with zeros.  A zeroblob uses a fixed amount of memory
** (just an integer to hold it size) while it is being processed.
** Zeroblobs are intended to serve as place-holders for BLOBs whose
** content is later written using 
** [sqlite3_blob_open | increment BLOB I/O] routines.
**
** The sqlite3_bind_*() routines must be called after
** [sqlite3_prepare_v2()] (and its variants) or [sqlite3_reset()] and
** before [sqlite3_step()].
** Bindings are not cleared by the [sqlite3_reset()] routine.
** Unbound parameters are interpreted as NULL.
**
** These routines return [SQLITE_OK] on success or an error code if
** anything goes wrong.  [SQLITE_RANGE] is returned if the parameter
** index is out of range.  [SQLITE_NOMEM] is returned if malloc fails.
** [SQLITE_MISUSE] is returned if these routines are called on a virtual
** machine that is the wrong state or which has already been finalized.
*/
int sqlite3_bind_blob(sqlite3_stmt*, int, const void*, int n, void(*)(void*));
int sqlite3_bind_double(sqlite3_stmt*, int, double);
int sqlite3_bind_int(sqlite3_stmt*, int, int);
int sqlite3_bind_int64(sqlite3_stmt*, int, sqlite_int64);
int sqlite3_bind_null(sqlite3_stmt*, int);
int sqlite3_bind_text(sqlite3_stmt*, int, const char*, int n, void(*)(void*));
int sqlite3_bind_text16(sqlite3_stmt*, int, const void*, int, void(*)(void*));
int sqlite3_bind_value(sqlite3_stmt*, int, const sqlite3_value*);
int sqlite3_bind_zeroblob(sqlite3_stmt*, int, int n);

/*
** CAPI3REF: Number Of Host Parameters
**
** Return the largest host parameter index in the precompiled statement given
** as the argument.  When the host parameters are of the forms like ":AAA"
** or "?", then they are assigned sequential increasing numbers beginning
** with one, so the value returned is the number of parameters.  However
** if the same host parameter name is used multiple times, each occurrance
** is given the same number, so the value returned in that case is the number
** of unique host parameter names.  If host parameters of the form "?NNN"
** are used (where NNN is an integer) then there might be gaps in the
** numbering and the value returned by this interface is the index of the
** host parameter with the largest index value.
*/
int sqlite3_bind_parameter_count(sqlite3_stmt*);

/*
** CAPI3REF: Name Of A Host Parameter
**
** This routine returns a pointer to the name of the n-th parameter in a 
** [sqlite3_stmt | prepared statement].
** Host parameters of the form ":AAA" or "@AAA" or "$VVV" have a name
** which is the string ":AAA" or "@AAA" or "$VVV".  
** In other words, the initial ":" or "$" or "@"
** is included as part of the name.
** Parameters of the form "?" or "?NNN" have no name.
**
** The first bound parameter has an index of 1, not 0.
**
** If the value n is out of range or if the n-th parameter is nameless,
** then NULL is returned.  The returned string is always in the
** UTF-8 encoding even if the named parameter was originally specified
** as UTF-16 in [sqlite3_prepare16()] or [sqlite3_prepare16_v2()].
*/
const char *sqlite3_bind_parameter_name(sqlite3_stmt*, int);

/*
** CAPI3REF: Index Of A Parameter With A Given Name
**
** This routine returns the index of a host parameter with the given name.
** The name must match exactly.  If no parameter with the given name is 
** found, return 0.  Parameter names must be UTF8.
*/
int sqlite3_bind_parameter_index(sqlite3_stmt*, const char *zName);

/*
** CAPI3REF: Reset All Bindings On A Prepared Statement
**
** Contrary to the intuition of many, [sqlite3_reset()] does not
** reset the [sqlite3_bind_blob | bindings] on a 
** [sqlite3_stmt | prepared statement].  Use this routine to
** reset all host parameters to NULL.
*/
int sqlite3_clear_bindings(sqlite3_stmt*);

/*
** CAPI3REF: Number Of Columns In A Result Set
**
** Return the number of columns in the result set returned by the 
** [sqlite3_stmt | compiled SQL statement]. This routine returns 0
** if pStmt is an SQL statement that does not return data (for 
** example an UPDATE).
*/
int sqlite3_column_count(sqlite3_stmt *pStmt);

/*
** CAPI3REF: Column Names In A Result Set
**
** These routines return the name assigned to a particular column
** in the result set of a SELECT statement.  The sqlite3_column_name()
** interface returns a pointer to a UTF8 string and sqlite3_column_name16()
** returns a pointer to a UTF16 string.  The first parameter is the
** [sqlite_stmt | prepared statement] that implements the SELECT statement.
** The second parameter is the column number.  The left-most column is
** number 0.
**
** The returned string pointer is valid until either the 
** [sqlite_stmt | prepared statement] is destroyed by [sqlite3_finalize()]
** or until the next call sqlite3_column_name() or sqlite3_column_name16()
** on the same column.
*/
const char *sqlite3_column_name(sqlite3_stmt*, int N);
const void *sqlite3_column_name16(sqlite3_stmt*, int N);

/*
** CAPI3REF: Source Of Data In A Query Result
**
** These routines provide a means to determine what column of what
** table in which database a result of a SELECT statement comes from.
** The name of the database or table or column can be returned as
** either a UTF8 or UTF16 string.  The _database_ routines return
** the database name, the _table_ routines return the table name, and
** the origin_ routines return the column name.
** The returned string is valid until
** the [sqlite3_stmt | prepared statement] is destroyed using
** [sqlite3_finalize()] or until the same information is requested
** again in a different encoding.
**
** The names returned are the original un-aliased names of the
** database, table, and column.
**
** The first argument to the following calls is a 
** [sqlite3_stmt | compiled SQL statement].
** These functions return information about the Nth column returned by 
** the statement, where N is the second function argument.
**
** If the Nth column returned by the statement is an expression
** or subquery and is not a column value, then all of these functions
** return NULL. Otherwise, they return the 
** name of the attached database, table and column that query result
** column was extracted from.
**
** As with all other SQLite APIs, those postfixed with "16" return UTF-16
** encoded strings, the other functions return UTF-8.
**
** These APIs are only available if the library was compiled with the 
** SQLITE_ENABLE_COLUMN_METADATA preprocessor symbol defined.
*/
const char *sqlite3_column_database_name(sqlite3_stmt*,int);
const void *sqlite3_column_database_name16(sqlite3_stmt*,int);
const char *sqlite3_column_table_name(sqlite3_stmt*,int);
const void *sqlite3_column_table_name16(sqlite3_stmt*,int);
const char *sqlite3_column_origin_name(sqlite3_stmt*,int);
const void *sqlite3_column_origin_name16(sqlite3_stmt*,int);

/*
** CAPI3REF: Declared Datatype Of A Query Result
**
** The first parameter is a [sqlite3_stmt | compiled SQL statement]. 
** If this statement is a SELECT statement and the Nth column of the 
** returned result set  of that SELECT is a table column (not an
** expression or subquery) then the declared type of the table
** column is returned. If the Nth column of the result set is an
** expression or subquery, then a NULL pointer is returned.
** The returned string is always UTF-8 encoded. For example, in
** the database schema:
**
** CREATE TABLE t1(c1 VARIANT);
**
** And the following statement compiled:
**
** SELECT c1 + 1, c1 FROM t1;
**
** Then this routine would return the string "VARIANT" for the second
** result column (i==1), and a NULL pointer for the first result column
** (i==0).
**
** SQLite uses dynamic run-time typing.  So just because a column
** is declared to contain a particular type does not mean that the
** data stored in that column is of the declared type.  SQLite is
** strongly typed, but the typing is dynamic not static.  Type
** is associated with individual values, not with the containers
** used to hold those values.
*/
const char *sqlite3_column_decltype(sqlite3_stmt *, int i);
const void *sqlite3_column_decltype16(sqlite3_stmt*,int);

/* 
** CAPI3REF:  Evaluate An SQL Statement
**
** After an [sqlite3_stmt | SQL statement] has been prepared with a call
** to either [sqlite3_prepare_v2()] or [sqlite3_prepare16_v2()] or to one of
** the legacy interfaces [sqlite3_prepare()] or [sqlite3_prepare16()],
** then this function must be called one or more times to evaluate the 
** statement.
**
** The details of the behavior of this sqlite3_step() interface depend
** on whether the statement was prepared using the newer "v2" interface
** [sqlite3_prepare_v2()] and [sqlite3_prepare16_v2()] or the older legacy
** interface [sqlite3_prepare()] and [sqlite3_prepare16()].  The use of the
** new "v2" interface is recommended for new applications but the legacy
** interface will continue to be supported.
**
** In the lagacy interface, the return value will be either [SQLITE_BUSY], 
** [SQLITE_DONE], [SQLITE_ROW], [SQLITE_ERROR], or [SQLITE_MISUSE].
** With the "v2" interface, any of the other [SQLITE_OK | result code]
** or [SQLITE_IOERR_READ | extended result code] might be returned as
** well.
**
** [SQLITE_BUSY] means that the database engine was unable to acquire the
** database locks it needs to do its job.  If the statement is a COMMIT
** or occurs outside of an explicit transaction, then you can retry the
** statement.  If the statement is not a COMMIT and occurs within a
** explicit transaction then you should rollback the transaction before
** continuing.
**
** [SQLITE_DONE] means that the statement has finished executing
** successfully.  sqlite3_step() should not be called again on this virtual
** machine without first calling [sqlite3_reset()] to reset the virtual
** machine back to its initial state.
**
** If the SQL statement being executed returns any data, then 
** [SQLITE_ROW] is returned each time a new row of data is ready
** for processing by the caller. The values may be accessed using
** the [sqlite3_column_int | column access functions].
** sqlite3_step() is called again to retrieve the next row of data.
** 
** [SQLITE_ERROR] means that a run-time error (such as a constraint
** violation) has occurred.  sqlite3_step() should not be called again on
** the VM. More information may be found by calling [sqlite3_errmsg()].
** With the legacy interface, a more specific error code (example:
** [SQLITE_INTERRUPT], [SQLITE_SCHEMA], [SQLITE_CORRUPT], and so forth)
** can be obtained by calling [sqlite3_reset()] on the
** [sqlite_stmt | prepared statement].  In the "v2" interface,
** the more specific error code is returned directly by sqlite3_step().
**
** [SQLITE_MISUSE] means that the this routine was called inappropriately.
** Perhaps it was called on a [sqlite_stmt | prepared statement] that has
** already been [sqlite3_finalize | finalized] or on one that had 
** previously returned [SQLITE_ERROR] or [SQLITE_DONE].  Or it could
** be the case that the same database connection is being used by two or
** more threads at the same moment in time.
**
** <b>Goofy Interface Alert:</b>
** In the legacy interface, 
** the sqlite3_step() API always returns a generic error code,
** [SQLITE_ERROR], following any error other than [SQLITE_BUSY]
** and [SQLITE_MISUSE].  You must call [sqlite3_reset()] or
** [sqlite3_finalize()] in order to find one of the specific
** [SQLITE_ERROR | result codes] that better describes the error.
** We admit that this is a goofy design.  The problem has been fixed
** with the "v2" interface.  If you prepare all of your SQL statements
** using either [sqlite3_prepare_v2()] or [sqlite3_prepare16_v2()] instead
** of the legacy [sqlite3_prepare()] and [sqlite3_prepare16()], then the 
** more specific [SQLITE_ERROR | result codes] are returned directly
** by sqlite3_step().  The use of the "v2" interface is recommended.
*/
SQLITE_API int sqlite3_step(sqlite3_stmt*);

/*
** CAPI3REF:
**
** Return the number of values in the current row of the result set.
**
** After a call to [sqlite3_step()] that returns [SQLITE_ROW], this routine
** will return the same value as the [sqlite3_column_count()] function.
** After [sqlite3_step()] has returned an [SQLITE_DONE], [SQLITE_BUSY], or
** a [SQLITE_ERROR | error code], or before [sqlite3_step()] has been 
** called on the [sqlite_stmt | prepared statement] for the first time,
** this routine returns zero.
*/
int sqlite3_data_count(sqlite3_stmt *pStmt);

/*
** CAPI3REF: Fundamental Datatypes
**
** Every value in SQLite has one of five fundamental datatypes:
**
** <ul>
** <li> 64-bit signed integer
** <li> 64-bit IEEE floating point number
** <li> string
** <li> BLOB
** <li> NULL
** </ul>
**
** These constants are codes for each of those types.
**
** Note that the SQLITE_TEXT constant was also used in SQLite version 2
** for a completely different meaning.  Software that links against both
** SQLite version 2 and SQLite version 3 should use SQLITE3_TEXT not
** SQLITE_TEXT.
*/
#define SQLITE_INTEGER  1
#define SQLITE_FLOAT    2
#define SQLITE_BLOB     4
#define SQLITE_NULL     5
#ifdef SQLITE_TEXT
# undef SQLITE_TEXT
#else
# define SQLITE_TEXT     3
#endif
#define SQLITE3_TEXT     3

/*
** CAPI3REF: Results Values From A Query
**
** These routines return information about the information
** in a single column of the current result row of a query.  In every
** case the first argument is a pointer to the 
** [sqlite3_stmt | SQL statement] that is being
** evaluate (the [sqlite_stmt*] that was returned from 
** [sqlite3_prepare_v2()] or one of its variants) and
** the second argument is the index of the column for which information 
** should be returned.  The left-most column has an index of 0.
**
** If the SQL statement is not currently point to a valid row, or if the
** the column index is out of range, the result is undefined.
**
** The sqlite3_column_type() routine returns 
** [SQLITE_INTEGER | datatype code] for the initial data type
** of the result column.  The returned value is one of [SQLITE_INTEGER],
** [SQLITE_FLOAT], [SQLITE_TEXT], [SQLITE_BLOB], or [SQLITE_NULL].  The value
** returned by sqlite3_column_type() is only meaningful if no type
** conversions have occurred as described below.  After a type conversion,
** the value returned by sqlite3_column_type() is undefined.  Future
** versions of SQLite may change the behavior of sqlite3_column_type()
** following a type conversion.
**
** If the result is a BLOB or UTF-8 string then the sqlite3_column_bytes() 
** routine returns the number of bytes in that BLOB or string.
** If the result is a UTF-16 string, then sqlite3_column_bytes() converts
** the string to UTF-8 and then returns the number of bytes.
** If the result is a numeric value then sqlite3_column_bytes() uses
** [sqlite3_snprintf()] to convert that value to a UTF-8 string and returns
** the number of bytes in that string.
** The value returned does not include the zero terminator at the end
** of the string.  For clarity: the value returned is the number of
** bytes in the string, not the number of characters.
**
** The sqlite3_column_bytes16() routine is similar to sqlite3_column_bytes()
** but leaves the result in UTF-16 instead of UTF-8.  
** The zero terminator is not included in this count.
**
** These routines attempt to convert the value where appropriate.  For
** example, if the internal representation is FLOAT and a text result
** is requested, [sqlite3_snprintf()] is used internally to do the conversion
** automatically.  The following table details the conversions that
** are applied:
**
** <blockquote>
** <table border="1">
** <tr><th> Internal <th> Requested <th> 
** <tr><th>  Type    <th>    Type   <th> Conversion
**
** <tr><td>  NULL    <td> INTEGER   <td> Result is 0
** <tr><td>  NULL    <td>  FLOAT    <td> Result is 0.0
** <tr><td>  NULL    <td>   TEXT    <td> Result is NULL pointer
** <tr><td>  NULL    <td>   BLOB    <td> Result is NULL pointer
** <tr><td> INTEGER  <td>  FLOAT    <td> Convert from integer to float
** <tr><td> INTEGER  <td>   TEXT    <td> ASCII rendering of the integer
** <tr><td> INTEGER  <td>   BLOB    <td> Same as for INTEGER->TEXT
** <tr><td>  FLOAT   <td> INTEGER   <td> Convert from float to integer
** <tr><td>  FLOAT   <td>   TEXT    <td> ASCII rendering of the float
** <tr><td>  FLOAT   <td>   BLOB    <td> Same as FLOAT->TEXT
** <tr><td>  TEXT    <td> INTEGER   <td> Use atoi()
** <tr><td>  TEXT    <td>  FLOAT    <td> Use atof()
** <tr><td>  TEXT    <td>   BLOB    <td> No change
** <tr><td>  BLOB    <td> INTEGER   <td> Convert to TEXT then use atoi()
** <tr><td>  BLOB    <td>  FLOAT    <td> Convert to TEXT then use atof()
** <tr><td>  BLOB    <td>   TEXT    <td> Add a zero terminator if needed
** </table>
** </blockquote>
**
** The table above makes reference to standard C library functions atoi()
** and atof().  SQLite does not really use these functions.  It has its
** on equavalent internal routines.  The atoi() and atof() names are
** used in the table for brevity and because they are familiar to most
** C programmers.
**
** Note that when type conversions occur, pointers returned by prior
** calls to sqlite3_column_blob(), sqlite3_column_text(), and/or
** sqlite3_column_text16() may be invalidated. 
** Type conversions and pointer invalidations might occur
** in the following cases:
**
** <ul>
** <li><p>  The initial content is a BLOB and sqlite3_column_text() 
**          or sqlite3_column_text16() is called.  A zero-terminator might
**          need to be added to the string.</p></li>
**
** <li><p>  The initial content is UTF-8 text and sqlite3_column_bytes16() or
**          sqlite3_column_text16() is called.  The content must be converted
**          to UTF-16.</p></li>
**
** <li><p>  The initial content is UTF-16 text and sqlite3_column_bytes() or
**          sqlite3_column_text() is called.  The content must be converted
**          to UTF-8.</p></li>
** </ul>
**
** Conversions between UTF-16be and UTF-16le are always done in place and do
** not invalidate a prior pointer, though of course the content of the buffer
** that the prior pointer points to will have been modified.  Other kinds
** of conversion are done in place when it is possible, but sometime it is
** not possible and in those cases prior pointers are invalidated.  
**
** The safest and easiest to remember policy is to invoke these routines
** in one of the following ways:
**
**  <ul>
**  <li>sqlite3_column_text() followed by sqlite3_column_bytes()</li>
**  <li>sqlite3_column_blob() followed by sqlite3_column_bytes()</li>
**  <li>sqlite3_column_text16() followed by sqlite3_column_bytes16()</li>
**  </ul>
**
** In other words, you should call sqlite3_column_text(), sqlite3_column_blob(),
** or sqlite3_column_text16() first to force the result into the desired
** format, then invoke sqlite3_column_bytes() or sqlite3_column_bytes16() to
** find the size of the result.  Do not mix call to sqlite3_column_text() or
** sqlite3_column_blob() with calls to sqlite3_column_bytes16().  And do not
** mix calls to sqlite3_column_text16() with calls to sqlite3_column_bytes().
*/
const void *sqlite3_column_blob(sqlite3_stmt*, int iCol);
int sqlite3_column_bytes(sqlite3_stmt*, int iCol);
int sqlite3_column_bytes16(sqlite3_stmt*, int iCol);
double sqlite3_column_double(sqlite3_stmt*, int iCol);
int sqlite3_column_int(sqlite3_stmt*, int iCol);
sqlite_int64 sqlite3_column_int64(sqlite3_stmt*, int iCol);
const unsigned char *sqlite3_column_text(sqlite3_stmt*, int iCol);
const void *sqlite3_column_text16(sqlite3_stmt*, int iCol);
int sqlite3_column_type(sqlite3_stmt*, int iCol);
sqlite3_value *sqlite3_column_value(sqlite3_stmt*, int iCol);

/*
** CAPI3REF: Destroy A Prepared Statement Object
**
** The sqlite3_finalize() function is called to delete a 
** [sqlite3_stmt | compiled SQL statement]. If the statement was
** executed successfully, or not executed at all, then SQLITE_OK is returned.
** If execution of the statement failed then an 
** [SQLITE_ERROR | error code] or [SQLITE_IOERR_READ | extended error code]
** is returned. 
**
** This routine can be called at any point during the execution of the
** [sqlite3_stmt | virtual machine].  If the virtual machine has not 
** completed execution when this routine is called, that is like
** encountering an error or an interrupt.  (See [sqlite3_interrupt()].) 
** Incomplete updates may be rolled back and transactions cancelled,  
** depending on the circumstances, and the 
** [SQLITE_ERROR | result code] returned will be [SQLITE_ABORT].
*/
SQLITE_API int sqlite3_finalize(sqlite3_stmt *pStmt);

/*
** CAPI3REF: Reset A Prepared Statement Object
**
** The sqlite3_reset() function is called to reset a 
** [sqlite_stmt | compiled SQL statement] object.
** back to it's initial state, ready to be re-executed.
** Any SQL statement variables that had values bound to them using
** the [sqlite3_bind_blob | sqlite3_bind_*() API] retain their values.
** Use [sqlite3_clear_bindings()] to reset the bindings.
*/
SQLITE_API int sqlite3_reset(sqlite3_stmt *pStmt);

/*
** CAPI3REF: Create Or Redefine SQL Functions
**
** The following two functions are used to add SQL functions or aggregates
** or to redefine the behavior of existing SQL functions or aggregates.  The
** difference only between the two is that the second parameter, the
** name of the (scalar) function or aggregate, is encoded in UTF-8 for
** sqlite3_create_function() and UTF-16 for sqlite3_create_function16().
**
** The first argument is the [sqlite3 | database handle] that holds the
** SQL function or aggregate is to be added or redefined. If a single
** program uses more than one database handle internally, then SQL
** functions or aggregates must be added individually to each database
** handle with which they will be used.
**
** The second parameter is the name of the SQL function to be created
** or redefined.
** The length of the name is limited to 255 bytes, exclusive of the 
** zero-terminator.  Note that the name length limit is in bytes, not
** characters.  Any attempt to create a function with a longer name
** will result in an SQLITE_ERROR error.
**
** The third parameter is the number of arguments that the SQL function or
** aggregate takes. If this parameter is negative, then the SQL function or
** aggregate may take any number of arguments.
**
** The fourth parameter, eTextRep, specifies what 
** [SQLITE_UTF8 | text encoding] this SQL function prefers for
** its parameters.  Any SQL function implementation should be able to work
** work with UTF-8, UTF-16le, or UTF-16be.  But some implementations may be
** more efficient with one encoding than another.  It is allowed to
** invoke sqlite_create_function() or sqlite3_create_function16() multiple
** times with the same function but with different values of eTextRep.
** When multiple implementations of the same function are available, SQLite
** will pick the one that involves the least amount of data conversion.
** If there is only a single implementation which does not care what
** text encoding is used, then the fourth argument should be
** [SQLITE_ANY].
**
** The fifth parameter is an arbitrary pointer.  The implementation
** of the function can gain access to this pointer using
** [sqlite_user_data()].
**
** The seventh, eighth and ninth parameters, xFunc, xStep and xFinal, are
** pointers to C-language functions that implement the SQL
** function or aggregate. A scalar SQL function requires an implementation of
** the xFunc callback only, NULL pointers should be passed as the xStep
** and xFinal parameters. An aggregate SQL function requires an implementation
** of xStep and xFinal and NULL should be passed for xFunc. To delete an
** existing SQL function or aggregate, pass NULL for all three function
** callback.
**
** It is permitted to register multiple implementations of the same
** functions with the same name but with either differing numbers of
** arguments or differing perferred text encodings.  SQLite will use
** the implementation most closely matches the way in which the
** SQL function is used.
*/
int sqlite3_create_function(
  sqlite3 *,
  const char *zFunctionName,
  int nArg,
  int eTextRep,
  void*,
  void (*xFunc)(sqlite3_context*,int,sqlite3_value**),
  void (*xStep)(sqlite3_context*,int,sqlite3_value**),
  void (*xFinal)(sqlite3_context*)
);
int sqlite3_create_function16(
  sqlite3*,
  const void *zFunctionName,
  int nArg,
  int eTextRep,
  void*,
  void (*xFunc)(sqlite3_context*,int,sqlite3_value**),
  void (*xStep)(sqlite3_context*,int,sqlite3_value**),
  void (*xFinal)(sqlite3_context*)
);

/*
** CAPI3REF: Text Encodings
**
** These constant define integer codes that represent the various
** text encodings supported by SQLite.
*/
#define SQLITE_UTF8           1
#define SQLITE_UTF16LE        2
#define SQLITE_UTF16BE        3
#define SQLITE_UTF16          4    /* Use native byte order */
#define SQLITE_ANY            5    /* sqlite3_create_function only */
#define SQLITE_UTF16_ALIGNED  8    /* sqlite3_create_collation only */

/*
** CAPI3REF: Obsolete Functions
**
** These functions are all now obsolete.  In order to maintain
** backwards compatibility with older code, we continue to support
** these functions.  However, new development projects should avoid
** the use of these functions.  To help encourage people to avoid
** using these functions, we are not going to tell you want they do.
*/
int sqlite3_aggregate_count(sqlite3_context*);
SQLITE_API int sqlite3_expired(sqlite3_stmt*);
int sqlite3_transfer_bindings(sqlite3_stmt*, sqlite3_stmt*);
int sqlite3_global_recover(void);


/*
** CAPI3REF: Obtaining SQL Function Parameter Values
**
** The C-language implementation of SQL functions and aggregates uses
** this set of interface routines to access the parameter values on
** the function or aggregate.
**
** The xFunc (for scalar functions) or xStep (for aggregates) parameters
** to [sqlite3_create_function()] and [sqlite3_create_function16()]
** define callbacks that implement the SQL functions and aggregates.
** The 4th parameter to these callbacks is an array of pointers to
** [sqlite3_value] objects.  There is one [sqlite3_value] object for
** each parameter to the SQL function.  These routines are used to
** extract values from the [sqlite3_value] objects.
**
** These routines work just like the corresponding 
** [sqlite3_column_blob | sqlite3_column_* routines] except that 
** these routines take a single [sqlite3_value*] pointer instead
** of an [sqlite3_stmt*] pointer and an integer column number.
**
** The sqlite3_value_text16() interface extracts a UTF16 string
** in the native byte-order of the host machine.  The
** sqlite3_value_text16be() and sqlite3_value_text16le() interfaces
** extract UTF16 strings as big-endian and little-endian respectively.
**
** The sqlite3_value_numeric_type() interface attempts to apply
** numeric affinity to the value.  This means that an attempt is
** made to convert the value to an integer or floating point.  If
** such a conversion is possible without loss of information (in order
** words if the value is original a string that looks like a number)
** then it is done.  Otherwise no conversion occurs.  The 
** [SQLITE_INTEGER | datatype] after conversion is returned.
**
** Please pay particular attention to the fact that the pointer that
** is returned from [sqlite3_value_blob()], [sqlite3_value_text()], or
** [sqlite3_value_text16()] can be invalidated by a subsequent call to
** [sqlite3_value_bytes()], [sqlite3_value_bytes16()], [sqlite_value_text()],
** or [sqlite3_value_text16()].  
*/
const void *sqlite3_value_blob(sqlite3_value*);
int sqlite3_value_bytes(sqlite3_value*);
int sqlite3_value_bytes16(sqlite3_value*);
double sqlite3_value_double(sqlite3_value*);
int sqlite3_value_int(sqlite3_value*);
sqlite_int64 sqlite3_value_int64(sqlite3_value*);
const unsigned char *sqlite3_value_text(sqlite3_value*);
const void *sqlite3_value_text16(sqlite3_value*);
const void *sqlite3_value_text16le(sqlite3_value*);
const void *sqlite3_value_text16be(sqlite3_value*);
int sqlite3_value_type(sqlite3_value*);
int sqlite3_value_numeric_type(sqlite3_value*);

/*
** CAPI3REF: Obtain Aggregate Function Context
**
** The implementation of aggregate SQL functions use this routine to allocate
** a structure for storing their state.  The first time this routine
** is called for a particular aggregate, a new structure of size nBytes
** is allocated, zeroed, and returned.  On subsequent calls (for the
** same aggregate instance) the same buffer is returned.  The implementation
** of the aggregate can use the returned buffer to accumulate data.
**
** The buffer allocated is freed automatically by SQLite whan the aggregate
** query concludes.
**
** The first parameter should be a copy of the 
** [sqlite3_context | SQL function context] that is the first
** parameter to the callback routine that implements the aggregate
** function.
*/
void *sqlite3_aggregate_context(sqlite3_context*, int nBytes);

/*
** CAPI3REF: User Data For Functions
**
** The pUserData parameter to the [sqlite3_create_function()]
** and [sqlite3_create_function16()] routines
** used to register user functions is available to
** the implementation of the function using this call.
*/
void *sqlite3_user_data(sqlite3_context*);

/*
** CAPI3REF: Function Auxiliary Data
**
** The following two functions may be used by scalar SQL functions to
** associate meta-data with argument values. If the same value is passed to
** multiple invocations of the same SQL function during query execution, under
** some circumstances the associated meta-data may be preserved. This may
** be used, for example, to add a regular-expression matching scalar
** function. The compiled version of the regular expression is stored as
** meta-data associated with the SQL value passed as the regular expression
** pattern.  The compiled regular expression can be reused on multiple
** invocations of the same function so that the original pattern string
** does not need to be recompiled on each invocation.
**
** The sqlite3_get_auxdata() interface returns a pointer to the meta-data
** associated with the Nth argument value to the current SQL function
** call, where N is the second parameter. If no meta-data has been set for
** that value, then a NULL pointer is returned.
**
** The sqlite3_set_auxdata() is used to associate meta-data with an SQL
** function argument. The third parameter is a pointer to the meta-data
** to be associated with the Nth user function argument value. The fourth
** parameter specifies a destructor that will be called on the meta-
** data pointer to release it when it is no longer required. If the 
** destructor is NULL, it is not invoked.
**
** In practice, meta-data is preserved between function calls for
** expressions that are constant at compile time. This includes literal
** values and SQL variables.
*/
void *sqlite3_get_auxdata(sqlite3_context*, int);
void sqlite3_set_auxdata(sqlite3_context*, int, void*, void (*)(void*));


/*
** CAPI3REF: Constants Defining Special Destructor Behavior
**
** These are special value for the destructor that is passed in as the
** final argument to routines like [sqlite3_result_blob()].  If the destructor
** argument is SQLITE_STATIC, it means that the content pointer is constant
** and will never change.  It does not need to be destroyed.  The 
** SQLITE_TRANSIENT value means that the content will likely change in
** the near future and that SQLite should make its own private copy of
** the content before returning.
**
** The typedef is necessary to work around problems in certain
** C++ compilers.  See ticket #2191.
*/
typedef void (*sqlite3_destructor_type)(void*);
#define SQLITE_STATIC      ((sqlite3_destructor_type)0)
#define SQLITE_TRANSIENT   ((sqlite3_destructor_type)-1)

/*
** CAPI3REF: Setting The Result Of An SQL Function
**
** These routines are used by the xFunc or xFinal callbacks that
** implement SQL functions and aggregates.  See
** [sqlite3_create_function()] and [sqlite3_create_function16()]
** for additional information.
**
** These functions work very much like the 
** [sqlite3_bind_blob | sqlite3_bind_*] family of functions used
** to bind values to host parameters in prepared statements.
** Refer to the
** [sqlite3_bind_blob | sqlite3_bind_* documentation] for
** additional information.
**
** The sqlite3_result_error() and sqlite3_result_error16() functions
** cause the implemented SQL function to throw an exception.  The
** parameter to sqlite3_result_error() or sqlite3_result_error16()
** is the text of an error message.
**
** The sqlite3_result_toobig() cause the function implementation
** to throw and error indicating that a string or BLOB is to long
** to represent.
*/
void sqlite3_result_blob(sqlite3_context*, const void*, int, void(*)(void*));
void sqlite3_result_double(sqlite3_context*, double);
void sqlite3_result_error(sqlite3_context*, const char*, int);
void sqlite3_result_error16(sqlite3_context*, const void*, int);
void sqlite3_result_error_toobig(sqlite3_context*);
void sqlite3_result_int(sqlite3_context*, int);
void sqlite3_result_int64(sqlite3_context*, sqlite_int64);
void sqlite3_result_null(sqlite3_context*);
void sqlite3_result_text(sqlite3_context*, const char*, int, void(*)(void*));
void sqlite3_result_text16(sqlite3_context*, const void*, int, void(*)(void*));
void sqlite3_result_text16le(sqlite3_context*, const void*, int,void(*)(void*));
void sqlite3_result_text16be(sqlite3_context*, const void*, int,void(*)(void*));
void sqlite3_result_value(sqlite3_context*, sqlite3_value*);
void sqlite3_result_zeroblob(sqlite3_context*, int n);

/*
** CAPI3REF: Define New Collating Sequences
**
** These functions are used to add new collation sequences to the
** [sqlite3*] handle specified as the first argument. 
**
** The name of the new collation sequence is specified as a UTF-8 string
** for sqlite3_create_collation() and sqlite3_create_collation_v2()
** and a UTF-16 string for sqlite3_create_collation16().  In all cases
** the name is passed as the second function argument.
**
** The third argument must be one of the constants [SQLITE_UTF8],
** [SQLITE_UTF16LE] or [SQLITE_UTF16BE], indicating that the user-supplied
** routine expects to be passed pointers to strings encoded using UTF-8,
** UTF-16 little-endian or UTF-16 big-endian respectively.
**
** A pointer to the user supplied routine must be passed as the fifth
** argument. If it is NULL, this is the same as deleting the collation
** sequence (so that SQLite cannot call it anymore). Each time the user
** supplied function is invoked, it is passed a copy of the void* passed as
** the fourth argument to sqlite3_create_collation() or
** sqlite3_create_collation16() as its first parameter.
**
** The remaining arguments to the user-supplied routine are two strings,
** each represented by a [length, data] pair and encoded in the encoding
** that was passed as the third argument when the collation sequence was
** registered. The user routine should return negative, zero or positive if
** the first string is less than, equal to, or greater than the second
** string. i.e. (STRING1 - STRING2).
**
** The sqlite3_create_collation_v2() works like sqlite3_create_collation()
** excapt that it takes an extra argument which is a destructor for
** the collation.  The destructor is called when the collation is
** destroyed and is passed a copy of the fourth parameter void* pointer
** of the sqlite3_create_collation_v2().  Collations are destroyed when
** they are overridden by later calls to the collation creation functions
** or when the [sqlite3*] database handle is closed using [sqlite3_close()].
**
** The sqlite3_create_collation_v2() interface is experimental and
** subject to change in future releases.  The other collation creation
** functions are stable.
*/
int sqlite3_create_collation(
  sqlite3*, 
  const char *zName, 
  int eTextRep, 
  void*,
  int(*xCompare)(void*,int,const void*,int,const void*)
);
int sqlite3_create_collation_v2(
  sqlite3*, 
  const char *zName, 
  int eTextRep, 
  void*,
  int(*xCompare)(void*,int,const void*,int,const void*),
  void(*xDestroy)(void*)
);
int sqlite3_create_collation16(
  sqlite3*, 
  const char *zName, 
  int eTextRep, 
  void*,
  int(*xCompare)(void*,int,const void*,int,const void*)
);

/*
** CAPI3REF: Collation Needed Callbacks
**
** To avoid having to register all collation sequences before a database
** can be used, a single callback function may be registered with the
** database handle to be called whenever an undefined collation sequence is
** required.
**
** If the function is registered using the sqlite3_collation_needed() API,
** then it is passed the names of undefined collation sequences as strings
** encoded in UTF-8. If sqlite3_collation_needed16() is used, the names
** are passed as UTF-16 in machine native byte order. A call to either
** function replaces any existing callback.
**
** When the callback is invoked, the first argument passed is a copy
** of the second argument to sqlite3_collation_needed() or
** sqlite3_collation_needed16(). The second argument is the database
** handle. The third argument is one of [SQLITE_UTF8], [SQLITE_UTF16BE], or
** [SQLITE_UTF16LE], indicating the most desirable form of the collation
** sequence function required. The fourth parameter is the name of the
** required collation sequence.
**
** The callback function should register the desired collation using
** [sqlite3_create_collation()], [sqlite3_create_collation16()], or
** [sqlite3_create_collation_v2()].
*/
int sqlite3_collation_needed(
  sqlite3*, 
  void*, 
  void(*)(void*,sqlite3*,int eTextRep,const char*)
);
int sqlite3_collation_needed16(
  sqlite3*, 
  void*,
  void(*)(void*,sqlite3*,int eTextRep,const void*)
);

/*
** Specify the key for an encrypted database.  This routine should be
** called right after sqlite3_open().
**
** The code to implement this API is not available in the public release
** of SQLite.
*/
SQLITE_API int sqlite3_key(
  sqlite3 *db,                   /* Database to be rekeyed */
  const void *pKey, int nKey     /* The key */
);

/*
** Change the key on an open database.  If the current database is not
** encrypted, this routine will encrypt it.  If pNew==0 or nNew==0, the
** database is decrypted.
**
** The code to implement this API is not available in the public release
** of SQLite.
*/
SQLITE_API int sqlite3_rekey(
  sqlite3 *db,                   /* Database to be rekeyed */
  const void *pKey, int nKey     /* The new key */
);

/*
** CAPI3REF:  Suspend Execution For A Short Time
**
** This function causes the current thread to suspend execution
** a number of milliseconds specified in its parameter.
**
** If the operating system does not support sleep requests with 
** millisecond time resolution, then the time will be rounded up to 
** the nearest second. The number of milliseconds of sleep actually 
** requested from the operating system is returned.
*/
SQLITE_API int sqlite3_sleep(int);

/*
** CAPI3REF:  Name Of The Folder Holding Temporary Files
**
** If this global variable is made to point to a string which is
** the name of a folder (a.ka. directory), then all temporary files
** created by SQLite will be placed in that directory.  If this variable
** is NULL pointer, then SQLite does a search for an appropriate temporary
** file directory.
**
** Once [sqlite3_open()] has been called, changing this variable will
** invalidate the current temporary database, if any.  Generally speaking,
** it is not safe to invoke this routine after [sqlite3_open()] has
** been called.
*/
extern char *sqlite3_temp_directory;

/*
** CAPI3REF:  Test To See If The Databse Is In Auto-Commit Mode
**
** Test to see whether or not the database connection is in autocommit
** mode.  Return TRUE if it is and FALSE if not.  Autocommit mode is on
** by default.  Autocommit is disabled by a BEGIN statement and reenabled
** by the next COMMIT or ROLLBACK.
*/
int sqlite3_get_autocommit(sqlite3*);

/*
** CAPI3REF:  Find The Database Handle Associated With A Prepared Statement
**
** Return the [sqlite3*] database handle to which a
** [sqlite3_stmt | prepared statement] belongs.
** This is the same database handle that was
** the first argument to the [sqlite3_prepare_v2()] or its variants
** that was used to create the statement in the first place.
*/
sqlite3 *sqlite3_db_handle(sqlite3_stmt*);


/*
** CAPI3REF: Commit And Rollback Notification Callbacks
**
** These routines
** register callback functions to be invoked whenever a transaction
** is committed or rolled back.  The pArg argument is passed through
** to the callback.  If the callback on a commit hook function 
** returns non-zero, then the commit is converted into a rollback.
**
** If another function was previously registered, its pArg value is returned.
** Otherwise NULL is returned.
**
** Registering a NULL function disables the callback.
**
** For the purposes of this API, a transaction is said to have been 
** rolled back if an explicit "ROLLBACK" statement is executed, or
** an error or constraint causes an implicit rollback to occur. The 
** callback is not invoked if a transaction is automatically rolled
** back because the database connection is closed.
**
** These are experimental interfaces and are subject to change.
*/
void *sqlite3_commit_hook(sqlite3*, int(*)(void*), void*);
void *sqlite3_rollback_hook(sqlite3*, void(*)(void *), void*);

/*
** CAPI3REF: Data Change Notification Callbacks
**
** Register a callback function with the database connection identified by the 
** first argument to be invoked whenever a row is updated, inserted or deleted.
** Any callback set by a previous call to this function for the same 
** database connection is overridden.
**
** The second argument is a pointer to the function to invoke when a 
** row is updated, inserted or deleted. The first argument to the callback is
** a copy of the third argument to sqlite3_update_hook(). The second callback 
** argument is one of SQLITE_INSERT, SQLITE_DELETE or SQLITE_UPDATE, depending
** on the operation that caused the callback to be invoked. The third and 
** fourth arguments to the callback contain pointers to the database and 
** table name containing the affected row. The final callback parameter is 
** the rowid of the row. In the case of an update, this is the rowid after 
** the update takes place.
**
** The update hook is not invoked when internal system tables are
** modified (i.e. sqlite_master and sqlite_sequence).
**
** If another function was previously registered, its pArg value is returned.
** Otherwise NULL is returned.
*/
void *sqlite3_update_hook(
  sqlite3*, 
  void(*)(void *,int ,char const *,char const *,sqlite_int64),
  void*
);

/*
** CAPI3REF:  Enable Or Disable Shared Pager Cache
**
** This routine enables or disables the sharing of the database cache
** and schema data structures between connections to the same database.
** Sharing is enabled if the argument is true and disabled if the argument
** is false.
**
** Cache sharing is enabled and disabled on a thread-by-thread basis.
** Each call to this routine enables or disables cache sharing only for
** connections created in the same thread in which this routine is called.
** There is no mechanism for sharing cache between database connections
** running in different threads.
**
** Sharing must be disabled prior to shutting down a thread or else
** the thread will leak memory.  Call this routine with an argument of
** 0 to turn off sharing.  Or use the sqlite3_thread_cleanup() API.
**
** This routine must not be called when any database connections
** are active in the current thread.  Enabling or disabling shared
** cache while there are active database connections will result
** in memory corruption.
**
** When the shared cache is enabled, the
** following routines must always be called from the same thread:
** [sqlite3_open()], [sqlite3_prepare_v2()], [sqlite3_step()],
** [sqlite3_reset()], [sqlite3_finalize()], and [sqlite3_close()].
** This is due to the fact that the shared cache makes use of
** thread-specific storage so that it will be available for sharing
** with other connections.
**
** Virtual tables cannot be used with a shared cache.  When shared
** cache is enabled, the sqlite3_create_module() API used to register
** virtual tables will always return an error.
**
** This routine returns [SQLITE_OK] if shared cache was
** enabled or disabled successfully.  An [SQLITE_ERROR | error code]
** is returned otherwise.
**
** Shared cache is disabled by default for backward compatibility.
*/
int sqlite3_enable_shared_cache(int);

/*
** CAPI3REF:  Attempt To Free Heap Memory
**
** Attempt to free N bytes of heap memory by deallocating non-essential
** memory allocations held by the database library (example: memory 
** used to cache database pages to improve performance).
**
** This function is not a part of standard builds.  It is only created
** if SQLite is compiled with the SQLITE_ENABLE_MEMORY_MANAGEMENT macro.
*/
int sqlite3_release_memory(int);

/*
** CAPI3REF:  Impose A Limit On Heap Size
**
** Place a "soft" limit on the amount of heap memory that may be allocated by
** SQLite within the current thread. If an internal allocation is requested 
** that would exceed the specified limit, [sqlite3_release_memory()] is invoked
** one or more times to free up some space before the allocation is made.
**
** The limit is called "soft", because if [sqlite3_release_memory()] cannot free
** sufficient memory to prevent the limit from being exceeded, the memory is
** allocated anyway and the current operation proceeds.
**
** Prior to shutting down a thread sqlite3_soft_heap_limit() must be set to 
** zero (the default) or else the thread will leak memory. Alternatively, use
** the [sqlite3_thread_cleanup()] API.
**
** A negative or zero value for N means that there is no soft heap limit and
** [sqlite3_release_memory()] will only be called when memory is exhaused.
** The default value for the soft heap limit is zero.
**
** SQLite makes a best effort to honor the soft heap limit.  But if it
** is unable to reduce memory usage below the soft limit, execution will
** continue without error or notification.  This is why the limit is 
** called a "soft" limit.  It is advisory only.
**
** This function is only available if the library was compiled with the 
** SQLITE_ENABLE_MEMORY_MANAGEMENT option set.
** memory-management has been enabled.
*/
void sqlite3_soft_heap_limit(int);

/*
** CAPI3REF:  Clean Up Thread Local Storage
**
** This routine makes sure that all thread-local storage has been
** deallocated for the current thread.
**
** This routine is not technically necessary.  All thread-local storage
** will be automatically deallocated once memory-management and
** shared-cache are disabled and the soft heap limit has been set
** to zero.  This routine is provided as a convenience for users who
** want to make absolutely sure they have not forgotten something
** prior to killing off a thread.
*/
void sqlite3_thread_cleanup(void);

/*
** CAPI3REF:  Extract Metadata About A Column Of A Table
**
** This routine
** returns meta-data about a specific column of a specific database
** table accessible using the connection handle passed as the first function 
** argument.
**
** The column is identified by the second, third and fourth parameters to 
** this function. The second parameter is either the name of the database
** (i.e. "main", "temp" or an attached database) containing the specified
** table or NULL. If it is NULL, then all attached databases are searched
** for the table using the same algorithm as the database engine uses to 
** resolve unqualified table references.
**
** The third and fourth parameters to this function are the table and column 
** name of the desired column, respectively. Neither of these parameters 
** may be NULL.
**
** Meta information is returned by writing to the memory locations passed as
** the 5th and subsequent parameters to this function. Any of these 
** arguments may be NULL, in which case the corresponding element of meta 
** information is ommitted.
**
** <pre>
** Parameter     Output Type      Description
** -----------------------------------
**
**   5th         const char*      Data type
**   6th         const char*      Name of the default collation sequence 
**   7th         int              True if the column has a NOT NULL constraint
**   8th         int              True if the column is part of the PRIMARY KEY
**   9th         int              True if the column is AUTOINCREMENT
** </pre>
**
**
** The memory pointed to by the character pointers returned for the 
** declaration type and collation sequence is valid only until the next 
** call to any sqlite API function.
**
** If the specified table is actually a view, then an error is returned.
**
** If the specified column is "rowid", "oid" or "_rowid_" and an 
** INTEGER PRIMARY KEY column has been explicitly declared, then the output 
** parameters are set for the explicitly declared column. If there is no
** explicitly declared IPK column, then the output parameters are set as 
** follows:
**
** <pre>
**     data type: "INTEGER"
**     collation sequence: "BINARY"
**     not null: 0
**     primary key: 1
**     auto increment: 0
** </pre>
**
** This function may load one or more schemas from database files. If an
** error occurs during this process, or if the requested table or column
** cannot be found, an SQLITE error code is returned and an error message
** left in the database handle (to be retrieved using sqlite3_errmsg()).
**
** This API is only available if the library was compiled with the
** SQLITE_ENABLE_COLUMN_METADATA preprocessor symbol defined.
*/
int sqlite3_table_column_metadata(
  sqlite3 *db,                /* Connection handle */
  const char *zDbName,        /* Database name or NULL */
  const char *zTableName,     /* Table name */
  const char *zColumnName,    /* Column name */
  char const **pzDataType,    /* OUTPUT: Declared data type */
  char const **pzCollSeq,     /* OUTPUT: Collation sequence name */
  int *pNotNull,              /* OUTPUT: True if NOT NULL constraint exists */
  int *pPrimaryKey,           /* OUTPUT: True if column part of PK */
  int *pAutoinc               /* OUTPUT: True if colums is auto-increment */
);

/*
** CAPI3REF: Load An Extension
**
** Attempt to load an SQLite extension library contained in the file
** zFile.  The entry point is zProc.  zProc may be 0 in which case the
** name of the entry point defaults to "sqlite3_extension_init".
**
** Return [SQLITE_OK] on success and [SQLITE_ERROR] if something goes wrong.
**
** If an error occurs and pzErrMsg is not 0, then fill *pzErrMsg with 
** error message text.  The calling function should free this memory
** by calling [sqlite3_free()].
**
** Extension loading must be enabled using [sqlite3_enable_load_extension()]
** prior to calling this API or an error will be returned.
*/
int sqlite3_load_extension(
  sqlite3 *db,          /* Load the extension into this database connection */
  const char *zFile,    /* Name of the shared library containing extension */
  const char *zProc,    /* Entry point.  Derived from zFile if 0 */
  char **pzErrMsg       /* Put error message here if not 0 */
);

/*
** CAPI3REF:  Enable Or Disable Extension Loading
**
** So as not to open security holes in older applications that are
** unprepared to deal with extension loading, and as a means of disabling
** extension loading while evaluating user-entered SQL, the following
** API is provided to turn the [sqlite3_load_extension()] mechanism on and
** off.  It is off by default.  See ticket #1863.
**
** Call this routine with onoff==1 to turn extension loading on
** and call it with onoff==0 to turn it back off again.
*/
int sqlite3_enable_load_extension(sqlite3 *db, int onoff);

/*
** CAPI3REF: Make Arrangements To Automatically Load An Extension
**
** Register an extension entry point that is automatically invoked
** whenever a new database connection is opened using
** [sqlite3_open()] or [sqlite3_open16()].
**
** This API can be invoked at program startup in order to register
** one or more statically linked extensions that will be available
** to all new database connections.
**
** Duplicate extensions are detected so calling this routine multiple
** times with the same extension is harmless.
**
** This routine stores a pointer to the extension in an array
** that is obtained from malloc().  If you run a memory leak
** checker on your program and it reports a leak because of this
** array, then invoke [sqlite3_automatic_extension_reset()] prior
** to shutdown to free the memory.
**
** Automatic extensions apply across all threads.
**
** This interface is experimental and is subject to change or
** removal in future releases of SQLite.
*/
int sqlite3_auto_extension(void *xEntryPoint);


/*
** CAPI3REF: Reset Automatic Extension Loading
**
** Disable all previously registered automatic extensions.  This
** routine undoes the effect of all prior [sqlite3_automatic_extension()]
** calls.
**
** This call disabled automatic extensions in all threads.
**
** This interface is experimental and is subject to change or
** removal in future releases of SQLite.
*/
void sqlite3_reset_auto_extension(void);


/*
****** EXPERIMENTAL - subject to change without notice **************
**
** The interface to the virtual-table mechanism is currently considered
** to be experimental.  The interface might change in incompatible ways.
** If this is a problem for you, do not use the interface at this time.
**
** When the virtual-table mechanism stablizes, we will declare the
** interface fixed, support it indefinitely, and remove this comment.
*/

/*
** Structures used by the virtual table interface
*/
typedef struct sqlite3_vtab sqlite3_vtab;
typedef struct sqlite3_index_info sqlite3_index_info;
typedef struct sqlite3_vtab_cursor sqlite3_vtab_cursor;
typedef struct sqlite3_module sqlite3_module;

/*
** A module is a class of virtual tables.  Each module is defined
** by an instance of the following structure.  This structure consists
** mostly of methods for the module.
*/
struct sqlite3_module {
  int iVersion;
  int (*xCreate)(sqlite3*, void *pAux,
               int argc, const char *const*argv,
               sqlite3_vtab **ppVTab, char**);
  int (*xConnect)(sqlite3*, void *pAux,
               int argc, const char *const*argv,
               sqlite3_vtab **ppVTab, char**);
  int (*xBestIndex)(sqlite3_vtab *pVTab, sqlite3_index_info*);
  int (*xDisconnect)(sqlite3_vtab *pVTab);
  int (*xDestroy)(sqlite3_vtab *pVTab);
  int (*xOpen)(sqlite3_vtab *pVTab, sqlite3_vtab_cursor **ppCursor);
  int (*xClose)(sqlite3_vtab_cursor*);
  int (*xFilter)(sqlite3_vtab_cursor*, int idxNum, const char *idxStr,
                int argc, sqlite3_value **argv);
  int (*xNext)(sqlite3_vtab_cursor*);
  int (*xEof)(sqlite3_vtab_cursor*);
  int (*xColumn)(sqlite3_vtab_cursor*, sqlite3_context*, int);
  int (*xRowid)(sqlite3_vtab_cursor*, sqlite_int64 *pRowid);
  int (*xUpdate)(sqlite3_vtab *, int, sqlite3_value **, sqlite_int64 *);
  int (*xBegin)(sqlite3_vtab *pVTab);
  int (*xSync)(sqlite3_vtab *pVTab);
  int (*xCommit)(sqlite3_vtab *pVTab);
  int (*xRollback)(sqlite3_vtab *pVTab);
  int (*xFindFunction)(sqlite3_vtab *pVtab, int nArg, const char *zName,
                       void (**pxFunc)(sqlite3_context*,int,sqlite3_value**),
                       void **ppArg);

  int (*xRename)(sqlite3_vtab *pVtab, const char *zNew);
};

/*
** The sqlite3_index_info structure and its substructures is used to
** pass information into and receive the reply from the xBestIndex
** method of an sqlite3_module.  The fields under **Inputs** are the
** inputs to xBestIndex and are read-only.  xBestIndex inserts its
** results into the **Outputs** fields.
**
** The aConstraint[] array records WHERE clause constraints of the
** form:
**
**         column OP expr
**
** Where OP is =, <, <=, >, or >=.  The particular operator is stored
** in aConstraint[].op.  The index of the column is stored in 
** aConstraint[].iColumn.  aConstraint[].usable is TRUE if the
** expr on the right-hand side can be evaluated (and thus the constraint
** is usable) and false if it cannot.
**
** The optimizer automatically inverts terms of the form "expr OP column"
** and makes other simplificatinos to the WHERE clause in an attempt to
** get as many WHERE clause terms into the form shown above as possible.
** The aConstraint[] array only reports WHERE clause terms in the correct
** form that refer to the particular virtual table being queried.
**
** Information about the ORDER BY clause is stored in aOrderBy[].
** Each term of aOrderBy records a column of the ORDER BY clause.
**
** The xBestIndex method must fill aConstraintUsage[] with information
** about what parameters to pass to xFilter.  If argvIndex>0 then
** the right-hand side of the corresponding aConstraint[] is evaluated
** and becomes the argvIndex-th entry in argv.  If aConstraintUsage[].omit
** is true, then the constraint is assumed to be fully handled by the
** virtual table and is not checked again by SQLite.
**
** The idxNum and idxPtr values are recorded and passed into xFilter.
** sqlite3_free() is used to free idxPtr if needToFreeIdxPtr is true.
**
** The orderByConsumed means that output from xFilter will occur in
** the correct order to satisfy the ORDER BY clause so that no separate
** sorting step is required.
**
** The estimatedCost value is an estimate of the cost of doing the
** particular lookup.  A full scan of a table with N entries should have
** a cost of N.  A binary search of a table of N entries should have a
** cost of approximately log(N).
*/
struct sqlite3_index_info {
  /* Inputs */
  const int nConstraint;     /* Number of entries in aConstraint */
  const struct sqlite3_index_constraint {
     int iColumn;              /* Column on left-hand side of constraint */
     unsigned char op;         /* Constraint operator */
     unsigned char usable;     /* True if this constraint is usable */
     int iTermOffset;          /* Used internally - xBestIndex should ignore */
  } *const aConstraint;      /* Table of WHERE clause constraints */
  const int nOrderBy;        /* Number of terms in the ORDER BY clause */
  const struct sqlite3_index_orderby {
     int iColumn;              /* Column number */
     unsigned char desc;       /* True for DESC.  False for ASC. */
  } *const aOrderBy;         /* The ORDER BY clause */

  /* Outputs */
  struct sqlite3_index_constraint_usage {
    int argvIndex;           /* if >0, constraint is part of argv to xFilter */
    unsigned char omit;      /* Do not code a test for this constraint */
  } *const aConstraintUsage;
  int idxNum;                /* Number used to identify the index */
  char *idxStr;              /* String, possibly obtained from sqlite3_malloc */
  int needToFreeIdxStr;      /* Free idxStr using sqlite3_free() if true */
  int orderByConsumed;       /* True if output is already ordered */
  double estimatedCost;      /* Estimated cost of using this index */
};
#define SQLITE_INDEX_CONSTRAINT_EQ    2
#define SQLITE_INDEX_CONSTRAINT_GT    4
#define SQLITE_INDEX_CONSTRAINT_LE    8
#define SQLITE_INDEX_CONSTRAINT_LT    16
#define SQLITE_INDEX_CONSTRAINT_GE    32
#define SQLITE_INDEX_CONSTRAINT_MATCH 64

/*
** This routine is used to register a new module name with an SQLite
** connection.  Module names must be registered before creating new
** virtual tables on the module, or before using preexisting virtual
** tables of the module.
*/
int sqlite3_create_module(
  sqlite3 *db,               /* SQLite connection to register module with */
  const char *zName,         /* Name of the module */
  const sqlite3_module *,    /* Methods for the module */
  void *                     /* Client data for xCreate/xConnect */
);

/*
** This routine is identical to the sqlite3_create_module() method above,
** except that it allows a destructor function to be specified. It is
** even more experimental than the rest of the virtual tables API.
*/
int sqlite3_create_module_v2(
  sqlite3 *db,               /* SQLite connection to register module with */
  const char *zName,         /* Name of the module */
  const sqlite3_module *,    /* Methods for the module */
  void *,                    /* Client data for xCreate/xConnect */
  void(*xDestroy)(void*)     /* Module destructor function */
);

/*
** Every module implementation uses a subclass of the following structure
** to describe a particular instance of the module.  Each subclass will
** be taylored to the specific needs of the module implementation.   The
** purpose of this superclass is to define certain fields that are common
** to all module implementations.
**
** Virtual tables methods can set an error message by assigning a
** string obtained from sqlite3_mprintf() to zErrMsg.  The method should
** take care that any prior string is freed by a call to sqlite3_free()
** prior to assigning a new string to zErrMsg.  After the error message
** is delivered up to the client application, the string will be automatically
** freed by sqlite3_free() and the zErrMsg field will be zeroed.  Note
** that sqlite3_mprintf() and sqlite3_free() are used on the zErrMsg field
** since virtual tables are commonly implemented in loadable extensions which
** do not have access to sqlite3MPrintf() or sqlite3Free().
*/
struct sqlite3_vtab {
  const sqlite3_module *pModule;  /* The module for this virtual table */
  int nRef;                       /* Used internally */
  char *zErrMsg;                  /* Error message from sqlite3_mprintf() */
  /* Virtual table implementations will typically add additional fields */
};

/* Every module implementation uses a subclass of the following structure
** to describe cursors that point into the virtual table and are used
** to loop through the virtual table.  Cursors are created using the
** xOpen method of the module.  Each module implementation will define
** the content of a cursor structure to suit its own needs.
**
** This superclass exists in order to define fields of the cursor that
** are common to all implementations.
*/
struct sqlite3_vtab_cursor {
  sqlite3_vtab *pVtab;      /* Virtual table of this cursor */
  /* Virtual table implementations will typically add additional fields */
};

/*
** The xCreate and xConnect methods of a module use the following API
** to declare the format (the names and datatypes of the columns) of
** the virtual tables they implement.
*/
int sqlite3_declare_vtab(sqlite3*, const char *zCreateTable);

/*
** Virtual tables can provide alternative implementations of functions
** using the xFindFunction method.  But global versions of those functions
** must exist in order to be overloaded.
**
** This API makes sure a global version of a function with a particular
** name and number of parameters exists.  If no such function exists
** before this API is called, a new function is created.  The implementation
** of the new function always causes an exception to be thrown.  So
** the new function is not good for anything by itself.  Its only
** purpose is to be a place-holder function that can be overloaded
** by virtual tables.
**
** This API should be considered part of the virtual table interface,
** which is experimental and subject to change.
*/
int sqlite3_overload_function(sqlite3*, const char *zFuncName, int nArg);

/*
** The interface to the virtual-table mechanism defined above (back up
** to a comment remarkably similar to this one) is currently considered
** to be experimental.  The interface might change in incompatible ways.
** If this is a problem for you, do not use the interface at this time.
**
** When the virtual-table mechanism stablizes, we will declare the
** interface fixed, support it indefinitely, and remove this comment.
**
****** EXPERIMENTAL - subject to change without notice **************
*/

/*
** CAPI3REF: A Handle To An Open BLOB
**
** An instance of the following opaque structure is used to 
** represent an blob-handle.  A blob-handle is created by
** [sqlite3_blob_open()] and destroyed by [sqlite3_blob_close()].
** The [sqlite3_blob_read()] and [sqlite3_blob_write()] interfaces
** can be used to read or write small subsections of the blob.
** The [sqltie3_blob_size()] interface returns the size of the
** blob in bytes.
*/
typedef struct sqlite3_blob sqlite3_blob;

/*
** CAPI3REF: Open A BLOB For Incremental I/O
**
** Open a handle to the blob located in row iRow,, column zColumn, 
** table zTable in database zDb. i.e. the same blob that would
** be selected by:
**
** <pre>
**     SELECT zColumn FROM zDb.zTable WHERE rowid = iRow;
** </pre>
**
** If the flags parameter is non-zero, the blob is opened for 
** read and write access. If it is zero, the blob is opened for read 
** access.
**
** On success, [SQLITE_OK] is returned and the new 
** [sqlite3_blob | blob handle] is written to *ppBlob.
** Otherwise an error code is returned and 
** any value written to *ppBlob should not be used by the caller.
** This function sets the database-handle error code and message
** accessible via [sqlite3_errcode()] and [sqlite3_errmsg()].
*/
int sqlite3_blob_open(
  sqlite3*,
  const char *zDb,
  const char *zTable,
  const char *zColumn,
  sqlite_int64 iRow,
  int flags,
  sqlite3_blob **ppBlob
);

/*
** CAPI3REF:  Close A BLOB Handle
**
** Close an open [sqlite3_blob | blob handle].
*/
int sqlite3_blob_close(sqlite3_blob *);

/*
** CAPI3REF:  Return The Size Of An Open BLOB
**
** Return the size in bytes of the blob accessible via the open 
** [sqlite3_blob | blob-handle] passed as an argument.
*/
int sqlite3_blob_bytes(sqlite3_blob *);

/*
** CAPI3REF:  Read Data From A BLOB Incrementally
**
** This function is used to read data from an open 
** [sqlite3_blob | blob-handle] into a caller supplied buffer.
** n bytes of data are copied into buffer
** z from the open blob, starting at offset iOffset.
**
** On success, SQLITE_OK is returned. Otherwise, an 
** [SQLITE_ERROR | SQLite error code] or an
** [SQLITE_IOERR_READ | extended error code] is returned.
*/
int sqlite3_blob_read(sqlite3_blob *, void *z, int n, int iOffset);

/*
** CAPI3REF:  Write Data Into A BLOB Incrementally
**
** This function is used to write data into an open 
** [sqlite3_blob | blob-handle] from a user supplied buffer.
** n bytes of data are copied from the buffer
** pointed to by z into the open blob, starting at offset iOffset.
**
** If the [sqlite3_blob | blob-handle] passed as the first argument
** was not opened for writing (the flags parameter to [sqlite3_blob_open()]
*** was zero), this function returns [SQLITE_READONLY].
**
** This function may only modify the contents of the blob, it is
** not possible to increase the size of a blob using this API. If
** offset iOffset is less than n bytes from the end of the blob, 
** [SQLITE_ERROR] is returned and no data is written.
**
** On success, SQLITE_OK is returned. Otherwise, an 
** [SQLITE_ERROR | SQLite error code] or an
** [SQLITE_IOERR_READ | extended error code] is returned.
*/
int sqlite3_blob_write(sqlite3_blob *, const void *z, int n, int iOffset);

/*
** Undo the hack that converts floating point types to integer for
** builds on processors without floating point support.
*/
#ifdef SQLITE_OMIT_FLOATING_POINT
# undef double
#endif

#if 0
}  /* End of the 'extern "C"' block */
#endif
#endif

/************** End of sqlite3.h *********************************************/
/************** Begin file date.c ********************************************/
/*
** 2003 October 31
**
** The author disclaims copyright to this source code.  In place of
** a legal notice, here is a blessing:
**
**    May you do good and not evil.
**    May you find forgiveness for yourself and forgive others.
**    May you share freely, never taking more than you give.
**
*************************************************************************
** This file contains the C functions that implement date and time
** functions for SQLite.  
**
** There is only one exported symbol in this file - the function
** sqlite3RegisterDateTimeFunctions() found at the bottom of the file.
** All other code has file scope.
**
** $Id: sqlite3.c,v 1.4 2007/07/21 08:51:43 rmsimpson Exp $
**
** SQLite processes all times and dates as Julian Day numbers.  The
** dates and times are stored as the number of days since noon
** in Greenwich on November 24, 4714 B.C. according to the Gregorian
** calendar system. 
**
** 1970-01-01 00:00:00 is JD 2440587.5
** 2000-01-01 00:00:00 is JD 2451544.5
**
** This implemention requires years to be expressed as a 4-digit number
** which means that only dates between 0000-01-01 and 9999-12-31 can
** be represented, even though julian day numbers allow a much wider
** range of dates.
**
** The Gregorian calendar system is used for all dates and times,
** even those that predate the Gregorian calendar.  Historians usually
** use the Julian calendar for dates prior to 1582-10-15 and for some
** dates afterwards, depending on locale.  Beware of this difference.
**
** The conversion algorithms are implemented based on descriptions
** in the following text:
**
**      Jean Meeus
**      Astronomical Algorithms, 2nd Edition, 1998
**      ISBM 0-943396-61-1
**      Willmann-Bell, Inc
**      Richmond, Virginia (USA)
*/
/************** Include sqliteInt.h in the middle of date.c ******************/
/************** Begin file sqliteInt.h ***************************************/
/*
** 2001 September 15
**
** The author disclaims copyright to this source code.  In place of
** a legal notice, here is a blessing:
**
**    May you do good and not evil.
**    May you find forgiveness for yourself and forgive others.
**    May you share freely, never taking more than you give.
**
*************************************************************************
** Internal interface definitions for SQLite.
**
** @(#) $Id: sqlite3.c,v 1.4 2007/07/21 08:51:43 rmsimpson Exp $
*/
#ifndef _SQLITEINT_H_
#define _SQLITEINT_H_
/************** Include sqliteLimit.h in the middle of sqliteInt.h ***********/
/************** Begin file sqliteLimit.h *************************************/
/*
** 2007 May 7
**
** The author disclaims copyright to this source code.  In place of
** a legal notice, here is a blessing:
**
**    May you do good and not evil.
**    May you find forgiveness for yourself and forgive others.
**    May you share freely, never taking more than you give.
**
*************************************************************************
** 
** This file defines various limits of what SQLite can process.
**
** @(#) $Id: sqlite3.c,v 1.4 2007/07/21 08:51:43 rmsimpson Exp $
*/

/*
** The maximum length of a TEXT or BLOB in bytes.   This also
** limits the size of a row in a table or index.
**
** The hard limit is the ability of a 32-bit signed integer
** to count the size: 2^31-1 or 2147483647.
*/
#ifndef SQLITE_MAX_LENGTH
# define SQLITE_MAX_LENGTH 1000000000
#endif

/*
** This is the maximum number of
**
**    * Columns in a table
**    * Columns in an index
**    * Columns in a view
**    * Terms in the SET clause of an UPDATE statement
**    * Terms in the result set of a SELECT statement
**    * Terms in the GROUP BY or ORDER BY clauses of a SELECT statement.
**    * Terms in the VALUES clause of an INSERT statement
**
** The hard upper limit here is 32676.  Most database people will
** tell you that in a well-normalized database, you usually should
** not have more than a dozen or so columns in any table.  And if
** that is the case, there is no point in having more than a few
** dozen values in any of the other situations described above.
*/
#ifndef SQLITE_MAX_COLUMN
# define SQLITE_MAX_COLUMN 2000
#endif

/*
** The maximum length of a single SQL statement in bytes.
** The hard limit here is the same as SQLITE_MAX_LENGTH.
*/
#ifndef SQLITE_MAX_SQL_LENGTH
# define SQLITE_MAX_SQL_LENGTH 1000000
#endif

/*
** The maximum depth of an expression tree. This is limited to 
** some extent by SQLITE_MAX_SQL_LENGTH. But sometime you might 
** want to place more severe limits on the complexity of an 
** expression. A value of 0 (the default) means do not enforce
** any limitation on expression tree depth.
*/
#ifndef SQLITE_MAX_EXPR_DEPTH
# define SQLITE_MAX_EXPR_DEPTH 1000
#endif

/*
** The maximum number of terms in a compound SELECT statement.
** The code generator for compound SELECT statements does one
** level of recursion for each term.  A stack overflow can result
** if the number of terms is too large.  In practice, most SQL
** never has more than 3 or 4 terms.  Use a value of 0 to disable
** any limit on the number of terms in a compount SELECT.
*/
#ifndef SQLITE_MAX_COMPOUND_SELECT
# define SQLITE_MAX_COMPOUND_SELECT 500
#endif

/*
** The maximum number of opcodes in a VDBE program.
** Not currently enforced.
*/
#ifndef SQLITE_MAX_VDBE_OP
# define SQLITE_MAX_VDBE_OP 25000
#endif

/*
** The maximum number of arguments to an SQL function.
*/
#ifndef SQLITE_MAX_FUNCTION_ARG
# define SQLITE_MAX_FUNCTION_ARG 100
#endif

/*
** The maximum number of in-memory pages to use for the main database
** table and for temporary tables.  The SQLITE_DEFAULT_CACHE_SIZE
*/
#ifndef SQLITE_DEFAULT_CACHE_SIZE
# define SQLITE_DEFAULT_CACHE_SIZE  2000
#endif
#ifndef SQLITE_DEFAULT_TEMP_CACHE_SIZE
# define SQLITE_DEFAULT_TEMP_CACHE_SIZE  500
#endif

/*
** The maximum number of attached databases.  This must be at least 2
** in order to support the main database file (0) and the file used to
** hold temporary tables (1).  And it must be less than 32 because
** we use a bitmask of databases with a u32 in places (for example
** the Parse.cookieMask field).
*/
#ifndef SQLITE_MAX_ATTACHED
# define SQLITE_MAX_ATTACHED 10
#endif


/*
** The maximum value of a ?nnn wildcard that the parser will accept.
*/
#ifndef SQLITE_MAX_VARIABLE_NUMBER
# define SQLITE_MAX_VARIABLE_NUMBER 999
#endif

/*
** The default size of a database page.
*/
#ifndef SQLITE_DEFAULT_PAGE_SIZE
# define SQLITE_DEFAULT_PAGE_SIZE 1024
#endif

/* Maximum page size.  The upper bound on this value is 32768.  This a limit
** imposed by the necessity of storing the value in a 2-byte unsigned integer
** and the fact that the page size must be a power of 2.
*/
#ifndef SQLITE_MAX_PAGE_SIZE
# define SQLITE_MAX_PAGE_SIZE 32768
#endif

/*
** Maximum number of pages in one database file.
**
** This is really just the default value for the max_page_count pragma.
** This value can be lowered (or raised) at run-time using that the
** max_page_count macro.
*/
#ifndef SQLITE_MAX_PAGE_COUNT
# define SQLITE_MAX_PAGE_COUNT 1073741823
#endif

/*
** Maximum length (in bytes) of the pattern in a LIKE or GLOB
** operator.
*/
#ifndef SQLITE_MAX_LIKE_PATTERN_LENGTH
# define SQLITE_MAX_LIKE_PATTERN_LENGTH 50000
#endif

/************** End of sqliteLimit.h *****************************************/
/************** Continuing where we left off in sqliteInt.h ******************/


#if defined(SQLITE_TCL) || defined(TCLSH)
# include <tcl.h>
#endif

/*
** Many people are failing to set -DNDEBUG=1 when compiling SQLite.
** Setting NDEBUG makes the code smaller and run faster.  So the following
** lines are added to automatically set NDEBUG unless the -DSQLITE_DEBUG=1
** option is set.  Thus NDEBUG becomes an opt-in rather than an opt-out
** feature.
*/
#if !defined(NDEBUG) && !defined(SQLITE_DEBUG) 
# define NDEBUG 1
#endif

/*
** These #defines should enable >2GB file support on Posix if the
** underlying operating system supports it.  If the OS lacks
** large file support, or if the OS is windows, these should be no-ops.
**
** Large file support can be disabled using the -DSQLITE_DISABLE_LFS switch
** on the compiler command line.  This is necessary if you are compiling
** on a recent machine (ex: RedHat 7.2) but you want your code to work
** on an older machine (ex: RedHat 6.0).  If you compile on RedHat 7.2
** without this option, LFS is enable.  But LFS does not exist in the kernel
** in RedHat 6.0, so the code won't work.  Hence, for maximum binary
** portability you should omit LFS.
**
** Similar is true for MacOS.  LFS is only supported on MacOS 9 and later.
*/
#ifndef SQLITE_DISABLE_LFS
# define _LARGE_FILE       1
# ifndef _FILE_OFFSET_BITS
#   define _FILE_OFFSET_BITS 64
# endif
# define _LARGEFILE_SOURCE 1
#endif

/************** Include hash.h in the middle of sqliteInt.h ******************/
/************** Begin file hash.h ********************************************/
/*
** 2001 September 22
**
** The author disclaims copyright to this source code.  In place of
** a legal notice, here is a blessing:
**
**    May you do good and not evil.
**    May you find forgiveness for yourself and forgive others.
**    May you share freely, never taking more than you give.
**
*************************************************************************
** This is the header file for the generic hash-table implemenation
** used in SQLite.
**
** $Id: sqlite3.c,v 1.4 2007/07/21 08:51:43 rmsimpson Exp $
*/
#ifndef _SQLITE_HASH_H_
#define _SQLITE_HASH_H_

/* Forward declarations of structures. */
typedef struct Hash Hash;
typedef struct HashElem HashElem;

/* A complete hash table is an instance of the following structure.
** The internals of this structure are intended to be opaque -- client
** code should not attempt to access or modify the fields of this structure
** directly.  Change this structure only by using the routines below.
** However, many of the "procedures" and "functions" for modifying and
** accessing this structure are really macros, so we can't really make
** this structure opaque.
*/
struct Hash {
  char keyClass;          /* SQLITE_HASH_INT, _POINTER, _STRING, _BINARY */
  char copyKey;           /* True if copy of key made on insert */
  int count;              /* Number of entries in this table */
  HashElem *first;        /* The first element of the array */
  void *(*xMalloc)(int);  /* malloc() function to use */
  void (*xFree)(void *);  /* free() function to use */
  int htsize;             /* Number of buckets in the hash table */
  struct _ht {            /* the hash table */
    int count;               /* Number of entries with this hash */
    HashElem *chain;         /* Pointer to first entry with this hash */
  } *ht;
};

/* Each element in the hash table is an instance of the following 
** structure.  All elements are stored on a single doubly-linked list.
**
** Again, this structure is intended to be opaque, but it can't really
** be opaque because it is used by macros.
*/
struct HashElem {
  HashElem *next, *prev;   /* Next and previous elements in the table */
  void *data;              /* Data associated with this element */
  void *pKey; int nKey;    /* Key associated with this element */
};

/*
** There are 4 different modes of operation for a hash table:
**
**   SQLITE_HASH_INT         nKey is used as the key and pKey is ignored.
**
**   SQLITE_HASH_POINTER     pKey is used as the key and nKey is ignored.
**
**   SQLITE_HASH_STRING      pKey points to a string that is nKey bytes long
**                           (including the null-terminator, if any).  Case
**                           is ignored in comparisons.
**
**   SQLITE_HASH_BINARY      pKey points to binary data nKey bytes long. 
**                           memcmp() is used to compare keys.
**
** A copy of the key is made for SQLITE_HASH_STRING and SQLITE_HASH_BINARY
** if the copyKey parameter to HashInit is 1.  
*/
/* #define SQLITE_HASH_INT       1 // NOT USED */
/* #define SQLITE_HASH_POINTER   2 // NOT USED */
#define SQLITE_HASH_STRING    3
#define SQLITE_HASH_BINARY    4

/*
** Access routines.  To delete, insert a NULL pointer.
*/
SQLITE_PRIVATE void sqlite3HashInit(Hash*, int keytype, int copyKey);
SQLITE_PRIVATE void *sqlite3HashInsert(Hash*, const void *pKey, int nKey, void *pData);
SQLITE_PRIVATE void *sqlite3HashFind(const Hash*, const void *pKey, int nKey);
SQLITE_PRIVATE void sqlite3HashClear(Hash*);

/*
** Macros for looping over all elements of a hash table.  The idiom is
** like this:
**
**   Hash h;
**   HashElem *p;
**   ...
**   for(p=sqliteHashFirst(&h); p; p=sqliteHashNext(p)){
**     SomeStructure *pData = sqliteHashData(p);
**     // do something with pData
**   }
*/
#define sqliteHashFirst(H)  ((H)->first)
#define sqliteHashNext(E)   ((E)->next)
#define sqliteHashData(E)   ((E)->data)
#define sqliteHashKey(E)    ((E)->pKey)
#define sqliteHashKeysize(E) ((E)->nKey)

/*
** Number of entries in a hash table
*/
#define sqliteHashCount(H)  ((H)->count)

#endif /* _SQLITE_HASH_H_ */

/************** End of hash.h ************************************************/
/************** Continuing where we left off in sqliteInt.h ******************/
/************** Include parse.h in the middle of sqliteInt.h *****************/
/************** Begin file parse.h *******************************************/
#define TK_SEMI                            1
#define TK_EXPLAIN                         2
#define TK_QUERY                           3
#define TK_PLAN                            4
#define TK_BEGIN                           5
#define TK_TRANSACTION                     6
#define TK_DEFERRED                        7
#define TK_IMMEDIATE                       8
#define TK_EXCLUSIVE                       9
#define TK_COMMIT                         10
#define TK_END                            11
#define TK_ROLLBACK                       12
#define TK_CREATE                         13
#define TK_TABLE                          14
#define TK_IF                             15
#define TK_NOT                            16
#define TK_EXISTS                         17
#define TK_TEMP                           18
#define TK_LP                             19
#define TK_RP                             20
#define TK_AS                             21
#define TK_COMMA                          22
#define TK_ID                             23
#define TK_ABORT                          24
#define TK_AFTER                          25
#define TK_ANALYZE                        26
#define TK_ASC                            27
#define TK_ATTACH                         28
#define TK_BEFORE                         29
#define TK_CASCADE                        30
#define TK_CAST                           31
#define TK_CONFLICT                       32
#define TK_DATABASE                       33
#define TK_DESC                           34
#define TK_DETACH                         35
#define TK_EACH                           36
#define TK_FAIL                           37
#define TK_FOR                            38
#define TK_IGNORE                         39
#define TK_INITIALLY                      40
#define TK_INSTEAD                        41
#define TK_LIKE_KW                        42
#define TK_MATCH                          43
#define TK_KEY                            44
#define TK_OF                             45
#define TK_OFFSET                         46
#define TK_PRAGMA                         47
#define TK_RAISE                          48
#define TK_REPLACE                        49
#define TK_RESTRICT                       50
#define TK_ROW                            51
#define TK_TRIGGER                        52
#define TK_VACUUM                         53
#define TK_VIEW                           54
#define TK_VIRTUAL                        55
#define TK_REINDEX                        56
#define TK_RENAME                         57
#define TK_CTIME_KW                       58
#define TK_ANY                            59
#define TK_OR                             60
#define TK_AND                            61
#define TK_IS                             62
#define TK_BETWEEN                        63
#define TK_IN                             64
#define TK_ISNULL                         65
#define TK_NOTNULL                        66
#define TK_NE                             67
#define TK_EQ                             68
#define TK_GT                             69
#define TK_LE                             70
#define TK_LT                             71
#define TK_GE                             72
#define TK_ESCAPE                         73
#define TK_BITAND                         74
#define TK_BITOR                          75
#define TK_LSHIFT                         76
#define TK_RSHIFT                         77
#define TK_PLUS                           78
#define TK_MINUS                          79
#define TK_STAR                           80
#define TK_SLASH                          81
#define TK_REM                            82
#define TK_CONCAT                         83
#define TK_COLLATE                        84
#define TK_UMINUS                         85
#define TK_UPLUS                          86
#define TK_BITNOT                         87
#define TK_STRING                         88
#define TK_JOIN_KW                        89
#define TK_CONSTRAINT                     90
#define TK_DEFAULT                        91
#define TK_NULL                           92
#define TK_PRIMARY                        93
#define TK_UNIQUE                         94
#define TK_CHECK                          95
#define TK_REFERENCES                     96
#define TK_AUTOINCR                       97
#define TK_ON                             98
#define TK_DELETE                         99
#define TK_UPDATE                         100
#define TK_INSERT                         101
#define TK_SET                            102
#define TK_DEFERRABLE                     103
#define TK_FOREIGN                        104
#define TK_DROP                           105
#define TK_UNION                          106
#define TK_ALL                            107
#define TK_EXCEPT                         108
#define TK_INTERSECT                      109
#define TK_SELECT                         110
#define TK_DISTINCT                       111
#define TK_DOT                            112
#define TK_FROM                           113
#define TK_JOIN                           114
#define TK_USING                          115
#define TK_ORDER                          116
#define TK_BY                             117
#define TK_GROUP                          118
#define TK_HAVING                         119
#define TK_LIMIT                          120
#define TK_WHERE                          121
#define TK_INTO                           122
#define TK_VALUES                         123
#define TK_INTEGER                        124
#define TK_FLOAT                          125
#define TK_BLOB                           126
#define TK_REGISTER                       127
#define TK_VARIABLE                       128
#define TK_CASE                           129
#define TK_WHEN                           130
#define TK_THEN                           131
#define TK_ELSE                           132
#define TK_INDEX                          133
#define TK_ALTER                          134
#define TK_TO                             135
#define TK_ADD                            136
#define TK_COLUMNKW                       137
#define TK_TO_TEXT                        138
#define TK_TO_BLOB                        139
#define TK_TO_NUMERIC                     140
#define TK_TO_INT                         141
#define TK_TO_REAL                        142
#define TK_END_OF_FILE                    143
#define TK_ILLEGAL                        144
#define TK_SPACE                          145
#define TK_UNCLOSED_STRING                146
#define TK_COMMENT                        147
#define TK_FUNCTION                       148
#define TK_COLUMN                         149
#define TK_AGG_FUNCTION                   150
#define TK_AGG_COLUMN                     151
#define TK_CONST_FUNC                     152

/************** End of parse.h ***********************************************/
/************** Continuing where we left off in sqliteInt.h ******************/
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <assert.h>
#include <stddef.h>

#define sqlite3_isnan(X)  ((X)!=(X))

/*
** If compiling for a processor that lacks floating point support,
** substitute integer for floating-point
*/
#ifdef SQLITE_OMIT_FLOATING_POINT
# define double sqlite_int64
# define LONGDOUBLE_TYPE sqlite_int64
# ifndef SQLITE_BIG_DBL
#   define SQLITE_BIG_DBL (0x7fffffffffffffff)
# endif
# define SQLITE_OMIT_DATETIME_FUNCS 1
# define SQLITE_OMIT_TRACE 1
# undef SQLITE_MIXED_ENDIAN_64BIT_FLOAT
#endif
#ifndef SQLITE_BIG_DBL
# define SQLITE_BIG_DBL (1e99)
#endif

/*
** OMIT_TEMPDB is set to 1 if SQLITE_OMIT_TEMPDB is defined, or 0
** afterward. Having this macro allows us to cause the C compiler 
** to omit code used by TEMP tables without messy #ifndef statements.
*/
#ifdef SQLITE_OMIT_TEMPDB
#define OMIT_TEMPDB 1
#else
#define OMIT_TEMPDB 0
#endif

/*
** If the following macro is set to 1, then NULL values are considered
** distinct when determining whether or not two entries are the same
** in a UNIQUE index.  This is the way PostgreSQL, Oracle, DB2, MySQL,
** OCELOT, and Firebird all work.  The SQL92 spec explicitly says this
** is the way things are suppose to work.
**
** If the following macro is set to 0, the NULLs are indistinct for
** a UNIQUE index.  In this mode, you can only have a single NULL entry
** for a column declared UNIQUE.  This is the way Informix and SQL Server
** work.
*/
#define NULL_DISTINCT_FOR_UNIQUE 1

/*
** The "file format" number is an integer that is incremented whenever
** the VDBE-level file format changes.  The following macros define the
** the default file format for new databases and the maximum file format
** that the library can read.
*/
#define SQLITE_MAX_FILE_FORMAT 4
#ifndef SQLITE_DEFAULT_FILE_FORMAT
# define SQLITE_DEFAULT_FILE_FORMAT 1
#endif

/*
** Provide a default value for TEMP_STORE in case it is not specified
** on the command-line
*/
#ifndef TEMP_STORE
# define TEMP_STORE 1
#endif

/*
** GCC does not define the offsetof() macro so we'll have to do it
** ourselves.
*/
#ifndef offsetof
#define offsetof(STRUCTURE,FIELD) ((int)((char*)&((STRUCTURE*)0)->FIELD))
#endif

/*
** Check to see if this machine uses EBCDIC.  (Yes, believe it or
** not, there are still machines out there that use EBCDIC.)
*/
#if 'A' == '\301'
# define SQLITE_EBCDIC 1
#else
# define SQLITE_ASCII 1
#endif

/*
** Integers of known sizes.  These typedefs might change for architectures
** where the sizes very.  Preprocessor macros are available so that the
** types can be conveniently redefined at compile-type.  Like this:
**
**         cc '-DUINTPTR_TYPE=long long int' ...
*/
#ifndef UINT32_TYPE
# define UINT32_TYPE unsigned int
#endif
#ifndef UINT16_TYPE
# define UINT16_TYPE unsigned short int
#endif
#ifndef INT16_TYPE
# define INT16_TYPE short int
#endif
#ifndef UINT8_TYPE
# define UINT8_TYPE unsigned char
#endif
#ifndef INT8_TYPE
# define INT8_TYPE signed char
#endif
#ifndef LONGDOUBLE_TYPE
# define LONGDOUBLE_TYPE long double
#endif
typedef sqlite_int64 i64;          /* 8-byte signed integer */
typedef sqlite_uint64 u64;         /* 8-byte unsigned integer */
typedef UINT32_TYPE u32;           /* 4-byte unsigned integer */
typedef UINT16_TYPE u16;           /* 2-byte unsigned integer */
typedef INT16_TYPE i16;            /* 2-byte signed integer */
typedef UINT8_TYPE u8;             /* 1-byte unsigned integer */
typedef UINT8_TYPE i8;             /* 1-byte signed integer */

/*
** Macros to determine whether the machine is big or little endian,
** evaluated at runtime.
*/
extern const int sqlite3one;
#if defined(i386) || defined(__i386__) || defined(_M_IX86)
# define SQLITE_BIGENDIAN    0
# define SQLITE_LITTLEENDIAN 1
# define SQLITE_UTF16NATIVE  SQLITE_UTF16LE
#else
# define SQLITE_BIGENDIAN    (*(char *)(&sqlite3one)==0)
# define SQLITE_LITTLEENDIAN (*(char *)(&sqlite3one)==1)
# define SQLITE_UTF16NATIVE (SQLITE_BIGENDIAN?SQLITE_UTF16BE:SQLITE_UTF16LE)
#endif

/*
** An instance of the following structure is used to store the busy-handler
** callback for a given sqlite handle. 
**
** The sqlite.busyHandler member of the sqlite struct contains the busy
** callback for the database handle. Each pager opened via the sqlite
** handle is passed a pointer to sqlite.busyHandler. The busy-handler
** callback is currently invoked only from within pager.c.
*/
typedef struct BusyHandler BusyHandler;
struct BusyHandler {
  int (*xFunc)(void *,int);  /* The busy callback */
  void *pArg;                /* First arg to busy callback */
  int nBusy;                 /* Incremented with each busy call */
};

/*
** Defer sourcing vdbe.h and btree.h until after the "u8" and 
** "BusyHandler typedefs.
*/
/************** Include vdbe.h in the middle of sqliteInt.h ******************/
/************** Begin file vdbe.h ********************************************/
/*
** 2001 September 15
**
** The author disclaims copyright to this source code.  In place of
** a legal notice, here is a blessing:
**
**    May you do good and not evil.
**    May you find forgiveness for yourself and forgive others.
**    May you share freely, never taking more than you give.
**
*************************************************************************
** Header file for the Virtual DataBase Engine (VDBE)
**
** This header defines the interface to the virtual database engine
** or VDBE.  The VDBE implements an abstract machine that runs a
** simple program to access and modify the underlying database.
**
** $Id: sqlite3.c,v 1.4 2007/07/21 08:51:43 rmsimpson Exp $
*/
#ifndef _SQLITE_VDBE_H_
#define _SQLITE_VDBE_H_

/*
** A single VDBE is an opaque structure named "Vdbe".  Only routines
** in the source file sqliteVdbe.c are allowed to see the insides
** of this structure.
*/
typedef struct Vdbe Vdbe;

/*
** A single instruction of the virtual machine has an opcode
** and as many as three operands.  The instruction is recorded
** as an instance of the following structure:
*/
struct VdbeOp {
  u8 opcode;          /* What operation to perform */
  int p1;             /* First operand */
  int p2;             /* Second parameter (often the jump destination) */
  char *p3;           /* Third parameter */
  int p3type;         /* One of the P3_xxx constants defined below */
#ifdef VDBE_PROFILE
  int cnt;            /* Number of times this instruction was executed */
  long long cycles;   /* Total time spend executing this instruction */
#endif
};
typedef struct VdbeOp VdbeOp;

/*
** A smaller version of VdbeOp used for the VdbeAddOpList() function because
** it takes up less space.
*/
struct VdbeOpList {
  u8 opcode;          /* What operation to perform */
  signed char p1;     /* First operand */
  short int p2;       /* Second parameter (often the jump destination) */
  char *p3;           /* Third parameter */
};
typedef struct VdbeOpList VdbeOpList;

/*
** Allowed values of VdbeOp.p3type
*/
#define P3_NOTUSED    0   /* The P3 parameter is not used */
#define P3_DYNAMIC  (-1)  /* Pointer to a string obtained from sqliteMalloc() */
#define P3_STATIC   (-2)  /* Pointer to a static string */
#define P3_COLLSEQ  (-4)  /* P3 is a pointer to a CollSeq structure */
#define P3_FUNCDEF  (-5)  /* P3 is a pointer to a FuncDef structure */
#define P3_KEYINFO  (-6)  /* P3 is a pointer to a KeyInfo structure */
#define P3_VDBEFUNC (-7)  /* P3 is a pointer to a VdbeFunc structure */
#define P3_MEM      (-8)  /* P3 is a pointer to a Mem*    structure */
#define P3_TRANSIENT (-9) /* P3 is a pointer to a transient string */
#define P3_VTAB     (-10) /* P3 is a pointer to an sqlite3_vtab structure */
#define P3_MPRINTF  (-11) /* P3 is a string obtained from sqlite3_mprintf() */

/* When adding a P3 argument using P3_KEYINFO, a copy of the KeyInfo structure
** is made.  That copy is freed when the Vdbe is finalized.  But if the
** argument is P3_KEYINFO_HANDOFF, the passed in pointer is used.  It still
** gets freed when the Vdbe is finalized so it still should be obtained
** from a single sqliteMalloc().  But no copy is made and the calling
** function should *not* try to free the KeyInfo.
*/
#define P3_KEYINFO_HANDOFF (-9)

/*
** The Vdbe.aColName array contains 5n Mem structures, where n is the 
** number of columns of data returned by the statement.
*/
#define COLNAME_NAME     0
#define COLNAME_DECLTYPE 1
#define COLNAME_DATABASE 2
#define COLNAME_TABLE    3
#define COLNAME_COLUMN   4
#define COLNAME_N        5      /* Number of COLNAME_xxx symbols */

/*
** The following macro converts a relative address in the p2 field
** of a VdbeOp structure into a negative number so that 
** sqlite3VdbeAddOpList() knows that the address is relative.  Calling
** the macro again restores the address.
*/
#define ADDR(X)  (-1-(X))

/*
** The makefile scans the vdbe.c source file and creates the "opcodes.h"
** header file that defines a number for each opcode used by the VDBE.
*/
/************** Include opcodes.h in the middle of vdbe.h ********************/
/************** Begin file opcodes.h *****************************************/
/* Automatically generated.  Do not edit */
/* See the mkopcodeh.awk script for details */
#define OP_MemLoad                              1
#define OP_VNext                                2
#define OP_HexBlob                            126   /* same as TK_BLOB     */
#define OP_Column                               3
#define OP_SetCookie                            4
#define OP_IfMemPos                             5
#define OP_Real                               125   /* same as TK_FLOAT    */
#define OP_Sequence                             6
#define OP_MoveGt                               7
#define OP_Ge                                  72   /* same as TK_GE       */
#define OP_RowKey                               8
#define OP_Eq                                  68   /* same as TK_EQ       */
#define OP_OpenWrite                            9
#define OP_NotNull                             66   /* same as TK_NOTNULL  */
#define OP_If                                  10
#define OP_ToInt                              141   /* same as TK_TO_INT   */
#define OP_String8                             88   /* same as TK_STRING   */
#define OP_Pop                                 11
#define OP_VRowid                              12
#define OP_CollSeq                             13
#define OP_OpenRead                            14
#define OP_Expire                              15
#define OP_AutoCommit                          17
#define OP_Gt                                  69   /* same as TK_GT       */
#define OP_IntegrityCk                         18
#define OP_Sort                                19
#define OP_Function                            20
#define OP_And                                 61   /* same as TK_AND      */
#define OP_Subtract                            79   /* same as TK_MINUS    */
#define OP_Noop                                21
#define OP_Return                              22
#define OP_Remainder                           82   /* same as TK_REM      */
#define OP_NewRowid                            23
#define OP_Multiply                            80   /* same as TK_STAR     */
#define OP_IfMemNeg                            24
#define OP_Variable                            25
#define OP_String                              26
#define OP_RealAffinity                        27
#define OP_VRename                             28
#define OP_ParseSchema                         29
#define OP_VOpen                               30
#define OP_Close                               31
#define OP_CreateIndex                         32
#define OP_IsUnique                            33
#define OP_NotFound                            34
#define OP_Int64                               35
#define OP_MustBeInt                           36
#define OP_Halt                                37
#define OP_Rowid                               38
#define OP_IdxLT                               39
#define OP_AddImm                              40
#define OP_Statement                           41
#define OP_RowData                             42
#define OP_MemMax                              43
#define OP_Push                                44
#define OP_Or                                  60   /* same as TK_OR       */
#define OP_NotExists                           45
#define OP_MemIncr                             46
#define OP_Gosub                               47
#define OP_Divide                              81   /* same as TK_SLASH    */
#define OP_Integer                             48
#define OP_ToNumeric                          140   /* same as TK_TO_NUMERIC*/
#define OP_MemInt                              49
#define OP_Prev                                50
#define OP_Concat                              83   /* same as TK_CONCAT   */
#define OP_BitAnd                              74   /* same as TK_BITAND   */
#define OP_VColumn                             51
#define OP_CreateTable                         52
#define OP_Last                                53
#define OP_IsNull                              65   /* same as TK_ISNULL   */
#define OP_IncrVacuum                          54
#define OP_IdxRowid                            55
#define OP_MakeIdxRec                          56
#define OP_ShiftRight                          77   /* same as TK_RSHIFT   */
#define OP_ResetCount                          57
#define OP_FifoWrite                           58
#define OP_Callback                            59
#define OP_ContextPush                         62
#define OP_DropTrigger                         63
#define OP_DropIndex                           64
#define OP_IdxGE                               73
#define OP_IdxDelete                           84
#define OP_Vacuum                              86
#define OP_MoveLe                              89
#define OP_IfNot                               90
#define OP_DropTable                           91
#define OP_MakeRecord                          92
#define OP_ToBlob                             139   /* same as TK_TO_BLOB  */
#define OP_Delete                              93
#define OP_AggFinal                            94
#define OP_ShiftLeft                           76   /* same as TK_LSHIFT   */
#define OP_Dup                                 95
#define OP_Goto                                96
#define OP_TableLock                           97
#define OP_FifoRead                            98
#define OP_Clear                               99
#define OP_IdxGT                              100
#define OP_MoveLt                             101
#define OP_Le                                  70   /* same as TK_LE       */
#define OP_VerifyCookie                       102
#define OP_AggStep                            103
#define OP_Pull                               104
#define OP_ToText                             138   /* same as TK_TO_TEXT  */
#define OP_Not                                 16   /* same as TK_NOT      */
#define OP_ToReal                             142   /* same as TK_TO_REAL  */
#define OP_SetNumColumns                      105
#define OP_AbsValue                           106
#define OP_Transaction                        107
#define OP_VFilter                            108
#define OP_Negative                            85   /* same as TK_UMINUS   */
#define OP_Ne                                  67   /* same as TK_NE       */
#define OP_VDestroy                           109
#define OP_ContextPop                         110
#define OP_BitOr                               75   /* same as TK_BITOR    */
#define OP_Next                               111
#define OP_IdxInsert                          112
#define OP_Distinct                           113
#define OP_Lt                                  71   /* same as TK_LT       */
#define OP_Insert                             114
#define OP_Destroy                            115
#define OP_ReadCookie                         116
#define OP_ForceInt                           117
#define OP_LoadAnalysis                       118
#define OP_Explain                            119
#define OP_IfMemZero                          120
#define OP_OpenPseudo                         121
#define OP_OpenEphemeral                      122
#define OP_Null                               123
#define OP_Blob                               124
#define OP_Add                                 78   /* same as TK_PLUS     */
#define OP_MemStore                           127
#define OP_Rewind                             128
#define OP_MoveGe                             129
#define OP_VBegin                             130
#define OP_VUpdate                            131
#define OP_BitNot                              87   /* same as TK_BITNOT   */
#define OP_VCreate                            132
#define OP_MemMove                            133
#define OP_MemNull                            134
#define OP_Found                              135
#define OP_NullRow                            136

/* The following opcode values are never used */
#define OP_NotUsed_137                        137

/* Opcodes that are guaranteed to never push a value onto the stack
** contain a 1 their corresponding position of the following mask
** set.  See the opcodeNoPush() function in vdbeaux.c  */
#define NOPUSH_MASK_0 0xeeb4
#define NOPUSH_MASK_1 0xf96b
#define NOPUSH_MASK_2 0xfbb6
#define NOPUSH_MASK_3 0xfe64
#define NOPUSH_MASK_4 0xffff
#define NOPUSH_MASK_5 0x6ef7
#define NOPUSH_MASK_6 0xfbfb
#define NOPUSH_MASK_7 0x8767
#define NOPUSH_MASK_8 0x7d9f
#define NOPUSH_MASK_9 0x0000

/************** End of opcodes.h *********************************************/
/************** Continuing where we left off in vdbe.h ***********************/

/*
** Prototypes for the VDBE interface.  See comments on the implementation
** for a description of what each of these routines does.
*/
SQLITE_PRIVATE Vdbe *sqlite3VdbeCreate(sqlite3*);
SQLITE_PRIVATE int sqlite3VdbeAddOp(Vdbe*,int,int,int);
SQLITE_PRIVATE int sqlite3VdbeOp3(Vdbe*,int,int,int,const char *zP3,int);
SQLITE_PRIVATE int sqlite3VdbeAddOpList(Vdbe*, int nOp, VdbeOpList const *aOp);
SQLITE_PRIVATE void sqlite3VdbeChangeP1(Vdbe*, int addr, int P1);
SQLITE_PRIVATE void sqlite3VdbeChangeP2(Vdbe*, int addr, int P2);
SQLITE_PRIVATE void sqlite3VdbeJumpHere(Vdbe*, int addr);
SQLITE_PRIVATE void sqlite3VdbeChangeToNoop(Vdbe*, int addr, int N);
SQLITE_PRIVATE void sqlite3VdbeChangeP3(Vdbe*, int addr, const char *zP1, int N);
SQLITE_PRIVATE VdbeOp *sqlite3VdbeGetOp(Vdbe*, int);
SQLITE_PRIVATE int sqlite3VdbeMakeLabel(Vdbe*);
SQLITE_PRIVATE void sqlite3VdbeDelete(Vdbe*);
SQLITE_PRIVATE void sqlite3VdbeMakeReady(Vdbe*,int,int,int,int);
SQLITE_PRIVATE int sqlite3VdbeFinalize(Vdbe*);
SQLITE_PRIVATE void sqlite3VdbeResolveLabel(Vdbe*, int);
SQLITE_PRIVATE int sqlite3VdbeCurrentAddr(Vdbe*);
#ifdef SQLITE_DEBUG
SQLITE_PRIVATE   void sqlite3VdbeTrace(Vdbe*,FILE*);
#endif
SQLITE_PRIVATE void sqlite3VdbeResetStepResult(Vdbe*);
SQLITE_PRIVATE int sqlite3VdbeReset(Vdbe*);
SQLITE_PRIVATE void sqlite3VdbeSetNumCols(Vdbe*,int);
SQLITE_PRIVATE int sqlite3VdbeSetColName(Vdbe*, int, int, const char *, int);
SQLITE_PRIVATE void sqlite3VdbeCountChanges(Vdbe*);
SQLITE_PRIVATE sqlite3 *sqlite3VdbeDb(Vdbe*);
SQLITE_PRIVATE void sqlite3VdbeSetSql(Vdbe*, const char *z, int n);
SQLITE_PRIVATE const char *sqlite3VdbeGetSql(Vdbe*);
SQLITE_PRIVATE void sqlite3VdbeSwap(Vdbe*,Vdbe*);

#ifndef NDEBUG
SQLITE_PRIVATE   void sqlite3VdbeComment(Vdbe*, const char*, ...);
# define VdbeComment(X)  sqlite3VdbeComment X
#else
# define VdbeComment(X)
#endif

#endif

/************** End of vdbe.h ************************************************/
/************** Continuing where we left off in sqliteInt.h ******************/
/************** Include btree.h in the middle of sqliteInt.h *****************/
/************** Begin file btree.h *******************************************/
/*
** 2001 September 15
**
** The author disclaims copyright to this source code.  In place of
** a legal notice, here is a blessing:
**
**    May you do good and not evil.
**    May you find forgiveness for yourself and forgive others.
**    May you share freely, never taking more than you give.
**
*************************************************************************
** This header file defines the interface that the sqlite B-Tree file
** subsystem.  See comments in the source code for a detailed description
** of what each interface routine does.
**
** @(#) $Id: sqlite3.c,v 1.4 2007/07/21 08:51:43 rmsimpson Exp $
*/
#ifndef _BTREE_H_
#define _BTREE_H_

/* TODO: This definition is just included so other modules compile. It
** needs to be revisited.
*/
#define SQLITE_N_BTREE_META 10

/*
** If defined as non-zero, auto-vacuum is enabled by default. Otherwise
** it must be turned on for each database using "PRAGMA auto_vacuum = 1".
*/
#ifndef SQLITE_DEFAULT_AUTOVACUUM
  #define SQLITE_DEFAULT_AUTOVACUUM 0
#endif

#define BTREE_AUTOVACUUM_NONE 0        /* Do not do auto-vacuum */
#define BTREE_AUTOVACUUM_FULL 1        /* Do full auto-vacuum */
#define BTREE_AUTOVACUUM_INCR 2        /* Incremental vacuum */

/*
** Forward declarations of structure
*/
typedef struct Btree Btree;
typedef struct BtCursor BtCursor;
typedef struct BtShared BtShared;


SQLITE_PRIVATE int sqlite3BtreeOpen(
  const char *zFilename,   /* Name of database file to open */
  sqlite3 *db,             /* Associated database connection */
  Btree **,                /* Return open Btree* here */
  int flags                /* Flags */
);

/* The flags parameter to sqlite3BtreeOpen can be the bitwise or of the
** following values.
**
** NOTE:  These values must match the corresponding PAGER_ values in
** pager.h.
*/
#define BTREE_OMIT_JOURNAL  1  /* Do not use journal.  No argument */
#define BTREE_NO_READLOCK   2  /* Omit readlocks on readonly files */
#define BTREE_MEMORY        4  /* In-memory DB.  No argument */

SQLITE_PRIVATE int sqlite3BtreeClose(Btree*);
SQLITE_PRIVATE int sqlite3BtreeSetBusyHandler(Btree*,BusyHandler*);
SQLITE_PRIVATE int sqlite3BtreeSetCacheSize(Btree*,int);
SQLITE_PRIVATE int sqlite3BtreeSetSafetyLevel(Btree*,int,int);
SQLITE_PRIVATE int sqlite3BtreeSyncDisabled(Btree*);
SQLITE_PRIVATE int sqlite3BtreeSetPageSize(Btree*,int,int);
SQLITE_PRIVATE int sqlite3BtreeGetPageSize(Btree*);
SQLITE_PRIVATE int sqlite3BtreeMaxPageCount(Btree*,int);
SQLITE_PRIVATE int sqlite3BtreeGetReserve(Btree*);
SQLITE_PRIVATE int sqlite3BtreeSetAutoVacuum(Btree *, int);
SQLITE_PRIVATE int sqlite3BtreeGetAutoVacuum(Btree *);
SQLITE_PRIVATE int sqlite3BtreeBeginTrans(Btree*,int);
SQLITE_PRIVATE int sqlite3BtreeCommitPhaseOne(Btree*, const char *zMaster);
SQLITE_PRIVATE int sqlite3BtreeCommitPhaseTwo(Btree*);
SQLITE_PRIVATE int sqlite3BtreeCommit(Btree*);
SQLITE_PRIVATE int sqlite3BtreeRollback(Btree*);
SQLITE_PRIVATE int sqlite3BtreeBeginStmt(Btree*);
SQLITE_PRIVATE int sqlite3BtreeCommitStmt(Btree*);
SQLITE_PRIVATE int sqlite3BtreeRollbackStmt(Btree*);
SQLITE_PRIVATE int sqlite3BtreeCreateTable(Btree*, int*, int flags);
SQLITE_PRIVATE int sqlite3BtreeIsInTrans(Btree*);
SQLITE_PRIVATE int sqlite3BtreeIsInStmt(Btree*);
SQLITE_PRIVATE int sqlite3BtreeIsInReadTrans(Btree*);
SQLITE_PRIVATE void *sqlite3BtreeSchema(Btree *, int, void(*)(void *));
SQLITE_PRIVATE int sqlite3BtreeSchemaLocked(Btree *);
SQLITE_PRIVATE int sqlite3BtreeLockTable(Btree *, int, u8);

SQLITE_PRIVATE const char *sqlite3BtreeGetFilename(Btree *);
SQLITE_PRIVATE const char *sqlite3BtreeGetDirname(Btree *);
SQLITE_PRIVATE const char *sqlite3BtreeGetJournalname(Btree *);
SQLITE_PRIVATE int sqlite3BtreeCopyFile(Btree *, Btree *);

SQLITE_PRIVATE int sqlite3BtreeIncrVacuum(Btree *);

/* The flags parameter to sqlite3BtreeCreateTable can be the bitwise OR
** of the following flags:
*/
#define BTREE_INTKEY     1    /* Table has only 64-bit signed integer keys */
#define BTREE_ZERODATA   2    /* Table has keys only - no data */
#define BTREE_LEAFDATA   4    /* Data stored in leaves only.  Implies INTKEY */

SQLITE_PRIVATE int sqlite3BtreeDropTable(Btree*, int, int*);
SQLITE_PRIVATE int sqlite3BtreeClearTable(Btree*, int);
SQLITE_PRIVATE int sqlite3BtreeGetMeta(Btree*, int idx, u32 *pValue);
SQLITE_PRIVATE int sqlite3BtreeUpdateMeta(Btree*, int idx, u32 value);

SQLITE_PRIVATE int sqlite3BtreeCursor(
  Btree*,                              /* BTree containing table to open */
  int iTable,                          /* Index of root page */
  int wrFlag,                          /* 1 for writing.  0 for read-only */
  int(*)(void*,int,const void*,int,const void*),  /* Key comparison function */
  void*,                               /* First argument to compare function */
  BtCursor **ppCursor                  /* Returned cursor */
);

SQLITE_PRIVATE int sqlite3BtreeCloseCursor(BtCursor*);
SQLITE_PRIVATE int sqlite3BtreeMoveto(BtCursor*,const void *pKey,i64 nKey,int bias,int *pRes);
SQLITE_PRIVATE int sqlite3BtreeDelete(BtCursor*);
SQLITE_PRIVATE int sqlite3BtreeInsert(BtCursor*, const void *pKey, i64 nKey,
                                  const void *pData, int nData,
                                  int nZero, int bias);
SQLITE_PRIVATE int sqlite3BtreeFirst(BtCursor*, int *pRes);
SQLITE_PRIVATE int sqlite3BtreeLast(BtCursor*, int *pRes);
SQLITE_PRIVATE int sqlite3BtreeNext(BtCursor*, int *pRes);
SQLITE_PRIVATE int sqlite3BtreeEof(BtCursor*);
SQLITE_PRIVATE int sqlite3BtreeFlags(BtCursor*);
SQLITE_PRIVATE int sqlite3BtreePrevious(BtCursor*, int *pRes);
SQLITE_PRIVATE int sqlite3BtreeKeySize(BtCursor*, i64 *pSize);
SQLITE_PRIVATE int sqlite3BtreeKey(BtCursor*, u32 offset, u32 amt, void*);
SQLITE_PRIVATE const void *sqlite3BtreeKeyFetch(BtCursor*, int *pAmt);
SQLITE_PRIVATE const void *sqlite3BtreeDataFetch(BtCursor*, int *pAmt);
SQLITE_PRIVATE int sqlite3BtreeDataSize(BtCursor*, u32 *pSize);
SQLITE_PRIVATE int sqlite3BtreeData(BtCursor*, u32 offset, u32 amt, void*);

SQLITE_PRIVATE char *sqlite3BtreeIntegrityCheck(Btree*, int *aRoot, int nRoot, int, int*);
SQLITE_PRIVATE struct Pager *sqlite3BtreePager(Btree*);

SQLITE_PRIVATE int sqlite3BtreePutData(BtCursor*, u32 offset, u32 amt, void*);
SQLITE_PRIVATE void sqlite3BtreeCacheOverflow(BtCursor *);

#ifdef SQLITE_TEST
SQLITE_PRIVATE int sqlite3BtreeCursorInfo(BtCursor*, int*, int);
SQLITE_PRIVATE void sqlite3BtreeCursorList(Btree*);
SQLITE_PRIVATE int sqlite3BtreePageDump(Btree*, int, int recursive);
#endif

#endif /* _BTREE_H_ */

/************** End of btree.h ***********************************************/
/************** Continuing where we left off in sqliteInt.h ******************/
/************** Include pager.h in the middle of sqliteInt.h *****************/
/************** Begin file pager.h *******************************************/
/*
** 2001 September 15
**
** The author disclaims copyright to this source code.  In place of
** a legal notice, here is a blessing:
**
**    May you do good and not evil.
**    May you find forgiveness for yourself and forgive others.
**    May you share freely, never taking more than you give.
**
*************************************************************************
** This header file defines the interface that the sqlite page cache
** subsystem.  The page cache subsystem reads and writes a file a page
** at a time and provides a journal for rollback.
**
** @(#) $Id: sqlite3.c,v 1.4 2007/07/21 08:51:43 rmsimpson Exp $
*/

#ifndef _PAGER_H_
#define _PAGER_H_

/*
** The type used to represent a page number.  The first page in a file
** is called page 1.  0 is used to represent "not a page".
*/
typedef unsigned int Pgno;

/*
** Each open file is managed by a separate instance of the "Pager" structure.
*/
typedef struct Pager Pager;

/*
** Handle type for pages.
*/
typedef struct PgHdr DbPage;

/*
** Allowed values for the flags parameter to sqlite3PagerOpen().
**
** NOTE: This values must match the corresponding BTREE_ values in btree.h.
*/
#define PAGER_OMIT_JOURNAL  0x0001    /* Do not use a rollback journal */
#define PAGER_NO_READLOCK   0x0002    /* Omit readlocks on readonly files */

/*
** Valid values for the second argument to sqlite3PagerLockingMode().
*/
#define PAGER_LOCKINGMODE_QUERY      -1
#define PAGER_LOCKINGMODE_NORMAL      0
#define PAGER_LOCKINGMODE_EXCLUSIVE   1

/*
** See source code comments for a detailed description of the following
** routines:
*/
SQLITE_PRIVATE int sqlite3PagerOpen(Pager **ppPager, const char *zFilename,
                     int nExtra, int flags);
SQLITE_PRIVATE void sqlite3PagerSetBusyhandler(Pager*, BusyHandler *pBusyHandler);
SQLITE_PRIVATE void sqlite3PagerSetDestructor(Pager*, void(*)(DbPage*,int));
SQLITE_PRIVATE void sqlite3PagerSetReiniter(Pager*, void(*)(DbPage*,int));
SQLITE_PRIVATE int sqlite3PagerSetPagesize(Pager*, int);
SQLITE_PRIVATE int sqlite3PagerMaxPageCount(Pager*, int);
SQLITE_PRIVATE int sqlite3PagerReadFileheader(Pager*, int, unsigned char*);
SQLITE_PRIVATE void sqlite3PagerSetCachesize(Pager*, int);
SQLITE_PRIVATE int sqlite3PagerClose(Pager *pPager);
SQLITE_PRIVATE int sqlite3PagerAcquire(Pager *pPager, Pgno pgno, DbPage **ppPage, int clrFlag);
#define sqlite3PagerGet(A,B,C) sqlite3PagerAcquire(A,B,C,0)
SQLITE_PRIVATE DbPage *sqlite3PagerLookup(Pager *pPager, Pgno pgno);
SQLITE_PRIVATE int sqlite3PagerRef(DbPage*);
SQLITE_PRIVATE int sqlite3PagerUnref(DbPage*);
SQLITE_PRIVATE int sqlite3PagerWrite(DbPage*);
SQLITE_PRIVATE int sqlite3PagerOverwrite(Pager *pPager, Pgno pgno, void*);
SQLITE_PRIVATE int sqlite3PagerPagecount(Pager*);
SQLITE_PRIVATE int sqlite3PagerTruncate(Pager*,Pgno);
SQLITE_PRIVATE int sqlite3PagerBegin(DbPage*, int exFlag);
SQLITE_PRIVATE int sqlite3PagerCommitPhaseOne(Pager*,const char *zMaster, Pgno);
SQLITE_PRIVATE int sqlite3PagerCommitPhaseTwo(Pager*);
SQLITE_PRIVATE int sqlite3PagerRollback(Pager*);
SQLITE_PRIVATE int sqlite3PagerIsreadonly(Pager*);
SQLITE_PRIVATE int sqlite3PagerStmtBegin(Pager*);
SQLITE_PRIVATE int sqlite3PagerStmtCommit(Pager*);
SQLITE_PRIVATE int sqlite3PagerStmtRollback(Pager*);
SQLITE_PRIVATE void sqlite3PagerDontRollback(DbPage*);
SQLITE_PRIVATE void sqlite3PagerDontWrite(DbPage*);
SQLITE_PRIVATE int sqlite3PagerRefcount(Pager*);
SQLITE_PRIVATE void sqlite3PagerSetSafetyLevel(Pager*,int,int);
SQLITE_PRIVATE const char *sqlite3PagerFilename(Pager*);
SQLITE_PRIVATE const char *sqlite3PagerDirname(Pager*);
SQLITE_PRIVATE const char *sqlite3PagerJournalname(Pager*);
SQLITE_PRIVATE int sqlite3PagerNosync(Pager*);
SQLITE_PRIVATE int sqlite3PagerMovepage(Pager*,DbPage*,Pgno);
SQLITE_PRIVATE void *sqlite3PagerGetData(DbPage *); 
SQLITE_PRIVATE void *sqlite3PagerGetExtra(DbPage *); 
SQLITE_PRIVATE int sqlite3PagerLockingMode(Pager *, int);

#if defined(SQLITE_ENABLE_MEMORY_MANAGEMENT) && !defined(SQLITE_OMIT_DISKIO)
SQLITE_PRIVATE   int sqlite3PagerReleaseMemory(int);
#endif

#ifdef SQLITE_HAS_CODEC
SQLITE_PRIVATE   void sqlite3PagerSetCodec(Pager*,void*(*)(void*,void*,Pgno,int),void*);
#endif

#if !defined(NDEBUG) || defined(SQLITE_TEST)
SQLITE_PRIVATE   Pgno sqlite3PagerPagenumber(DbPage*);
SQLITE_PRIVATE   int sqlite3PagerIswriteable(DbPage*);
#endif

#if defined(SQLITE_DEBUG) || defined(SQLITE_TEST)
SQLITE_PRIVATE   int sqlite3PagerLockstate(Pager*);
#endif

#ifdef SQLITE_TEST
SQLITE_PRIVATE   int *sqlite3PagerStats(Pager*);
SQLITE_PRIVATE   void sqlite3PagerRefdump(Pager*);
  int pager3_refinfo_enable;
#endif

#ifdef SQLITE_TEST
void disable_simulated_io_errors(void);
void enable_simulated_io_errors(void);
#else
# define disable_simulated_io_errors()
# define enable_simulated_io_errors()
#endif

#endif /* _PAGER_H_ */

/************** End of pager.h ***********************************************/
/************** Continuing where we left off in sqliteInt.h ******************/

#ifdef SQLITE_MEMDEBUG
/*
** The following global variables are used for testing and debugging
** only.  They only work if SQLITE_MEMDEBUG is defined.
*/
extern int sqlite3_nMalloc;      /* Number of sqliteMalloc() calls */
extern int sqlite3_nFree;        /* Number of sqliteFree() calls */
extern int sqlite3_iMallocFail;  /* Fail sqliteMalloc() after this many calls */
extern int sqlite3_iMallocReset; /* Set iMallocFail to this when it reaches 0 */

extern void *sqlite3_pFirst;         /* Pointer to linked list of allocations */
extern int sqlite3_nMaxAlloc;        /* High water mark of ThreadData.nAlloc */
extern int sqlite3_mallocDisallowed; /* assert() in sqlite3Malloc() if set */
extern int sqlite3_isFail;           /* True if all malloc calls should fail */
extern const char *sqlite3_zFile;    /* Filename to associate debug info with */
extern int sqlite3_iLine;            /* Line number for debug info */

#define ENTER_MALLOC (sqlite3_zFile = __FILE__, sqlite3_iLine = __LINE__)
#define sqliteMalloc(x)          (ENTER_MALLOC, sqlite3Malloc(x,1))
#define sqliteMallocRaw(x)       (ENTER_MALLOC, sqlite3MallocRaw(x,1))
#define sqliteRealloc(x,y)       (ENTER_MALLOC, sqlite3Realloc(x,y))
#define sqliteStrDup(x)          (ENTER_MALLOC, sqlite3StrDup(x))
#define sqliteStrNDup(x,y)       (ENTER_MALLOC, sqlite3StrNDup(x,y))
#define sqliteReallocOrFree(x,y) (ENTER_MALLOC, sqlite3ReallocOrFree(x,y))

#else

#define ENTER_MALLOC 0
#define sqliteMalloc(x)          sqlite3Malloc(x,1)
#define sqliteMallocRaw(x)       sqlite3MallocRaw(x,1)
#define sqliteRealloc(x,y)       sqlite3Realloc(x,y)
#define sqliteStrDup(x)          sqlite3StrDup(x)
#define sqliteStrNDup(x,y)       sqlite3StrNDup(x,y)
#define sqliteReallocOrFree(x,y) sqlite3ReallocOrFree(x,y)

#endif

/* Variable sqlite3_mallocHasFailed is set to true after a malloc() 
** failure occurs. 
**
** The sqlite3MallocFailed() macro returns true if a malloc has failed
** in this thread since the last call to sqlite3ApiExit(), or false 
** otherwise.
*/
extern int sqlite3_mallocHasFailed;
#define sqlite3MallocFailed() (sqlite3_mallocHasFailed && sqlite3OsInMutex(1))

#define sqliteFree(x)          sqlite3FreeX(x)
#define sqliteAllocSize(x)     sqlite3AllocSize(x)

/*
** An instance of this structure might be allocated to store information
** specific to a single thread.
*/
struct ThreadData {
  int dummy;               /* So that this structure is never empty */

#ifdef SQLITE_ENABLE_MEMORY_MANAGEMENT
  int nSoftHeapLimit;      /* Suggested max mem allocation.  No limit if <0 */
  int nAlloc;              /* Number of bytes currently allocated */
  Pager *pPager;           /* Linked list of all pagers in this thread */
#endif

#ifndef SQLITE_OMIT_SHARED_CACHE
  u8 useSharedData;        /* True if shared pagers and schemas are enabled */
  BtShared *pBtree;        /* Linked list of all currently open BTrees */
#endif
};

/*
** Name of the master database table.  The master database table
** is a special table that holds the names and attributes of all
** user tables and indices.
*/
#define MASTER_NAME       "sqlite_master"
#define TEMP_MASTER_NAME  "sqlite_temp_master"

/*
** The root-page of the master database table.
*/
#define MASTER_ROOT       1

/*
** The name of the schema table.
*/
#define SCHEMA_TABLE(x)  ((!OMIT_TEMPDB)&&(x==1)?TEMP_MASTER_NAME:MASTER_NAME)

/*
** A convenience macro that returns the number of elements in
** an array.
*/
#define ArraySize(X)    (sizeof(X)/sizeof(X[0]))

/*
** Forward references to structures
*/
typedef struct AggInfo AggInfo;
typedef struct AuthContext AuthContext;
typedef struct CollSeq CollSeq;
typedef struct Column Column;
typedef struct Db Db;
typedef struct Schema Schema;
typedef struct Expr Expr;
typedef struct ExprList ExprList;
typedef struct FKey FKey;
typedef struct FuncDef FuncDef;
typedef struct IdList IdList;
typedef struct Index Index;
typedef struct KeyClass KeyClass;
typedef struct KeyInfo KeyInfo;
typedef struct Module Module;
typedef struct NameContext NameContext;
typedef struct Parse Parse;
typedef struct Select Select;
typedef struct SrcList SrcList;
typedef struct ThreadData ThreadData;
typedef struct Table Table;
typedef struct TableLock TableLock;
typedef struct Token Token;
typedef struct TriggerStack TriggerStack;
typedef struct TriggerStep TriggerStep;
typedef struct Trigger Trigger;
typedef struct WhereInfo WhereInfo;
typedef struct WhereLevel WhereLevel;

/************** Include os.h in the middle of sqliteInt.h ********************/
/************** Begin file os.h **********************************************/
/*
** 2001 September 16
**
** The author disclaims copyright to this source code.  In place of
** a legal notice, here is a blessing:
**
**    May you do good and not evil.
**    May you find forgiveness for yourself and forgive others.
**    May you share freely, never taking more than you give.
**
******************************************************************************
**
** This header file (together with is companion C source-code file
** "os.c") attempt to abstract the underlying operating system so that
** the SQLite library will work on both POSIX and windows systems.
*/
#ifndef _SQLITE_OS_H_
#define _SQLITE_OS_H_

/*
** Figure out if we are dealing with Unix, Windows, or some other
** operating system.
*/
#if defined(OS_OTHER)
# if OS_OTHER==1
#   undef OS_UNIX
#   define OS_UNIX 0
#   undef OS_WIN
#   define OS_WIN 0
#   undef OS_OS2
#   define OS_OS2 0
# else
#   undef OS_OTHER
# endif
#endif
#if !defined(OS_UNIX) && !defined(OS_OTHER)
# define OS_OTHER 0
# ifndef OS_WIN
#   if defined(_WIN32) || defined(WIN32) || defined(__CYGWIN__) || defined(__MINGW32__) || defined(__BORLANDC__)
#     define OS_WIN 1
#     define OS_UNIX 0
#     define OS_OS2 0
#   elif defined(__EMX__) || defined(_OS2) || defined(OS2) || defined(_OS2_) || defined(__OS2__)
#     define OS_WIN 0
#     define OS_UNIX 0
#     define OS_OS2 1
#   else
#     define OS_WIN 0
#     define OS_UNIX 1
#     define OS_OS2 0
#  endif
# else
#  define OS_UNIX 0
#  define OS_OS2 0
# endif
#else
# ifndef OS_WIN
#  define OS_WIN 0
# endif
#endif


/*
** Define the maximum size of a temporary filename
*/
#if OS_WIN
# include <windows.h>
# define SQLITE_TEMPNAME_SIZE (MAX_PATH+50)
#elif OS_OS2
# if (__GNUC__ > 3 || __GNUC__ == 3 && __GNUC_MINOR__ >= 3) && defined(OS2_HIGH_MEMORY)
#  include <os2safe.h> /* has to be included before os2.h for linking to work */
# endif
# define INCL_DOSDATETIME
# define INCL_DOSFILEMGR
# define INCL_DOSERRORS
# define INCL_DOSMISC
# define INCL_DOSPROCESS
# define INCL_DOSMODULEMGR
# include <os2.h>
# define SQLITE_TEMPNAME_SIZE (CCHMAXPATHCOMP)
#else
# define SQLITE_TEMPNAME_SIZE 200
#endif

/* If the SET_FULLSYNC macro is not defined above, then make it
** a no-op
*/
#ifndef SET_FULLSYNC
# define SET_FULLSYNC(x,y)
#endif

/*
** The default size of a disk sector
*/
#ifndef SQLITE_DEFAULT_SECTOR_SIZE
# define SQLITE_DEFAULT_SECTOR_SIZE 512
#endif

/*
** Temporary files are named starting with this prefix followed by 16 random
** alphanumeric characters, and no file extension. They are stored in the
** OS's standard temporary file directory, and are deleted prior to exit.
** If sqlite is being embedded in another program, you may wish to change the
** prefix to reflect your program's name, so that if your program exits
** prematurely, old temporary files can be easily identified. This can be done
** using -DTEMP_FILE_PREFIX=myprefix_ on the compiler command line.
**
** 2006-10-31:  The default prefix used to be "sqlite_".  But then
** Mcafee started using SQLite in their anti-virus product and it
** started putting files with the "sqlite" name in the c:/temp folder.
** This annoyed many windows users.  Those users would then do a 
** Google search for "sqlite", find the telephone numbers of the
** developers and call to wake them up at night and complain.
** For this reason, the default name prefix is changed to be "sqlite" 
** spelled backwards.  So the temp files are still identified, but
** anybody smart enough to figure out the code is also likely smart
** enough to know that calling the developer will not help get rid
** of the file.
*/
#ifndef TEMP_FILE_PREFIX
# define TEMP_FILE_PREFIX "etilqs_"
#endif

/*
** Define the interfaces for Unix, Windows, and OS/2.
*/
#if OS_UNIX
#define sqlite3OsOpenReadWrite      sqlite3UnixOpenReadWrite
#define sqlite3OsOpenExclusive      sqlite3UnixOpenExclusive
#define sqlite3OsOpenReadOnly       sqlite3UnixOpenReadOnly
#define sqlite3OsDelete             sqlite3UnixDelete
#define sqlite3OsFileExists         sqlite3UnixFileExists
#define sqlite3OsFullPathname       sqlite3UnixFullPathname
#define sqlite3OsIsDirWritable      sqlite3UnixIsDirWritable
#define sqlite3OsSyncDirectory      sqlite3UnixSyncDirectory
#define sqlite3OsTempFileName       sqlite3UnixTempFileName
#define sqlite3OsRandomSeed         sqlite3UnixRandomSeed
#define sqlite3OsSleep              sqlite3UnixSleep
#define sqlite3OsCurrentTime        sqlite3UnixCurrentTime
#define sqlite3OsEnterMutex         sqlite3UnixEnterMutex
#define sqlite3OsLeaveMutex         sqlite3UnixLeaveMutex
#define sqlite3OsInMutex            sqlite3UnixInMutex
#define sqlite3OsThreadSpecificData sqlite3UnixThreadSpecificData
#define sqlite3OsMalloc             sqlite3GenericMalloc
#define sqlite3OsRealloc            sqlite3GenericRealloc
#define sqlite3OsFree               sqlite3GenericFree
#define sqlite3OsAllocationSize     sqlite3GenericAllocationSize
#define sqlite3OsDlopen             sqlite3UnixDlopen
#define sqlite3OsDlsym              sqlite3UnixDlsym
#define sqlite3OsDlclose            sqlite3UnixDlclose
#endif
#if OS_WIN
#define sqlite3OsOpenReadWrite      sqlite3WinOpenReadWrite
#define sqlite3OsOpenExclusive      sqlite3WinOpenExclusive
#define sqlite3OsOpenReadOnly       sqlite3WinOpenReadOnly
#define sqlite3OsDelete             sqlite3WinDelete
#define sqlite3OsFileExists         sqlite3WinFileExists
#define sqlite3OsFullPathname       sqlite3WinFullPathname
#define sqlite3OsIsDirWritable      sqlite3WinIsDirWritable
#define sqlite3OsSyncDirectory      sqlite3WinSyncDirectory
#define sqlite3OsTempFileName       sqlite3WinTempFileName
#define sqlite3OsRandomSeed         sqlite3WinRandomSeed
#define sqlite3OsSleep              sqlite3WinSleep
#define sqlite3OsCurrentTime        sqlite3WinCurrentTime
#define sqlite3OsEnterMutex         sqlite3WinEnterMutex
#define sqlite3OsLeaveMutex         sqlite3WinLeaveMutex
#define sqlite3OsInMutex            sqlite3WinInMutex
#define sqlite3OsThreadSpecificData sqlite3WinThreadSpecificData
#define sqlite3OsMalloc             sqlite3GenericMalloc
#define sqlite3OsRealloc            sqlite3GenericRealloc
#define sqlite3OsFree               sqlite3GenericFree
#define sqlite3OsAllocationSize     sqlite3GenericAllocationSize
#define sqlite3OsDlopen             sqlite3WinDlopen
#define sqlite3OsDlsym              sqlite3WinDlsym
#define sqlite3OsDlclose            sqlite3WinDlclose
#endif
#if OS_OS2
#define sqlite3OsOpenReadWrite      sqlite3Os2OpenReadWrite
#define sqlite3OsOpenExclusive      sqlite3Os2OpenExclusive
#define sqlite3OsOpenReadOnly       sqlite3Os2OpenReadOnly
#define sqlite3OsDelete             sqlite3Os2Delete
#define sqlite3OsFileExists         sqlite3Os2FileExists
#define sqlite3OsFullPathname       sqlite3Os2FullPathname
#define sqlite3OsIsDirWritable      sqlite3Os2IsDirWritable
#define sqlite3OsSyncDirectory      sqlite3Os2SyncDirectory
#define sqlite3OsTempFileName       sqlite3Os2TempFileName
#define sqlite3OsRandomSeed         sqlite3Os2RandomSeed
#define sqlite3OsSleep              sqlite3Os2Sleep
#define sqlite3OsCurrentTime        sqlite3Os2CurrentTime
#define sqlite3OsEnterMutex         sqlite3Os2EnterMutex
#define sqlite3OsLeaveMutex         sqlite3Os2LeaveMutex
#define sqlite3OsInMutex            sqlite3Os2InMutex
#define sqlite3OsThreadSpecificData sqlite3Os2ThreadSpecificData
#define sqlite3OsMalloc             sqlite3GenericMalloc
#define sqlite3OsRealloc            sqlite3GenericRealloc
#define sqlite3OsFree               sqlite3GenericFree
#define sqlite3OsAllocationSize     sqlite3GenericAllocationSize
#define sqlite3OsDlopen             sqlite3Os2Dlopen
#define sqlite3OsDlsym              sqlite3Os2Dlsym
#define sqlite3OsDlclose            sqlite3Os2Dlclose
#endif




/*
** If using an alternative OS interface, then we must have an "os_other.h"
** header file available for that interface.  Presumably the "os_other.h"
** header file contains #defines similar to those above.
*/
#if OS_OTHER
# include "os_other.h"
#endif



/*
** Forward declarations
*/
typedef struct OsFile OsFile;
typedef struct IoMethod IoMethod;

/*
** An instance of the following structure contains pointers to all
** methods on an OsFile object.
*/
struct IoMethod {
  int (*xClose)(OsFile**);
  int (*xOpenDirectory)(OsFile*, const char*);
  int (*xRead)(OsFile*, void*, int amt);
  int (*xWrite)(OsFile*, const void*, int amt);
  int (*xSeek)(OsFile*, i64 offset);
  int (*xTruncate)(OsFile*, i64 size);
  int (*xSync)(OsFile*, int);
  void (*xSetFullSync)(OsFile *id, int setting);
  int (*xFileHandle)(OsFile *id);
  int (*xFileSize)(OsFile*, i64 *pSize);
  int (*xLock)(OsFile*, int);
  int (*xUnlock)(OsFile*, int);
  int (*xLockState)(OsFile *id);
  int (*xCheckReservedLock)(OsFile *id);
  int (*xSectorSize)(OsFile *id);
};

/*
** The OsFile object describes an open disk file in an OS-dependent way.
** The version of OsFile defined here is a generic version.  Each OS
** implementation defines its own subclass of this structure that contains
** additional information needed to handle file I/O.  But the pMethod
** entry (pointing to the virtual function table) always occurs first
** so that we can always find the appropriate methods.
*/
struct OsFile {
  IoMethod const *pMethod;
};

/*
** The following values may be passed as the second argument to
** sqlite3OsLock(). The various locks exhibit the following semantics:
**
** SHARED:    Any number of processes may hold a SHARED lock simultaneously.
** RESERVED:  A single process may hold a RESERVED lock on a file at
**            any time. Other processes may hold and obtain new SHARED locks.
** PENDING:   A single process may hold a PENDING lock on a file at
**            any one time. Existing SHARED locks may persist, but no new
**            SHARED locks may be obtained by other processes.
** EXCLUSIVE: An EXCLUSIVE lock precludes all other locks.
**
** PENDING_LOCK may not be passed directly to sqlite3OsLock(). Instead, a
** process that requests an EXCLUSIVE lock may actually obtain a PENDING
** lock. This can be upgraded to an EXCLUSIVE lock by a subsequent call to
** sqlite3OsLock().
*/
#define NO_LOCK         0
#define SHARED_LOCK     1
#define RESERVED_LOCK   2
#define PENDING_LOCK    3
#define EXCLUSIVE_LOCK  4

/*
** File Locking Notes:  (Mostly about windows but also some info for Unix)
**
** We cannot use LockFileEx() or UnlockFileEx() on Win95/98/ME because
** those functions are not available.  So we use only LockFile() and
** UnlockFile().
**
** LockFile() prevents not just writing but also reading by other processes.
** A SHARED_LOCK is obtained by locking a single randomly-chosen 
** byte out of a specific range of bytes. The lock byte is obtained at 
** random so two separate readers can probably access the file at the 
** same time, unless they are unlucky and choose the same lock byte.
** An EXCLUSIVE_LOCK is obtained by locking all bytes in the range.
** There can only be one writer.  A RESERVED_LOCK is obtained by locking
** a single byte of the file that is designated as the reserved lock byte.
** A PENDING_LOCK is obtained by locking a designated byte different from
** the RESERVED_LOCK byte.
**
** On WinNT/2K/XP systems, LockFileEx() and UnlockFileEx() are available,
** which means we can use reader/writer locks.  When reader/writer locks
** are used, the lock is placed on the same range of bytes that is used
** for probabilistic locking in Win95/98/ME.  Hence, the locking scheme
** will support two or more Win95 readers or two or more WinNT readers.
** But a single Win95 reader will lock out all WinNT readers and a single
** WinNT reader will lock out all other Win95 readers.
**
** The following #defines specify the range of bytes used for locking.
** SHARED_SIZE is the number of bytes available in the pool from which
** a random byte is selected for a shared lock.  The pool of bytes for
** shared locks begins at SHARED_FIRST. 
**
** These #defines are available in sqlite_aux.h so that adaptors for
** connecting SQLite to other operating systems can use the same byte
** ranges for locking.  In particular, the same locking strategy and
** byte ranges are used for Unix.  This leaves open the possiblity of having
** clients on win95, winNT, and unix all talking to the same shared file
** and all locking correctly.  To do so would require that samba (or whatever
** tool is being used for file sharing) implements locks correctly between
** windows and unix.  I'm guessing that isn't likely to happen, but by
** using the same locking range we are at least open to the possibility.
**
** Locking in windows is manditory.  For this reason, we cannot store
** actual data in the bytes used for locking.  The pager never allocates
** the pages involved in locking therefore.  SHARED_SIZE is selected so
** that all locks will fit on a single page even at the minimum page size.
** PENDING_BYTE defines the beginning of the locks.  By default PENDING_BYTE
** is set high so that we don't have to allocate an unused page except
** for very large databases.  But one should test the page skipping logic 
** by setting PENDING_BYTE low and running the entire regression suite.
**
** Changing the value of PENDING_BYTE results in a subtly incompatible
** file format.  Depending on how it is changed, you might not notice
** the incompatibility right away, even running a full regression test.
** The default location of PENDING_BYTE is the first byte past the
** 1GB boundary.
**
*/
#ifndef SQLITE_TEST
#define PENDING_BYTE      0x40000000  /* First byte past the 1GB boundary */
#else
extern unsigned int sqlite3_pending_byte;
#define PENDING_BYTE sqlite3_pending_byte
#endif

#define RESERVED_BYTE     (PENDING_BYTE+1)
#define SHARED_FIRST      (PENDING_BYTE+2)
#define SHARED_SIZE       510

/*
** Prototypes for operating system interface routines.
*/
SQLITE_PRIVATE int sqlite3OsClose(OsFile**);
SQLITE_PRIVATE int sqlite3OsOpenDirectory(OsFile*, const char*);
SQLITE_PRIVATE int sqlite3OsRead(OsFile*, void*, int amt);
SQLITE_PRIVATE int sqlite3OsWrite(OsFile*, const void*, int amt);
SQLITE_PRIVATE int sqlite3OsSeek(OsFile*, i64 offset);
SQLITE_PRIVATE int sqlite3OsTruncate(OsFile*, i64 size);
SQLITE_PRIVATE int sqlite3OsSync(OsFile*, int);
SQLITE_PRIVATE void sqlite3OsSetFullSync(OsFile *id, int setting);
SQLITE_PRIVATE int sqlite3OsFileSize(OsFile*, i64 *pSize);
SQLITE_PRIVATE int sqlite3OsLock(OsFile*, int);
SQLITE_PRIVATE int sqlite3OsUnlock(OsFile*, int);
SQLITE_PRIVATE int sqlite3OsCheckReservedLock(OsFile *id);
SQLITE_PRIVATE int sqlite3OsOpenReadWrite(const char*, OsFile**, int*);
SQLITE_PRIVATE int sqlite3OsOpenExclusive(const char*, OsFile**, int);
SQLITE_PRIVATE int sqlite3OsOpenReadOnly(const char*, OsFile**);
SQLITE_PRIVATE int sqlite3OsDelete(const char*);
SQLITE_PRIVATE int sqlite3OsFileExists(const char*);
SQLITE_PRIVATE char *sqlite3OsFullPathname(const char*);
SQLITE_PRIVATE int sqlite3OsIsDirWritable(char*);
SQLITE_PRIVATE int sqlite3OsSyncDirectory(const char*);
SQLITE_PRIVATE int sqlite3OsSectorSize(OsFile *id);
SQLITE_PRIVATE int sqlite3OsTempFileName(char*);
SQLITE_PRIVATE int sqlite3OsRandomSeed(char*);
SQLITE_PRIVATE int sqlite3OsSleep(int ms);
SQLITE_PRIVATE int sqlite3OsCurrentTime(double*);
SQLITE_PRIVATE void sqlite3OsEnterMutex(void);
SQLITE_PRIVATE void sqlite3OsLeaveMutex(void);
SQLITE_PRIVATE int sqlite3OsInMutex(int);
SQLITE_PRIVATE ThreadData *sqlite3OsThreadSpecificData(int);
SQLITE_PRIVATE void *sqlite3OsMalloc(int);
SQLITE_PRIVATE void *sqlite3OsRealloc(void *, int);
SQLITE_PRIVATE void sqlite3OsFree(void *);
SQLITE_PRIVATE int sqlite3OsAllocationSize(void *);
SQLITE_PRIVATE void *sqlite3OsDlopen(const char*);
SQLITE_PRIVATE void *sqlite3OsDlsym(void*, const char*);
SQLITE_PRIVATE int sqlite3OsDlclose(void*);

#if defined(SQLITE_TEST) || defined(SQLITE_DEBUG)
SQLITE_PRIVATE   int sqlite3OsFileHandle(OsFile *id);
SQLITE_PRIVATE   int sqlite3OsLockState(OsFile *id);
#endif

/*
** If the SQLITE_ENABLE_REDEF_IO macro is defined, then the OS-layer
** interface routines are not called directly but are invoked using
** pointers to functions.  This allows the implementation of various
** OS-layer interface routines to be modified at run-time.  There are
** obscure but legitimate reasons for wanting to do this.  But for
** most users, a direct call to the underlying interface is preferable
** so the the redefinable I/O interface is turned off by default.
*/
#ifdef SQLITE_ENABLE_REDEF_IO

/*
** When redefinable I/O is enabled, a single global instance of the
** following structure holds pointers to the routines that SQLite 
** uses to talk with the underlying operating system.  Modify this
** structure (before using any SQLite API!) to accomodate perculiar
** operating system interfaces or behaviors.
*/
struct sqlite3OsVtbl {
  int (*xOpenReadWrite)(const char*, OsFile**, int*);
  int (*xOpenExclusive)(const char*, OsFile**, int);
  int (*xOpenReadOnly)(const char*, OsFile**);

  int (*xDelete)(const char*);
  int (*xFileExists)(const char*);
  char *(*xFullPathname)(const char*);
  int (*xIsDirWritable)(char*);
  int (*xSyncDirectory)(const char*);
  int (*xTempFileName)(char*);

  int (*xRandomSeed)(char*);
  int (*xSleep)(int ms);
  int (*xCurrentTime)(double*);

  void (*xEnterMutex)(void);
  void (*xLeaveMutex)(void);
  int (*xInMutex)(int);
  ThreadData *(*xThreadSpecificData)(int);

  void *(*xMalloc)(int);
  void *(*xRealloc)(void *, int);
  void (*xFree)(void *);
  int (*xAllocationSize)(void *);

  void *(*xDlopen)(const char*);
  void *(*xDlsym)(void*, const char*);
  int (*xDlclose)(void*);
};

/* Macro used to comment out routines that do not exists when there is
** no disk I/O or extension loading
*/
#ifdef SQLITE_OMIT_DISKIO
# define IF_DISKIO(X)  0
#else
# define IF_DISKIO(X)  X
#endif
#ifdef SQLITE_OMIT_LOAD_EXTENSION
# define IF_DLOPEN(X)  0
#else
# define IF_DLOPEN(X)  X
#endif


#if defined(_SQLITE_OS_C_) || defined(SQLITE_AMALGAMATION)
  /*
  ** The os.c file implements the global virtual function table.
  ** We have to put this file here because the initializers
  ** (ex: sqlite3OsRandomSeed) are macros that are about to be
  ** redefined.
  */
  struct sqlite3OsVtbl sqlite3Os = {
    IF_DISKIO( sqlite3OsOpenReadWrite ),
    IF_DISKIO( sqlite3OsOpenExclusive ),
    IF_DISKIO( sqlite3OsOpenReadOnly ),
    IF_DISKIO( sqlite3OsDelete ),
    IF_DISKIO( sqlite3OsFileExists ),
    IF_DISKIO( sqlite3OsFullPathname ),
    IF_DISKIO( sqlite3OsIsDirWritable ),
    IF_DISKIO( sqlite3OsSyncDirectory ),
    IF_DISKIO( sqlite3OsTempFileName ),
    sqlite3OsRandomSeed,
    sqlite3OsSleep,
    sqlite3OsCurrentTime,
    sqlite3OsEnterMutex,
    sqlite3OsLeaveMutex,
    sqlite3OsInMutex,
    sqlite3OsThreadSpecificData,
    sqlite3OsMalloc,
    sqlite3OsRealloc,
    sqlite3OsFree,
    sqlite3OsAllocationSize,
    IF_DLOPEN( sqlite3OsDlopen ),
    IF_DLOPEN( sqlite3OsDlsym ),
    IF_DLOPEN( sqlite3OsDlclose ),
  };
#else
  /*
  ** Files other than os.c just reference the global virtual function table. 
  */
  extern struct sqlite3OsVtbl sqlite3Os;
#endif /* _SQLITE_OS_C_ */


/* This additional API routine is available with redefinable I/O */
struct sqlite3OsVtbl *sqlite3_os_switch(void);


/*
** Redefine the OS interface to go through the virtual function table
** rather than calling routines directly.
*/
#undef sqlite3OsOpenReadWrite
#undef sqlite3OsOpenExclusive
#undef sqlite3OsOpenReadOnly
#undef sqlite3OsDelete
#undef sqlite3OsFileExists
#undef sqlite3OsFullPathname
#undef sqlite3OsIsDirWritable
#undef sqlite3OsSyncDirectory
#undef sqlite3OsTempFileName
#undef sqlite3OsRandomSeed
#undef sqlite3OsSleep
#undef sqlite3OsCurrentTime
#undef sqlite3OsEnterMutex
#undef sqlite3OsLeaveMutex
#undef sqlite3OsInMutex
#undef sqlite3OsThreadSpecificData
#undef sqlite3OsMalloc
#undef sqlite3OsRealloc
#undef sqlite3OsFree
#undef sqlite3OsAllocationSize
#define sqlite3OsOpenReadWrite      sqlite3Os.xOpenReadWrite
#define sqlite3OsOpenExclusive      sqlite3Os.xOpenExclusive
#define sqlite3OsOpenReadOnly       sqlite3Os.xOpenReadOnly
#define sqlite3OsDelete             sqlite3Os.xDelete
#define sqlite3OsFileExists         sqlite3Os.xFileExists
#define sqlite3OsFullPathname       sqlite3Os.xFullPathname
#define sqlite3OsIsDirWritable      sqlite3Os.xIsDirWritable
#define sqlite3OsSyncDirectory      sqlite3Os.xSyncDirectory
#define sqlite3OsTempFileName       sqlite3Os.xTempFileName
#define sqlite3OsRandomSeed         sqlite3Os.xRandomSeed
#define sqlite3OsSleep              sqlite3Os.xSleep
#define sqlite3OsCurrentTime        sqlite3Os.xCurrentTime
#define sqlite3OsEnterMutex         sqlite3Os.xEnterMutex
#define sqlite3OsLeaveMutex         sqlite3Os.xLeaveMutex
#define sqlite3OsInMutex            sqlite3Os.xInMutex
#define sqlite3OsThreadSpecificData sqlite3Os.xThreadSpecificData
#define sqlite3OsMalloc             sqlite3Os.xMalloc
#define sqlite3OsRealloc            sqlite3Os.xRealloc
#define sqlite3OsFree               sqlite3Os.xFree
#define sqlite3OsAllocationSize     sqlite3Os.xAllocationSize

#endif /* SQLITE_ENABLE_REDEF_IO */

#endif /* _SQLITE_OS_H_ */

/************** End of os.h **************************************************/
/************** Continuing where we left off in sqliteInt.h ******************/

/*
** Each database file to be accessed by the system is an instance
** of the following structure.  There are normally two of these structures
** in the sqlite.aDb[] array.  aDb[0] is the main database file and
** aDb[1] is the database file used to hold temporary tables.  Additional
** databases may be attached.
*/
struct Db {
  char *zName;         /* Name of this database */
  Btree *pBt;          /* The B*Tree structure for this database file */
  u8 inTrans;          /* 0: not writable.  1: Transaction.  2: Checkpoint */
  u8 safety_level;     /* How aggressive at synching data to disk */
  void *pAux;               /* Auxiliary data.  Usually NULL */
  void (*xFreeAux)(void*);  /* Routine to free pAux */
  Schema *pSchema;     /* Pointer to database schema (possibly shared) */
};

/*
** An instance of the following structure stores a database schema.
**
** If there are no virtual tables configured in this schema, the
** Schema.db variable is set to NULL. After the first virtual table
** has been added, it is set to point to the database connection 
** used to create the connection. Once a virtual table has been
** added to the Schema structure and the Schema.db variable populated, 
** only that database connection may use the Schema to prepare 
** statements.
*/
struct Schema {
  int schema_cookie;   /* Database schema version number for this file */
  Hash tblHash;        /* All tables indexed by name */
  Hash idxHash;        /* All (named) indices indexed by name */
  Hash trigHash;       /* All triggers indexed by name */
  Hash aFKey;          /* Foreign keys indexed by to-table */
  Table *pSeqTab;      /* The sqlite_sequence table used by AUTOINCREMENT */
  u8 file_format;      /* Schema format version for this file */
  u8 enc;              /* Text encoding used by this database */
  u16 flags;           /* Flags associated with this schema */
  int cache_size;      /* Number of pages to use in the cache */
#ifndef SQLITE_OMIT_VIRTUALTABLE
  sqlite3 *db;         /* "Owner" connection. See comment above */
#endif
};

/*
** These macros can be used to test, set, or clear bits in the 
** Db.flags field.
*/
#define DbHasProperty(D,I,P)     (((D)->aDb[I].pSchema->flags&(P))==(P))
#define DbHasAnyProperty(D,I,P)  (((D)->aDb[I].pSchema->flags&(P))!=0)
#define DbSetProperty(D,I,P)     (D)->aDb[I].pSchema->flags|=(P)
#define DbClearProperty(D,I,P)   (D)->aDb[I].pSchema->flags&=~(P)

/*
** Allowed values for the DB.flags field.
**
** The DB_SchemaLoaded flag is set after the database schema has been
** read into internal hash tables.
**
** DB_UnresetViews means that one or more views have column names that
** have been filled out.  If the schema changes, these column names might
** changes and so the view will need to be reset.
*/
#define DB_SchemaLoaded    0x0001  /* The schema has been loaded */
#define DB_UnresetViews    0x0002  /* Some views have defined column names */
#define DB_Empty           0x0004  /* The file is empty (length 0 bytes) */


/*
** Each database is an instance of the following structure.
**
** The sqlite.lastRowid records the last insert rowid generated by an
** insert statement.  Inserts on views do not affect its value.  Each
** trigger has its own context, so that lastRowid can be updated inside
** triggers as usual.  The previous value will be restored once the trigger
** exits.  Upon entering a before or instead of trigger, lastRowid is no
** longer (since after version 2.8.12) reset to -1.
**
** The sqlite.nChange does not count changes within triggers and keeps no
** context.  It is reset at start of sqlite3_exec.
** The sqlite.lsChange represents the number of changes made by the last
** insert, update, or delete statement.  It remains constant throughout the
** length of a statement and is then updated by OP_SetCounts.  It keeps a
** context stack just like lastRowid so that the count of changes
** within a trigger is not seen outside the trigger.  Changes to views do not
** affect the value of lsChange.
** The sqlite.csChange keeps track of the number of current changes (since
** the last statement) and is used to update sqlite_lsChange.
**
** The member variables sqlite.errCode, sqlite.zErrMsg and sqlite.zErrMsg16
** store the most recent error code and, if applicable, string. The
** internal function sqlite3Error() is used to set these variables
** consistently.
*/
struct sqlite3 {
  int nDb;                      /* Number of backends currently in use */
  Db *aDb;                      /* All backends */
  int flags;                    /* Miscellanous flags. See below */
  int errCode;                  /* Most recent error code (SQLITE_*) */
  int errMask;                  /* & result codes with this before returning */
  u8 autoCommit;                /* The auto-commit flag. */
  u8 temp_store;                /* 1: file 2: memory 0: default */
  int nTable;                   /* Number of tables in the database */
  CollSeq *pDfltColl;           /* The default collating sequence (BINARY) */
  i64 lastRowid;                /* ROWID of most recent insert (see above) */
  i64 priorNewRowid;            /* Last randomly generated ROWID */
  int magic;                    /* Magic number for detect library misuse */
  int nChange;                  /* Value returned by sqlite3_changes() */
  int nTotalChange;             /* Value returned by sqlite3_total_changes() */
  struct sqlite3InitInfo {      /* Information used during initialization */
    int iDb;                    /* When back is being initialized */
    int newTnum;                /* Rootpage of table being initialized */
    u8 busy;                    /* TRUE if currently initializing */
  } init;
  int nExtension;               /* Number of loaded extensions */
  void **aExtension;            /* Array of shared libraray handles */
  struct Vdbe *pVdbe;           /* List of active virtual machines */
  int activeVdbeCnt;            /* Number of vdbes currently executing */
  void (*xTrace)(void*,const char*);        /* Trace function */
  void *pTraceArg;                          /* Argument to the trace function */
  void (*xProfile)(void*,const char*,u64);  /* Profiling function */
  void *pProfileArg;                        /* Argument to profile function */
  void *pCommitArg;                 /* Argument to xCommitCallback() */   
  int (*xCommitCallback)(void*);    /* Invoked at every commit. */
  void *pRollbackArg;               /* Argument to xRollbackCallback() */   
  void (*xRollbackCallback)(void*); /* Invoked at every commit. */
  void *pUpdateArg;
  void (*xUpdateCallback)(void*,int, const char*,const char*,sqlite_int64);
  void(*xCollNeeded)(void*,sqlite3*,int eTextRep,const char*);
  void(*xCollNeeded16)(void*,sqlite3*,int eTextRep,const void*);
  void *pCollNeededArg;
  sqlite3_value *pErr;          /* Most recent error message */
  char *zErrMsg;                /* Most recent error message (UTF-8 encoded) */
  char *zErrMsg16;              /* Most recent error message (UTF-16 encoded) */
  union {
    int isInterrupted;          /* True if sqlite3_interrupt has been called */
    double notUsed1;            /* Spacer */
  } u1;
#ifndef SQLITE_OMIT_AUTHORIZATION
  int (*xAuth)(void*,int,const char*,const char*,const char*,const char*);
                                /* Access authorization function */
  void *pAuthArg;               /* 1st argument to the access auth function */
#endif
#ifndef SQLITE_OMIT_PROGRESS_CALLBACK
  int (*xProgress)(void *);     /* The progress callback */
  void *pProgressArg;           /* Argument to the progress callback */
  int nProgressOps;             /* Number of opcodes for progress callback */
#endif
#ifndef SQLITE_OMIT_VIRTUALTABLE
  Hash aModule;                 /* populated by sqlite3_create_module() */
  Table *pVTab;                 /* vtab with active Connect/Create method */
  sqlite3_vtab **aVTrans;       /* Virtual tables with open transactions */
  int nVTrans;                  /* Allocated size of aVTrans */
#endif
  Hash aFunc;                   /* All functions that can be in SQL exprs */
  Hash aCollSeq;                /* All collating sequences */
  BusyHandler busyHandler;      /* Busy callback */
  int busyTimeout;              /* Busy handler timeout, in msec */
  Db aDbStatic[2];              /* Static space for the 2 default backends */
#ifdef SQLITE_SSE
  sqlite3_stmt *pFetch;         /* Used by SSE to fetch stored statements */
#endif
  u8 dfltLockMode;              /* Default locking-mode for attached dbs */
};

/*
** A macro to discover the encoding of a database.
*/
#define ENC(db) ((db)->aDb[0].pSchema->enc)

/*
** Possible values for the sqlite.flags and or Db.flags fields.
**
** On sqlite.flags, the SQLITE_InTrans value means that we have
** executed a BEGIN.  On Db.flags, SQLITE_InTrans means a statement
** transaction is active on that particular database file.
*/
#define SQLITE_VdbeTrace      0x00000001  /* True to trace VDBE execution */
#define SQLITE_InTrans        0x00000008  /* True if in a transaction */
#define SQLITE_InternChanges  0x00000010  /* Uncommitted Hash table changes */
#define SQLITE_FullColNames   0x00000020  /* Show full column names on SELECT */
#define SQLITE_ShortColNames  0x00000040  /* Show short columns names */
#define SQLITE_CountRows      0x00000080  /* Count rows changed by INSERT, */
                                          /*   DELETE, or UPDATE and return */
                                          /*   the count using a callback. */
#define SQLITE_NullCallback   0x00000100  /* Invoke the callback once if the */
                                          /*   result set is empty */
#define SQLITE_SqlTrace       0x00000200  /* Debug print SQL as it executes */
#define SQLITE_VdbeListing    0x00000400  /* Debug listings of VDBE programs */
#define SQLITE_WriteSchema    0x00000800  /* OK to update SQLITE_MASTER */
#define SQLITE_NoReadlock     0x00001000  /* Readlocks are omitted when 
                                          ** accessing read-only databases */
#define SQLITE_IgnoreChecks   0x00002000  /* Do not enforce check constraints */
#define SQLITE_ReadUncommitted 0x00004000 /* For shared-cache mode */
#define SQLITE_LegacyFileFmt  0x00008000  /* Create new databases in format 1 */
#define SQLITE_FullFSync      0x00010000  /* Use full fsync on the backend */
#define SQLITE_LoadExtension  0x00020000  /* Enable load_extension */

#define SQLITE_RecoveryMode   0x00040000  /* Ignore schema errors */

/*
** Possible values for the sqlite.magic field.
** The numbers are obtained at random and have no special meaning, other
** than being distinct from one another.
*/
#define SQLITE_MAGIC_OPEN     0xa029a697  /* Database is open */
#define SQLITE_MAGIC_CLOSED   0x9f3c2d33  /* Database is closed */
#define SQLITE_MAGIC_BUSY     0xf03b7906  /* Database currently in use */
#define SQLITE_MAGIC_ERROR    0xb5357930  /* An SQLITE_MISUSE error occurred */

/*
** Each SQL function is defined by an instance of the following
** structure.  A pointer to this structure is stored in the sqlite.aFunc
** hash table.  When multiple functions have the same name, the hash table
** points to a linked list of these structures.
*/
struct FuncDef {
  i16 nArg;            /* Number of arguments.  -1 means unlimited */
  u8 iPrefEnc;         /* Preferred text encoding (SQLITE_UTF8, 16LE, 16BE) */
  u8 needCollSeq;      /* True if sqlite3GetFuncCollSeq() might be called */
  u8 flags;            /* Some combination of SQLITE_FUNC_* */
  void *pUserData;     /* User data parameter */
  FuncDef *pNext;      /* Next function with same name */
  void (*xFunc)(sqlite3_context*,int,sqlite3_value**); /* Regular function */
  void (*xStep)(sqlite3_context*,int,sqlite3_value**); /* Aggregate step */
  void (*xFinalize)(sqlite3_context*);                /* Aggregate finializer */
  char zName[1];       /* SQL name of the function.  MUST BE LAST */
};

/*
** Each SQLite module (virtual table definition) is defined by an
** instance of the following structure, stored in the sqlite3.aModule
** hash table.
*/
struct Module {
  const sqlite3_module *pModule;       /* Callback pointers */
  const char *zName;                   /* Name passed to create_module() */
  void *pAux;                          /* pAux passed to create_module() */
  void (*xDestroy)(void *);            /* Module destructor function */
};

/*
** Possible values for FuncDef.flags
*/
#define SQLITE_FUNC_LIKE   0x01  /* Candidate for the LIKE optimization */
#define SQLITE_FUNC_CASE   0x02  /* Case-sensitive LIKE-type function */
#define SQLITE_FUNC_EPHEM  0x04  /* Ephermeral.  Delete with VDBE */

/*
** information about each column of an SQL table is held in an instance
** of this structure.
*/
struct Column {
  char *zName;     /* Name of this column */
  Expr *pDflt;     /* Default value of this column */
  char *zType;     /* Data type for this column */
  char *zColl;     /* Collating sequence.  If NULL, use the default */
  u8 notNull;      /* True if there is a NOT NULL constraint */
  u8 isPrimKey;    /* True if this column is part of the PRIMARY KEY */
  char affinity;   /* One of the SQLITE_AFF_... values */
#ifndef SQLITE_OMIT_VIRTUALTABLE
  u8 isHidden;     /* True if this column is 'hidden' */
#endif
};

/*
** A "Collating Sequence" is defined by an instance of the following
** structure. Conceptually, a collating sequence consists of a name and
** a comparison routine that defines the order of that sequence.
**
** There may two seperate implementations of the collation function, one
** that processes text in UTF-8 encoding (CollSeq.xCmp) and another that
** processes text encoded in UTF-16 (CollSeq.xCmp16), using the machine
** native byte order. When a collation sequence is invoked, SQLite selects
** the version that will require the least expensive encoding
** translations, if any.
**
** The CollSeq.pUser member variable is an extra parameter that passed in
** as the first argument to the UTF-8 comparison function, xCmp.
** CollSeq.pUser16 is the equivalent for the UTF-16 comparison function,
** xCmp16.
**
** If both CollSeq.xCmp and CollSeq.xCmp16 are NULL, it means that the
** collating sequence is undefined.  Indices built on an undefined
** collating sequence may not be read or written.
*/
struct CollSeq {
  char *zName;          /* Name of the collating sequence, UTF-8 encoded */
  u8 enc;               /* Text encoding handled by xCmp() */
  u8 type;              /* One of the SQLITE_COLL_... values below */
  void *pUser;          /* First argument to xCmp() */
  int (*xCmp)(void*,int, const void*, int, const void*);
  void (*xDel)(void*);  /* Destructor for pUser */
};

/*
** Allowed values of CollSeq flags:
*/
#define SQLITE_COLL_BINARY  1  /* The default memcmp() collating sequence */
#define SQLITE_COLL_NOCASE  2  /* The built-in NOCASE collating sequence */
#define SQLITE_COLL_REVERSE 3  /* The built-in REVERSE collating sequence */
#define SQLITE_COLL_USER    0  /* Any other user-defined collating sequence */

/*
** A sort order can be either ASC or DESC.
*/
#define SQLITE_SO_ASC       0  /* Sort in ascending order */
#define SQLITE_SO_DESC      1  /* Sort in ascending order */

/*
** Column affinity types.
**
** These used to have mnemonic name like 'i' for SQLITE_AFF_INTEGER and
** 't' for SQLITE_AFF_TEXT.  But we can save a little space and improve
** the speed a little by number the values consecutively.  
**
** But rather than start with 0 or 1, we begin with 'a'.  That way,
** when multiple affinity types are concatenated into a string and
** used as the P3 operand, they will be more readable.
**
** Note also that the numeric types are grouped together so that testing
** for a numeric type is a single comparison.
*/
#define SQLITE_AFF_TEXT     'a'
#define SQLITE_AFF_NONE     'b'
#define SQLITE_AFF_NUMERIC  'c'
#define SQLITE_AFF_INTEGER  'd'
#define SQLITE_AFF_REAL     'e'

#define sqlite3IsNumericAffinity(X)  ((X)>=SQLITE_AFF_NUMERIC)

/*
** Each SQL table is represented in memory by an instance of the
** following structure.
**
** Table.zName is the name of the table.  The case of the original
** CREATE TABLE statement is stored, but case is not significant for
** comparisons.
**
** Table.nCol is the number of columns in this table.  Table.aCol is a
** pointer to an array of Column structures, one for each column.
**
** If the table has an INTEGER PRIMARY KEY, then Table.iPKey is the index of
** the column that is that key.   Otherwise Table.iPKey is negative.  Note
** that the datatype of the PRIMARY KEY must be INTEGER for this field to
** be set.  An INTEGER PRIMARY KEY is used as the rowid for each row of
** the table.  If a table has no INTEGER PRIMARY KEY, then a random rowid
** is generated for each row of the table.  Table.hasPrimKey is true if
** the table has any PRIMARY KEY, INTEGER or otherwise.
**
** Table.tnum is the page number for the root BTree page of the table in the
** database file.  If Table.iDb is the index of the database table backend
** in sqlite.aDb[].  0 is for the main database and 1 is for the file that
** holds temporary tables and indices.  If Table.isEphem
** is true, then the table is stored in a file that is automatically deleted
** when the VDBE cursor to the table is closed.  In this case Table.tnum 
** refers VDBE cursor number that holds the table open, not to the root
** page number.  Transient tables are used to hold the results of a
** sub-query that appears instead of a real table name in the FROM clause 
** of a SELECT statement.
*/
struct Table {
  char *zName;     /* Name of the table */
  int nCol;        /* Number of columns in this table */
  Column *aCol;    /* Information about each column */
  int iPKey;       /* If not less then 0, use aCol[iPKey] as the primary key */
  Index *pIndex;   /* List of SQL indexes on this table. */
  int tnum;        /* Root BTree node for this table (see note above) */
  Select *pSelect; /* NULL for tables.  Points to definition if a view. */
  int nRef;          /* Number of pointers to this Table */
  Trigger *pTrigger; /* List of SQL triggers on this table */
  FKey *pFKey;       /* Linked list of all foreign keys in this table */
  char *zColAff;     /* String defining the affinity of each column */
#ifndef SQLITE_OMIT_CHECK
  Expr *pCheck;      /* The AND of all CHECK constraints */
#endif
#ifndef SQLITE_OMIT_ALTERTABLE
  int addColOffset;  /* Offset in CREATE TABLE statement to add a new column */
#endif
  u8 readOnly;     /* True if this table should not be written by the user */
  u8 isEphem;      /* True if created using OP_OpenEphermeral */
  u8 hasPrimKey;   /* True if there exists a primary key */
  u8 keyConf;      /* What to do in case of uniqueness conflict on iPKey */
  u8 autoInc;      /* True if the integer primary key is autoincrement */
#ifndef SQLITE_OMIT_VIRTUALTABLE
  u8 isVirtual;             /* True if this is a virtual table */
  u8 isCommit;              /* True once the CREATE TABLE has been committed */
  Module *pMod;             /* Pointer to the implementation of the module */
  sqlite3_vtab *pVtab;      /* Pointer to the module instance */
  int nModuleArg;           /* Number of arguments to the module */
  char **azModuleArg;       /* Text of all module args. [0] is module name */
#endif
  Schema *pSchema;
};

/*
** Test to see whether or not a table is a virtual table.  This is
** done as a macro so that it will be optimized out when virtual
** table support is omitted from the build.
*/
#ifndef SQLITE_OMIT_VIRTUALTABLE
#  define IsVirtual(X)      ((X)->isVirtual)
#  define IsHiddenColumn(X) ((X)->isHidden)
#else
#  define IsVirtual(X)      0
#  define IsHiddenColumn(X) 0
#endif

/*
** Each foreign key constraint is an instance of the following structure.
**
** A foreign key is associated with two tables.  The "from" table is
** the table that contains the REFERENCES clause that creates the foreign
** key.  The "to" table is the table that is named in the REFERENCES clause.
** Consider this example:
**
**     CREATE TABLE ex1(
**       a INTEGER PRIMARY KEY,
**       b INTEGER CONSTRAINT fk1 REFERENCES ex2(x)
**     );
**
** For foreign key "fk1", the from-table is "ex1" and the to-table is "ex2".
**
** Each REFERENCES clause generates an instance of the following structure
** which is attached to the from-table.  The to-table need not exist when
** the from-table is created.  The existance of the to-table is not checked
** until an attempt is made to insert data into the from-table.
**
** The sqlite.aFKey hash table stores pointers to this structure
** given the name of a to-table.  For each to-table, all foreign keys
** associated with that table are on a linked list using the FKey.pNextTo
** field.
*/
struct FKey {
  Table *pFrom;     /* The table that constains the REFERENCES clause */
  FKey *pNextFrom;  /* Next foreign key in pFrom */
  char *zTo;        /* Name of table that the key points to */
  FKey *pNextTo;    /* Next foreign key that points to zTo */
  int nCol;         /* Number of columns in this key */
  struct sColMap {  /* Mapping of columns in pFrom to columns in zTo */
    int iFrom;         /* Index of column in pFrom */
    char *zCol;        /* Name of column in zTo.  If 0 use PRIMARY KEY */
  } *aCol;          /* One entry for each of nCol column s */
  u8 isDeferred;    /* True if constraint checking is deferred till COMMIT */
  u8 updateConf;    /* How to resolve conflicts that occur on UPDATE */
  u8 deleteConf;    /* How to resolve conflicts that occur on DELETE */
  u8 insertConf;    /* How to resolve conflicts that occur on INSERT */
};

/*
** SQLite supports many different ways to resolve a contraint
** error.  ROLLBACK processing means that a constraint violation
** causes the operation in process to fail and for the current transaction
** to be rolled back.  ABORT processing means the operation in process
** fails and any prior changes from that one operation are backed out,
** but the transaction is not rolled back.  FAIL processing means that
** the operation in progress stops and returns an error code.  But prior
** changes due to the same operation are not backed out and no rollback
** occurs.  IGNORE means that the particular row that caused the constraint
** error is not inserted or updated.  Processing continues and no error
** is returned.  REPLACE means that preexisting database rows that caused
** a UNIQUE constraint violation are removed so that the new insert or
** update can proceed.  Processing continues and no error is reported.
**
** RESTRICT, SETNULL, and CASCADE actions apply only to foreign keys.
** RESTRICT is the same as ABORT for IMMEDIATE foreign keys and the
** same as ROLLBACK for DEFERRED keys.  SETNULL means that the foreign
** key is set to NULL.  CASCADE means that a DELETE or UPDATE of the
** referenced table row is propagated into the row that holds the
** foreign key.
** 
** The following symbolic values are used to record which type
** of action to take.
*/
#define OE_None     0   /* There is no constraint to check */
#define OE_Rollback 1   /* Fail the operation and rollback the transaction */
#define OE_Abort    2   /* Back out changes but do no rollback transaction */
#define OE_Fail     3   /* Stop the operation but leave all prior changes */
#define OE_Ignore   4   /* Ignore the error. Do not do the INSERT or UPDATE */
#define OE_Replace  5   /* Delete existing record, then do INSERT or UPDATE */

#define OE_Restrict 6   /* OE_Abort for IMMEDIATE, OE_Rollback for DEFERRED */
#define OE_SetNull  7   /* Set the foreign key value to NULL */
#define OE_SetDflt  8   /* Set the foreign key value to its default */
#define OE_Cascade  9   /* Cascade the changes */

#define OE_Default  99  /* Do whatever the default action is */


/*
** An instance of the following structure is passed as the first
** argument to sqlite3VdbeKeyCompare and is used to control the 
** comparison of the two index keys.
**
** If the KeyInfo.incrKey value is true and the comparison would
** otherwise be equal, then return a result as if the second key
** were larger.
*/
struct KeyInfo {
  u8 enc;             /* Text encoding - one of the TEXT_Utf* values */
  u8 incrKey;         /* Increase 2nd key by epsilon before comparison */
  int nField;         /* Number of entries in aColl[] */
  u8 *aSortOrder;     /* If defined an aSortOrder[i] is true, sort DESC */
  CollSeq *aColl[1];  /* Collating sequence for each term of the key */
};

/*
** Each SQL index is represented in memory by an
** instance of the following structure.
**
** The columns of the table that are to be indexed are described
** by the aiColumn[] field of this structure.  For example, suppose
** we have the following table and index:
**
**     CREATE TABLE Ex1(c1 int, c2 int, c3 text);
**     CREATE INDEX Ex2 ON Ex1(c3,c1);
**
** In the Table structure describing Ex1, nCol==3 because there are
** three columns in the table.  In the Index structure describing
** Ex2, nColumn==2 since 2 of the 3 columns of Ex1 are indexed.
** The value of aiColumn is {2, 0}.  aiColumn[0]==2 because the 
** first column to be indexed (c3) has an index of 2 in Ex1.aCol[].
** The second column to be indexed (c1) has an index of 0 in
** Ex1.aCol[], hence Ex2.aiColumn[1]==0.
**
** The Index.onError field determines whether or not the indexed columns
** must be unique and what to do if they are not.  When Index.onError=OE_None,
** it means this is not a unique index.  Otherwise it is a unique index
** and the value of Index.onError indicate the which conflict resolution 
** algorithm to employ whenever an attempt is made to insert a non-unique
** element.
*/
struct Index {
  char *zName;     /* Name of this index */
  int nColumn;     /* Number of columns in the table used by this index */
  int *aiColumn;   /* Which columns are used by this index.  1st is 0 */
  unsigned *aiRowEst; /* Result of ANALYZE: Est. rows selected by each column */
  Table *pTable;   /* The SQL table being indexed */
  int tnum;        /* Page containing root of this index in database file */
  u8 onError;      /* OE_Abort, OE_Ignore, OE_Replace, or OE_None */
  u8 autoIndex;    /* True if is automatically created (ex: by UNIQUE) */
  char *zColAff;   /* String defining the affinity of each column */
  Index *pNext;    /* The next index associated with the same table */
  Schema *pSchema; /* Schema containing this index */
  u8 *aSortOrder;  /* Array of size Index.nColumn. True==DESC, False==ASC */
  char **azColl;   /* Array of collation sequence names for index */
};

/*
** Each token coming out of the lexer is an instance of
** this structure.  Tokens are also used as part of an expression.
**
** Note if Token.z==0 then Token.dyn and Token.n are undefined and
** may contain random values.  Do not make any assuptions about Token.dyn
** and Token.n when Token.z==0.
*/
struct Token {
  const unsigned char *z; /* Text of the token.  Not NULL-terminated! */
  unsigned dyn  : 1;      /* True for malloced memory, false for static */
  unsigned n    : 31;     /* Number of characters in this token */
};

/*
** An instance of this structure contains information needed to generate
** code for a SELECT that contains aggregate functions.
**
** If Expr.op==TK_AGG_COLUMN or TK_AGG_FUNCTION then Expr.pAggInfo is a
** pointer to this structure.  The Expr.iColumn field is the index in
** AggInfo.aCol[] or AggInfo.aFunc[] of information needed to generate
** code for that node.
**
** AggInfo.pGroupBy and AggInfo.aFunc.pExpr point to fields within the
** original Select structure that describes the SELECT statement.  These
** fields do not need to be freed when deallocating the AggInfo structure.
*/
struct AggInfo {
  u8 directMode;          /* Direct rendering mode means take data directly
                          ** from source tables rather than from accumulators */
  u8 useSortingIdx;       /* In direct mode, reference the sorting index rather
                          ** than the source table */
  int sortingIdx;         /* Cursor number of the sorting index */
  ExprList *pGroupBy;     /* The group by clause */
  int nSortingColumn;     /* Number of columns in the sorting index */
  struct AggInfo_col {    /* For each column used in source tables */
    Table *pTab;             /* Source table */
    int iTable;              /* Cursor number of the source table */
    int iColumn;             /* Column number within the source table */
    int iSorterColumn;       /* Column number in the sorting index */
    int iMem;                /* Memory location that acts as accumulator */
    Expr *pExpr;             /* The original expression */
  } *aCol;
  int nColumn;            /* Number of used entries in aCol[] */
  int nColumnAlloc;       /* Number of slots allocated for aCol[] */
  int nAccumulator;       /* Number of columns that show through to the output.
                          ** Additional columns are used only as parameters to
                          ** aggregate functions */
  struct AggInfo_func {   /* For each aggregate function */
    Expr *pExpr;             /* Expression encoding the function */
    FuncDef *pFunc;          /* The aggregate function implementation */
    int iMem;                /* Memory location that acts as accumulator */
    int iDistinct;           /* Ephermeral table used to enforce DISTINCT */
  } *aFunc;
  int nFunc;              /* Number of entries in aFunc[] */
  int nFuncAlloc;         /* Number of slots allocated for aFunc[] */
};

/*
** Each node of an expression in the parse tree is an instance
** of this structure.
**
** Expr.op is the opcode.  The integer parser token codes are reused
** as opcodes here.  For example, the parser defines TK_GE to be an integer
** code representing the ">=" operator.  This same integer code is reused
** to represent the greater-than-or-equal-to operator in the expression
** tree.
**
** Expr.pRight and Expr.pLeft are subexpressions.  Expr.pList is a list
** of argument if the expression is a function.
**
** Expr.token is the operator token for this node.  For some expressions
** that have subexpressions, Expr.token can be the complete text that gave
** rise to the Expr.  In the latter case, the token is marked as being
** a compound token.
**
** An expression of the form ID or ID.ID refers to a column in a table.
** For such expressions, Expr.op is set to TK_COLUMN and Expr.iTable is
** the integer cursor number of a VDBE cursor pointing to that table and
** Expr.iColumn is the column number for the specific column.  If the
** expression is used as a result in an aggregate SELECT, then the
** value is also stored in the Expr.iAgg column in the aggregate so that
** it can be accessed after all aggregates are computed.
**
** If the expression is a function, the Expr.iTable is an integer code
** representing which function.  If the expression is an unbound variable
** marker (a question mark character '?' in the original SQL) then the
** Expr.iTable holds the index number for that variable.
**
** If the expression is a subquery then Expr.iColumn holds an integer
** register number containing the result of the subquery.  If the
** subquery gives a constant result, then iTable is -1.  If the subquery
** gives a different answer at different times during statement processing
** then iTable is the address of a subroutine that computes the subquery.
**
** The Expr.pSelect field points to a SELECT statement.  The SELECT might
** be the right operand of an IN operator.  Or, if a scalar SELECT appears
** in an expression the opcode is TK_SELECT and Expr.pSelect is the only
** operand.
**
** If the Expr is of type OP_Column, and the table it is selecting from
** is a disk table or the "old.*" pseudo-table, then pTab points to the
** corresponding table definition.
*/
struct Expr {
  u8 op;                 /* Operation performed by this node */
  char affinity;         /* The affinity of the column or 0 if not a column */
  u16 flags;             /* Various flags.  See below */
  CollSeq *pColl;        /* The collation type of the column or 0 */
  Expr *pLeft, *pRight;  /* Left and right subnodes */
  ExprList *pList;       /* A list of expressions used as function arguments
                         ** or in "<expr> IN (<expr-list)" */
  Token token;           /* An operand token */
  Token span;            /* Complete text of the expression */
  int iTable, iColumn;   /* When op==TK_COLUMN, then this expr node means the
                         ** iColumn-th field of the iTable-th table. */
  AggInfo *pAggInfo;     /* Used by TK_AGG_COLUMN and TK_AGG_FUNCTION */
  int iAgg;              /* Which entry in pAggInfo->aCol[] or ->aFunc[] */
  int iRightJoinTable;   /* If EP_FromJoin, the right table of the join */
  Select *pSelect;       /* When the expression is a sub-select.  Also the
                         ** right side of "<expr> IN (<select>)" */
  Table *pTab;           /* Table for OP_Column expressions. */
  Schema *pSchema;
#if SQLITE_MAX_EXPR_DEPTH>0
  int nHeight;           /* Height of the tree headed by this node */
#endif
};

/*
** The following are the meanings of bits in the Expr.flags field.
*/
#define EP_FromJoin     0x01  /* Originated in ON or USING clause of a join */
#define EP_Agg          0x02  /* Contains one or more aggregate functions */
#define EP_Resolved     0x04  /* IDs have been resolved to COLUMNs */
#define EP_Error        0x08  /* Expression contains one or more errors */
#define EP_Distinct     0x10  /* Aggregate function with DISTINCT keyword */
#define EP_VarSelect    0x20  /* pSelect is correlated, not constant */
#define EP_Dequoted     0x40  /* True if the string has been dequoted */
#define EP_InfixFunc    0x80  /* True for an infix function: LIKE, GLOB, etc */
#define EP_ExpCollate  0x100  /* Collating sequence specified explicitly */

/*
** These macros can be used to test, set, or clear bits in the 
** Expr.flags field.
*/
#define ExprHasProperty(E,P)     (((E)->flags&(P))==(P))
#define ExprHasAnyProperty(E,P)  (((E)->flags&(P))!=0)
#define ExprSetProperty(E,P)     (E)->flags|=(P)
#define ExprClearProperty(E,P)   (E)->flags&=~(P)

/*
** A list of expressions.  Each expression may optionally have a
** name.  An expr/name combination can be used in several ways, such
** as the list of "expr AS ID" fields following a "SELECT" or in the
** list of "ID = expr" items in an UPDATE.  A list of expressions can
** also be used as the argument to a function, in which case the a.zName
** field is not used.
*/
struct ExprList {
  int nExpr;             /* Number of expressions on the list */
  int nAlloc;            /* Number of entries allocated below */
  int iECursor;          /* VDBE Cursor associated with this ExprList */
  struct ExprList_item {
    Expr *pExpr;           /* The list of expressions */
    char *zName;           /* Token associated with this expression */
    u8 sortOrder;          /* 1 for DESC or 0 for ASC */
    u8 isAgg;              /* True if this is an aggregate like count(*) */
    u8 done;               /* A flag to indicate when processing is finished */
  } *a;                  /* One entry for each expression */
};

/*
** An instance of this structure can hold a simple list of identifiers,
** such as the list "a,b,c" in the following statements:
**
**      INSERT INTO t(a,b,c) VALUES ...;
**      CREATE INDEX idx ON t(a,b,c);
**      CREATE TRIGGER trig BEFORE UPDATE ON t(a,b,c) ...;
**
** The IdList.a.idx field is used when the IdList represents the list of
** column names after a table name in an INSERT statement.  In the statement
**
**     INSERT INTO t(a,b,c) ...
**
** If "a" is the k-th column of table "t", then IdList.a[0].idx==k.
*/
struct IdList {
  struct IdList_item {
    char *zName;      /* Name of the identifier */
    int idx;          /* Index in some Table.aCol[] of a column named zName */
  } *a;
  int nId;         /* Number of identifiers on the list */
  int nAlloc;      /* Number of entries allocated for a[] below */
};

/*
** The bitmask datatype defined below is used for various optimizations.
**
** Changing this from a 64-bit to a 32-bit type limits the number of
** tables in a join to 32 instead of 64.  But it also reduces the size
** of the library by 738 bytes on ix86.
*/
typedef u64 Bitmask;

/*
** The following structure describes the FROM clause of a SELECT statement.
** Each table or subquery in the FROM clause is a separate element of
** the SrcList.a[] array.
**
** With the addition of multiple database support, the following structure
** can also be used to describe a particular table such as the table that
** is modified by an INSERT, DELETE, or UPDATE statement.  In standard SQL,
** such a table must be a simple name: ID.  But in SQLite, the table can
** now be identified by a database name, a dot, then the table name: ID.ID.
**
** The jointype starts out showing the join type between the current table
** and the next table on the list.  The parser builds the list this way.
** But sqlite3SrcListShiftJoinType() later shifts the jointypes so that each
** jointype expresses the join between the table and the previous table.
*/
struct SrcList {
  i16 nSrc;        /* Number of tables or subqueries in the FROM clause */
  i16 nAlloc;      /* Number of entries allocated in a[] below */
  struct SrcList_item {
    char *zDatabase;  /* Name of database holding this table */
    char *zName;      /* Name of the table */
    char *zAlias;     /* The "B" part of a "A AS B" phrase.  zName is the "A" */
    Table *pTab;      /* An SQL table corresponding to zName */
    Select *pSelect;  /* A SELECT statement used in place of a table name */
    u8 isPopulated;   /* Temporary table associated with SELECT is populated */
    u8 jointype;      /* Type of join between this able and the previous */
    int iCursor;      /* The VDBE cursor number used to access this table */
    Expr *pOn;        /* The ON clause of a join */
    IdList *pUsing;   /* The USING clause of a join */
    Bitmask colUsed;  /* Bit N (1<<N) set if column N or pTab is used */
  } a[1];             /* One entry for each identifier on the list */
};

/*
** Permitted values of the SrcList.a.jointype field
*/
#define JT_INNER     0x0001    /* Any kind of inner or cross join */
#define JT_CROSS     0x0002    /* Explicit use of the CROSS keyword */
#define JT_NATURAL   0x0004    /* True for a "natural" join */
#define JT_LEFT      0x0008    /* Left outer join */
#define JT_RIGHT     0x0010    /* Right outer join */
#define JT_OUTER     0x0020    /* The "OUTER" keyword is present */
#define JT_ERROR     0x0040    /* unknown or unsupported join type */

/*
** For each nested loop in a WHERE clause implementation, the WhereInfo
** structure contains a single instance of this structure.  This structure
** is intended to be private the the where.c module and should not be
** access or modified by other modules.
**
** The pIdxInfo and pBestIdx fields are used to help pick the best
** index on a virtual table.  The pIdxInfo pointer contains indexing
** information for the i-th table in the FROM clause before reordering.
** All the pIdxInfo pointers are freed by whereInfoFree() in where.c.
** The pBestIdx pointer is a copy of pIdxInfo for the i-th table after
** FROM clause ordering.  This is a little confusing so I will repeat
** it in different words.  WhereInfo.a[i].pIdxInfo is index information 
** for WhereInfo.pTabList.a[i].  WhereInfo.a[i].pBestInfo is the
** index information for the i-th loop of the join.  pBestInfo is always
** either NULL or a copy of some pIdxInfo.  So for cleanup it is 
** sufficient to free all of the pIdxInfo pointers.
** 
*/
struct WhereLevel {
  int iFrom;            /* Which entry in the FROM clause */
  int flags;            /* Flags associated with this level */
  int iMem;             /* First memory cell used by this level */
  int iLeftJoin;        /* Memory cell used to implement LEFT OUTER JOIN */
  Index *pIdx;          /* Index used.  NULL if no index */
  int iTabCur;          /* The VDBE cursor used to access the table */
  int iIdxCur;          /* The VDBE cursor used to acesss pIdx */
  int brk;              /* Jump here to break out of the loop */
  int nxt;              /* Jump here to start the next IN combination */
  int cont;             /* Jump here to continue with the next loop cycle */
  int top;              /* First instruction of interior of the loop */
  int op, p1, p2;       /* Opcode used to terminate the loop */
  int nEq;              /* Number of == or IN constraints on this loop */
  int nIn;              /* Number of IN operators constraining this loop */
  struct InLoop {
    int iCur;              /* The VDBE cursor used by this IN operator */
    int topAddr;           /* Top of the IN loop */
  } *aInLoop;           /* Information about each nested IN operator */
  sqlite3_index_info *pBestIdx;  /* Index information for this level */

  /* The following field is really not part of the current level.  But
  ** we need a place to cache index information for each table in the
  ** FROM clause and the WhereLevel structure is a convenient place.
  */
  sqlite3_index_info *pIdxInfo;  /* Index info for n-th source table */
};

/*
** The WHERE clause processing routine has two halves.  The
** first part does the start of the WHERE loop and the second
** half does the tail of the WHERE loop.  An instance of
** this structure is returned by the first half and passed
** into the second half to give some continuity.
*/
struct WhereInfo {
  Parse *pParse;
  SrcList *pTabList;   /* List of tables in the join */
  int iTop;            /* The very beginning of the WHERE loop */
  int iContinue;       /* Jump here to continue with next record */
  int iBreak;          /* Jump here to break out of the loop */
  int nLevel;          /* Number of nested loop */
  sqlite3_index_info **apInfo;  /* Array of pointers to index info structures */
  WhereLevel a[1];     /* Information about each nest loop in the WHERE */
};

/*
** A NameContext defines a context in which to resolve table and column
** names.  The context consists of a list of tables (the pSrcList) field and
** a list of named expression (pEList).  The named expression list may
** be NULL.  The pSrc corresponds to the FROM clause of a SELECT or
** to the table being operated on by INSERT, UPDATE, or DELETE.  The
** pEList corresponds to the result set of a SELECT and is NULL for
** other statements.
**
** NameContexts can be nested.  When resolving names, the inner-most 
** context is searched first.  If no match is found, the next outer
** context is checked.  If there is still no match, the next context
** is checked.  This process continues until either a match is found
** or all contexts are check.  When a match is found, the nRef member of
** the context containing the match is incremented. 
**
** Each subquery gets a new NameContext.  The pNext field points to the
** NameContext in the parent query.  Thus the process of scanning the
** NameContext list corresponds to searching through successively outer
** subqueries looking for a match.
*/
struct NameContext {
  Parse *pParse;       /* The parser */
  SrcList *pSrcList;   /* One or more tables used to resolve names */
  ExprList *pEList;    /* Optional list of named expressions */
  int nRef;            /* Number of names resolved by this context */
  int nErr;            /* Number of errors encountered while resolving names */
  u8 allowAgg;         /* Aggregate functions allowed here */
  u8 hasAgg;           /* True if aggregates are seen */
  u8 isCheck;          /* True if resolving names in a CHECK constraint */
  int nDepth;          /* Depth of subquery recursion. 1 for no recursion */
  AggInfo *pAggInfo;   /* Information about aggregates at this level */
  NameContext *pNext;  /* Next outer name context.  NULL for outermost */
};

/*
** An instance of the following structure contains all information
** needed to generate code for a single SELECT statement.
**
** nLimit is set to -1 if there is no LIMIT clause.  nOffset is set to 0.
** If there is a LIMIT clause, the parser sets nLimit to the value of the
** limit and nOffset to the value of the offset (or 0 if there is not
** offset).  But later on, nLimit and nOffset become the memory locations
** in the VDBE that record the limit and offset counters.
**
** addrOpenEphm[] entries contain the address of OP_OpenEphemeral opcodes.
** These addresses must be stored so that we can go back and fill in
** the P3_KEYINFO and P2 parameters later.  Neither the KeyInfo nor
** the number of columns in P2 can be computed at the same time
** as the OP_OpenEphm instruction is coded because not
** enough information about the compound query is known at that point.
** The KeyInfo for addrOpenTran[0] and [1] contains collating sequences
** for the result set.  The KeyInfo for addrOpenTran[2] contains collating
** sequences for the ORDER BY clause.
*/
struct Select {
  ExprList *pEList;      /* The fields of the result */
  u8 op;                 /* One of: TK_UNION TK_ALL TK_INTERSECT TK_EXCEPT */
  u8 isDistinct;         /* True if the DISTINCT keyword is present */
  u8 isResolved;         /* True once sqlite3SelectResolve() has run. */
  u8 isAgg;              /* True if this is an aggregate query */
  u8 usesEphm;           /* True if uses an OpenEphemeral opcode */
  u8 disallowOrderBy;    /* Do not allow an ORDER BY to be attached if TRUE */
  char affinity;         /* MakeRecord with this affinity for SRT_Set */
  SrcList *pSrc;         /* The FROM clause */
  Expr *pWhere;          /* The WHERE clause */
  ExprList *pGroupBy;    /* The GROUP BY clause */
  Expr *pHaving;         /* The HAVING clause */
  ExprList *pOrderBy;    /* The ORDER BY clause */
  Select *pPrior;        /* Prior select in a compound select statement */
  Select *pRightmost;    /* Right-most select in a compound select statement */
  Expr *pLimit;          /* LIMIT expression. NULL means not used. */
  Expr *pOffset;         /* OFFSET expression. NULL means not used. */
  int iLimit, iOffset;   /* Memory registers holding LIMIT & OFFSET counters */
  int addrOpenEphm[3];   /* OP_OpenEphem opcodes related to this select */
};

/*
** The results of a select can be distributed in several ways.
*/
#define SRT_Union        1  /* Store result as keys in an index */
#define SRT_Except       2  /* Remove result from a UNION index */
#define SRT_Discard      3  /* Do not save the results anywhere */

/* The ORDER BY clause is ignored for all of the above */
#define IgnorableOrderby(X) (X<=SRT_Discard)

#define SRT_Callback     4  /* Invoke a callback with each row of result */
#define SRT_Mem          5  /* Store result in a memory cell */
#define SRT_Set          6  /* Store non-null results as keys in an index */
#define SRT_Table        7  /* Store result as data with an automatic rowid */
#define SRT_EphemTab     8  /* Create transient tab and store like SRT_Table */
#define SRT_Subroutine   9  /* Call a subroutine to handle results */
#define SRT_Exists      10  /* Store 1 if the result is not empty */

/*
** An SQL parser context.  A copy of this structure is passed through
** the parser and down into all the parser action routine in order to
** carry around information that is global to the entire parse.
**
** The structure is divided into two parts.  When the parser and code
** generate call themselves recursively, the first part of the structure
** is constant but the second part is reset at the beginning and end of
** each recursion.
**
** The nTableLock and aTableLock variables are only used if the shared-cache 
** feature is enabled (if sqlite3Tsd()->useSharedData is true). They are
** used to store the set of table-locks required by the statement being
** compiled. Function sqlite3TableLock() is used to add entries to the
** list.
*/
struct Parse {
  sqlite3 *db;         /* The main database structure */
  int rc;              /* Return code from execution */
  char *zErrMsg;       /* An error message */
  Vdbe *pVdbe;         /* An engine for executing database bytecode */
  u8 colNamesSet;      /* TRUE after OP_ColumnName has been issued to pVdbe */
  u8 nameClash;        /* A permanent table name clashes with temp table name */
  u8 checkSchema;      /* Causes schema cookie check after an error */
  u8 nested;           /* Number of nested calls to the parser/code generator */
  u8 parseError;       /* True after a parsing error.  Ticket #1794 */
  int nErr;            /* Number of errors seen */
  int nTab;            /* Number of previously allocated VDBE cursors */
  int nMem;            /* Number of memory cells used so far */
  int nSet;            /* Number of sets used so far */
  int ckOffset;        /* Stack offset to data used by CHECK constraints */
  u32 writeMask;       /* Start a write transaction on these databases */
  u32 cookieMask;      /* Bitmask of schema verified databases */
  int cookieGoto;      /* Address of OP_Goto to cookie verifier subroutine */
  int cookieValue[SQLITE_MAX_ATTACHED+2];  /* Values of cookies to verify */
#ifndef SQLITE_OMIT_SHARED_CACHE
  int nTableLock;        /* Number of locks in aTableLock */
  TableLock *aTableLock; /* Required table locks for shared-cache mode */
#endif

  /* Above is constant between recursions.  Below is reset before and after
  ** each recursion */

  int nVar;            /* Number of '?' variables seen in the SQL so far */
  int nVarExpr;        /* Number of used slots in apVarExpr[] */
  int nVarExprAlloc;   /* Number of allocated slots in apVarExpr[] */
  Expr **apVarExpr;    /* Pointers to :aaa and $aaaa wildcard expressions */
  u8 explain;          /* True if the EXPLAIN flag is found on the query */
  Token sErrToken;     /* The token at which the error occurred */
  Token sNameToken;    /* Token with unqualified schema object name */
  Token sLastToken;    /* The last token parsed */
  const char *zSql;    /* All SQL text */
  const char *zTail;   /* All SQL text past the last semicolon parsed */
  Table *pNewTable;    /* A table being constructed by CREATE TABLE */
  Trigger *pNewTrigger;     /* Trigger under construct by a CREATE TRIGGER */
  TriggerStack *trigStack;  /* Trigger actions being coded */
  const char *zAuthContext; /* The 6th parameter to db->xAuth callbacks */
#ifndef SQLITE_OMIT_VIRTUALTABLE
  Token sArg;                /* Complete text of a module argument */
  u8 declareVtab;            /* True if inside sqlite3_declare_vtab() */
  Table *pVirtualLock;       /* Require virtual table lock on this table */
#endif
#if SQLITE_MAX_EXPR_DEPTH>0
  int nHeight;            /* Expression tree height of current sub-select */
#endif
};

#ifdef SQLITE_OMIT_VIRTUALTABLE
  #define IN_DECLARE_VTAB 0
#else
  #define IN_DECLARE_VTAB (pParse->declareVtab)
#endif

/*
** An instance of the following structure can be declared on a stack and used
** to save the Parse.zAuthContext value so that it can be restored later.
*/
struct AuthContext {
  const char *zAuthContext;   /* Put saved Parse.zAuthContext here */
  Parse *pParse;              /* The Parse structure */
};

/*
** Bitfield flags for P2 value in OP_Insert and OP_Delete
*/
#define OPFLAG_NCHANGE   1    /* Set to update db->nChange */
#define OPFLAG_LASTROWID 2    /* Set to update db->lastRowid */
#define OPFLAG_ISUPDATE  4    /* This OP_Insert is an sql UPDATE */
#define OPFLAG_APPEND    8    /* This is likely to be an append */

/*
 * Each trigger present in the database schema is stored as an instance of
 * struct Trigger. 
 *
 * Pointers to instances of struct Trigger are stored in two ways.
 * 1. In the "trigHash" hash table (part of the sqlite3* that represents the 
 *    database). This allows Trigger structures to be retrieved by name.
 * 2. All triggers associated with a single table form a linked list, using the
 *    pNext member of struct Trigger. A pointer to the first element of the
 *    linked list is stored as the "pTrigger" member of the associated
 *    struct Table.
 *
 * The "step_list" member points to the first element of a linked list
 * containing the SQL statements specified as the trigger program.
 */
struct Trigger {
  char *name;             /* The name of the trigger                        */
  char *table;            /* The table or view to which the trigger applies */
  u8 op;                  /* One of TK_DELETE, TK_UPDATE, TK_INSERT         */
  u8 tr_tm;               /* One of TRIGGER_BEFORE, TRIGGER_AFTER */
  Expr *pWhen;            /* The WHEN clause of the expresion (may be NULL) */
  IdList *pColumns;       /* If this is an UPDATE OF <column-list> trigger,
                             the <column-list> is stored here */
  Token nameToken;        /* Token containing zName. Use during parsing only */
  Schema *pSchema;        /* Schema containing the trigger */
  Schema *pTabSchema;     /* Schema containing the table */
  TriggerStep *step_list; /* Link list of trigger program steps             */
  Trigger *pNext;         /* Next trigger associated with the table */
};

/*
** A trigger is either a BEFORE or an AFTER trigger.  The following constants
** determine which. 
**
** If there are multiple triggers, you might of some BEFORE and some AFTER.
** In that cases, the constants below can be ORed together.
*/
#define TRIGGER_BEFORE  1
#define TRIGGER_AFTER   2

/*
 * An instance of struct TriggerStep is used to store a single SQL statement
 * that is a part of a trigger-program. 
 *
 * Instances of struct TriggerStep are stored in a singly linked list (linked
 * using the "pNext" member) referenced by the "step_list" member of the 
 * associated struct Trigger instance. The first element of the linked list is
 * the first step of the trigger-program.
 * 
 * The "op" member indicates whether this is a "DELETE", "INSERT", "UPDATE" or
 * "SELECT" statement. The meanings of the other members is determined by the 
 * value of "op" as follows:
 *
 * (op == TK_INSERT)
 * orconf    -> stores the ON CONFLICT algorithm
 * pSelect   -> If this is an INSERT INTO ... SELECT ... statement, then
 *              this stores a pointer to the SELECT statement. Otherwise NULL.
 * target    -> A token holding the name of the table to insert into.
 * pExprList -> If this is an INSERT INTO ... VALUES ... statement, then
 *              this stores values to be inserted. Otherwise NULL.
 * pIdList   -> If this is an INSERT INTO ... (<column-names>) VALUES ... 
 *              statement, then this stores the column-names to be
 *              inserted into.
 *
 * (op == TK_DELETE)
 * target    -> A token holding the name of the table to delete from.
 * pWhere    -> The WHERE clause of the DELETE statement if one is specified.
 *              Otherwise NULL.
 * 
 * (op == TK_UPDATE)
 * target    -> A token holding the name of the table to update rows of.
 * pWhere    -> The WHERE clause of the UPDATE statement if one is specified.
 *              Otherwise NULL.
 * pExprList -> A list of the columns to update and the expressions to update
 *              them to. See sqlite3Update() documentation of "pChanges"
 *              argument.
 * 
 */
struct TriggerStep {
  int op;              /* One of TK_DELETE, TK_UPDATE, TK_INSERT, TK_SELECT */
  int orconf;          /* OE_Rollback etc. */
  Trigger *pTrig;      /* The trigger that this step is a part of */

  Select *pSelect;     /* Valid for SELECT and sometimes 
			  INSERT steps (when pExprList == 0) */
  Token target;        /* Valid for DELETE, UPDATE, INSERT steps */
  Expr *pWhere;        /* Valid for DELETE, UPDATE steps */
  ExprList *pExprList; /* Valid for UPDATE statements and sometimes 
			   INSERT steps (when pSelect == 0)         */
  IdList *pIdList;     /* Valid for INSERT statements only */
  TriggerStep *pNext;  /* Next in the link-list */
  TriggerStep *pLast;  /* Last element in link-list. Valid for 1st elem only */
};

/*
 * An instance of struct TriggerStack stores information required during code
 * generation of a single trigger program. While the trigger program is being
 * coded, its associated TriggerStack instance is pointed to by the
 * "pTriggerStack" member of the Parse structure.
 *
 * The pTab member points to the table that triggers are being coded on. The 
 * newIdx member contains the index of the vdbe cursor that points at the temp
 * table that stores the new.* references. If new.* references are not valid
 * for the trigger being coded (for example an ON DELETE trigger), then newIdx
 * is set to -1. The oldIdx member is analogous to newIdx, for old.* references.
 *
 * The ON CONFLICT policy to be used for the trigger program steps is stored 
 * as the orconf member. If this is OE_Default, then the ON CONFLICT clause 
 * specified for individual triggers steps is used.
 *
 * struct TriggerStack has a "pNext" member, to allow linked lists to be
 * constructed. When coding nested triggers (triggers fired by other triggers)
 * each nested trigger stores its parent trigger's TriggerStack as the "pNext" 
 * pointer. Once the nested trigger has been coded, the pNext value is restored
 * to the pTriggerStack member of the Parse stucture and coding of the parent
 * trigger continues.
 *
 * Before a nested trigger is coded, the linked list pointed to by the 
 * pTriggerStack is scanned to ensure that the trigger is not about to be coded
 * recursively. If this condition is detected, the nested trigger is not coded.
 */
struct TriggerStack {
  Table *pTab;         /* Table that triggers are currently being coded on */
  int newIdx;          /* Index of vdbe cursor to "new" temp table */
  int oldIdx;          /* Index of vdbe cursor to "old" temp table */
  int orconf;          /* Current orconf policy */
  int ignoreJump;      /* where to jump to for a RAISE(IGNORE) */
  Trigger *pTrigger;   /* The trigger currently being coded */
  TriggerStack *pNext; /* Next trigger down on the trigger stack */
};

/*
** The following structure contains information used by the sqliteFix...
** routines as they walk the parse tree to make database references
** explicit.  
*/
typedef struct DbFixer DbFixer;
struct DbFixer {
  Parse *pParse;      /* The parsing context.  Error messages written here */
  const char *zDb;    /* Make sure all objects are contained in this database */
  const char *zType;  /* Type of the container - used for error messages */
  const Token *pName; /* Name of the container - used for error messages */
};

/*
** A pointer to this structure is used to communicate information
** from sqlite3Init and OP_ParseSchema into the sqlite3InitCallback.
*/
typedef struct {
  sqlite3 *db;        /* The database being initialized */
  int iDb;            /* 0 for main database.  1 for TEMP, 2.. for ATTACHed */
  char **pzErrMsg;    /* Error message stored here */
  int rc;             /* Result code stored here */
} InitData;

/*
 * This global flag is set for performance testing of triggers. When it is set
 * SQLite will perform the overhead of building new and old trigger references 
 * even when no triggers exist
 */
extern int sqlite3_always_code_trigger_setup;

/*
** A lookup table used by the SQLITE_READ_UTF8 macro.  The definition
** is in utf.c.
*/
extern const unsigned char sqlite3UtfTrans1[];

/*
** Macros for reading UTF8 characters.
**
** SQLITE_READ_UTF8(x,c) reads a single UTF8 value out of x and writes
** that value into c.  The type of x must be unsigned char*.  The type
** of c must be unsigned int.
**
** SQLITE_SKIP_UTF8(x) advances x forward by one character.  The type of
** x must be unsigned char*.
**
** Notes On Invalid UTF-8:
**
**  *  These macros never allow a 7-bit character (0x00 through 0x7f) to
**     be encoded as a multi-byte character.  Any multi-byte character that
**     attempts to encode a value between 0x00 and 0x7f is rendered as 0xfffd.
**
**  *  These macros never allow a UTF16 surrogate value to be encoded.
**     If a multi-byte character attempts to encode a value between
**     0xd800 and 0xe000 then it is rendered as 0xfffd.
**
**  *  Bytes in the range of 0x80 through 0xbf which occur as the first
**     byte of a character are interpreted as single-byte characters
**     and rendered as themselves even though they are technically
**     invalid characters.
**
**  *  These routines accept an infinite number of different UTF8 encodings
**     for unicode values 0x80 and greater.  They do not change over-length
**     encodings to 0xfffd as some systems recommend.
** 
*/
#define SQLITE_READ_UTF8(zIn, c) {                     \
  c = *(zIn++);                                        \
  if( c>=0xc0 ){                                       \
    c = sqlite3UtfTrans1[c-0xc0];                      \
    while( (*zIn & 0xc0)==0x80 ){                      \
      c = (c<<6) + (0x3f & *(zIn++));                  \
    }                                                  \
    if( c<0x80                                         \
        || (c&0xFFFFF800)==0xD800                      \
        || (c&0xFFFFFFFE)==0xFFFE ){  c = 0xFFFD; }    \
  }                                                    \
}
#define SQLITE_SKIP_UTF8(zIn) {                        \
  if( (*(zIn++))>=0xc0 ){                              \
    while( (*zIn & 0xc0)==0x80 ){ zIn++; }             \
  }                                                    \
}




/*
** The SQLITE_CORRUPT_BKPT macro can be either a constant (for production
** builds) or a function call (for debugging).  If it is a function call,
** it allows the operator to set a breakpoint at the spot where database
** corruption is first detected.
*/
#ifdef SQLITE_DEBUG
SQLITE_PRIVATE   int sqlite3Corrupt(void);
# define SQLITE_CORRUPT_BKPT sqlite3Corrupt()
#else
# define SQLITE_CORRUPT_BKPT SQLITE_CORRUPT
#endif

/*
** Internal function prototypes
*/
SQLITE_PRIVATE int sqlite3StrICmp(const char *, const char *);
SQLITE_PRIVATE int sqlite3StrNICmp(const char *, const char *, int);
SQLITE_PRIVATE int sqlite3IsNumber(const char*, int*, u8);

SQLITE_PRIVATE void *sqlite3Malloc(int,int);
SQLITE_PRIVATE void *sqlite3MallocRaw(int,int);
SQLITE_PRIVATE void *sqlite3Realloc(void*,int);
SQLITE_PRIVATE char *sqlite3StrDup(const char*);
SQLITE_PRIVATE char *sqlite3StrNDup(const char*, int);
# define sqlite3CheckMemory(a,b)
SQLITE_PRIVATE void *sqlite3ReallocOrFree(void*,int);
SQLITE_PRIVATE void sqlite3FreeX(void*);
SQLITE_PRIVATE void *sqlite3MallocX(int);
#ifdef SQLITE_ENABLE_MEMORY_MANAGEMENT
SQLITE_PRIVATE   int sqlite3AllocSize(void *);
#endif

SQLITE_PRIVATE char *sqlite3MPrintf(const char*, ...);
SQLITE_PRIVATE char *sqlite3VMPrintf(const char*, va_list);
#if defined(SQLITE_TEST) || defined(SQLITE_DEBUG)
SQLITE_PRIVATE   void sqlite3DebugPrintf(const char*, ...);
SQLITE_PRIVATE   void *sqlite3TextToPtr(const char*);
#endif
SQLITE_PRIVATE void sqlite3SetString(char **, ...);
SQLITE_PRIVATE void sqlite3ErrorMsg(Parse*, const char*, ...);
SQLITE_PRIVATE void sqlite3ErrorClear(Parse*);
SQLITE_PRIVATE void sqlite3Dequote(char*);
SQLITE_PRIVATE void sqlite3DequoteExpr(Expr*);
SQLITE_PRIVATE int sqlite3KeywordCode(const unsigned char*, int);
SQLITE_PRIVATE int sqlite3RunParser(Parse*, const char*, char **);
SQLITE_PRIVATE void sqlite3FinishCoding(Parse*);
SQLITE_PRIVATE Expr *sqlite3Expr(int, Expr*, Expr*, const Token*);
SQLITE_PRIVATE Expr *sqlite3ExprOrFree(int, Expr*, Expr*, const Token*);
SQLITE_PRIVATE Expr *sqlite3RegisterExpr(Parse*,Token*);
SQLITE_PRIVATE Expr *sqlite3ExprAnd(Expr*, Expr*);
SQLITE_PRIVATE void sqlite3ExprSpan(Expr*,Token*,Token*);
SQLITE_PRIVATE Expr *sqlite3ExprFunction(ExprList*, Token*);
SQLITE_PRIVATE void sqlite3ExprAssignVarNumber(Parse*, Expr*);
SQLITE_PRIVATE void sqlite3ExprDelete(Expr*);
SQLITE_PRIVATE ExprList *sqlite3ExprListAppend(ExprList*,Expr*,Token*);
SQLITE_PRIVATE void sqlite3ExprListDelete(ExprList*);
SQLITE_PRIVATE int sqlite3Init(sqlite3*, char**);
SQLITE_PRIVATE int sqlite3InitCallback(void*, int, char**, char**);
SQLITE_PRIVATE void sqlite3Pragma(Parse*,Token*,Token*,Token*,int);
SQLITE_PRIVATE void sqlite3ResetInternalSchema(sqlite3*, int);
SQLITE_PRIVATE void sqlite3BeginParse(Parse*,int);
SQLITE_PRIVATE void sqlite3CommitInternalChanges(sqlite3*);
SQLITE_PRIVATE Table *sqlite3ResultSetOfSelect(Parse*,char*,Select*);
SQLITE_PRIVATE void sqlite3OpenMasterTable(Parse *, int);
SQLITE_PRIVATE void sqlite3StartTable(Parse*,Token*,Token*,int,int,int,int);
SQLITE_PRIVATE void sqlite3AddColumn(Parse*,Token*);
SQLITE_PRIVATE void sqlite3AddNotNull(Parse*, int);
SQLITE_PRIVATE void sqlite3AddPrimaryKey(Parse*, ExprList*, int, int, int);
SQLITE_PRIVATE void sqlite3AddCheckConstraint(Parse*, Expr*);
SQLITE_PRIVATE void sqlite3AddColumnType(Parse*,Token*);
SQLITE_PRIVATE void sqlite3AddDefaultValue(Parse*,Expr*);
SQLITE_PRIVATE void sqlite3AddCollateType(Parse*, const char*, int);
SQLITE_PRIVATE void sqlite3EndTable(Parse*,Token*,Token*,Select*);

SQLITE_PRIVATE void sqlite3CreateView(Parse*,Token*,Token*,Token*,Select*,int,int);

#if !defined(SQLITE_OMIT_VIEW) || !defined(SQLITE_OMIT_VIRTUALTABLE)
SQLITE_PRIVATE   int sqlite3ViewGetColumnNames(Parse*,Table*);
#else
# define sqlite3ViewGetColumnNames(A,B) 0
#endif

SQLITE_PRIVATE void sqlite3DropTable(Parse*, SrcList*, int, int);
SQLITE_PRIVATE void sqlite3DeleteTable(Table*);
SQLITE_PRIVATE void sqlite3Insert(Parse*, SrcList*, ExprList*, Select*, IdList*, int);
SQLITE_PRIVATE void *sqlite3ArrayAllocate(void*,int,int,int*,int*,int*);
SQLITE_PRIVATE IdList *sqlite3IdListAppend(IdList*, Token*);
SQLITE_PRIVATE int sqlite3IdListIndex(IdList*,const char*);
SQLITE_PRIVATE SrcList *sqlite3SrcListAppend(SrcList*, Token*, Token*);
SQLITE_PRIVATE SrcList *sqlite3SrcListAppendFromTerm(SrcList*, Token*, Token*, Token*,
                                      Select*, Expr*, IdList*);
SQLITE_PRIVATE void sqlite3SrcListShiftJoinType(SrcList*);
SQLITE_PRIVATE void sqlite3SrcListAssignCursors(Parse*, SrcList*);
SQLITE_PRIVATE void sqlite3IdListDelete(IdList*);
SQLITE_PRIVATE void sqlite3SrcListDelete(SrcList*);
SQLITE_PRIVATE void sqlite3CreateIndex(Parse*,Token*,Token*,SrcList*,ExprList*,int,Token*,
                        Token*, int, int);
SQLITE_PRIVATE void sqlite3DropIndex(Parse*, SrcList*, int);
SQLITE_PRIVATE int sqlite3Select(Parse*, Select*, int, int, Select*, int, int*, char *aff);
SQLITE_PRIVATE Select *sqlite3SelectNew(ExprList*,SrcList*,Expr*,ExprList*,Expr*,ExprList*,
                        int,Expr*,Expr*);
SQLITE_PRIVATE void sqlite3SelectDelete(Select*);
SQLITE_PRIVATE Table *sqlite3SrcListLookup(Parse*, SrcList*);
SQLITE_PRIVATE int sqlite3IsReadOnly(Parse*, Table*, int);
SQLITE_PRIVATE void sqlite3OpenTable(Parse*, int iCur, int iDb, Table*, int);
SQLITE_PRIVATE void sqlite3DeleteFrom(Parse*, SrcList*, Expr*);
SQLITE_PRIVATE void sqlite3Update(Parse*, SrcList*, ExprList*, Expr*, int);
SQLITE_PRIVATE WhereInfo *sqlite3WhereBegin(Parse*, SrcList*, Expr*, ExprList**);
SQLITE_PRIVATE void sqlite3WhereEnd(WhereInfo*);
SQLITE_PRIVATE void sqlite3ExprCodeGetColumn(Vdbe*, Table*, int, int);
SQLITE_PRIVATE void sqlite3ExprCode(Parse*, Expr*);
SQLITE_PRIVATE void sqlite3ExprCodeAndCache(Parse*, Expr*);
SQLITE_PRIVATE int sqlite3ExprCodeExprList(Parse*, ExprList*);
SQLITE_PRIVATE void sqlite3ExprIfTrue(Parse*, Expr*, int, int);
SQLITE_PRIVATE void sqlite3ExprIfFalse(Parse*, Expr*, int, int);
SQLITE_PRIVATE Table *sqlite3FindTable(sqlite3*,const char*, const char*);
SQLITE_PRIVATE Table *sqlite3LocateTable(Parse*,const char*, const char*);
SQLITE_PRIVATE Index *sqlite3FindIndex(sqlite3*,const char*, const char*);
SQLITE_PRIVATE void sqlite3UnlinkAndDeleteTable(sqlite3*,int,const char*);
SQLITE_PRIVATE void sqlite3UnlinkAndDeleteIndex(sqlite3*,int,const char*);
SQLITE_PRIVATE void sqlite3Vacuum(Parse*);
SQLITE_PRIVATE int sqlite3RunVacuum(char**, sqlite3*);
SQLITE_PRIVATE char *sqlite3NameFromToken(Token*);
SQLITE_PRIVATE int sqlite3ExprCompare(Expr*, Expr*);
int sqliteFuncId(Token*);
SQLITE_PRIVATE int sqlite3ExprResolveNames(NameContext *, Expr *);
SQLITE_PRIVATE int sqlite3ExprAnalyzeAggregates(NameContext*, Expr*);
SQLITE_PRIVATE int sqlite3ExprAnalyzeAggList(NameContext*,ExprList*);
SQLITE_PRIVATE Vdbe *sqlite3GetVdbe(Parse*);
SQLITE_PRIVATE Expr *sqlite3CreateIdExpr(const char*);
SQLITE_PRIVATE void sqlite3Randomness(int, void*);
SQLITE_PRIVATE void sqlite3RollbackAll(sqlite3*);
SQLITE_PRIVATE void sqlite3CodeVerifySchema(Parse*, int);
SQLITE_PRIVATE void sqlite3BeginTransaction(Parse*, int);
SQLITE_PRIVATE void sqlite3CommitTransaction(Parse*);
SQLITE_PRIVATE void sqlite3RollbackTransaction(Parse*);
SQLITE_PRIVATE int sqlite3ExprIsConstant(Expr*);
SQLITE_PRIVATE int sqlite3ExprIsConstantNotJoin(Expr*);
SQLITE_PRIVATE int sqlite3ExprIsConstantOrFunction(Expr*);
SQLITE_PRIVATE int sqlite3ExprIsInteger(Expr*, int*);
SQLITE_PRIVATE int sqlite3IsRowid(const char*);
SQLITE_PRIVATE void sqlite3GenerateRowDelete(sqlite3*, Vdbe*, Table*, int, int);
SQLITE_PRIVATE void sqlite3GenerateRowIndexDelete(Vdbe*, Table*, int, char*);
SQLITE_PRIVATE void sqlite3GenerateIndexKey(Vdbe*, Index*, int);
SQLITE_PRIVATE void sqlite3GenerateConstraintChecks(Parse*,Table*,int,char*,int,int,int,int);
SQLITE_PRIVATE void sqlite3CompleteInsertion(Parse*, Table*, int, char*, int, int, int, int);
SQLITE_PRIVATE void sqlite3OpenTableAndIndices(Parse*, Table*, int, int);
SQLITE_PRIVATE void sqlite3BeginWriteOperation(Parse*, int, int);
SQLITE_PRIVATE Expr *sqlite3ExprDup(Expr*);
SQLITE_PRIVATE void sqlite3TokenCopy(Token*, Token*);
SQLITE_PRIVATE ExprList *sqlite3ExprListDup(ExprList*);
SQLITE_PRIVATE SrcList *sqlite3SrcListDup(SrcList*);
SQLITE_PRIVATE IdList *sqlite3IdListDup(IdList*);
SQLITE_PRIVATE Select *sqlite3SelectDup(Select*);
SQLITE_PRIVATE FuncDef *sqlite3FindFunction(sqlite3*,const char*,int,int,u8,int);
SQLITE_PRIVATE void sqlite3RegisterBuiltinFunctions(sqlite3*);
SQLITE_PRIVATE void sqlite3RegisterDateTimeFunctions(sqlite3*);
SQLITE_PRIVATE int sqlite3SafetyOn(sqlite3*);
SQLITE_PRIVATE int sqlite3SafetyOff(sqlite3*);
SQLITE_PRIVATE int sqlite3SafetyCheck(sqlite3*);
SQLITE_PRIVATE void sqlite3ChangeCookie(sqlite3*, Vdbe*, int);

#ifndef SQLITE_OMIT_TRIGGER
SQLITE_PRIVATE   void sqlite3BeginTrigger(Parse*, Token*,Token*,int,int,IdList*,SrcList*,
                           Expr*,int, int);
SQLITE_PRIVATE   void sqlite3FinishTrigger(Parse*, TriggerStep*, Token*);
SQLITE_PRIVATE   void sqlite3DropTrigger(Parse*, SrcList*, int);
SQLITE_PRIVATE   void sqlite3DropTriggerPtr(Parse*, Trigger*);
SQLITE_PRIVATE   int sqlite3TriggersExist(Parse*, Table*, int, ExprList*);
SQLITE_PRIVATE   int sqlite3CodeRowTrigger(Parse*, int, ExprList*, int, Table *, int, int, 
                           int, int);
  void sqliteViewTriggers(Parse*, Table*, Expr*, int, ExprList*);
SQLITE_PRIVATE   void sqlite3DeleteTriggerStep(TriggerStep*);
SQLITE_PRIVATE   TriggerStep *sqlite3TriggerSelectStep(Select*);
SQLITE_PRIVATE   TriggerStep *sqlite3TriggerInsertStep(Token*, IdList*, ExprList*,Select*,int);
SQLITE_PRIVATE   TriggerStep *sqlite3TriggerUpdateStep(Token*, ExprList*, Expr*, int);
SQLITE_PRIVATE   TriggerStep *sqlite3TriggerDeleteStep(Token*, Expr*);
SQLITE_PRIVATE   void sqlite3DeleteTrigger(Trigger*);
SQLITE_PRIVATE   void sqlite3UnlinkAndDeleteTrigger(sqlite3*,int,const char*);
#else
# define sqlite3TriggersExist(A,B,C,D,E,F) 0
# define sqlite3DeleteTrigger(A)
# define sqlite3DropTriggerPtr(A,B)
# define sqlite3UnlinkAndDeleteTrigger(A,B,C)
# define sqlite3CodeRowTrigger(A,B,C,D,E,F,G,H,I) 0
#endif

SQLITE_PRIVATE int sqlite3JoinType(Parse*, Token*, Token*, Token*);
SQLITE_PRIVATE void sqlite3CreateForeignKey(Parse*, ExprList*, Token*, ExprList*, int);
SQLITE_PRIVATE void sqlite3DeferForeignKey(Parse*, int);
#ifndef SQLITE_OMIT_AUTHORIZATION
SQLITE_PRIVATE   void sqlite3AuthRead(Parse*,Expr*,SrcList*);
SQLITE_PRIVATE   int sqlite3AuthCheck(Parse*,int, const char*, const char*, const char*);
SQLITE_PRIVATE   void sqlite3AuthContextPush(Parse*, AuthContext*, const char*);
SQLITE_PRIVATE   void sqlite3AuthContextPop(AuthContext*);
#else
# define sqlite3AuthRead(a,b,c)
# define sqlite3AuthCheck(a,b,c,d,e)    SQLITE_OK
# define sqlite3AuthContextPush(a,b,c)
# define sqlite3AuthContextPop(a)  ((void)(a))
#endif
SQLITE_PRIVATE void sqlite3Attach(Parse*, Expr*, Expr*, Expr*);
SQLITE_PRIVATE void sqlite3Detach(Parse*, Expr*);
SQLITE_PRIVATE int sqlite3BtreeFactory(const sqlite3 *db, const char *zFilename,
                       int omitJournal, int nCache, Btree **ppBtree);
SQLITE_PRIVATE int sqlite3FixInit(DbFixer*, Parse*, int, const char*, const Token*);
SQLITE_PRIVATE int sqlite3FixSrcList(DbFixer*, SrcList*);
SQLITE_PRIVATE int sqlite3FixSelect(DbFixer*, Select*);
SQLITE_PRIVATE int sqlite3FixExpr(DbFixer*, Expr*);
SQLITE_PRIVATE int sqlite3FixExprList(DbFixer*, ExprList*);
SQLITE_PRIVATE int sqlite3FixTriggerStep(DbFixer*, TriggerStep*);
SQLITE_PRIVATE int sqlite3AtoF(const char *z, double*);
SQLITE_API char *sqlite3_snprintf(int,char*,const char*,...);
SQLITE_PRIVATE int sqlite3GetInt32(const char *, int*);
SQLITE_PRIVATE int sqlite3FitsIn64Bits(const char *);
SQLITE_PRIVATE int sqlite3Utf16ByteLen(const void *pData, int nChar);
SQLITE_PRIVATE int sqlite3Utf8CharLen(const char *pData, int nByte);
SQLITE_PRIVATE u32 sqlite3ReadUtf8(const unsigned char *);
SQLITE_PRIVATE int sqlite3PutVarint(unsigned char *, u64);
SQLITE_PRIVATE int sqlite3GetVarint(const unsigned char *, u64 *);
SQLITE_PRIVATE int sqlite3GetVarint32(const unsigned char *, u32 *);
SQLITE_PRIVATE int sqlite3VarintLen(u64 v);
SQLITE_PRIVATE void sqlite3IndexAffinityStr(Vdbe *, Index *);
SQLITE_PRIVATE void sqlite3TableAffinityStr(Vdbe *, Table *);
SQLITE_PRIVATE char sqlite3CompareAffinity(Expr *pExpr, char aff2);
SQLITE_PRIVATE int sqlite3IndexAffinityOk(Expr *pExpr, char idx_affinity);
SQLITE_PRIVATE char sqlite3ExprAffinity(Expr *pExpr);
SQLITE_PRIVATE int sqlite3Atoi64(const char*, i64*);
SQLITE_PRIVATE void sqlite3Error(sqlite3*, int, const char*,...);
SQLITE_PRIVATE void *sqlite3HexToBlob(const char *z);
SQLITE_PRIVATE int sqlite3TwoPartName(Parse *, Token *, Token *, Token **);
SQLITE_PRIVATE const char *sqlite3ErrStr(int);
SQLITE_PRIVATE int sqlite3ReadSchema(Parse *pParse);
SQLITE_PRIVATE CollSeq *sqlite3FindCollSeq(sqlite3*,u8 enc, const char *,int,int);
SQLITE_PRIVATE CollSeq *sqlite3LocateCollSeq(Parse *pParse, const char *zName, int nName);
SQLITE_PRIVATE CollSeq *sqlite3ExprCollSeq(Parse *pParse, Expr *pExpr);
SQLITE_PRIVATE Expr *sqlite3ExprSetColl(Parse *pParse, Expr *, Token *);
SQLITE_PRIVATE int sqlite3CheckCollSeq(Parse *, CollSeq *);
SQLITE_PRIVATE int sqlite3CheckObjectName(Parse *, const char *);
SQLITE_PRIVATE void sqlite3VdbeSetChanges(sqlite3 *, int);
SQLITE_PRIVATE void sqlite3Utf16Substr(sqlite3_context *,int,sqlite3_value **);

SQLITE_PRIVATE const void *sqlite3ValueText(sqlite3_value*, u8);
SQLITE_PRIVATE int sqlite3ValueBytes(sqlite3_value*, u8);
SQLITE_PRIVATE void sqlite3ValueSetStr(sqlite3_value*, int, const void *,u8, void(*)(void*));
SQLITE_PRIVATE void sqlite3ValueFree(sqlite3_value*);
SQLITE_PRIVATE sqlite3_value *sqlite3ValueNew(void);
SQLITE_PRIVATE char *sqlite3Utf16to8(const void*, int);
SQLITE_PRIVATE int sqlite3ValueFromExpr(Expr *, u8, u8, sqlite3_value **);
SQLITE_PRIVATE void sqlite3ValueApplyAffinity(sqlite3_value *, u8, u8);
extern const unsigned char sqlite3UpperToLower[];
SQLITE_PRIVATE void sqlite3RootPageMoved(Db*, int, int);
SQLITE_PRIVATE void sqlite3Reindex(Parse*, Token*, Token*);
SQLITE_PRIVATE void sqlite3AlterFunctions(sqlite3*);
SQLITE_PRIVATE void sqlite3AlterRenameTable(Parse*, SrcList*, Token*);
SQLITE_PRIVATE int sqlite3GetToken(const unsigned char *, int *);
SQLITE_PRIVATE void sqlite3NestedParse(Parse*, const char*, ...);
SQLITE_PRIVATE void sqlite3ExpirePreparedStatements(sqlite3*);
SQLITE_PRIVATE void sqlite3CodeSubselect(Parse *, Expr *);
SQLITE_PRIVATE int sqlite3SelectResolve(Parse *, Select *, NameContext *);
SQLITE_PRIVATE void sqlite3ColumnDefault(Vdbe *, Table *, int);
SQLITE_PRIVATE void sqlite3AlterFinishAddColumn(Parse *, Token *);
SQLITE_PRIVATE void sqlite3AlterBeginAddColumn(Parse *, SrcList *);
SQLITE_PRIVATE CollSeq *sqlite3GetCollSeq(sqlite3*, CollSeq *, const char *, int);
SQLITE_PRIVATE char sqlite3AffinityType(const Token*);
SQLITE_PRIVATE void sqlite3Analyze(Parse*, Token*, Token*);
SQLITE_PRIVATE int sqlite3InvokeBusyHandler(BusyHandler*);
SQLITE_PRIVATE int sqlite3FindDb(sqlite3*, Token*);
SQLITE_PRIVATE int sqlite3AnalysisLoad(sqlite3*,int iDB);
SQLITE_PRIVATE void sqlite3DefaultRowEst(Index*);
SQLITE_PRIVATE void sqlite3RegisterLikeFunctions(sqlite3*, int);
SQLITE_PRIVATE int sqlite3IsLikeFunction(sqlite3*,Expr*,int*,char*);
SQLITE_PRIVATE ThreadData *sqlite3ThreadData(void);
SQLITE_PRIVATE const ThreadData *sqlite3ThreadDataReadOnly(void);
SQLITE_PRIVATE void sqlite3ReleaseThreadData(void);
SQLITE_PRIVATE void sqlite3AttachFunctions(sqlite3 *);
SQLITE_PRIVATE void sqlite3MinimumFileFormat(Parse*, int, int);
SQLITE_PRIVATE void sqlite3SchemaFree(void *);
SQLITE_PRIVATE Schema *sqlite3SchemaGet(Btree *);
SQLITE_PRIVATE int sqlite3SchemaToIndex(sqlite3 *db, Schema *);
SQLITE_PRIVATE KeyInfo *sqlite3IndexKeyinfo(Parse *, Index *);
SQLITE_PRIVATE int sqlite3CreateFunc(sqlite3 *, const char *, int, int, void *, 
  void (*)(sqlite3_context*,int,sqlite3_value **),
  void (*)(sqlite3_context*,int,sqlite3_value **), void (*)(sqlite3_context*));
SQLITE_PRIVATE int sqlite3ApiExit(sqlite3 *db, int);
SQLITE_PRIVATE void sqlite3FailedMalloc(void);
SQLITE_PRIVATE void sqlite3AbortOtherActiveVdbes(sqlite3 *, Vdbe *);
SQLITE_PRIVATE int sqlite3OpenTempDatabase(Parse *);

#ifndef SQLITE_OMIT_LOAD_EXTENSION
SQLITE_PRIVATE   void sqlite3CloseExtensions(sqlite3*);
SQLITE_PRIVATE   int sqlite3AutoLoadExtensions(sqlite3*);
#else
# define sqlite3CloseExtensions(X)
# define sqlite3AutoLoadExtensions(X)  SQLITE_OK
#endif

#ifndef SQLITE_OMIT_SHARED_CACHE
SQLITE_PRIVATE   void sqlite3TableLock(Parse *, int, int, u8, const char *);
#else
  #define sqlite3TableLock(v,w,x,y,z)
#endif

#ifdef SQLITE_TEST
SQLITE_PRIVATE   int sqlite3Utf8To8(unsigned char*);
#endif

#ifdef SQLITE_MEMDEBUG
SQLITE_PRIVATE   void sqlite3MallocDisallow(void);
SQLITE_PRIVATE   void sqlite3MallocAllow(void);
SQLITE_PRIVATE   int sqlite3TestMallocFail(void);
#else
  #define sqlite3TestMallocFail() 0
  #define sqlite3MallocDisallow()
  #define sqlite3MallocAllow()
#endif

#ifdef SQLITE_ENABLE_MEMORY_MANAGEMENT
SQLITE_PRIVATE   void *sqlite3ThreadSafeMalloc(int);
SQLITE_PRIVATE   void sqlite3ThreadSafeFree(void *);
#else
  #define sqlite3ThreadSafeMalloc sqlite3MallocX
  #define sqlite3ThreadSafeFree sqlite3FreeX
#endif

#ifdef SQLITE_OMIT_VIRTUALTABLE
#  define sqlite3VtabClear(X)
#  define sqlite3VtabSync(X,Y) (Y)
#  define sqlite3VtabRollback(X)
#  define sqlite3VtabCommit(X)
#else
SQLITE_PRIVATE    void sqlite3VtabClear(Table*);
SQLITE_PRIVATE    int sqlite3VtabSync(sqlite3 *db, int rc);
SQLITE_PRIVATE    int sqlite3VtabRollback(sqlite3 *db);
SQLITE_PRIVATE    int sqlite3VtabCommit(sqlite3 *db);
#endif
SQLITE_PRIVATE void sqlite3VtabLock(sqlite3_vtab*);
SQLITE_PRIVATE void sqlite3VtabUnlock(sqlite3*, sqlite3_vtab*);
SQLITE_PRIVATE void sqlite3VtabBeginParse(Parse*, Token*, Token*, Token*);
SQLITE_PRIVATE void sqlite3VtabFinishParse(Parse*, Token*);
SQLITE_PRIVATE void sqlite3VtabArgInit(Parse*);
SQLITE_PRIVATE void sqlite3VtabArgExtend(Parse*, Token*);
SQLITE_PRIVATE int sqlite3VtabCallCreate(sqlite3*, int, const char *, char **);
SQLITE_PRIVATE int sqlite3VtabCallConnect(Parse*, Table*);
SQLITE_PRIVATE int sqlite3VtabCallDestroy(sqlite3*, int, const char *);
SQLITE_PRIVATE int sqlite3VtabBegin(sqlite3 *, sqlite3_vtab *);
SQLITE_PRIVATE FuncDef *sqlite3VtabOverloadFunction(FuncDef*, int nArg, Expr*);
SQLITE_PRIVATE void sqlite3InvalidFunction(sqlite3_context*,int,sqlite3_value**);
SQLITE_PRIVATE int sqlite3Reprepare(Vdbe*);
SQLITE_PRIVATE void sqlite3ExprListCheckLength(Parse*, ExprList*, int, const char*);
CollSeq* sqlite3BinaryCompareCollSeq(Parse *, Expr *, Expr *);

#if SQLITE_MAX_EXPR_DEPTH>0
SQLITE_PRIVATE   void sqlite3ExprSetHeight(Expr *);
SQLITE_PRIVATE   int sqlite3SelectExprHeight(Select *);
#else
  #define sqlite3ExprSetHeight(x)
#endif

SQLITE_PRIVATE u32 sqlite3Get2byte(const u8*);
SQLITE_PRIVATE u32 sqlite3Get4byte(const u8*);
SQLITE_PRIVATE void sqlite3Put2byte(u8*, u32);
SQLITE_PRIVATE void sqlite3Put4byte(u8*, u32);

#ifdef SQLITE_SSE
#include "sseInt.h"
#endif

#ifdef SQLITE_DEBUG
SQLITE_PRIVATE   void sqlite3ParserTrace(FILE*, char *);
#endif

/*
** If the SQLITE_ENABLE IOTRACE exists then the global variable
** sqlite3_io_trace is a pointer to a printf-like routine used to
** print I/O tracing messages. 
*/
#ifdef SQLITE_ENABLE_IOTRACE
# define IOTRACE(A)  if( sqlite3_io_trace ){ sqlite3_io_trace A; }
SQLITE_PRIVATE   void sqlite3VdbeIOTraceSql(Vdbe*);
#else
# define IOTRACE(A)
# define sqlite3VdbeIOTraceSql(X)
#endif
extern void (*sqlite3_io_trace)(const char*,...);

#endif

/************** End of sqliteInt.h *******************************************/
/************** Continuing where we left off in date.c ***********************/
#include <ctype.h>
#include <time.h>

#ifndef SQLITE_OMIT_DATETIME_FUNCS

/*
** A structure for holding a single date and time.
*/
typedef struct DateTime DateTime;
struct DateTime {
  double rJD;      /* The julian day number */
  int Y, M, D;     /* Year, month, and day */
  int h, m;        /* Hour and minutes */
  int tz;          /* Timezone offset in minutes */
  double s;        /* Seconds */
  char validYMD;   /* True if Y,M,D are valid */
  char validHMS;   /* True if h,m,s are valid */
  char validJD;    /* True if rJD is valid */
  char validTZ;    /* True if tz is valid */
};


/*
** Convert zDate into one or more integers.  Additional arguments
** come in groups of 5 as follows:
**
**       N       number of digits in the integer
**       min     minimum allowed value of the integer
**       max     maximum allowed value of the integer
**       nextC   first character after the integer
**       pVal    where to write the integers value.
**
** Conversions continue until one with nextC==0 is encountered.
** The function returns the number of successful conversions.
*/
static int getDigits(const char *zDate, ...){
  va_list ap;
  int val;
  int N;
  int min;
  int max;
  int nextC;
  int *pVal;
  int cnt = 0;
  va_start(ap, zDate);
  do{
    N = va_arg(ap, int);
    min = va_arg(ap, int);
    max = va_arg(ap, int);
    nextC = va_arg(ap, int);
    pVal = va_arg(ap, int*);
    val = 0;
    while( N-- ){
      if( !isdigit(*(u8*)zDate) ){
        goto end_getDigits;
      }
      val = val*10 + *zDate - '0';
      zDate++;
    }
    if( val<min || val>max || (nextC!=0 && nextC!=*zDate) ){
      goto end_getDigits;
    }
    *pVal = val;
    zDate++;
    cnt++;
  }while( nextC );
end_getDigits:
  va_end(ap);
  return cnt;
}

/*
** Read text from z[] and convert into a floating point number.  Return
** the number of digits converted.
*/
#define getValue sqlite3AtoF

/*
** Parse a timezone extension on the end of a date-time.
** The extension is of the form:
**
**        (+/-)HH:MM
**
** If the parse is successful, write the number of minutes
** of change in *pnMin and return 0.  If a parser error occurs,
** return 0.
**
** A missing specifier is not considered an error.
*/
static int parseTimezone(const char *zDate, DateTime *p){
  int sgn = 0;
  int nHr, nMn;
  while( isspace(*(u8*)zDate) ){ zDate++; }
  p->tz = 0;
  if( *zDate=='-' ){
    sgn = -1;
  }else if( *zDate=='+' ){
    sgn = +1;
  }else{
    return *zDate!=0;
  }
  zDate++;
  if( getDigits(zDate, 2, 0, 14, ':', &nHr, 2, 0, 59, 0, &nMn)!=2 ){
    return 1;
  }
  zDate += 5;
  p->tz = sgn*(nMn + nHr*60);
  while( isspace(*(u8*)zDate) ){ zDate++; }
  return *zDate!=0;
}

/*
** Parse times of the form HH:MM or HH:MM:SS or HH:MM:SS.FFFF.
** The HH, MM, and SS must each be exactly 2 digits.  The
** fractional seconds FFFF can be one or more digits.
**
** Return 1 if there is a parsing error and 0 on success.
*/
static int parseHhMmSs(const char *zDate, DateTime *p){
  int h, m, s;
  double ms = 0.0;
  if( getDigits(zDate, 2, 0, 24, ':', &h, 2, 0, 59, 0, &m)!=2 ){
    return 1;
  }
  zDate += 5;
  if( *zDate==':' ){
    zDate++;
    if( getDigits(zDate, 2, 0, 59, 0, &s)!=1 ){
      return 1;
    }
    zDate += 2;
    if( *zDate=='.' && isdigit((u8)zDate[1]) ){
      double rScale = 1.0;
      zDate++;
      while( isdigit(*(u8*)zDate) ){
        ms = ms*10.0 + *zDate - '0';
        rScale *= 10.0;
        zDate++;
      }
      ms /= rScale;
    }
  }else{
    s = 0;
  }
  p->validJD = 0;
  p->validHMS = 1;
  p->h = h;
  p->m = m;
  p->s = s + ms;
  if( parseTimezone(zDate, p) ) return 1;
  p->validTZ = p->tz!=0;
  return 0;
}

/*
** Convert from YYYY-MM-DD HH:MM:SS to julian day.  We always assume
** that the YYYY-MM-DD is according to the Gregorian calendar.
**
** Reference:  Meeus page 61
*/
static void computeJD(DateTime *p){
  int Y, M, D, A, B, X1, X2;

  if( p->validJD ) return;
  if( p->validYMD ){
    Y = p->Y;
    M = p->M;
    D = p->D;
  }else{
    Y = 2000;  /* If no YMD specified, assume 2000-Jan-01 */
    M = 1;
    D = 1;
  }
  if( M<=2 ){
    Y--;
    M += 12;
  }
  A = Y/100;
  B = 2 - A + (A/4);
  X1 = 365.25*(Y+4716);
  X2 = 30.6001*(M+1);
  p->rJD = X1 + X2 + D + B - 1524.5;
  p->validJD = 1;
  if( p->validHMS ){
    p->rJD += (p->h*3600.0 + p->m*60.0 + p->s)/86400.0;
    if( p->validTZ ){
      p->rJD -= p->tz*60/86400.0;
      p->validYMD = 0;
      p->validHMS = 0;
      p->validTZ = 0;
    }
  }
}

/*
** Parse dates of the form
**
**     YYYY-MM-DD HH:MM:SS.FFF
**     YYYY-MM-DD HH:MM:SS
**     YYYY-MM-DD HH:MM
**     YYYY-MM-DD
**
** Write the result into the DateTime structure and return 0
** on success and 1 if the input string is not a well-formed
** date.
*/
static int parseYyyyMmDd(const char *zDate, DateTime *p){
  int Y, M, D, neg;

  if( zDate[0]=='-' ){
    zDate++;
    neg = 1;
  }else{
    neg = 0;
  }
  if( getDigits(zDate,4,0,9999,'-',&Y,2,1,12,'-',&M,2,1,31,0,&D)!=3 ){
    return 1;
  }
  zDate += 10;
  while( isspace(*(u8*)zDate) || 'T'==*(u8*)zDate ){ zDate++; }
  if( parseHhMmSs(zDate, p)==0 ){
    /* We got the time */
  }else if( *zDate==0 ){
    p->validHMS = 0;
  }else{
    return 1;
  }
  p->validJD = 0;
  p->validYMD = 1;
  p->Y = neg ? -Y : Y;
  p->M = M;
  p->D = D;
  if( p->validTZ ){
    computeJD(p);
  }
  return 0;
}

/*
** Attempt to parse the given string into a Julian Day Number.  Return
** the number of errors.
**
** The following are acceptable forms for the input string:
**
**      YYYY-MM-DD HH:MM:SS.FFF  +/-HH:MM
**      DDDD.DD 
**      now
**
** In the first form, the +/-HH:MM is always optional.  The fractional
** seconds extension (the ".FFF") is optional.  The seconds portion
** (":SS.FFF") is option.  The year and date can be omitted as long
** as there is a time string.  The time string can be omitted as long
** as there is a year and date.
*/
static int parseDateOrTime(const char *zDate, DateTime *p){
  memset(p, 0, sizeof(*p));
  if( parseYyyyMmDd(zDate,p)==0 ){
    return 0;
  }else if( parseHhMmSs(zDate, p)==0 ){
    return 0;
  }else if( sqlite3StrICmp(zDate,"now")==0){
    double r;
    sqlite3OsCurrentTime(&r);
    p->rJD = r;
    p->validJD = 1;
    return 0;
  }else if( sqlite3IsNumber(zDate, 0, SQLITE_UTF8) ){
    getValue(zDate, &p->rJD);
    p->validJD = 1;
    return 0;
  }
  return 1;
}

/*
** Compute the Year, Month, and Day from the julian day number.
*/
static void computeYMD(DateTime *p){
  int Z, A, B, C, D, E, X1;
  if( p->validYMD ) return;
  if( !p->validJD ){
    p->Y = 2000;
    p->M = 1;
    p->D = 1;
  }else{
    Z = p->rJD + 0.5;
    A = (Z - 1867216.25)/36524.25;
    A = Z + 1 + A - (A/4);
    B = A + 1524;
    C = (B - 122.1)/365.25;
    D = 365.25*C;
    E = (B-D)/30.6001;
    X1 = 30.6001*E;
    p->D = B - D - X1;
    p->M = E<14 ? E-1 : E-13;
    p->Y = p->M>2 ? C - 4716 : C - 4715;
  }
  p->validYMD = 1;
}

/*
** Compute the Hour, Minute, and Seconds from the julian day number.
*/
static void computeHMS(DateTime *p){
  int Z, s;
  if( p->validHMS ) return;
  computeJD(p);
  Z = p->rJD + 0.5;
  s = (p->rJD + 0.5 - Z)*86400000.0 + 0.5;
  p->s = 0.001*s;
  s = p->s;
  p->s -= s;
  p->h = s/3600;
  s -= p->h*3600;
  p->m = s/60;
  p->s += s - p->m*60;
  p->validHMS = 1;
}

/*
** Compute both YMD and HMS
*/
static void computeYMD_HMS(DateTime *p){
  computeYMD(p);
  computeHMS(p);
}

/*
** Clear the YMD and HMS and the TZ
*/
static void clearYMD_HMS_TZ(DateTime *p){
  p->validYMD = 0;
  p->validHMS = 0;
  p->validTZ = 0;
}

/*
** Compute the difference (in days) between localtime and UTC (a.k.a. GMT)
** for the time value p where p is in UTC.
*/
static double localtimeOffset(DateTime *p){
  DateTime x, y;
  time_t t;
  x = *p;
  computeYMD_HMS(&x);
  if( x.Y<1971 || x.Y>=2038 ){
    x.Y = 2000;
    x.M = 1;
    x.D = 1;
    x.h = 0;
    x.m = 0;
    x.s = 0.0;
  } else {
    int s = x.s + 0.5;
    x.s = s;
  }
  x.tz = 0;
  x.validJD = 0;
  computeJD(&x);
  t = (x.rJD-2440587.5)*86400.0 + 0.5;
#ifdef HAVE_LOCALTIME_R
  {
    struct tm sLocal;
    localtime_r(&t, &sLocal);
    y.Y = sLocal.tm_year + 1900;
    y.M = sLocal.tm_mon + 1;
    y.D = sLocal.tm_mday;
    y.h = sLocal.tm_hour;
    y.m = sLocal.tm_min;
    y.s = sLocal.tm_sec;
  }
#else
  {
    struct tm *pTm;
    sqlite3OsEnterMutex();
    pTm = localtime(&t);
    y.Y = pTm->tm_year + 1900;
    y.M = pTm->tm_mon + 1;
    y.D = pTm->tm_mday;
    y.h = pTm->tm_hour;
    y.m = pTm->tm_min;
    y.s = pTm->tm_sec;
    sqlite3OsLeaveMutex();
  }
#endif
  y.validYMD = 1;
  y.validHMS = 1;
  y.validJD = 0;
  y.validTZ = 0;
  computeJD(&y);
  return y.rJD - x.rJD;
}

/*
** Process a modifier to a date-time stamp.  The modifiers are
** as follows:
**
**     NNN days
**     NNN hours
**     NNN minutes
**     NNN.NNNN seconds
**     NNN months
**     NNN years
**     start of month
**     start of year
**     start of week
**     start of day
**     weekday N
**     unixepoch
**     localtime
**     utc
**
** Return 0 on success and 1 if there is any kind of error.
*/
static int parseModifier(const char *zMod, DateTime *p){
  int rc = 1;
  int n;
  double r;
  char *z, zBuf[30];
  z = zBuf;
  for(n=0; n<sizeof(zBuf)-1 && zMod[n]; n++){
    z[n] = tolower(zMod[n]);
  }
  z[n] = 0;
  switch( z[0] ){
    case 'l': {
      /*    localtime
      **
      ** Assuming the current time value is UTC (a.k.a. GMT), shift it to
      ** show local time.
      */
      if( strcmp(z, "localtime")==0 ){
        computeJD(p);
        p->rJD += localtimeOffset(p);
        clearYMD_HMS_TZ(p);
        rc = 0;
      }
      break;
    }
    case 'u': {
      /*
      **    unixepoch
      **
      ** Treat the current value of p->rJD as the number of
      ** seconds since 1970.  Convert to a real julian day number.
      */
      if( strcmp(z, "unixepoch")==0 && p->validJD ){
        p->rJD = p->rJD/86400.0 + 2440587.5;
        clearYMD_HMS_TZ(p);
        rc = 0;
      }else if( strcmp(z, "utc")==0 ){
        double c1;
        computeJD(p);
        c1 = localtimeOffset(p);
        p->rJD -= c1;
        clearYMD_HMS_TZ(p);
        p->rJD += c1 - localtimeOffset(p);
        rc = 0;
      }
      break;
    }
    case 'w': {
      /*
      **    weekday N
      **
      ** Move the date to the same time on the next occurrence of
      ** weekday N where 0==Sunday, 1==Monday, and so forth.  If the
      ** date is already on the appropriate weekday, this is a no-op.
      */
      if( strncmp(z, "weekday ", 8)==0 && getValue(&z[8],&r)>0
                 && (n=r)==r && n>=0 && r<7 ){
        int Z;
        computeYMD_HMS(p);
        p->validTZ = 0;
        p->validJD = 0;
        computeJD(p);
        Z = p->rJD + 1.5;
        Z %= 7;
        if( Z>n ) Z -= 7;
        p->rJD += n - Z;
        clearYMD_HMS_TZ(p);
        rc = 0;
      }
      break;
    }
    case 's': {
      /*
      **    start of TTTTT
      **
      ** Move the date backwards to the beginning of the current day,
      ** or month or year.
      */
      if( strncmp(z, "start of ", 9)!=0 ) break;
      z += 9;
      computeYMD(p);
      p->validHMS = 1;
      p->h = p->m = 0;
      p->s = 0.0;
      p->validTZ = 0;
      p->validJD = 0;
      if( strcmp(z,"month")==0 ){
        p->D = 1;
        rc = 0;
      }else if( strcmp(z,"year")==0 ){
        computeYMD(p);
        p->M = 1;
        p->D = 1;
        rc = 0;
      }else if( strcmp(z,"day")==0 ){
        rc = 0;
      }
      break;
    }
    case '+':
    case '-':
    case '0':
    case '1':
    case '2':
    case '3':
    case '4':
    case '5':
    case '6':
    case '7':
    case '8':
    case '9': {
      n = getValue(z, &r);
      assert( n>=1 );
      if( z[n]==':' ){
        /* A modifier of the form (+|-)HH:MM:SS.FFF adds (or subtracts) the
        ** specified number of hours, minutes, seconds, and fractional seconds
        ** to the time.  The ".FFF" may be omitted.  The ":SS.FFF" may be
        ** omitted.
        */
        const char *z2 = z;
        DateTime tx;
        int day;
        if( !isdigit(*(u8*)z2) ) z2++;
        memset(&tx, 0, sizeof(tx));
        if( parseHhMmSs(z2, &tx) ) break;
        computeJD(&tx);
        tx.rJD -= 0.5;
        day = (int)tx.rJD;
        tx.rJD -= day;
        if( z[0]=='-' ) tx.rJD = -tx.rJD;
        computeJD(p);
        clearYMD_HMS_TZ(p);
        p->rJD += tx.rJD;
        rc = 0;
        break;
      }
      z += n;
      while( isspace(*(u8*)z) ) z++;
      n = strlen(z);
      if( n>10 || n<3 ) break;
      if( z[n-1]=='s' ){ z[n-1] = 0; n--; }
      computeJD(p);
      rc = 0;
      if( n==3 && strcmp(z,"day")==0 ){
        p->rJD += r;
      }else if( n==4 && strcmp(z,"hour")==0 ){
        p->rJD += r/24.0;
      }else if( n==6 && strcmp(z,"minute")==0 ){
        p->rJD += r/(24.0*60.0);
      }else if( n==6 && strcmp(z,"second")==0 ){
        p->rJD += r/(24.0*60.0*60.0);
      }else if( n==5 && strcmp(z,"month")==0 ){
        int x, y;
        computeYMD_HMS(p);
        p->M += r;
        x = p->M>0 ? (p->M-1)/12 : (p->M-12)/12;
        p->Y += x;
        p->M -= x*12;
        p->validJD = 0;
        computeJD(p);
        y = r;
        if( y!=r ){
          p->rJD += (r - y)*30.0;
        }
      }else if( n==4 && strcmp(z,"year")==0 ){
        computeYMD_HMS(p);
        p->Y += r;
        p->validJD = 0;
        computeJD(p);
      }else{
        rc = 1;
      }
      clearYMD_HMS_TZ(p);
      break;
    }
    default: {
      break;
    }
  }
  return rc;
}

/*
** Process time function arguments.  argv[0] is a date-time stamp.
** argv[1] and following are modifiers.  Parse them all and write
** the resulting time into the DateTime structure p.  Return 0
** on success and 1 if there are any errors.
*/
static int isDate(int argc, sqlite3_value **argv, DateTime *p){
  int i;
  const unsigned char *z;
  if( argc==0 ) return 1;
  if( (z = sqlite3_value_text(argv[0]))==0 || parseDateOrTime((char*)z, p) ){
    return 1;
  }
  for(i=1; i<argc; i++){
    if( (z = sqlite3_value_text(argv[i]))==0 || parseModifier((char*)z, p) ){
      return 1;
    }
  }
  return 0;
}


/*
** The following routines implement the various date and time functions
** of SQLite.
*/

/*
**    julianday( TIMESTRING, MOD, MOD, ...)
**
** Return the julian day number of the date specified in the arguments
*/
static void juliandayFunc(
  sqlite3_context *context,
  int argc,
  sqlite3_value **argv
){
  DateTime x;
  if( isDate(argc, argv, &x)==0 ){
    computeJD(&x);
    sqlite3_result_double(context, x.rJD);
  }
}

/*
**    datetime( TIMESTRING, MOD, MOD, ...)
**
** Return YYYY-MM-DD HH:MM:SS
*/
static void datetimeFunc(
  sqlite3_context *context,
  int argc,
  sqlite3_value **argv
){
  DateTime x;
  if( isDate(argc, argv, &x)==0 ){
    char zBuf[100];
    computeYMD_HMS(&x);
    sqlite3_snprintf(sizeof(zBuf), zBuf, "%04d-%02d-%02d %02d:%02d:%02d",
                     x.Y, x.M, x.D, x.h, x.m, (int)(x.s));
    sqlite3_result_text(context, zBuf, -1, SQLITE_TRANSIENT);
  }
}

/*
**    time( TIMESTRING, MOD, MOD, ...)
**
** Return HH:MM:SS
*/
static void timeFunc(
  sqlite3_context *context,
  int argc,
  sqlite3_value **argv
){
  DateTime x;
  if( isDate(argc, argv, &x)==0 ){
    char zBuf[100];
    computeHMS(&x);
    sqlite3_snprintf(sizeof(zBuf), zBuf, "%02d:%02d:%02d", x.h, x.m, (int)x.s);
    sqlite3_result_text(context, zBuf, -1, SQLITE_TRANSIENT);
  }
}

/*
**    date( TIMESTRING, MOD, MOD, ...)
**
** Return YYYY-MM-DD
*/
static void dateFunc(
  sqlite3_context *context,
  int argc,
  sqlite3_value **argv
){
  DateTime x;
  if( isDate(argc, argv, &x)==0 ){
    char zBuf[100];
    computeYMD(&x);
    sqlite3_snprintf(sizeof(zBuf), zBuf, "%04d-%02d-%02d", x.Y, x.M, x.D);
    sqlite3_result_text(context, zBuf, -1, SQLITE_TRANSIENT);
  }
}

/*
**    strftime( FORMAT, TIMESTRING, MOD, MOD, ...)
**
** Return a string described by FORMAT.  Conversions as follows:
**
**   %d  day of month
**   %f  ** fractional seconds  SS.SSS
**   %H  hour 00-24
**   %j  day of year 000-366
**   %J  ** Julian day number
**   %m  month 01-12
**   %M  minute 00-59
**   %s  seconds since 1970-01-01
**   %S  seconds 00-59
**   %w  day of week 0-6  sunday==0
**   %W  week of year 00-53
**   %Y  year 0000-9999
**   %%  %
*/
static void strftimeFunc(
  sqlite3_context *context,
  int argc,
  sqlite3_value **argv
){
  DateTime x;
  u64 n;
  int i, j;
  char *z;
  const char *zFmt = (const char*)sqlite3_value_text(argv[0]);
  char zBuf[100];
  if( zFmt==0 || isDate(argc-1, argv+1, &x) ) return;
  for(i=0, n=1; zFmt[i]; i++, n++){
    if( zFmt[i]=='%' ){
      switch( zFmt[i+1] ){
        case 'd':
        case 'H':
        case 'm':
        case 'M':
        case 'S':
        case 'W':
          n++;
          /* fall thru */
        case 'w':
        case '%':
          break;
        case 'f':
          n += 8;
          break;
        case 'j':
          n += 3;
          break;
        case 'Y':
          n += 8;
          break;
        case 's':
        case 'J':
          n += 50;
          break;
        default:
          return;  /* ERROR.  return a NULL */
      }
      i++;
    }
  }
  if( n<sizeof(zBuf) ){
    z = zBuf;
  }else if( n>SQLITE_MAX_LENGTH ){
    sqlite3_result_error_toobig(context);
    return;
  }else{
    z = sqliteMalloc( n );
    if( z==0 ) return;
  }
  computeJD(&x);
  computeYMD_HMS(&x);
  for(i=j=0; zFmt[i]; i++){
    if( zFmt[i]!='%' ){
      z[j++] = zFmt[i];
    }else{
      i++;
      switch( zFmt[i] ){
        case 'd':  sqlite3_snprintf(3, &z[j],"%02d",x.D); j+=2; break;
        case 'f': {
          double s = x.s;
          if( s>59.999 ) s = 59.999;
          sqlite3_snprintf(7, &z[j],"%06.3f", s);
          j += strlen(&z[j]);
          break;
        }
        case 'H':  sqlite3_snprintf(3, &z[j],"%02d",x.h); j+=2; break;
        case 'W': /* Fall thru */
        case 'j': {
          int nDay;             /* Number of days since 1st day of year */
          DateTime y = x;
          y.validJD = 0;
          y.M = 1;
          y.D = 1;
          computeJD(&y);
          nDay = x.rJD - y.rJD + 0.5;
          if( zFmt[i]=='W' ){
            int wd;   /* 0=Monday, 1=Tuesday, ... 6=Sunday */
            wd = ((int)(x.rJD+0.5)) % 7;
            sqlite3_snprintf(3, &z[j],"%02d",(nDay+7-wd)/7);
            j += 2;
          }else{
            sqlite3_snprintf(4, &z[j],"%03d",nDay+1);
            j += 3;
          }
          break;
        }
        case 'J': {
          sqlite3_snprintf(20, &z[j],"%.16g",x.rJD);
          j+=strlen(&z[j]);
          break;
        }
        case 'm':  sqlite3_snprintf(3, &z[j],"%02d",x.M); j+=2; break;
        case 'M':  sqlite3_snprintf(3, &z[j],"%02d",x.m); j+=2; break;
        case 's': {
          sqlite3_snprintf(30,&z[j],"%d",
                           (int)((x.rJD-2440587.5)*86400.0 + 0.5));
          j += strlen(&z[j]);
          break;
        }
        case 'S':  sqlite3_snprintf(3,&z[j],"%02d",(int)x.s); j+=2; break;
        case 'w':  z[j++] = (((int)(x.rJD+1.5)) % 7) + '0'; break;
        case 'Y':  sqlite3_snprintf(5,&z[j],"%04d",x.Y); j+=strlen(&z[j]);break;
        case '%':  z[j++] = '%'; break;
      }
    }
  }
  z[j] = 0;
  sqlite3_result_text(context, z, -1, SQLITE_TRANSIENT);
  if( z!=zBuf ){
    sqliteFree(z);
  }
}

/*
** current_time()
**
** This function returns the same value as time('now').
*/
static void ctimeFunc(
  sqlite3_context *context,
  int argc,
  sqlite3_value **argv
){
  sqlite3_value *pVal = sqlite3ValueNew();
  if( pVal ){
    sqlite3ValueSetStr(pVal, -1, "now", SQLITE_UTF8, SQLITE_STATIC);
    timeFunc(context, 1, &pVal);
    sqlite3ValueFree(pVal);
  }
}

/*
** current_date()
**
** This function returns the same value as date('now').
*/
static void cdateFunc(
  sqlite3_context *context,
  int argc,
  sqlite3_value **argv
){
  sqlite3_value *pVal = sqlite3ValueNew();
  if( pVal ){
    sqlite3ValueSetStr(pVal, -1, "now", SQLITE_UTF8, SQLITE_STATIC);
    dateFunc(context, 1, &pVal);
    sqlite3ValueFree(pVal);
  }
}

/*
** current_timestamp()
**
** This function returns the same value as datetime('now').
*/
static void ctimestampFunc(
  sqlite3_context *context,
  int argc,
  sqlite3_value **argv
){
  sqlite3_value *pVal = sqlite3ValueNew();
  if( pVal ){
    sqlite3ValueSetStr(pVal, -1, "now", SQLITE_UTF8, SQLITE_STATIC);
    datetimeFunc(context, 1, &pVal);
    sqlite3ValueFree(pVal);
  }
}
#endif /* !defined(SQLITE_OMIT_DATETIME_FUNCS) */

#ifdef SQLITE_OMIT_DATETIME_FUNCS
/*
** If the library is compiled to omit the full-scale date and time
** handling (to get a smaller binary), the following minimal version
** of the functions current_time(), current_date() and current_timestamp()
** are included instead. This is to support column declarations that
** include "DEFAULT CURRENT_TIME" etc.
**
** This function uses the C-library functions time(), gmtime()
** and strftime(). The format string to pass to strftime() is supplied
** as the user-data for the function.
*/
static void currentTimeFunc(
  sqlite3_context *context,
  int argc,
  sqlite3_value **argv
){
  time_t t;
  char *zFormat = (char *)sqlite3_user_data(context);
  char zBuf[20];

  time(&t);
#ifdef SQLITE_TEST
  {
    extern int sqlite3_current_time;  /* See os_XXX.c */
    if( sqlite3_current_time ){
      t = sqlite3_current_time;
    }
  }
#endif

#ifdef HAVE_GMTIME_R
  {
    struct tm sNow;
    gmtime_r(&t, &sNow);
    strftime(zBuf, 20, zFormat, &sNow);
  }
#else
  {
    struct tm *pTm;
    sqlite3OsEnterMutex();
    pTm = gmtime(&t);
    strftime(zBuf, 20, zFormat, pTm);
    sqlite3OsLeaveMutex();
  }
#endif

  sqlite3_result_text(context, zBuf, -1, SQLITE_TRANSIENT);
}
#endif

/*
** This function registered all of the above C functions as SQL
** functions.  This should be the only routine in this file with
** external linkage.
*/
SQLITE_PRIVATE void sqlite3RegisterDateTimeFunctions(sqlite3 *db){
#ifndef SQLITE_OMIT_DATETIME_FUNCS
  static const struct {
     char *zName;
     int nArg;
     void (*xFunc)(sqlite3_context*,int,sqlite3_value**);
  } aFuncs[] = {
    { "julianday", -1, juliandayFunc   },
    { "date",      -1, dateFunc        },
    { "time",      -1, timeFunc        },
    { "datetime",  -1, datetimeFunc    },
    { "strftime",  -1, strftimeFunc    },
    { "current_time",       0, ctimeFunc      },
    { "current_timestamp",  0, ctimestampFunc },
    { "current_date",       0, cdateFunc      },
  };
  int i;

  for(i=0; i<sizeof(aFuncs)/sizeof(aFuncs[0]); i++){
    sqlite3CreateFunc(db, aFuncs[i].zName, aFuncs[i].nArg,
        SQLITE_UTF8, 0, aFuncs[i].xFunc, 0, 0);
  }
#else
  static const struct {
     char *zName;
     char *zFormat;
  } aFuncs[] = {
    { "current_time", "%H:%M:%S" },
    { "current_date", "%Y-%m-%d" },
    { "current_timestamp", "%Y-%m-%d %H:%M:%S" }
  };
  int i;

  for(i=0; i<sizeof(aFuncs)/sizeof(aFuncs[0]); i++){
    sqlite3CreateFunc(db, aFuncs[i].zName, 0, SQLITE_UTF8, 
        aFuncs[i].zFormat, currentTimeFunc, 0, 0);
  }
#endif
}

/************** End of date.c ************************************************/
/************** Begin file os.c **********************************************/
/*
** 2005 November 29
**
** The author disclaims copyright to this source code.  In place of
** a legal notice, here is a blessing:
**
**    May you do good and not evil.
**    May you find forgiveness for yourself and forgive others.
**    May you share freely, never taking more than you give.
**
******************************************************************************
**
** This file contains OS interface code that is common to all
** architectures.
*/
#define _SQLITE_OS_C_ 1
#undef _SQLITE_OS_C_

/*
** The following routines are convenience wrappers around methods
** of the OsFile object.  This is mostly just syntactic sugar.  All
** of this would be completely automatic if SQLite were coded using
** C++ instead of plain old C.
*/
SQLITE_PRIVATE int sqlite3OsClose(OsFile **pId){
  OsFile *id;
  if( pId!=0 && (id = *pId)!=0 ){
    return id->pMethod->xClose(pId);
  }else{
    return SQLITE_OK;
  }
}
SQLITE_PRIVATE int sqlite3OsOpenDirectory(OsFile *id, const char *zName){
  return id->pMethod->xOpenDirectory(id, zName);
}
SQLITE_PRIVATE int sqlite3OsRead(OsFile *id, void *pBuf, int amt){
  return id->pMethod->xRead(id, pBuf, amt);
}
SQLITE_PRIVATE int sqlite3OsWrite(OsFile *id, const void *pBuf, int amt){
  return id->pMethod->xWrite(id, pBuf, amt);
}
SQLITE_PRIVATE int sqlite3OsSeek(OsFile *id, i64 offset){
  return id->pMethod->xSeek(id, offset);
}
SQLITE_PRIVATE int sqlite3OsTruncate(OsFile *id, i64 size){
  return id->pMethod->xTruncate(id, size);
}
SQLITE_PRIVATE int sqlite3OsSync(OsFile *id, int fullsync){
  return id->pMethod->xSync(id, fullsync);
}
SQLITE_PRIVATE void sqlite3OsSetFullSync(OsFile *id, int value){
  id->pMethod->xSetFullSync(id, value);
}
SQLITE_PRIVATE int sqlite3OsFileSize(OsFile *id, i64 *pSize){
  return id->pMethod->xFileSize(id, pSize);
}
SQLITE_PRIVATE int sqlite3OsLock(OsFile *id, int lockType){
  return id->pMethod->xLock(id, lockType);
}
SQLITE_PRIVATE int sqlite3OsUnlock(OsFile *id, int lockType){
  return id->pMethod->xUnlock(id, lockType);
}
SQLITE_PRIVATE int sqlite3OsCheckReservedLock(OsFile *id){
  return id->pMethod->xCheckReservedLock(id);
}
SQLITE_PRIVATE int sqlite3OsSectorSize(OsFile *id){
  int (*xSectorSize)(OsFile*) = id->pMethod->xSectorSize;
  return xSectorSize ? xSectorSize(id) : SQLITE_DEFAULT_SECTOR_SIZE;
}

#if defined(SQLITE_TEST) || defined(SQLITE_DEBUG)
  /* These methods are currently only used for testing and debugging. */
  int sqlite3OsFileHandle(OsFile *id){
    return id->pMethod->xFileHandle(id);
  }
  int sqlite3OsLockState(OsFile *id){
    return id->pMethod->xLockState(id);
  }
#endif

#ifdef SQLITE_ENABLE_REDEF_IO
/*
** A function to return a pointer to the virtual function table.
** This routine really does not accomplish very much since the
** virtual function table is a global variable and anybody who
** can call this function can just as easily access the variable
** for themselves.  Nevertheless, we include this routine for
** backwards compatibility with an earlier redefinable I/O
** interface design.
*/
struct sqlite3OsVtbl *sqlite3_os_switch(void){
  return &sqlite3Os;
}
#endif

/************** End of os.c **************************************************/
/************** Begin file malloc.c ******************************************/
/*
** 2001 September 15
**
** The author disclaims copyright to this source code.  In place of
** a legal notice, here is a blessing:
**
**    May you do good and not evil.
**    May you find forgiveness for yourself and forgive others.
**    May you share freely, never taking more than you give.
**
*************************************************************************
** Memory allocation functions used throughout sqlite.
**
**
** $Id: sqlite3.c,v 1.4 2007/07/21 08:51:43 rmsimpson Exp $
*/

/*
** MALLOC WRAPPER ARCHITECTURE
**
** The sqlite code accesses dynamic memory allocation/deallocation by invoking
** the following six APIs (which may be implemented as macros).
**
**     sqlite3Malloc()
**     sqlite3MallocRaw()
**     sqlite3Realloc()
**     sqlite3ReallocOrFree()
**     sqlite3Free()
**     sqlite3AllocSize()
**
** The function sqlite3FreeX performs the same task as sqlite3Free and is
** guaranteed to be a real function. The same holds for sqlite3MallocX
**
** The above APIs are implemented in terms of the functions provided in the
** operating-system interface. The OS interface is never accessed directly
** by code outside of this file.
**
**     sqlite3OsMalloc()
**     sqlite3OsRealloc()
**     sqlite3OsFree()
**     sqlite3OsAllocationSize()
**
** Functions sqlite3MallocRaw() and sqlite3Realloc() may invoke 
** sqlite3_release_memory() if a call to sqlite3OsMalloc() or
** sqlite3OsRealloc() fails (or if the soft-heap-limit for the thread is
** exceeded). Function sqlite3Malloc() usually invokes
** sqlite3MallocRaw().
**
** MALLOC TEST WRAPPER ARCHITECTURE
**
** The test wrapper provides extra test facilities to ensure the library 
** does not leak memory and handles the failure of the underlying OS level
** allocation system correctly. It is only present if the library is 
** compiled with the SQLITE_MEMDEBUG macro set.
**
**     * Guardposts to detect overwrites.
**     * Ability to cause a specific Malloc() or Realloc() to fail.
**     * Audit outstanding memory allocations (i.e check for leaks).
*/

#define MAX(x,y) ((x)>(y)?(x):(y))

#if defined(SQLITE_ENABLE_MEMORY_MANAGEMENT) && !defined(SQLITE_OMIT_DISKIO)
/*
** Set the soft heap-size limit for the current thread. Passing a negative
** value indicates no limit.
*/
void sqlite3_soft_heap_limit(int n){
  ThreadData *pTd = sqlite3ThreadData();
  if( pTd ){
    pTd->nSoftHeapLimit = n;
  }
  sqlite3ReleaseThreadData();
}

/*
** Release memory held by SQLite instances created by the current thread.
*/
int sqlite3_release_memory(int n){
  return sqlite3PagerReleaseMemory(n);
}
#else
/* If SQLITE_ENABLE_MEMORY_MANAGEMENT is not defined, then define a version
** of sqlite3_release_memory() to be used by other code in this file.
** This is done for no better reason than to reduce the number of 
** pre-processor #ifndef statements.
*/
#define sqlite3_release_memory(x) 0    /* 0 == no memory freed */
#endif

#ifdef SQLITE_MEMDEBUG
/*--------------------------------------------------------------------------
** Begin code for memory allocation system test layer.
**
** Memory debugging is turned on by defining the SQLITE_MEMDEBUG macro.
**
** SQLITE_MEMDEBUG==1    -> Fence-posting only (thread safe) 
** SQLITE_MEMDEBUG==2    -> Fence-posting + linked list of allocations (not ts)
** SQLITE_MEMDEBUG==3    -> Above + backtraces (not thread safe, req. glibc)
*/

/* Figure out whether or not to store backtrace() information for each malloc.
** The backtrace() function is only used if SQLITE_MEMDEBUG is set to 2 or 
** greater and glibc is in use. If we don't want to use backtrace(), then just
** define it as an empty macro and set the amount of space reserved to 0.
*/
#if defined(__GLIBC__) && SQLITE_MEMDEBUG>2
  extern int backtrace(void **, int);
  #define TESTALLOC_STACKSIZE 128
  #define TESTALLOC_STACKFRAMES ((TESTALLOC_STACKSIZE-8)/sizeof(void*))
#else
  #define backtrace(x, y)
  #define TESTALLOC_STACKSIZE 0
  #define TESTALLOC_STACKFRAMES 0
#endif

/*
** Number of 32-bit guard words.  This should probably be a multiple of
** 2 since on 64-bit machines we want the value returned by sqliteMalloc()
** to be 8-byte aligned.
*/
#ifndef TESTALLOC_NGUARD
# define TESTALLOC_NGUARD 2
#endif

/*
** Size reserved for storing file-name along with each malloc()ed blob.
*/
#define TESTALLOC_FILESIZE 64

/*
** Size reserved for storing the user string. Each time a Malloc() or Realloc()
** call succeeds, up to TESTALLOC_USERSIZE bytes of the string pointed to by
** sqlite3_malloc_id are stored along with the other test system metadata.
*/
#define TESTALLOC_USERSIZE 64
const char *sqlite3_malloc_id = 0;

/*
** Blocks used by the test layer have the following format:
**
**        <sizeof(void *) pNext pointer>
**        <sizeof(void *) pPrev pointer>
**        <TESTALLOC_NGUARD 32-bit guard words>
**            <The application level allocation>
**        <TESTALLOC_NGUARD 32-bit guard words>
**        <32-bit line number>
**        <TESTALLOC_FILESIZE bytes containing null-terminated file name>
**        <TESTALLOC_STACKSIZE bytes of backtrace() output>
*/ 

#define TESTALLOC_OFFSET_GUARD1(p)    (sizeof(void *) * 2)
#define TESTALLOC_OFFSET_DATA(p) ( \
  TESTALLOC_OFFSET_GUARD1(p) + sizeof(u32) * TESTALLOC_NGUARD \
)
#define TESTALLOC_OFFSET_GUARD2(p) ( \
  TESTALLOC_OFFSET_DATA(p) + sqlite3OsAllocationSize(p) - TESTALLOC_OVERHEAD \
)
#define TESTALLOC_OFFSET_LINENUMBER(p) ( \
  TESTALLOC_OFFSET_GUARD2(p) + sizeof(u32) * TESTALLOC_NGUARD \
)
#define TESTALLOC_OFFSET_FILENAME(p) ( \
  TESTALLOC_OFFSET_LINENUMBER(p) + sizeof(u32) \
)
#define TESTALLOC_OFFSET_USER(p) ( \
  TESTALLOC_OFFSET_FILENAME(p) + TESTALLOC_FILESIZE \
)
#define TESTALLOC_OFFSET_STACK(p) ( \
  TESTALLOC_OFFSET_USER(p) + TESTALLOC_USERSIZE + 8 - \
  (TESTALLOC_OFFSET_USER(p) % 8) \
)

#define TESTALLOC_OVERHEAD ( \
  sizeof(void *)*2 +                   /* pPrev and pNext pointers */   \
  TESTALLOC_NGUARD*sizeof(u32)*2 +              /* Guard words */       \
  sizeof(u32) + TESTALLOC_FILESIZE +   /* File and line number */       \
  TESTALLOC_USERSIZE +                 /* User string */                \
  TESTALLOC_STACKSIZE                  /* backtrace() stack */          \
)


/*
** For keeping track of the number of mallocs and frees.   This
** is used to check for memory leaks.  The iMallocFail and iMallocReset
** values are used to simulate malloc() failures during testing in 
** order to verify that the library correctly handles an out-of-memory
** condition.
*/
int sqlite3_nMalloc;         /* Number of sqliteMalloc() calls */
int sqlite3_nFree;           /* Number of sqliteFree() calls */
int sqlite3_memUsed;         /* TODO Total memory obtained from malloc */
int sqlite3_memMax;          /* TODO Mem usage high-water mark */
int sqlite3_iMallocFail;     /* Fail sqliteMalloc() after this many calls */
int sqlite3_iMallocReset = -1; /* When iMallocFail reaches 0, set to this */

void *sqlite3_pFirst = 0;         /* Pointer to linked list of allocations */
int sqlite3_nMaxAlloc = 0;        /* High water mark of ThreadData.nAlloc */
int sqlite3_mallocDisallowed = 0; /* assert() in sqlite3Malloc() if set */
int sqlite3_isFail = 0;           /* True if all malloc calls should fail */
const char *sqlite3_zFile = 0;    /* Filename to associate debug info with */
int sqlite3_iLine = 0;            /* Line number for debug info */
int sqlite3_mallocfail_trace = 0; /* Print a msg on malloc fail if true */

/*
** Check for a simulated memory allocation failure.  Return true if
** the failure should be simulated.  Return false to proceed as normal.
*/
SQLITE_PRIVATE int sqlite3TestMallocFail(){
  if( sqlite3_isFail ){
    return 1;
  }
  if( sqlite3_iMallocFail>=0 ){
    sqlite3_iMallocFail--;
    if( sqlite3_iMallocFail==0 ){
      sqlite3_iMallocFail = sqlite3_iMallocReset;
      sqlite3_isFail = 1;
      if( sqlite3_mallocfail_trace ){
         sqlite3DebugPrintf("###_malloc_fails_###\n");
      }
      return 1;
    }
  }
  return 0;
}

/*
** The argument is a pointer returned by sqlite3OsMalloc() or xRealloc().
** assert() that the first and last (TESTALLOC_NGUARD*4) bytes are set to the
** values set by the applyGuards() function.
*/
static void checkGuards(u32 *p)
{
  int i;
  char *zAlloc = (char *)p;
  char *z;

  /* First set of guard words */
  z = &zAlloc[TESTALLOC_OFFSET_GUARD1(p)];
  for(i=0; i<TESTALLOC_NGUARD; i++){
    assert(((u32 *)z)[i]==0xdead1122);
  }

  /* Second set of guard words */
  z = &zAlloc[TESTALLOC_OFFSET_GUARD2(p)];
  for(i=0; i<TESTALLOC_NGUARD; i++){
    u32 guard = 0;
    memcpy(&guard, &z[i*sizeof(u32)], sizeof(u32));
    assert(guard==0xdead3344);
  }
}

/*
** The argument is a pointer returned by sqlite3OsMalloc() or Realloc(). The
** first and last (TESTALLOC_NGUARD*4) bytes are set to known values for use as 
** guard-posts.
*/
static void applyGuards(u32 *p)
{
  int i;
  char *z;
  char *zAlloc = (char *)p;

  /* First set of guard words */
  z = &zAlloc[TESTALLOC_OFFSET_GUARD1(p)];
  for(i=0; i<TESTALLOC_NGUARD; i++){
    ((u32 *)z)[i] = 0xdead1122;
  }

  /* Second set of guard words */
  z = &zAlloc[TESTALLOC_OFFSET_GUARD2(p)];
  for(i=0; i<TESTALLOC_NGUARD; i++){
    static const int guard = 0xdead3344;
    memcpy(&z[i*sizeof(u32)], &guard, sizeof(u32));
  }

  /* Line number */
  z = &((char *)z)[TESTALLOC_NGUARD*sizeof(u32)];             /* Guard words */
  z = &zAlloc[TESTALLOC_OFFSET_LINENUMBER(p)];
  memcpy(z, &sqlite3_iLine, sizeof(u32));

  /* File name */
  z = &zAlloc[TESTALLOC_OFFSET_FILENAME(p)];
  strncpy(z, sqlite3_zFile, TESTALLOC_FILESIZE);
  z[TESTALLOC_FILESIZE - 1] = '\0';

  /* User string */
  z = &zAlloc[TESTALLOC_OFFSET_USER(p)];
  z[0] = 0;
  if( sqlite3_malloc_id ){
    strncpy(z, sqlite3_malloc_id, TESTALLOC_USERSIZE);
    z[TESTALLOC_USERSIZE-1] = 0;
  }

  /* backtrace() stack */
  z = &zAlloc[TESTALLOC_OFFSET_STACK(p)];
  backtrace((void **)z, TESTALLOC_STACKFRAMES);

  /* Sanity check to make sure checkGuards() is working */
  checkGuards(p);
}

/*
** The argument is a malloc()ed pointer as returned by the test-wrapper.
** Return a pointer to the Os level allocation.
*/
static void *getOsPointer(void *p)
{
  char *z = (char *)p;
  return (void *)(&z[-1 * TESTALLOC_OFFSET_DATA(p)]);
}


#if SQLITE_MEMDEBUG>1
/*
** The argument points to an Os level allocation. Link it into the threads list
** of allocations.
*/
static void linkAlloc(void *p){
  void **pp = (void **)p;
  pp[0] = 0;
  pp[1] = sqlite3_pFirst;
  if( sqlite3_pFirst ){
    ((void **)sqlite3_pFirst)[0] = p;
  }
  sqlite3_pFirst = p;
}

/*
** The argument points to an Os level allocation. Unlinke it from the threads
** list of allocations.
*/
static void unlinkAlloc(void *p)
{
  void **pp = (void **)p;
  if( p==sqlite3_pFirst ){
    assert(!pp[0]);
    assert(!pp[1] || ((void **)(pp[1]))[0]==p);
    sqlite3_pFirst = pp[1];
    if( sqlite3_pFirst ){
      ((void **)sqlite3_pFirst)[0] = 0;
    }
  }else{
    void **pprev = pp[0];
    void **pnext = pp[1];
    assert(pprev);
    assert(pprev[1]==p);
    pprev[1] = (void *)pnext;
    if( pnext ){
      assert(pnext[0]==p);
      pnext[0] = (void *)pprev;
    }
  }
}

/*
** Pointer p is a pointer to an OS level allocation that has just been
** realloc()ed. Set the list pointers that point to this entry to it's new
** location.
*/
static void relinkAlloc(void *p)
{
  void **pp = (void **)p;
  if( pp[0] ){
    ((void **)(pp[0]))[1] = p;
  }else{
    sqlite3_pFirst = p;
  }
  if( pp[1] ){
    ((void **)(pp[1]))[0] = p;
  }
}
#else
#define linkAlloc(x)
#define relinkAlloc(x)
#define unlinkAlloc(x)
#endif

/*
** This function sets the result of the Tcl interpreter passed as an argument
** to a list containing an entry for each currently outstanding call made to 
** sqliteMalloc and friends by the current thread. Each list entry is itself a
** list, consisting of the following (in order):
**
**     * The number of bytes allocated
**     * The __FILE__ macro at the time of the sqliteMalloc() call.
**     * The __LINE__ macro ...
**     * The value of the sqlite3_malloc_id variable ...
**     * The output of backtrace() (if available) ...
**
** Todo: We could have a version of this function that outputs to stdout, 
** to debug memory leaks when Tcl is not available.
*/
#if defined(TCLSH) && defined(SQLITE_DEBUG) && SQLITE_MEMDEBUG>1
SQLITE_PRIVATE int sqlite3OutstandingMallocs(Tcl_Interp *interp){
  void *p;
  Tcl_Obj *pRes = Tcl_NewObj();
  Tcl_IncrRefCount(pRes);


  for(p=sqlite3_pFirst; p; p=((void **)p)[1]){
    Tcl_Obj *pEntry = Tcl_NewObj();
    Tcl_Obj *pStack = Tcl_NewObj();
    char *z;
    u32 iLine;
    int nBytes = sqlite3OsAllocationSize(p) - TESTALLOC_OVERHEAD;
    char *zAlloc = (char *)p;
    int i;

    Tcl_ListObjAppendElement(0, pEntry, Tcl_NewIntObj(nBytes));

    z = &zAlloc[TESTALLOC_OFFSET_FILENAME(p)];
    Tcl_ListObjAppendElement(0, pEntry, Tcl_NewStringObj(z, -1));

    z = &zAlloc[TESTALLOC_OFFSET_LINENUMBER(p)];
    memcpy(&iLine, z, sizeof(u32));
    Tcl_ListObjAppendElement(0, pEntry, Tcl_NewIntObj(iLine));

    z = &zAlloc[TESTALLOC_OFFSET_USER(p)];
    Tcl_ListObjAppendElement(0, pEntry, Tcl_NewStringObj(z, -1));

    z = &zAlloc[TESTALLOC_OFFSET_STACK(p)];
    for(i=0; i<TESTALLOC_STACKFRAMES; i++){
      char zHex[128];
      sqlite3_snprintf(sizeof(zHex), zHex, "%p", ((void **)z)[i]);
      Tcl_ListObjAppendElement(0, pStack, Tcl_NewStringObj(zHex, -1));
    }

    Tcl_ListObjAppendElement(0, pEntry, pStack);
    Tcl_ListObjAppendElement(0, pRes, pEntry);
  }

  Tcl_ResetResult(interp);
  Tcl_SetObjResult(interp, pRes);
  Tcl_DecrRefCount(pRes);
  return TCL_OK;
}
#endif

/*
** This is the test layer's wrapper around sqlite3OsMalloc().
*/
static void * OSMALLOC(int n){
  sqlite3OsEnterMutex();
#ifdef SQLITE_ENABLE_MEMORY_MANAGEMENT
  sqlite3_nMaxAlloc = 
      MAX(sqlite3_nMaxAlloc, sqlite3ThreadDataReadOnly()->nAlloc);
#endif
  assert( !sqlite3_mallocDisallowed );
  if( !sqlite3TestMallocFail() ){
    u32 *p;
    p = (u32 *)sqlite3OsMalloc(n + TESTALLOC_OVERHEAD);
    assert(p);
    sqlite3_nMalloc++;
    applyGuards(p);
    linkAlloc(p);
    sqlite3OsLeaveMutex();
    return (void *)(&p[TESTALLOC_NGUARD + 2*sizeof(void *)/sizeof(u32)]);
  }
  sqlite3OsLeaveMutex();
  return 0;
}

static int OSSIZEOF(void *p){
  if( p ){
    u32 *pOs = (u32 *)getOsPointer(p);
    return sqlite3OsAllocationSize(pOs) - TESTALLOC_OVERHEAD;
  }
  return 0;
}

/*
** This is the test layer's wrapper around sqlite3OsFree(). The argument is a
** pointer to the space allocated for the application to use.
*/
static void OSFREE(void *pFree){
  u32 *p;         /* Pointer to the OS-layer allocation */
  sqlite3OsEnterMutex();
  p = (u32 *)getOsPointer(pFree);
  checkGuards(p);
  unlinkAlloc(p);
  memset(pFree, 0x55, OSSIZEOF(pFree));
  sqlite3OsFree(p);
  sqlite3_nFree++;
  sqlite3OsLeaveMutex();
}

/*
** This is the test layer's wrapper around sqlite3OsRealloc().
*/
static void * OSREALLOC(void *pRealloc, int n){
#ifdef SQLITE_ENABLE_MEMORY_MANAGEMENT
  sqlite3_nMaxAlloc = 
      MAX(sqlite3_nMaxAlloc, sqlite3ThreadDataReadOnly()->nAlloc);
#endif
  assert( !sqlite3_mallocDisallowed );
  if( !sqlite3TestMallocFail() ){
    u32 *p = (u32 *)getOsPointer(pRealloc);
    checkGuards(p);
    p = sqlite3OsRealloc(p, n + TESTALLOC_OVERHEAD);
    applyGuards(p);
    relinkAlloc(p);
    return (void *)(&p[TESTALLOC_NGUARD + 2*sizeof(void *)/sizeof(u32)]);
  }
  return 0;
}

static void OSMALLOC_FAILED(){
  sqlite3_isFail = 0;
}

#else
/* Define macros to call the sqlite3OsXXX interface directly if 
** the SQLITE_MEMDEBUG macro is not defined.
*/
#define OSMALLOC(x)        sqlite3OsMalloc(x)
#define OSREALLOC(x,y)     sqlite3OsRealloc(x,y)
#define OSFREE(x)          sqlite3OsFree(x)
#define OSSIZEOF(x)        sqlite3OsAllocationSize(x)
#define OSMALLOC_FAILED()

#endif  /* SQLITE_MEMDEBUG */
/*
** End code for memory allocation system test layer.
**--------------------------------------------------------------------------*/

/*
** This routine is called when we are about to allocate n additional bytes
** of memory.  If the new allocation will put is over the soft allocation
** limit, then invoke sqlite3_release_memory() to try to release some
** memory before continuing with the allocation.
**
** This routine also makes sure that the thread-specific-data (TSD) has
** be allocated.  If it has not and can not be allocated, then return
** false.  The updateMemoryUsedCount() routine below will deallocate
** the TSD if it ought to be.
**
** If SQLITE_ENABLE_MEMORY_MANAGEMENT is not defined, this routine is
** a no-op
*/ 
#ifdef SQLITE_ENABLE_MEMORY_MANAGEMENT
static int enforceSoftLimit(int n){
  ThreadData *pTsd = sqlite3ThreadData();
  if( pTsd==0 ){
    return 0;
  }
  assert( pTsd->nAlloc>=0 );
  if( n>0 && pTsd->nSoftHeapLimit>0 ){
    while( pTsd->nAlloc+n>pTsd->nSoftHeapLimit && sqlite3_release_memory(n) ){}
  }
  return 1;
}
#else
# define enforceSoftLimit(X)  1
#endif

/*
** Update the count of total outstanding memory that is held in
** thread-specific-data (TSD).  If after this update the TSD is
** no longer being used, then deallocate it.
**
** If SQLITE_ENABLE_MEMORY_MANAGEMENT is not defined, this routine is
** a no-op
*/
#ifdef SQLITE_ENABLE_MEMORY_MANAGEMENT
static void updateMemoryUsedCount(int n){
  ThreadData *pTsd = sqlite3ThreadData();
  if( pTsd ){
    pTsd->nAlloc += n;
    assert( pTsd->nAlloc>=0 );
    if( pTsd->nAlloc==0 && pTsd->nSoftHeapLimit==0 ){
      sqlite3ReleaseThreadData();
    }
  }
}
#else
#define updateMemoryUsedCount(x)  /* no-op */
#endif

/*
** Allocate and return N bytes of uninitialised memory by calling
** sqlite3OsMalloc(). If the Malloc() call fails, attempt to free memory 
** by calling sqlite3_release_memory().
*/
SQLITE_PRIVATE void *sqlite3MallocRaw(int n, int doMemManage){
  void *p = 0;
  if( n>0 && !sqlite3MallocFailed() && (!doMemManage || enforceSoftLimit(n)) ){
    while( (p = OSMALLOC(n))==0 && sqlite3_release_memory(n) ){}
    if( !p ){
      sqlite3FailedMalloc();
      OSMALLOC_FAILED();
    }else if( doMemManage ){
      updateMemoryUsedCount(OSSIZEOF(p));
    }
  }
  return p;
}

/*
** Resize the allocation at p to n bytes by calling sqlite3OsRealloc(). The
** pointer to the new allocation is returned.  If the Realloc() call fails,
** attempt to free memory by calling sqlite3_release_memory().
*/
SQLITE_PRIVATE void *sqlite3Realloc(void *p, int n){
  if( sqlite3MallocFailed() ){
    return 0;
  }

  if( !p ){
    return sqlite3Malloc(n, 1);
  }else{
    void *np = 0;
#ifdef SQLITE_ENABLE_MEMORY_MANAGEMENT
    int origSize = OSSIZEOF(p);
#endif
    if( enforceSoftLimit(n - origSize) ){
      while( (np = OSREALLOC(p, n))==0 && sqlite3_release_memory(n) ){}
      if( !np ){
        sqlite3FailedMalloc();
        OSMALLOC_FAILED();
      }else{
        updateMemoryUsedCount(OSSIZEOF(np) - origSize);
      }
    }
    return np;
  }
}

/*
** Free the memory pointed to by p. p must be either a NULL pointer or a 
** value returned by a previous call to sqlite3Malloc() or sqlite3Realloc().
*/
SQLITE_PRIVATE void sqlite3FreeX(void *p){
  if( p ){
    updateMemoryUsedCount(0 - OSSIZEOF(p));
    OSFREE(p);
  }
}

/*
** A version of sqliteMalloc() that is always a function, not a macro.
** Currently, this is used only to alloc to allocate the parser engine.
*/
SQLITE_PRIVATE void *sqlite3MallocX(int n){
  return sqliteMalloc(n);
}

/*
** sqlite3Malloc
** sqlite3ReallocOrFree
**
** These two are implemented as wrappers around sqlite3MallocRaw(), 
** sqlite3Realloc() and sqlite3Free().
*/ 
SQLITE_PRIVATE void *sqlite3Malloc(int n, int doMemManage){
  void *p = sqlite3MallocRaw(n, doMemManage);
  if( p ){
    memset(p, 0, n);
  }
  return p;
}
SQLITE_PRIVATE void *sqlite3ReallocOrFree(void *p, int n){
  void *pNew;
  pNew = sqlite3Realloc(p, n);
  if( !pNew ){
    sqlite3FreeX(p);
  }
  return pNew;
}

/*
** sqlite3ThreadSafeMalloc() and sqlite3ThreadSafeFree() are used in those
** rare scenarios where sqlite may allocate memory in one thread and free
** it in another. They are exactly the same as sqlite3Malloc() and 
** sqlite3Free() except that:
**
**   * The allocated memory is not included in any calculations with 
**     respect to the soft-heap-limit, and
**
**   * sqlite3ThreadSafeMalloc() must be matched with ThreadSafeFree(),
**     not sqlite3Free(). Calling sqlite3Free() on memory obtained from
**     ThreadSafeMalloc() will cause an error somewhere down the line.
*/
#ifdef SQLITE_ENABLE_MEMORY_MANAGEMENT
SQLITE_PRIVATE void *sqlite3ThreadSafeMalloc(int n){
  (void)ENTER_MALLOC;
  return sqlite3Malloc(n, 0);
}
SQLITE_PRIVATE void sqlite3ThreadSafeFree(void *p){
  (void)ENTER_MALLOC;
  if( p ){
    OSFREE(p);
  }
}
#endif


/*
** Return the number of bytes allocated at location p. p must be either 
** a NULL pointer (in which case 0 is returned) or a pointer returned by 
** sqlite3Malloc(), sqlite3Realloc() or sqlite3ReallocOrFree().
**
** The number of bytes allocated does not include any overhead inserted by 
** any malloc() wrapper functions that may be called. So the value returned
** is the number of bytes that were available to SQLite using pointer p, 
** regardless of how much memory was actually allocated.
*/
#ifdef SQLITE_ENABLE_MEMORY_MANAGEMENT
SQLITE_PRIVATE int sqlite3AllocSize(void *p){
  return OSSIZEOF(p);
}
#endif

/*
** Make a copy of a string in memory obtained from sqliteMalloc(). These 
** functions call sqlite3MallocRaw() directly instead of sqliteMalloc(). This
** is because when memory debugging is turned on, these two functions are 
** called via macros that record the current file and line number in the
** ThreadData structure.
*/
SQLITE_PRIVATE char *sqlite3StrDup(const char *z){
  char *zNew;
  int n;
  if( z==0 ) return 0;
  n = strlen(z)+1;
  zNew = sqlite3MallocRaw(n, 1);
  if( zNew ) memcpy(zNew, z, n);
  return zNew;
}
SQLITE_PRIVATE char *sqlite3StrNDup(const char *z, int n){
  char *zNew;
  if( z==0 ) return 0;
  zNew = sqlite3MallocRaw(n+1, 1);
  if( zNew ){
    memcpy(zNew, z, n);
    zNew[n] = 0;
  }
  return zNew;
}

/*
** Create a string from the 2nd and subsequent arguments (up to the
** first NULL argument), store the string in memory obtained from
** sqliteMalloc() and make the pointer indicated by the 1st argument
** point to that string.  The 1st argument must either be NULL or 
** point to memory obtained from sqliteMalloc().
*/
SQLITE_PRIVATE void sqlite3SetString(char **pz, ...){
  va_list ap;
  int nByte;
  const char *z;
  char *zResult;

  assert( pz!=0 );
  nByte = 1;
  va_start(ap, pz);
  while( (z = va_arg(ap, const char*))!=0 ){
    nByte += strlen(z);
  }
  va_end(ap);
  sqliteFree(*pz);
  *pz = zResult = sqliteMallocRaw( nByte );
  if( zResult==0 ){
    return;
  }
  *zResult = 0;
  va_start(ap, pz);
  while( (z = va_arg(ap, const char*))!=0 ){
    int n = strlen(z);
    memcpy(zResult, z, n);
    zResult += n;
  }
  zResult[0] = 0;
  va_end(ap);
}


/*
** This function must be called before exiting any API function (i.e. 
** returning control to the user) that has called sqlite3Malloc or
** sqlite3Realloc.
**
** The returned value is normally a copy of the second argument to this
** function. However, if a malloc() failure has occured since the previous
** invocation SQLITE_NOMEM is returned instead. 
**
** If the first argument, db, is not NULL and a malloc() error has occured,
** then the connection error-code (the value returned by sqlite3_errcode())
** is set to SQLITE_NOMEM.
*/
int sqlite3_mallocHasFailed = 0;
SQLITE_PRIVATE int sqlite3ApiExit(sqlite3* db, int rc){
  if( sqlite3MallocFailed() ){
    sqlite3_mallocHasFailed = 0;
    sqlite3OsLeaveMutex();
    sqlite3Error(db, SQLITE_NOMEM, 0);
    rc = SQLITE_NOMEM;
  }
  return rc & (db ? db->errMask : 0xff);
}

/* 
** Set the "malloc has failed" condition to true for this thread.
*/
SQLITE_PRIVATE void sqlite3FailedMalloc(){
  if( !sqlite3MallocFailed() ){
    sqlite3OsEnterMutex();
    assert( sqlite3_mallocHasFailed==0 );
    sqlite3_mallocHasFailed = 1;
  }
}

#ifdef SQLITE_MEMDEBUG
/*
** This function sets a flag in the thread-specific-data structure that will
** cause an assert to fail if sqliteMalloc() or sqliteRealloc() is called.
*/
SQLITE_PRIVATE void sqlite3MallocDisallow(){
  assert( sqlite3_mallocDisallowed>=0 );
  sqlite3_mallocDisallowed++;
}

/*
** This function clears the flag set in the thread-specific-data structure set
** by sqlite3MallocDisallow().
*/
SQLITE_PRIVATE void sqlite3MallocAllow(){
  assert( sqlite3_mallocDisallowed>0 );
  sqlite3_mallocDisallowed--;
}
#endif

/************** End of malloc.c **********************************************/
/************** Begin file printf.c ******************************************/
/*
** The "printf" code that follows dates from the 1980's.  It is in
** the public domain.  The original comments are included here for
** completeness.  They are very out-of-date but might be useful as
** an historical reference.  Most of the "enhancements" have been backed
** out so that the functionality is now the same as standard printf().
**
**************************************************************************
**
** The following modules is an enhanced replacement for the "printf" subroutines
** found in the standard C library.  The following enhancements are
** supported:
**
**      +  Additional functions.  The standard set of "printf" functions
**         includes printf, fprintf, sprintf, vprintf, vfprintf, and
**         vsprintf.  This module adds the following:
**
**           *  snprintf -- Works like sprintf, but has an extra argument
**                          which is the size of the buffer written to.
**
**           *  mprintf --  Similar to sprintf.  Writes output to memory
**                          obtained from malloc.
**
**           *  xprintf --  Calls a function to dispose of output.
**
**           *  nprintf --  No output, but returns the number of characters
**                          that would have been output by printf.
**
**           *  A v- version (ex: vsnprintf) of every function is also
**              supplied.
**
**      +  A few extensions to the formatting notation are supported:
**
**           *  The "=" flag (similar to "-") causes the output to be
**              be centered in the appropriately sized field.
**
**           *  The %b field outputs an integer in binary notation.
**
**           *  The %c field now accepts a precision.  The character output
**              is repeated by the number of times the precision specifies.
**
**           *  The %' field works like %c, but takes as its character the
**              next character of the format string, instead of the next
**              argument.  For example,  printf("%.78'-")  prints 78 minus
**              signs, the same as  printf("%.78c",'-').
**
**      +  When compiled using GCC on a SPARC, this version of printf is
**         faster than the library printf for SUN OS 4.1.
**
**      +  All functions are fully reentrant.
**
*/
#include <math.h>

/*
** Conversion types fall into various categories as defined by the
** following enumeration.
*/
#define etRADIX       1 /* Integer types.  %d, %x, %o, and so forth */
#define etFLOAT       2 /* Floating point.  %f */
#define etEXP         3 /* Exponentional notation. %e and %E */
#define etGENERIC     4 /* Floating or exponential, depending on exponent. %g */
#define etSIZE        5 /* Return number of characters processed so far. %n */
#define etSTRING      6 /* Strings. %s */
#define etDYNSTRING   7 /* Dynamically allocated strings. %z */
#define etPERCENT     8 /* Percent symbol. %% */
#define etCHARX       9 /* Characters. %c */
/* The rest are extensions, not normally found in printf() */
#define etCHARLIT    10 /* Literal characters.  %' */
#define etSQLESCAPE  11 /* Strings with '\'' doubled.  %q */
#define etSQLESCAPE2 12 /* Strings with '\'' doubled and enclosed in '',
                          NULL pointers replaced by SQL NULL.  %Q */
#define etTOKEN      13 /* a pointer to a Token structure */
#define etSRCLIST    14 /* a pointer to a SrcList */
#define etPOINTER    15 /* The %p conversion */
#define etSQLESCAPE3 16 /* %w -> Strings with '\"' doubled */


/*
** An "etByte" is an 8-bit unsigned value.
*/
typedef unsigned char etByte;

/*
** Each builtin conversion character (ex: the 'd' in "%d") is described
** by an instance of the following structure
*/
typedef struct et_info {   /* Information about each format field */
  char fmttype;            /* The format field code letter */
  etByte base;             /* The base for radix conversion */
  etByte flags;            /* One or more of FLAG_ constants below */
  etByte type;             /* Conversion paradigm */
  etByte charset;          /* Offset into aDigits[] of the digits string */
  etByte prefix;           /* Offset into aPrefix[] of the prefix string */
} et_info;

/*
** Allowed values for et_info.flags
*/
#define FLAG_SIGNED  1     /* True if the value to convert is signed */
#define FLAG_INTERN  2     /* True if for internal use only */
#define FLAG_STRING  4     /* Allow infinity precision */


/*
** The following table is searched linearly, so it is good to put the
** most frequently used conversion types first.
*/
static const char aDigits[] = "0123456789ABCDEF0123456789abcdef";
static const char aPrefix[] = "-x0\000X0";
static const et_info fmtinfo[] = {
  {  'd', 10, 1, etRADIX,      0,  0 },
  {  's',  0, 4, etSTRING,     0,  0 },
  {  'g',  0, 1, etGENERIC,    30, 0 },
  {  'z',  0, 6, etDYNSTRING,  0,  0 },
  {  'q',  0, 4, etSQLESCAPE,  0,  0 },
  {  'Q',  0, 4, etSQLESCAPE2, 0,  0 },
  {  'w',  0, 4, etSQLESCAPE3, 0,  0 },
  {  'c',  0, 0, etCHARX,      0,  0 },
  {  'o',  8, 0, etRADIX,      0,  2 },
  {  'u', 10, 0, etRADIX,      0,  0 },
  {  'x', 16, 0, etRADIX,      16, 1 },
  {  'X', 16, 0, etRADIX,      0,  4 },
#ifndef SQLITE_OMIT_FLOATING_POINT
  {  'f',  0, 1, etFLOAT,      0,  0 },
  {  'e',  0, 1, etEXP,        30, 0 },
  {  'E',  0, 1, etEXP,        14, 0 },
  {  'G',  0, 1, etGENERIC,    14, 0 },
#endif
  {  'i', 10, 1, etRADIX,      0,  0 },
  {  'n',  0, 0, etSIZE,       0,  0 },
  {  '%',  0, 0, etPERCENT,    0,  0 },
  {  'p', 16, 0, etPOINTER,    0,  1 },
  {  'T',  0, 2, etTOKEN,      0,  0 },
  {  'S',  0, 2, etSRCLIST,    0,  0 },
};
#define etNINFO  (sizeof(fmtinfo)/sizeof(fmtinfo[0]))

/*
** If SQLITE_OMIT_FLOATING_POINT is defined, then none of the floating point
** conversions will work.
*/
#ifndef SQLITE_OMIT_FLOATING_POINT
/*
** "*val" is a double such that 0.1 <= *val < 10.0
** Return the ascii code for the leading digit of *val, then
** multiply "*val" by 10.0 to renormalize.
**
** Example:
**     input:     *val = 3.14159
**     output:    *val = 1.4159    function return = '3'
**
** The counter *cnt is incremented each time.  After counter exceeds
** 16 (the number of significant digits in a 64-bit float) '0' is
** always returned.
*/
static int et_getdigit(LONGDOUBLE_TYPE *val, int *cnt){
  int digit;
  LONGDOUBLE_TYPE d;
  if( (*cnt)++ >= 16 ) return '0';
  digit = (int)*val;
  d = digit;
  digit += '0';
  *val = (*val - d)*10.0;
  return digit;
}
#endif /* SQLITE_OMIT_FLOATING_POINT */

/*
** On machines with a small stack size, you can redefine the
** SQLITE_PRINT_BUF_SIZE to be less than 350.  But beware - for
** smaller values some %f conversions may go into an infinite loop.
*/
#ifndef SQLITE_PRINT_BUF_SIZE
# define SQLITE_PRINT_BUF_SIZE 350
#endif
#define etBUFSIZE SQLITE_PRINT_BUF_SIZE  /* Size of the output buffer */

/*
** The root program.  All variations call this core.
**
** INPUTS:
**   func   This is a pointer to a function taking three arguments
**            1. A pointer to anything.  Same as the "arg" parameter.
**            2. A pointer to the list of characters to be output
**               (Note, this list is NOT null terminated.)
**            3. An integer number of characters to be output.
**               (Note: This number might be zero.)
**
**   arg    This is the pointer to anything which will be passed as the
**          first argument to "func".  Use it for whatever you like.
**
**   fmt    This is the format string, as in the usual print.
**
**   ap     This is a pointer to a list of arguments.  Same as in
**          vfprint.
**
** OUTPUTS:
**          The return value is the total number of characters sent to
**          the function "func".  Returns -1 on a error.
**
** Note that the order in which automatic variables are declared below
** seems to make a big difference in determining how fast this beast
** will run.
*/
static int vxprintf(
  void (*func)(void*,const char*,int),     /* Consumer of text */
  void *arg,                         /* First argument to the consumer */
  int useExtended,                   /* Allow extended %-conversions */
  const char *fmt,                   /* Format string */
  va_list ap                         /* arguments */
){
  int c;                     /* Next character in the format string */
  char *bufpt;               /* Pointer to the conversion buffer */
  int precision;             /* Precision of the current field */
  int length;                /* Length of the field */
  int idx;                   /* A general purpose loop counter */
  int count;                 /* Total number of characters output */
  int width;                 /* Width of the current field */
  etByte flag_leftjustify;   /* True if "-" flag is present */
  etByte flag_plussign;      /* True if "+" flag is present */
  etByte flag_blanksign;     /* True if " " flag is present */
  etByte flag_alternateform; /* True if "#" flag is present */
  etByte flag_altform2;      /* True if "!" flag is present */
  etByte flag_zeropad;       /* True if field width constant starts with zero */
  etByte flag_long;          /* True if "l" flag is present */
  etByte flag_longlong;      /* True if the "ll" flag is present */
  etByte done;               /* Loop termination flag */
  sqlite_uint64 longvalue;   /* Value for integer types */
  LONGDOUBLE_TYPE realvalue; /* Value for real types */
  const et_info *infop;      /* Pointer to the appropriate info structure */
  char buf[etBUFSIZE];       /* Conversion buffer */
  char prefix;               /* Prefix character.  "+" or "-" or " " or '\0'. */
  etByte errorflag = 0;      /* True if an error is encountered */
  etByte xtype;              /* Conversion paradigm */
  char *zExtra;              /* Extra memory used for etTCLESCAPE conversions */
  static const char spaces[] =
   "                                                                         ";
#define etSPACESIZE (sizeof(spaces)-1)
#ifndef SQLITE_OMIT_FLOATING_POINT
  int  exp, e2;              /* exponent of real numbers */
  double rounder;            /* Used for rounding floating point values */
  etByte flag_dp;            /* True if decimal point should be shown */
  etByte flag_rtz;           /* True if trailing zeros should be removed */
  etByte flag_exp;           /* True to force display of the exponent */
  int nsd;                   /* Number of significant digits returned */
#endif

  func(arg,"",0);
  count = length = 0;
  bufpt = 0;
  for(; (c=(*fmt))!=0; ++fmt){
    if( c!='%' ){
      int amt;
      bufpt = (char *)fmt;
      amt = 1;
      while( (c=(*++fmt))!='%' && c!=0 ) amt++;
      (*func)(arg,bufpt,amt);
      count += amt;
      if( c==0 ) break;
    }
    if( (c=(*++fmt))==0 ){
      errorflag = 1;
      (*func)(arg,"%",1);
      count++;
      break;
    }
    /* Find out what flags are present */
    flag_leftjustify = flag_plussign = flag_blanksign = 
     flag_alternateform = flag_altform2 = flag_zeropad = 0;
    done = 0;
    do{
      switch( c ){
        case '-':   flag_leftjustify = 1;     break;
        case '+':   flag_plussign = 1;        break;
        case ' ':   flag_blanksign = 1;       break;
        case '#':   flag_alternateform = 1;   break;
        case '!':   flag_altform2 = 1;        break;
        case '0':   flag_zeropad = 1;         break;
        default:    done = 1;                 break;
      }
    }while( !done && (c=(*++fmt))!=0 );
    /* Get the field width */
    width = 0;
    if( c=='*' ){
      width = va_arg(ap,int);
      if( width<0 ){
        flag_leftjustify = 1;
        width = -width;
      }
      c = *++fmt;
    }else{
      while( c>='0' && c<='9' ){
        width = width*10 + c - '0';
        c = *++fmt;
      }
    }
    if( width > etBUFSIZE-10 ){
      width = etBUFSIZE-10;
    }
    /* Get the precision */
    if( c=='.' ){
      precision = 0;
      c = *++fmt;
      if( c=='*' ){
        precision = va_arg(ap,int);
        if( precision<0 ) precision = -precision;
        c = *++fmt;
      }else{
        while( c>='0' && c<='9' ){
          precision = precision*10 + c - '0';
          c = *++fmt;
        }
      }
    }else{
      precision = -1;
    }
    /* Get the conversion type modifier */
    if( c=='l' ){
      flag_long = 1;
      c = *++fmt;
      if( c=='l' ){
        flag_longlong = 1;
        c = *++fmt;
      }else{
        flag_longlong = 0;
      }
    }else{
      flag_long = flag_longlong = 0;
    }
    /* Fetch the info entry for the field */
    infop = 0;
    for(idx=0; idx<etNINFO; idx++){
      if( c==fmtinfo[idx].fmttype ){
        infop = &fmtinfo[idx];
        if( useExtended || (infop->flags & FLAG_INTERN)==0 ){
          xtype = infop->type;
        }else{
          return -1;
        }
        break;
      }
    }
    zExtra = 0;
    if( infop==0 ){
      return -1;
    }


    /* Limit the precision to prevent overflowing buf[] during conversion */
    if( precision>etBUFSIZE-40 && (infop->flags & FLAG_STRING)==0 ){
      precision = etBUFSIZE-40;
    }

    /*
    ** At this point, variables are initialized as follows:
    **
    **   flag_alternateform          TRUE if a '#' is present.
    **   flag_altform2               TRUE if a '!' is present.
    **   flag_plussign               TRUE if a '+' is present.
    **   flag_leftjustify            TRUE if a '-' is present or if the
    **                               field width was negative.
    **   flag_zeropad                TRUE if the width began with 0.
    **   flag_long                   TRUE if the letter 'l' (ell) prefixed
    **                               the conversion character.
    **   flag_longlong               TRUE if the letter 'll' (ell ell) prefixed
    **                               the conversion character.
    **   flag_blanksign              TRUE if a ' ' is present.
    **   width                       The specified field width.  This is
    **                               always non-negative.  Zero is the default.
    **   precision                   The specified precision.  The default
    **                               is -1.
    **   xtype                       The class of the conversion.
    **   infop                       Pointer to the appropriate info struct.
    */
    switch( xtype ){
      case etPOINTER:
        flag_longlong = sizeof(char*)==sizeof(i64);
        flag_long = sizeof(char*)==sizeof(long int);
        /* Fall through into the next case */
      case etRADIX:
        if( infop->flags & FLAG_SIGNED ){
          i64 v;
          if( flag_longlong )   v = va_arg(ap,i64);
          else if( flag_long )  v = va_arg(ap,long int);
          else                  v = va_arg(ap,int);
          if( v<0 ){
            longvalue = -v;
            prefix = '-';
          }else{
            longvalue = v;
            if( flag_plussign )        prefix = '+';
            else if( flag_blanksign )  prefix = ' ';
            else                       prefix = 0;
          }
        }else{
          if( flag_longlong )   longvalue = va_arg(ap,u64);
          else if( flag_long )  longvalue = va_arg(ap,unsigned long int);
          else                  longvalue = va_arg(ap,unsigned int);
          prefix = 0;
        }
        if( longvalue==0 ) flag_alternateform = 0;
        if( flag_zeropad && precision<width-(prefix!=0) ){
          precision = width-(prefix!=0);
        }
        bufpt = &buf[etBUFSIZE-1];
        {
          register const char *cset;      /* Use registers for speed */
          register int base;
          cset = &aDigits[infop->charset];
          base = infop->base;
          do{                                           /* Convert to ascii */
            *(--bufpt) = cset[longvalue%base];
            longvalue = longvalue/base;
          }while( longvalue>0 );
        }
        length = &buf[etBUFSIZE-1]-bufpt;
        for(idx=precision-length; idx>0; idx--){
          *(--bufpt) = '0';                             /* Zero pad */
        }
        if( prefix ) *(--bufpt) = prefix;               /* Add sign */
        if( flag_alternateform && infop->prefix ){      /* Add "0" or "0x" */
          const char *pre;
          char x;
          pre = &aPrefix[infop->prefix];
          if( *bufpt!=pre[0] ){
            for(; (x=(*pre))!=0; pre++) *(--bufpt) = x;
          }
        }
        length = &buf[etBUFSIZE-1]-bufpt;
        break;
      case etFLOAT:
      case etEXP:
      case etGENERIC:
        realvalue = va_arg(ap,double);
#ifndef SQLITE_OMIT_FLOATING_POINT
        if( precision<0 ) precision = 6;         /* Set default precision */
        if( precision>etBUFSIZE/2-10 ) precision = etBUFSIZE/2-10;
        if( realvalue<0.0 ){
          realvalue = -realvalue;
          prefix = '-';
        }else{
          if( flag_plussign )          prefix = '+';
          else if( flag_blanksign )    prefix = ' ';
          else                         prefix = 0;
        }
        if( xtype==etGENERIC && precision>0 ) precision--;
#if 0
        /* Rounding works like BSD when the constant 0.4999 is used.  Wierd! */
        for(idx=precision, rounder=0.4999; idx>0; idx--, rounder*=0.1);
#else
        /* It makes more sense to use 0.5 */
        for(idx=precision, rounder=0.5; idx>0; idx--, rounder*=0.1){}
#endif
        if( xtype==etFLOAT ) realvalue += rounder;
        /* Normalize realvalue to within 10.0 > realvalue >= 1.0 */
        exp = 0;
        if( sqlite3_isnan(realvalue) ){
          bufpt = "NaN";
          length = 3;
          break;
        }
        if( realvalue>0.0 ){
          while( realvalue>=1e32 && exp<=350 ){ realvalue *= 1e-32; exp+=32; }
          while( realvalue>=1e8 && exp<=350 ){ realvalue *= 1e-8; exp+=8; }
          while( realvalue>=10.0 && exp<=350 ){ realvalue *= 0.1; exp++; }
          while( realvalue<1e-8 && exp>=-350 ){ realvalue *= 1e8; exp-=8; }
          while( realvalue<1.0 && exp>=-350 ){ realvalue *= 10.0; exp--; }
          if( exp>350 || exp<-350 ){
            if( prefix=='-' ){
              bufpt = "-Inf";
            }else if( prefix=='+' ){
              bufpt = "+Inf";
            }else{
              bufpt = "Inf";
            }
            length = strlen(bufpt);
            break;
          }
        }
        bufpt = buf;
        /*
        ** If the field type is etGENERIC, then convert to either etEXP
        ** or etFLOAT, as appropriate.
        */
        flag_exp = xtype==etEXP;
        if( xtype!=etFLOAT ){
          realvalue += rounder;
          if( realvalue>=10.0 ){ realvalue *= 0.1; exp++; }
        }
        if( xtype==etGENERIC ){
          flag_rtz = !flag_alternateform;
          if( exp<-4 || exp>precision ){
            xtype = etEXP;
          }else{
            precision = precision - exp;
            xtype = etFLOAT;
          }
        }else{
          flag_rtz = 0;
        }
        if( xtype==etEXP ){
          e2 = 0;
        }else{
          e2 = exp;
        }
        nsd = 0;
        flag_dp = (precision>0) | flag_alternateform | flag_altform2;
        /* The sign in front of the number */
        if( prefix ){
          *(bufpt++) = prefix;
        }
        /* Digits prior to the decimal point */
        if( e2<0 ){
          *(bufpt++) = '0';
        }else{
          for(; e2>=0; e2--){
            *(bufpt++) = et_getdigit(&realvalue,&nsd);
          }
        }
        /* The decimal point */
        if( flag_dp ){
          *(bufpt++) = '.';
        }
        /* "0" digits after the decimal point but before the first
        ** significant digit of the number */
        for(e2++; e2<0 && precision>0; precision--, e2++){
          *(bufpt++) = '0';
        }
        /* Significant digits after the decimal point */
        while( (precision--)>0 ){
          *(bufpt++) = et_getdigit(&realvalue,&nsd);
        }
        /* Remove trailing zeros and the "." if no digits follow the "." */
        if( flag_rtz && flag_dp ){
          while( bufpt[-1]=='0' ) *(--bufpt) = 0;
          assert( bufpt>buf );
          if( bufpt[-1]=='.' ){
            if( flag_altform2 ){
              *(bufpt++) = '0';
            }else{
              *(--bufpt) = 0;
            }
          }
        }
        /* Add the "eNNN" suffix */
        if( flag_exp || (xtype==etEXP && exp) ){
          *(bufpt++) = aDigits[infop->charset];
          if( exp<0 ){
            *(bufpt++) = '-'; exp = -exp;
          }else{
            *(bufpt++) = '+';
          }
          if( exp>=100 ){
            *(bufpt++) = (exp/100)+'0';                /* 100's digit */
            exp %= 100;
          }
          *(bufpt++) = exp/10+'0';                     /* 10's digit */
          *(bufpt++) = exp%10+'0';                     /* 1's digit */
        }
        *bufpt = 0;

        /* The converted number is in buf[] and zero terminated. Output it.
        ** Note that the number is in the usual order, not reversed as with
        ** integer conversions. */
        length = bufpt-buf;
        bufpt = buf;

        /* Special case:  Add leading zeros if the flag_zeropad flag is
        ** set and we are not left justified */
        if( flag_zeropad && !flag_leftjustify && length < width){
          int i;
          int nPad = width - length;
          for(i=width; i>=nPad; i--){
            bufpt[i] = bufpt[i-nPad];
          }
          i = prefix!=0;
          while( nPad-- ) bufpt[i++] = '0';
          length = width;
        }
#endif
        break;
      case etSIZE:
        *(va_arg(ap,int*)) = count;
        length = width = 0;
        break;
      case etPERCENT:
        buf[0] = '%';
        bufpt = buf;
        length = 1;
        break;
      case etCHARLIT:
      case etCHARX:
        c = buf[0] = (xtype==etCHARX ? va_arg(ap,int) : *++fmt);
        if( precision>=0 ){
          for(idx=1; idx<precision; idx++) buf[idx] = c;
          length = precision;
        }else{
          length =1;
        }
        bufpt = buf;
        break;
      case etSTRING:
      case etDYNSTRING:
        bufpt = va_arg(ap,char*);
        if( bufpt==0 ){
          bufpt = "";
        }else if( xtype==etDYNSTRING ){
          zExtra = bufpt;
        }
        length = strlen(bufpt);
        if( precision>=0 && precision<length ) length = precision;
        break;
      case etSQLESCAPE:
      case etSQLESCAPE2:
      case etSQLESCAPE3: {
        int i, j, n, ch, isnull;
        int needQuote;
        char q = ((xtype==etSQLESCAPE3)?'"':'\'');   /* Quote character */
        char *escarg = va_arg(ap,char*);
        isnull = escarg==0;
        if( isnull ) escarg = (xtype==etSQLESCAPE2 ? "NULL" : "(NULL)");
        for(i=n=0; (ch=escarg[i])!=0; i++){
          if( ch==q )  n++;
        }
        needQuote = !isnull && xtype==etSQLESCAPE2;
        n += i + 1 + needQuote*2;
        if( n>etBUFSIZE ){
          bufpt = zExtra = sqliteMalloc( n );
          if( bufpt==0 ) return -1;
        }else{
          bufpt = buf;
        }
        j = 0;
        if( needQuote ) bufpt[j++] = q;
        for(i=0; (ch=escarg[i])!=0; i++){
          bufpt[j++] = ch;
          if( ch==q ) bufpt[j++] = ch;
        }
        if( needQuote ) bufpt[j++] = q;
        bufpt[j] = 0;
        length = j;
        /* The precision is ignored on %q and %Q */
        /* if( precision>=0 && precision<length ) length = precision; */
        break;
      }
      case etTOKEN: {
        Token *pToken = va_arg(ap, Token*);
        if( pToken && pToken->z ){
          (*func)(arg, (char*)pToken->z, pToken->n);
        }
        length = width = 0;
        break;
      }
      case etSRCLIST: {
        SrcList *pSrc = va_arg(ap, SrcList*);
        int k = va_arg(ap, int);
        struct SrcList_item *pItem = &pSrc->a[k];
        assert( k>=0 && k<pSrc->nSrc );
        if( pItem->zDatabase && pItem->zDatabase[0] ){
          (*func)(arg, pItem->zDatabase, strlen(pItem->zDatabase));
          (*func)(arg, ".", 1);
        }
        (*func)(arg, pItem->zName, strlen(pItem->zName));
        length = width = 0;
        break;
      }
    }/* End switch over the format type */
    /*
    ** The text of the conversion is pointed to by "bufpt" and is
    ** "length" characters long.  The field width is "width".  Do
    ** the output.
    */
    if( !flag_leftjustify ){
      register int nspace;
      nspace = width-length;
      if( nspace>0 ){
        count += nspace;
        while( nspace>=etSPACESIZE ){
          (*func)(arg,spaces,etSPACESIZE);
          nspace -= etSPACESIZE;
        }
        if( nspace>0 ) (*func)(arg,spaces,nspace);
      }
    }
    if( length>0 ){
      (*func)(arg,bufpt,length);
      count += length;
    }
    if( flag_leftjustify ){
      register int nspace;
      nspace = width-length;
      if( nspace>0 ){
        count += nspace;
        while( nspace>=etSPACESIZE ){
          (*func)(arg,spaces,etSPACESIZE);
          nspace -= etSPACESIZE;
        }
        if( nspace>0 ) (*func)(arg,spaces,nspace);
      }
    }
    if( zExtra ){
      sqliteFree(zExtra);
    }
  }/* End for loop over the format string */
  return errorflag ? -1 : count;
} /* End of function */


/* This structure is used to store state information about the
** write to memory that is currently in progress.
*/
struct sgMprintf {
  char *zBase;     /* A base allocation */
  char *zText;     /* The string collected so far */
  int  nChar;      /* Length of the string so far */
  int  nTotal;     /* Output size if unconstrained */
  int  nAlloc;     /* Amount of space allocated in zText */
  void *(*xRealloc)(void*,int);  /* Function used to realloc memory */
};

/* 
** This function implements the callback from vxprintf. 
**
** This routine add nNewChar characters of text in zNewText to
** the sgMprintf structure pointed to by "arg".
*/
static void mout(void *arg, const char *zNewText, int nNewChar){
  struct sgMprintf *pM = (struct sgMprintf*)arg;
  pM->nTotal += nNewChar;
  if( pM->nChar + nNewChar + 1 > pM->nAlloc ){
    if( pM->xRealloc==0 ){
      nNewChar =  pM->nAlloc - pM->nChar - 1;
    }else{
      int nAlloc = pM->nChar + nNewChar*2 + 1;
      if( pM->zText==pM->zBase ){
        pM->zText = pM->xRealloc(0, nAlloc);
        if( pM->zText && pM->nChar ){
          memcpy(pM->zText, pM->zBase, pM->nChar);
        }
      }else{
        char *zNew;
        zNew = pM->xRealloc(pM->zText, nAlloc);
        if( zNew ){
          pM->zText = zNew;
        }else{
          return;
        }
      }
      pM->nAlloc = nAlloc;
    }
  }
  if( pM->zText ){
    if( nNewChar>0 ){
      memcpy(&pM->zText[pM->nChar], zNewText, nNewChar);
      pM->nChar += nNewChar;
    }
    pM->zText[pM->nChar] = 0;
  }
}

/*
** This routine is a wrapper around xprintf() that invokes mout() as
** the consumer.  
*/
static char *base_vprintf(
  void *(*xRealloc)(void*,int),   /* Routine to realloc memory. May be NULL */
  int useInternal,                /* Use internal %-conversions if true */
  char *zInitBuf,                 /* Initially write here, before mallocing */
  int nInitBuf,                   /* Size of zInitBuf[] */
  const char *zFormat,            /* format string */
  va_list ap                      /* arguments */
){
  struct sgMprintf sM;
  sM.zBase = sM.zText = zInitBuf;
  sM.nChar = sM.nTotal = 0;
  sM.nAlloc = nInitBuf;
  sM.xRealloc = xRealloc;
  vxprintf(mout, &sM, useInternal, zFormat, ap);
  if( xRealloc ){
    if( sM.zText==sM.zBase ){
      sM.zText = xRealloc(0, sM.nChar+1);
      if( sM.zText ){
        memcpy(sM.zText, sM.zBase, sM.nChar+1);
      }
    }else if( sM.nAlloc>sM.nChar+10 ){
      char *zNew = xRealloc(sM.zText, sM.nChar+1);
      if( zNew ){
        sM.zText = zNew;
      }
    }
  }
  return sM.zText;
}

/*
** Realloc that is a real function, not a macro.
*/
static void *printf_realloc(void *old, int size){
  return sqliteRealloc(old,size);
}

/*
** Print into memory obtained from sqliteMalloc().  Use the internal
** %-conversion extensions.
*/
SQLITE_PRIVATE char *sqlite3VMPrintf(const char *zFormat, va_list ap){
  char zBase[SQLITE_PRINT_BUF_SIZE];
  return base_vprintf(printf_realloc, 1, zBase, sizeof(zBase), zFormat, ap);
}

/*
** Print into memory obtained from sqliteMalloc().  Use the internal
** %-conversion extensions.
*/
SQLITE_PRIVATE char *sqlite3MPrintf(const char *zFormat, ...){
  va_list ap;
  char *z;
  char zBase[SQLITE_PRINT_BUF_SIZE];
  va_start(ap, zFormat);
  z = base_vprintf(printf_realloc, 1, zBase, sizeof(zBase), zFormat, ap);
  va_end(ap);
  return z;
}

/*
** Print into memory obtained from sqlite3_malloc().  Omit the internal
** %-conversion extensions.
*/
SQLITE_API char *sqlite3_vmprintf(const char *zFormat, va_list ap){
  char zBase[SQLITE_PRINT_BUF_SIZE];
  return base_vprintf(sqlite3_realloc, 0, zBase, sizeof(zBase), zFormat, ap);
}

/*
** Print into memory obtained from sqlite3_malloc()().  Omit the internal
** %-conversion extensions.
*/
SQLITE_API char *sqlite3_mprintf(const char *zFormat, ...){
  va_list ap;
  char *z;
  va_start(ap, zFormat);
  z = sqlite3_vmprintf(zFormat, ap);
  va_end(ap);
  return z;
}

/*
** sqlite3_snprintf() works like snprintf() except that it ignores the
** current locale settings.  This is important for SQLite because we
** are not able to use a "," as the decimal point in place of "." as
** specified by some locales.
*/
SQLITE_API char *sqlite3_snprintf(int n, char *zBuf, const char *zFormat, ...){
  char *z;
  va_list ap;

  if( n<=0 ){
    return zBuf;
  }
  zBuf[0] = 0;
  va_start(ap,zFormat);
  z = base_vprintf(0, 0, zBuf, n, zFormat, ap);
  va_end(ap);
  return z;
}

#if defined(SQLITE_TEST) || defined(SQLITE_DEBUG) || defined(SQLITE_MEMDEBUG)
/*
** A version of printf() that understands %lld.  Used for debugging.
** The printf() built into some versions of windows does not understand %lld
** and segfaults if you give it a long long int.
*/
SQLITE_PRIVATE void sqlite3DebugPrintf(const char *zFormat, ...){
  extern int getpid(void);
  va_list ap;
  char zBuf[500];
  va_start(ap, zFormat);
  base_vprintf(0, 0, zBuf, sizeof(zBuf), zFormat, ap);
  va_end(ap);
  fprintf(stdout,"%s", zBuf);
  fflush(stdout);
}
#endif

/************** End of printf.c **********************************************/
/************** Begin file random.c ******************************************/
/*
** 2001 September 15
**
** The author disclaims copyright to this source code.  In place of
** a legal notice, here is a blessing:
**
**    May you do good and not evil.
**    May you find forgiveness for yourself and forgive others.
**    May you share freely, never taking more than you give.
**
*************************************************************************
** This file contains code to implement a pseudo-random number
** generator (PRNG) for SQLite.
**
** Random numbers are used by some of the database backends in order
** to generate random integer keys for tables or random filenames.
**
** $Id: sqlite3.c,v 1.4 2007/07/21 08:51:43 rmsimpson Exp $
*/


/*
** Get a single 8-bit random value from the RC4 PRNG.  The Mutex
** must be held while executing this routine.
**
** Why not just use a library random generator like lrand48() for this?
** Because the OP_NewRowid opcode in the VDBE depends on having a very
** good source of random numbers.  The lrand48() library function may
** well be good enough.  But maybe not.  Or maybe lrand48() has some
** subtle problems on some systems that could cause problems.  It is hard
** to know.  To minimize the risk of problems due to bad lrand48()
** implementations, SQLite uses this random number generator based
** on RC4, which we know works very well.
**
** (Later):  Actually, OP_NewRowid does not depend on a good source of
** randomness any more.  But we will leave this code in all the same.
*/
static int randomByte(void){
  unsigned char t;

  /* All threads share a single random number generator.
  ** This structure is the current state of the generator.
  */
  static struct {
    unsigned char isInit;          /* True if initialized */
    unsigned char i, j;            /* State variables */
    unsigned char s[256];          /* State variables */
  } prng;

  /* Initialize the state of the random number generator once,
  ** the first time this routine is called.  The seed value does
  ** not need to contain a lot of randomness since we are not
  ** trying to do secure encryption or anything like that...
  **
  ** Nothing in this file or anywhere else in SQLite does any kind of
  ** encryption.  The RC4 algorithm is being used as a PRNG (pseudo-random
  ** number generator) not as an encryption device.
  */
  if( !prng.isInit ){
    int i;
    char k[256];
    prng.j = 0;
    prng.i = 0;
    sqlite3OsRandomSeed(k);
    for(i=0; i<256; i++){
      prng.s[i] = i;
    }
    for(i=0; i<256; i++){
      prng.j += prng.s[i] + k[i];
      t = prng.s[prng.j];
      prng.s[prng.j] = prng.s[i];
      prng.s[i] = t;
    }
    prng.isInit = 1;
  }

  /* Generate and return single random byte
  */
  prng.i++;
  t = prng.s[prng.i];
  prng.j += t;
  prng.s[prng.i] = prng.s[prng.j];
  prng.s[prng.j] = t;
  t += prng.s[prng.i];
  return prng.s[t];
}

/*
** Return N random bytes.
*/
SQLITE_PRIVATE void sqlite3Randomness(int N, void *pBuf){
  unsigned char *zBuf = pBuf;
  sqlite3OsEnterMutex();
  while( N-- ){
    *(zBuf++) = randomByte();
  }
  sqlite3OsLeaveMutex();
}

/************** End of random.c **********************************************/
/************** Begin file utf.c *********************************************/
/*
** 2004 April 13
**
** The author disclaims copyright to this source code.  In place of
** a legal notice, here is a blessing:
**
**    May you do good and not evil.
**    May you find forgiveness for yourself and forgive others.
**    May you share freely, never taking more than you give.
**
*************************************************************************
** This file contains routines used to translate between UTF-8, 
** UTF-16, UTF-16BE, and UTF-16LE.
**
** $Id: sqlite3.c,v 1.4 2007/07/21 08:51:43 rmsimpson Exp $
**
** Notes on UTF-8:
**
**   Byte-0    Byte-1    Byte-2    Byte-3    Value
**  0xxxxxxx                                 00000000 00000000 0xxxxxxx
**  110yyyyy  10xxxxxx                       00000000 00000yyy yyxxxxxx
**  1110zzzz  10yyyyyy  10xxxxxx             00000000 zzzzyyyy yyxxxxxx
**  11110uuu  10uuzzzz  10yyyyyy  10xxxxxx   000uuuuu zzzzyyyy yyxxxxxx
**
**
** Notes on UTF-16:  (with wwww+1==uuuuu)
**
**      Word-0               Word-1          Value
**  110110ww wwzzzzyy   110111yy yyxxxxxx    000uuuuu zzzzyyyy yyxxxxxx
**  zzzzyyyy yyxxxxxx                        00000000 zzzzyyyy yyxxxxxx
**
**
** BOM or Byte Order Mark:
**     0xff 0xfe   little-endian utf-16 follows
**     0xfe 0xff   big-endian utf-16 follows
**
*/
/************** Include vdbeInt.h in the middle of utf.c *********************/
/************** Begin file vdbeInt.h *****************************************/
/*
** 2003 September 6
**
** The author disclaims copyright to this source code.  In place of
** a legal notice, here is a blessing:
**
**    May you do good and not evil.
**    May you find forgiveness for yourself and forgive others.
**    May you share freely, never taking more than you give.
**
*************************************************************************
** This is the header file for information that is private to the
** VDBE.  This information used to all be at the top of the single
** source code file "vdbe.c".  When that file became too big (over
** 6000 lines long) it was split up into several smaller files and
** this header information was factored out.
*/
#ifndef _VDBEINT_H_
#define _VDBEINT_H_

/*
** intToKey() and keyToInt() used to transform the rowid.  But with
** the latest versions of the design they are no-ops.
*/
#define keyToInt(X)   (X)
#define intToKey(X)   (X)

/*
** The makefile scans the vdbe.c source file and creates the following
** array of string constants which are the names of all VDBE opcodes.  This
** array is defined in a separate source code file named opcode.c which is
** automatically generated by the makefile.
*/
extern const char *const sqlite3OpcodeNames[];

/*
** SQL is translated into a sequence of instructions to be
** executed by a virtual machine.  Each instruction is an instance
** of the following structure.
*/
typedef struct VdbeOp Op;

/*
** Boolean values
*/
typedef unsigned char Bool;

/*
** A cursor is a pointer into a single BTree within a database file.
** The cursor can seek to a BTree entry with a particular key, or
** loop over all entries of the Btree.  You can also insert new BTree
** entries or retrieve the key or data from the entry that the cursor
** is currently pointing to.
** 
** Every cursor that the virtual machine has open is represented by an
** instance of the following structure.
**
** If the Cursor.isTriggerRow flag is set it means that this cursor is
** really a single row that represents the NEW or OLD pseudo-table of
** a row trigger.  The data for the row is stored in Cursor.pData and
** the rowid is in Cursor.iKey.
*/
struct Cursor {
  BtCursor *pCursor;    /* The cursor structure of the backend */
  int iDb;              /* Index of cursor database in db->aDb[] (or -1) */
  i64 lastRowid;        /* Last rowid from a Next or NextIdx operation */
  i64 nextRowid;        /* Next rowid returned by OP_NewRowid */
  Bool zeroed;          /* True if zeroed out and ready for reuse */
  Bool rowidIsValid;    /* True if lastRowid is valid */
  Bool atFirst;         /* True if pointing to first entry */
  Bool useRandomRowid;  /* Generate new record numbers semi-randomly */
  Bool nullRow;         /* True if pointing to a row with no data */
  Bool nextRowidValid;  /* True if the nextRowid field is valid */
  Bool pseudoTable;     /* This is a NEW or OLD pseudo-tables of a trigger */
  Bool deferredMoveto;  /* A call to sqlite3BtreeMoveto() is needed */
  Bool isTable;         /* True if a table requiring integer keys */
  Bool isIndex;         /* True if an index containing keys only - no data */
  u8 bogusIncrKey;      /* Something for pIncrKey to point to if pKeyInfo==0 */
  i64 movetoTarget;     /* Argument to the deferred sqlite3BtreeMoveto() */
  Btree *pBt;           /* Separate file holding temporary table */
  int nData;            /* Number of bytes in pData */
  char *pData;          /* Data for a NEW or OLD pseudo-table */
  i64 iKey;             /* Key for the NEW or OLD pseudo-table row */
  u8 *pIncrKey;         /* Pointer to pKeyInfo->incrKey */
  KeyInfo *pKeyInfo;    /* Info about index keys needed by index cursors */
  int nField;           /* Number of fields in the header */
  i64 seqCount;         /* Sequence counter */
  sqlite3_vtab_cursor *pVtabCursor;  /* The cursor for a virtual table */
  const sqlite3_module *pModule;     /* Module for cursor pVtabCursor */

  /* Cached information about the header for the data record that the
  ** cursor is currently pointing to.  Only valid if cacheValid is true.
  ** aRow might point to (ephemeral) data for the current row, or it might
  ** be NULL.
  */
  int cacheStatus;      /* Cache is valid if this matches Vdbe.cacheCtr */
  int payloadSize;      /* Total number of bytes in the record */
  u32 *aType;           /* Type values for all entries in the record */
  u32 *aOffset;         /* Cached offsets to the start of each columns data */
  u8 *aRow;             /* Data for the current row, if all on one page */
};
typedef struct Cursor Cursor;

/*
** Number of bytes of string storage space available to each stack
** layer without having to malloc.  NBFS is short for Number of Bytes
** For Strings.
*/
#define NBFS 32

/*
** A value for Cursor.cacheValid that means the cache is always invalid.
*/
#define CACHE_STALE 0

/*
** Internally, the vdbe manipulates nearly all SQL values as Mem
** structures. Each Mem struct may cache multiple representations (string,
** integer etc.) of the same value.  A value (and therefore Mem structure)
** has the following properties:
**
** Each value has a manifest type. The manifest type of the value stored
** in a Mem struct is returned by the MemType(Mem*) macro. The type is
** one of SQLITE_NULL, SQLITE_INTEGER, SQLITE_REAL, SQLITE_TEXT or
** SQLITE_BLOB.
*/
struct Mem {
  union {
    i64 i;              /* Integer value. Or FuncDef* when flags==MEM_Agg */
    FuncDef *pDef;      /* Used only when flags==MEM_Agg */
  } u;
  double r;           /* Real value */
  char *z;            /* String or BLOB value */
  int n;              /* Number of characters in string value, including '\0' */
  u16 flags;          /* Some combination of MEM_Null, MEM_Str, MEM_Dyn, etc. */
  u8  type;           /* One of SQLITE_NULL, SQLITE_TEXT, SQLITE_INTEGER, etc */
  u8  enc;            /* SQLITE_UTF8, SQLITE_UTF16BE, SQLITE_UTF16LE */
  void (*xDel)(void *);  /* If not null, call this function to delete Mem.z */
  char zShort[NBFS];  /* Space for short strings */
};
typedef struct Mem Mem;

/* One or more of the following flags are set to indicate the validOK
** representations of the value stored in the Mem struct.
**
** If the MEM_Null flag is set, then the value is an SQL NULL value.
** No other flags may be set in this case.
**
** If the MEM_Str flag is set then Mem.z points at a string representation.
** Usually this is encoded in the same unicode encoding as the main
** database (see below for exceptions). If the MEM_Term flag is also
** set, then the string is nul terminated. The MEM_Int and MEM_Real 
** flags may coexist with the MEM_Str flag.
**
** Multiple of these values can appear in Mem.flags.  But only one
** at a time can appear in Mem.type.
*/
#define MEM_Null      0x0001   /* Value is NULL */
#define MEM_Str       0x0002   /* Value is a string */
#define MEM_Int       0x0004   /* Value is an integer */
#define MEM_Real      0x0008   /* Value is a real number */
#define MEM_Blob      0x0010   /* Value is a BLOB */

/* Whenever Mem contains a valid string or blob representation, one of
** the following flags must be set to determine the memory management
** policy for Mem.z.  The MEM_Term flag tells us whether or not the
** string is \000 or \u0000 terminated
*/
#define MEM_Term      0x0020   /* String rep is nul terminated */
#define MEM_Dyn       0x0040   /* Need to call sqliteFree() on Mem.z */
#define MEM_Static    0x0080   /* Mem.z points to a static string */
#define MEM_Ephem     0x0100   /* Mem.z points to an ephemeral string */
#define MEM_Short     0x0200   /* Mem.z points to Mem.zShort */
#define MEM_Agg       0x0400   /* Mem.z points to an agg function context */
#define MEM_Zero      0x0800   /* Mem.i contains count of 0s appended to blob */

#ifdef SQLITE_OMIT_INCRBLOB
  #undef MEM_Zero
  #define MEM_Zero 0x0000
#endif


/* A VdbeFunc is just a FuncDef (defined in sqliteInt.h) that contains
** additional information about auxiliary information bound to arguments
** of the function.  This is used to implement the sqlite3_get_auxdata()
** and sqlite3_set_auxdata() APIs.  The "auxdata" is some auxiliary data
** that can be associated with a constant argument to a function.  This
** allows functions such as "regexp" to compile their constant regular
** expression argument once and reused the compiled code for multiple
** invocations.
*/
struct VdbeFunc {
  FuncDef *pFunc;               /* The definition of the function */
  int nAux;                     /* Number of entries allocated for apAux[] */
  struct AuxData {
    void *pAux;                   /* Aux data for the i-th argument */
    void (*xDelete)(void *);      /* Destructor for the aux data */
  } apAux[1];                   /* One slot for each function argument */
};
typedef struct VdbeFunc VdbeFunc;

/*
** The "context" argument for a installable function.  A pointer to an
** instance of this structure is the first argument to the routines used
** implement the SQL functions.
**
** There is a typedef for this structure in sqlite.h.  So all routines,
** even the public interface to SQLite, can use a pointer to this structure.
** But this file is the only place where the internal details of this
** structure are known.
**
** This structure is defined inside of vdbeInt.h because it uses substructures
** (Mem) which are only defined there.
*/
struct sqlite3_context {
  FuncDef *pFunc;       /* Pointer to function information.  MUST BE FIRST */
  VdbeFunc *pVdbeFunc;  /* Auxilary data, if created. */
  Mem s;                /* The return value is stored here */
  Mem *pMem;            /* Memory cell used to store aggregate context */
  u8 isError;           /* Set to true for an error */
  CollSeq *pColl;       /* Collating sequence */
};

/*
** A Set structure is used for quick testing to see if a value
** is part of a small set.  Sets are used to implement code like
** this:
**            x.y IN ('hi','hoo','hum')
*/
typedef struct Set Set;
struct Set {
  Hash hash;             /* A set is just a hash table */
  HashElem *prev;        /* Previously accessed hash elemen */
};

/*
** A FifoPage structure holds a single page of valves.  Pages are arranged
** in a list.
*/
typedef struct FifoPage FifoPage;
struct FifoPage {
  int nSlot;         /* Number of entries aSlot[] */
  int iWrite;        /* Push the next value into this entry in aSlot[] */
  int iRead;         /* Read the next value from this entry in aSlot[] */
  FifoPage *pNext;   /* Next page in the fifo */
  i64 aSlot[1];      /* One or more slots for rowid values */
};

/*
** The Fifo structure is typedef-ed in vdbeInt.h.  But the implementation
** of that structure is private to this file.
**
** The Fifo structure describes the entire fifo.  
*/
typedef struct Fifo Fifo;
struct Fifo {
  int nEntry;         /* Total number of entries */
  FifoPage *pFirst;   /* First page on the list */
  FifoPage *pLast;    /* Last page on the list */
};

/*
** A Context stores the last insert rowid, the last statement change count,
** and the current statement change count (i.e. changes since last statement).
** The current keylist is also stored in the context.
** Elements of Context structure type make up the ContextStack, which is
** updated by the ContextPush and ContextPop opcodes (used by triggers).
** The context is pushed before executing a trigger a popped when the
** trigger finishes.
*/
typedef struct Context Context;
struct Context {
  i64 lastRowid;    /* Last insert rowid (sqlite3.lastRowid) */
  int nChange;      /* Statement changes (Vdbe.nChanges)     */
  Fifo sFifo;       /* Records that will participate in a DELETE or UPDATE */
};

/*
** An instance of the virtual machine.  This structure contains the complete
** state of the virtual machine.
**
** The "sqlite3_stmt" structure pointer that is returned by sqlite3_compile()
** is really a pointer to an instance of this structure.
**
** The Vdbe.inVtabMethod variable is set to non-zero for the duration of
** any virtual table method invocations made by the vdbe program. It is
** set to 2 for xDestroy method calls and 1 for all other methods. This
** variable is used for two purposes: to allow xDestroy methods to execute
** "DROP TABLE" statements and to prevent some nasty side effects of
** malloc failure when SQLite is invoked recursively by a virtual table 
** method function.
*/
struct Vdbe {
  sqlite3 *db;        /* The whole database */
  Vdbe *pPrev,*pNext; /* Linked list of VDBEs with the same Vdbe.db */
  int nOp;            /* Number of instructions in the program */
  int nOpAlloc;       /* Number of slots allocated for aOp[] */
  Op *aOp;            /* Space to hold the virtual machine's program */
  int nLabel;         /* Number of labels used */
  int nLabelAlloc;    /* Number of slots allocated in aLabel[] */
  int *aLabel;        /* Space to hold the labels */
  Mem *aStack;        /* The operand stack, except string values */
  Mem *pTos;          /* Top entry in the operand stack */
  Mem **apArg;        /* Arguments to currently executing user function */
  Mem *aColName;      /* Column names to return */
  int nCursor;        /* Number of slots in apCsr[] */
  Cursor **apCsr;     /* One element of this array for each open cursor */
  int nVar;           /* Number of entries in aVar[] */
  Mem *aVar;          /* Values for the OP_Variable opcode. */
  char **azVar;       /* Name of variables */
  int okVar;          /* True if azVar[] has been initialized */
  int magic;              /* Magic number for sanity checking */
  int nMem;               /* Number of memory locations currently allocated */
  Mem *aMem;              /* The memory locations */
  int nCallback;          /* Number of callbacks invoked so far */
  int cacheCtr;           /* Cursor row cache generation counter */
  Fifo sFifo;             /* A list of ROWIDs */
  int contextStackTop;    /* Index of top element in the context stack */
  int contextStackDepth;  /* The size of the "context" stack */
  Context *contextStack;  /* Stack used by opcodes ContextPush & ContextPop*/
  int pc;                 /* The program counter */
  int rc;                 /* Value to return */
  unsigned uniqueCnt;     /* Used by OP_MakeRecord when P2!=0 */
  int errorAction;        /* Recovery action to do in case of an error */
  int inTempTrans;        /* True if temp database is transactioned */
  int returnStack[100];   /* Return address stack for OP_Gosub & OP_Return */
  int returnDepth;        /* Next unused element in returnStack[] */
  int nResColumn;         /* Number of columns in one row of the result set */
  char **azResColumn;     /* Values for one row of result */ 
  int popStack;           /* Pop the stack this much on entry to VdbeExec() */
  char *zErrMsg;          /* Error message written here */
  u8 resOnStack;          /* True if there are result values on the stack */
  u8 explain;             /* True if EXPLAIN present on SQL command */
  u8 changeCntOn;         /* True to update the change-counter */
  u8 aborted;             /* True if ROLLBACK in another VM causes an abort */
  u8 expired;             /* True if the VM needs to be recompiled */
  u8 minWriteFileFormat;  /* Minimum file format for writable database files */
  u8 inVtabMethod;        /* See comments above */
  int nChange;            /* Number of db changes made since last reset */
  i64 startTime;          /* Time when query started - used for profiling */
  int nSql;             /* Number of bytes in zSql */
  char *zSql;           /* Text of the SQL statement that generated this */
#ifdef SQLITE_DEBUG
  FILE *trace;        /* Write an execution trace here, if not NULL */
#endif
  int openedStatement;  /* True if this VM has opened a statement journal */
#ifdef SQLITE_SSE
  int fetchId;          /* Statement number used by sqlite3_fetch_statement */
  int lru;              /* Counter used for LRU cache replacement */
#endif
};

/*
** The following are allowed values for Vdbe.magic
*/
#define VDBE_MAGIC_INIT     0x26bceaa5    /* Building a VDBE program */
#define VDBE_MAGIC_RUN      0xbdf20da3    /* VDBE is ready to execute */
#define VDBE_MAGIC_HALT     0x519c2973    /* VDBE has completed execution */
#define VDBE_MAGIC_DEAD     0xb606c3c8    /* The VDBE has been deallocated */

/*
** Function prototypes
*/
SQLITE_PRIVATE void sqlite3VdbeFreeCursor(Vdbe *, Cursor*);
void sqliteVdbePopStack(Vdbe*,int);
SQLITE_PRIVATE int sqlite3VdbeCursorMoveto(Cursor*);
#if defined(SQLITE_DEBUG) || defined(VDBE_PROFILE)
SQLITE_PRIVATE void sqlite3VdbePrintOp(FILE*, int, Op*);
#endif
SQLITE_PRIVATE int sqlite3VdbeSerialTypeLen(u32);
SQLITE_PRIVATE u32 sqlite3VdbeSerialType(Mem*, int);
SQLITE_PRIVATE int sqlite3VdbeSerialPut(unsigned char*, int, Mem*, int);
SQLITE_PRIVATE int sqlite3VdbeSerialGet(const unsigned char*, u32, Mem*);
SQLITE_PRIVATE void sqlite3VdbeDeleteAuxData(VdbeFunc*, int);

int sqlite2BtreeKeyCompare(BtCursor *, const void *, int, int, int *);
SQLITE_PRIVATE int sqlite3VdbeIdxKeyCompare(Cursor*, int , const unsigned char*, int*);
SQLITE_PRIVATE int sqlite3VdbeIdxRowid(BtCursor *, i64 *);
SQLITE_PRIVATE int sqlite3MemCompare(const Mem*, const Mem*, const CollSeq*);
SQLITE_PRIVATE int sqlite3VdbeRecordCompare(void*,int,const void*,int, const void*);
SQLITE_PRIVATE int sqlite3VdbeIdxRowidLen(const u8*);
SQLITE_PRIVATE int sqlite3VdbeExec(Vdbe*);
SQLITE_PRIVATE int sqlite3VdbeList(Vdbe*);
SQLITE_PRIVATE int sqlite3VdbeHalt(Vdbe*);
SQLITE_PRIVATE int sqlite3VdbeChangeEncoding(Mem *, int);
SQLITE_PRIVATE int sqlite3VdbeMemTooBig(Mem*);
SQLITE_PRIVATE int sqlite3VdbeMemCopy(Mem*, const Mem*);
SQLITE_PRIVATE void sqlite3VdbeMemShallowCopy(Mem*, const Mem*, int);
SQLITE_PRIVATE int sqlite3VdbeMemMove(Mem*, Mem*);
SQLITE_PRIVATE int sqlite3VdbeMemNulTerminate(Mem*);
SQLITE_PRIVATE int sqlite3VdbeMemSetStr(Mem*, const char*, int, u8, void(*)(void*));
SQLITE_PRIVATE void sqlite3VdbeMemSetInt64(Mem*, i64);
SQLITE_PRIVATE void sqlite3VdbeMemSetDouble(Mem*, double);
SQLITE_PRIVATE void sqlite3VdbeMemSetNull(Mem*);
SQLITE_PRIVATE void sqlite3VdbeMemSetZeroBlob(Mem*,int);
SQLITE_PRIVATE int sqlite3VdbeMemMakeWriteable(Mem*);
SQLITE_PRIVATE int sqlite3VdbeMemDynamicify(Mem*);
SQLITE_PRIVATE int sqlite3VdbeMemStringify(Mem*, int);
SQLITE_PRIVATE i64 sqlite3VdbeIntValue(Mem*);
SQLITE_PRIVATE int sqlite3VdbeMemIntegerify(Mem*);
SQLITE_PRIVATE double sqlite3VdbeRealValue(Mem*);
SQLITE_PRIVATE void sqlite3VdbeIntegerAffinity(Mem*);
SQLITE_PRIVATE int sqlite3VdbeMemRealify(Mem*);
SQLITE_PRIVATE int sqlite3VdbeMemNumerify(Mem*);
SQLITE_PRIVATE int sqlite3VdbeMemFromBtree(BtCursor*,int,int,int,Mem*);
SQLITE_PRIVATE void sqlite3VdbeMemRelease(Mem *p);
SQLITE_PRIVATE int sqlite3VdbeMemFinalize(Mem*, FuncDef*);
#ifndef NDEBUG
SQLITE_PRIVATE   void sqlite3VdbeMemSanity(Mem*);
SQLITE_PRIVATE   int sqlite3VdbeOpcodeNoPush(u8);
#endif
SQLITE_PRIVATE int sqlite3VdbeMemTranslate(Mem*, u8);
#ifdef SQLITE_DEBUG
SQLITE_PRIVATE   void sqlite3VdbePrintSql(Vdbe*);
SQLITE_PRIVATE   void sqlite3VdbeMemPrettyPrint(Mem *pMem, char *zBuf);
#endif
SQLITE_PRIVATE int sqlite3VdbeMemHandleBom(Mem *pMem);
SQLITE_PRIVATE void sqlite3VdbeFifoInit(Fifo*);
SQLITE_PRIVATE int sqlite3VdbeFifoPush(Fifo*, i64);
SQLITE_PRIVATE int sqlite3VdbeFifoPop(Fifo*, i64*);
SQLITE_PRIVATE void sqlite3VdbeFifoClear(Fifo*);

#ifndef SQLITE_OMIT_INCRBLOB
SQLITE_PRIVATE   int sqlite3VdbeMemExpandBlob(Mem *);
#else
  #define sqlite3VdbeMemExpandBlob(x) SQLITE_OK
#endif

#endif /* !defined(_VDBEINT_H_) */

/************** End of vdbeInt.h *********************************************/
/************** Continuing where we left off in utf.c ************************/

/*
** The following constant value is used by the SQLITE_BIGENDIAN and
** SQLITE_LITTLEENDIAN macros.
*/
const int sqlite3one = 1;

/*
** This lookup table is used to help decode the first byte of
** a multi-byte UTF8 character.
*/
const unsigned char sqlite3UtfTrans1[] = {
  0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07,
  0x08, 0x09, 0x0a, 0x0b, 0x0c, 0x0d, 0x0e, 0x0f,
  0x10, 0x11, 0x12, 0x13, 0x14, 0x15, 0x16, 0x17,
  0x18, 0x19, 0x1a, 0x1b, 0x1c, 0x1d, 0x1e, 0x1f,
  0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07,
  0x08, 0x09, 0x0a, 0x0b, 0x0c, 0x0d, 0x0e, 0x0f,
  0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07,
  0x00, 0x01, 0x02, 0x03, 0x00, 0x01, 0x00, 0x00,
};

#define WRITE_UTF8(zOut, c) {                          \
  if( c<0x00080 ){                                     \
    *zOut++ = (c&0xFF);                                \
  }                                                    \
  else if( c<0x00800 ){                                \
    *zOut++ = 0xC0 + ((c>>6)&0x1F);                    \
    *zOut++ = 0x80 + (c & 0x3F);                       \
  }                                                    \
  else if( c<0x10000 ){                                \
    *zOut++ = 0xE0 + ((c>>12)&0x0F);                   \
    *zOut++ = 0x80 + ((c>>6) & 0x3F);                  \
    *zOut++ = 0x80 + (c & 0x3F);                       \
  }else{                                               \
    *zOut++ = 0xF0 + ((c>>18) & 0x07);                 \
    *zOut++ = 0x80 + ((c>>12) & 0x3F);                 \
    *zOut++ = 0x80 + ((c>>6) & 0x3F);                  \
    *zOut++ = 0x80 + (c & 0x3F);                       \
  }                                                    \
}

#define WRITE_UTF16LE(zOut, c) {                                \
  if( c<=0xFFFF ){                                              \
    *zOut++ = (c&0x00FF);                                       \
    *zOut++ = ((c>>8)&0x00FF);                                  \
  }else{                                                        \
    *zOut++ = (((c>>10)&0x003F) + (((c-0x10000)>>10)&0x00C0));  \
    *zOut++ = (0x00D8 + (((c-0x10000)>>18)&0x03));              \
    *zOut++ = (c&0x00FF);                                       \
    *zOut++ = (0x00DC + ((c>>8)&0x03));                         \
  }                                                             \
}

#define WRITE_UTF16BE(zOut, c) {                                \
  if( c<=0xFFFF ){                                              \
    *zOut++ = ((c>>8)&0x00FF);                                  \
    *zOut++ = (c&0x00FF);                                       \
  }else{                                                        \
    *zOut++ = (0x00D8 + (((c-0x10000)>>18)&0x03));              \
    *zOut++ = (((c>>10)&0x003F) + (((c-0x10000)>>10)&0x00C0));  \
    *zOut++ = (0x00DC + ((c>>8)&0x03));                         \
    *zOut++ = (c&0x00FF);                                       \
  }                                                             \
}

#define READ_UTF16LE(zIn, c){                                         \
  c = (*zIn++);                                                       \
  c += ((*zIn++)<<8);                                                 \
  if( c>=0xD800 && c<0xE000 ){                                       \
    int c2 = (*zIn++);                                                \
    c2 += ((*zIn++)<<8);                                              \
    c = (c2&0x03FF) + ((c&0x003F)<<10) + (((c&0x03C0)+0x0040)<<10);   \
    if( (c & 0xFFFF0000)==0 ) c = 0xFFFD;                             \
  }                                                                   \
}

#define READ_UTF16BE(zIn, c){                                         \
  c = ((*zIn++)<<8);                                                  \
  c += (*zIn++);                                                      \
  if( c>=0xD800 && c<0xE000 ){                                       \
    int c2 = ((*zIn++)<<8);                                           \
    c2 += (*zIn++);                                                   \
    c = (c2&0x03FF) + ((c&0x003F)<<10) + (((c&0x03C0)+0x0040)<<10);   \
    if( (c & 0xFFFF0000)==0 ) c = 0xFFFD;                             \
  }                                                                   \
}

/*
** If the TRANSLATE_TRACE macro is defined, the value of each Mem is
** printed on stderr on the way into and out of sqlite3VdbeMemTranslate().
*/ 
/* #define TRANSLATE_TRACE 1 */

#ifndef SQLITE_OMIT_UTF16
/*
** This routine transforms the internal text encoding used by pMem to
** desiredEnc. It is an error if the string is already of the desired
** encoding, or if *pMem does not contain a string value.
*/
SQLITE_PRIVATE int sqlite3VdbeMemTranslate(Mem *pMem, u8 desiredEnc){
  unsigned char zShort[NBFS]; /* Temporary short output buffer */
  int len;                    /* Maximum length of output string in bytes */
  unsigned char *zOut;                  /* Output buffer */
  unsigned char *zIn;                   /* Input iterator */
  unsigned char *zTerm;                 /* End of input */
  unsigned char *z;                     /* Output iterator */
  unsigned int c;

  assert( pMem->flags&MEM_Str );
  assert( pMem->enc!=desiredEnc );
  assert( pMem->enc!=0 );
  assert( pMem->n>=0 );

#if defined(TRANSLATE_TRACE) && defined(SQLITE_DEBUG)
  {
    char zBuf[100];
    sqlite3VdbeMemPrettyPrint(pMem, zBuf);
    fprintf(stderr, "INPUT:  %s\n", zBuf);
  }
#endif

  /* If the translation is between UTF-16 little and big endian, then 
  ** all that is required is to swap the byte order. This case is handled
  ** differently from the others.
  */
  if( pMem->enc!=SQLITE_UTF8 && desiredEnc!=SQLITE_UTF8 ){
    u8 temp;
    int rc;
    rc = sqlite3VdbeMemMakeWriteable(pMem);
    if( rc!=SQLITE_OK ){
      assert( rc==SQLITE_NOMEM );
      return SQLITE_NOMEM;
    }
    zIn = (u8*)pMem->z;
    zTerm = &zIn[pMem->n];
    while( zIn<zTerm ){
      temp = *zIn;
      *zIn = *(zIn+1);
      zIn++;
      *zIn++ = temp;
    }
    pMem->enc = desiredEnc;
    goto translate_out;
  }

  /* Set len to the maximum number of bytes required in the output buffer. */
  if( desiredEnc==SQLITE_UTF8 ){
    /* When converting from UTF-16, the maximum growth results from
    ** translating a 2-byte character to a 4-byte UTF-8 character.
    ** A single byte is required for the output string
    ** nul-terminator.
    */
    len = pMem->n * 2 + 1;
  }else{
    /* When converting from UTF-8 to UTF-16 the maximum growth is caused
    ** when a 1-byte UTF-8 character is translated into a 2-byte UTF-16
    ** character. Two bytes are required in the output buffer for the
    ** nul-terminator.
    */
    len = pMem->n * 2 + 2;
  }

  /* Set zIn to point at the start of the input buffer and zTerm to point 1
  ** byte past the end.
  **
  ** Variable zOut is set to point at the output buffer. This may be space
  ** obtained from malloc(), or Mem.zShort, if it large enough and not in
  ** use, or the zShort array on the stack (see above).
  */
  zIn = (u8*)pMem->z;
  zTerm = &zIn[pMem->n];
  if( len>NBFS ){
    zOut = sqliteMallocRaw(len);
    if( !zOut ) return SQLITE_NOMEM;
  }else{
    zOut = zShort;
  }
  z = zOut;

  if( pMem->enc==SQLITE_UTF8 ){
    unsigned int iExtra = 0xD800;

    if( 0==(pMem->flags&MEM_Term) && zTerm>zIn && (zTerm[-1]&0x80) ){
      /* This UTF8 string is not nul-terminated, and the last byte is
      ** not a character in the ascii range (codpoints 0..127). This
      ** means the SQLITE_READ_UTF8() macro might read past the end
      ** of the allocated buffer.
      **
      ** There are four possibilities:
      **
      **   1. The last byte is the first byte of a non-ASCII character,
      **
      **   2. The final N bytes of the input string are continuation bytes
      **      and immediately preceding them is the first byte of a 
      **      non-ASCII character.
      **
      **   3. The final N bytes of the input string are continuation bytes
      **      and immediately preceding them is a byte that encodes a 
      **      character in the ASCII range.
      **
      **   4. The entire string consists of continuation characters.
      **
      ** Cases (3) and (4) require no special handling. The SQLITE_READ_UTF8()
      ** macro will not overread the buffer in these cases.
      */
      unsigned char *zExtra = &zTerm[-1];
      while( zExtra>zIn && (zExtra[0]&0xC0)==0x80 ){
        zExtra--;
      }

      if( (zExtra[0]&0xC0)==0xC0 ){
        /* Make a copy of the last character encoding in the input string.
        ** Then make sure it is nul-terminated and use SQLITE_READ_UTF8()
        ** to decode the codepoint. Store the codepoint in variable iExtra,
        ** it will be appended to the output string later.
        */
        unsigned char *zFree = 0;
        unsigned char zBuf[16];
        int nExtra = (pMem->n+zIn-zExtra);
        zTerm = zExtra;
        if( nExtra>15 ){
          zExtra = sqliteMallocRaw(nExtra+1);
          if( !zExtra ){
            return SQLITE_NOMEM;
          }
          zFree = zExtra;
        }else{
          zExtra = zBuf;
        }
        memcpy(zExtra, zTerm, nExtra);
        zExtra[nExtra] = '\0';
        SQLITE_READ_UTF8(zExtra, iExtra);
        sqliteFree(zFree);
      }
    }

    if( desiredEnc==SQLITE_UTF16LE ){
      /* UTF-8 -> UTF-16 Little-endian */
      while( zIn<zTerm ){
        SQLITE_READ_UTF8(zIn, c); 
        WRITE_UTF16LE(z, c);
      }
      if( iExtra!=0xD800 ){
        WRITE_UTF16LE(z, iExtra);
      }
    }else{
      assert( desiredEnc==SQLITE_UTF16BE );
      /* UTF-8 -> UTF-16 Big-endian */
      while( zIn<zTerm ){
        SQLITE_READ_UTF8(zIn, c); 
        WRITE_UTF16BE(z, c);
      }
      if( iExtra!=0xD800 ){
        WRITE_UTF16BE(z, iExtra);
      }
    }
    pMem->n = z - zOut;
    *z++ = 0;
  }else{
    assert( desiredEnc==SQLITE_UTF8 );
    if( pMem->enc==SQLITE_UTF16LE ){
      /* UTF-16 Little-endian -> UTF-8 */
      while( zIn<zTerm ){
        READ_UTF16LE(zIn, c); 
        WRITE_UTF8(z, c);
      }
    }else{
      /* UTF-16 Little-endian -> UTF-8 */
      while( zIn<zTerm ){
        READ_UTF16BE(zIn, c); 
        WRITE_UTF8(z, c);
      }
    }
    pMem->n = z - zOut;
  }
  *z = 0;
  assert( (pMem->n+(desiredEnc==SQLITE_UTF8?1:2))<=len );

  sqlite3VdbeMemRelease(pMem);
  pMem->flags &= ~(MEM_Static|MEM_Dyn|MEM_Ephem|MEM_Short);
  pMem->enc = desiredEnc;
  if( zOut==zShort ){
    memcpy(pMem->zShort, zOut, len);
    zOut = (u8*)pMem->zShort;
    pMem->flags |= (MEM_Term|MEM_Short);
  }else{
    pMem->flags |= (MEM_Term|MEM_Dyn);
  }
  pMem->z = (char*)zOut;

translate_out:
#if defined(TRANSLATE_TRACE) && defined(SQLITE_DEBUG)
  {
    char zBuf[100];
    sqlite3VdbeMemPrettyPrint(pMem, zBuf);
    fprintf(stderr, "OUTPUT: %s\n", zBuf);
  }
#endif
  return SQLITE_OK;
}

/*
** This routine checks for a byte-order mark at the beginning of the 
** UTF-16 string stored in *pMem. If one is present, it is removed and
** the encoding of the Mem adjusted. This routine does not do any
** byte-swapping, it just sets Mem.enc appropriately.
**
** The allocation (static, dynamic etc.) and encoding of the Mem may be
** changed by this function.
*/
SQLITE_PRIVATE int sqlite3VdbeMemHandleBom(Mem *pMem){
  int rc = SQLITE_OK;
  u8 bom = 0;

  if( pMem->n<0 || pMem->n>1 ){
    u8 b1 = *(u8 *)pMem->z;
    u8 b2 = *(((u8 *)pMem->z) + 1);
    if( b1==0xFE && b2==0xFF ){
      bom = SQLITE_UTF16BE;
    }
    if( b1==0xFF && b2==0xFE ){
      bom = SQLITE_UTF16LE;
    }
  }
  
  if( bom ){
    /* This function is called as soon as a string is stored in a Mem*,
    ** from within sqlite3VdbeMemSetStr(). At that point it is not possible
    ** for the string to be stored in Mem.zShort, or for it to be stored
    ** in dynamic memory with no destructor.
    */
    assert( !(pMem->flags&MEM_Short) );
    assert( !(pMem->flags&MEM_Dyn) || pMem->xDel );
    if( pMem->flags & MEM_Dyn ){
      void (*xDel)(void*) = pMem->xDel;
      char *z = pMem->z;
      pMem->z = 0;
      pMem->xDel = 0;
      rc = sqlite3VdbeMemSetStr(pMem, &z[2], pMem->n-2, bom, SQLITE_TRANSIENT);
      xDel(z);
    }else{
      rc = sqlite3VdbeMemSetStr(pMem, &pMem->z[2], pMem->n-2, bom, 
          SQLITE_TRANSIENT);
    }
  }
  return rc;
}
#endif /* SQLITE_OMIT_UTF16 */

/*
** pZ is a UTF-8 encoded unicode string. If nByte is less than zero,
** return the number of unicode characters in pZ up to (but not including)
** the first 0x00 byte. If nByte is not less than zero, return the
** number of unicode characters in the first nByte of pZ (or up to 
** the first 0x00, whichever comes first).
*/
SQLITE_PRIVATE int sqlite3Utf8CharLen(const char *zIn, int nByte){
  int r = 0;
  const u8 *z = (const u8*)zIn;
  const u8 *zTerm;
  if( nByte>=0 ){
    zTerm = &z[nByte];
  }else{
    zTerm = (const u8*)(-1);
  }
  assert( z<=zTerm );
  while( *z!=0 && z<zTerm ){
    SQLITE_SKIP_UTF8(z);
    r++;
  }
  return r;
}

#ifndef SQLITE_OMIT_UTF16
/*
** Convert a UTF-16 string in the native encoding into a UTF-8 string.
** Memory to hold the UTF-8 string is obtained from malloc and must be
** freed by the calling function.
**
** NULL is returned if there is an allocation error.
*/
SQLITE_PRIVATE char *sqlite3Utf16to8(const void *z, int nByte){
  Mem m;
  memset(&m, 0, sizeof(m));
  sqlite3VdbeMemSetStr(&m, z, nByte, SQLITE_UTF16NATIVE, SQLITE_STATIC);
  sqlite3VdbeChangeEncoding(&m, SQLITE_UTF8);
  assert( (m.flags & MEM_Term)!=0 || sqlite3MallocFailed() );
  assert( (m.flags & MEM_Str)!=0 || sqlite3MallocFailed() );
  return (m.flags & MEM_Dyn)!=0 ? m.z : sqliteStrDup(m.z);
}

/*
** pZ is a UTF-16 encoded unicode string. If nChar is less than zero,
** return the number of bytes up to (but not including), the first pair
** of consecutive 0x00 bytes in pZ. If nChar is not less than zero,
** then return the number of bytes in the first nChar unicode characters
** in pZ (or up until the first pair of 0x00 bytes, whichever comes first).
*/
SQLITE_PRIVATE int sqlite3Utf16ByteLen(const void *zIn, int nChar){
  unsigned int c = 1;
  char const *z = zIn;
  int n = 0;
  if( SQLITE_UTF16NATIVE==SQLITE_UTF16BE ){
    /* Using an "if (SQLITE_UTF16NATIVE==SQLITE_UTF16BE)" construct here
    ** and in other parts of this file means that at one branch will
    ** not be covered by coverage testing on any single host. But coverage
    ** will be complete if the tests are run on both a little-endian and 
    ** big-endian host. Because both the UTF16NATIVE and SQLITE_UTF16BE
    ** macros are constant at compile time the compiler can determine
    ** which branch will be followed. It is therefore assumed that no runtime
    ** penalty is paid for this "if" statement.
    */
    while( c && ((nChar<0) || n<nChar) ){
      READ_UTF16BE(z, c);
      n++;
    }
  }else{
    while( c && ((nChar<0) || n<nChar) ){
      READ_UTF16LE(z, c);
      n++;
    }
  }
  return (z-(char const *)zIn)-((c==0)?2:0);
}

#if defined(SQLITE_TEST)
/*
** Translate UTF-8 to UTF-8.
**
** This has the effect of making sure that the string is well-formed
** UTF-8.  Miscoded characters are removed.
**
** The translation is done in-place (since it is impossible for the
** correct UTF-8 encoding to be longer than a malformed encoding).
*/
SQLITE_PRIVATE int sqlite3Utf8To8(unsigned char *zIn){
  unsigned char *zOut = zIn;
  unsigned char *zStart = zIn;
  int c;

  while(1){
    SQLITE_READ_UTF8(zIn, c);
    if( c==0 ) break;
    if( c!=0xfffd ){
      WRITE_UTF8(zOut, c);
    }
  }
  *zOut = 0;
  return zOut - zStart;
}
#endif

#if defined(SQLITE_TEST)
/*
** This routine is called from the TCL test function "translate_selftest".
** It checks that the primitives for serializing and deserializing
** characters in each encoding are inverses of each other.
*/
SQLITE_PRIVATE void sqlite3UtfSelfTest(){
  unsigned int i, t;
  unsigned char zBuf[20];
  unsigned char *z;
  int n;
  unsigned int c;

  for(i=0; i<0x00110000; i++){
    z = zBuf;
    WRITE_UTF8(z, i);
    n = z-zBuf;
    z[0] = 0;
    z = zBuf;
    SQLITE_READ_UTF8(z, c);
    t = i;
    if( i>=0xD800 && i<=0xDFFF ) t = 0xFFFD;
    if( (i&0xFFFFFFFE)==0xFFFE ) t = 0xFFFD;
    assert( c==t );
    assert( (z-zBuf)==n );
  }
  for(i=0; i<0x00110000; i++){
    if( i>=0xD800 && i<0xE000 ) continue;
    z = zBuf;
    WRITE_UTF16LE(z, i);
    n = z-zBuf;
    z[0] = 0;
    z = zBuf;
    READ_UTF16LE(z, c);
    assert( c==i );
    assert( (z-zBuf)==n );
  }
  for(i=0; i<0x00110000; i++){
    if( i>=0xD800 && i<0xE000 ) continue;
    z = zBuf;
    WRITE_UTF16BE(z, i);
    n = z-zBuf;
    z[0] = 0;
    z = zBuf;
    READ_UTF16BE(z, c);
    assert( c==i );
    assert( (z-zBuf)==n );
  }
}
#endif /* SQLITE_TEST */
#endif /* SQLITE_OMIT_UTF16 */

/************** End of utf.c *************************************************/
/************** Begin file util.c ********************************************/
/*
** 2001 September 15
**
** The author disclaims copyright to this source code.  In place of
** a legal notice, here is a blessing:
**
**    May you do good and not evil.
**    May you find forgiveness for yourself and forgive others.
**    May you share freely, never taking more than you give.
**
*************************************************************************
** Utility functions used throughout sqlite.
**
** This file contains functions for allocating memory, comparing
** strings, and stuff like that.
**
** $Id: sqlite3.c,v 1.4 2007/07/21 08:51:43 rmsimpson Exp $
*/


/*
** Set the most recent error code and error string for the sqlite
** handle "db". The error code is set to "err_code".
**
** If it is not NULL, string zFormat specifies the format of the
** error string in the style of the printf functions: The following
** format characters are allowed:
**
**      %s      Insert a string
**      %z      A string that should be freed after use
**      %d      Insert an integer
**      %T      Insert a token
**      %S      Insert the first element of a SrcList
**
** zFormat and any string tokens that follow it are assumed to be
** encoded in UTF-8.
**
** To clear the most recent error for sqlite handle "db", sqlite3Error
** should be called with err_code set to SQLITE_OK and zFormat set
** to NULL.
*/
SQLITE_PRIVATE void sqlite3Error(sqlite3 *db, int err_code, const char *zFormat, ...){
  if( db && (db->pErr || (db->pErr = sqlite3ValueNew())!=0) ){
    db->errCode = err_code;
    if( zFormat ){
      char *z;
      va_list ap;
      va_start(ap, zFormat);
      z = sqlite3VMPrintf(zFormat, ap);
      va_end(ap);
      sqlite3ValueSetStr(db->pErr, -1, z, SQLITE_UTF8, sqlite3FreeX);
    }else{
      sqlite3ValueSetStr(db->pErr, 0, 0, SQLITE_UTF8, SQLITE_STATIC);
    }
  }
}

/*
** Add an error message to pParse->zErrMsg and increment pParse->nErr.
** The following formatting characters are allowed:
**
**      %s      Insert a string
**      %z      A string that should be freed after use
**      %d      Insert an integer
**      %T      Insert a token
**      %S      Insert the first element of a SrcList
**
** This function should be used to report any error that occurs whilst
** compiling an SQL statement (i.e. within sqlite3_prepare()). The
** last thing the sqlite3_prepare() function does is copy the error
** stored by this function into the database handle using sqlite3Error().
** Function sqlite3Error() should be used during statement execution
** (sqlite3_step() etc.).
*/
SQLITE_PRIVATE void sqlite3ErrorMsg(Parse *pParse, const char *zFormat, ...){
  va_list ap;
  pParse->nErr++;
  sqliteFree(pParse->zErrMsg);
  va_start(ap, zFormat);
  pParse->zErrMsg = sqlite3VMPrintf(zFormat, ap);
  va_end(ap);
  if( pParse->rc==SQLITE_OK ){
    pParse->rc = SQLITE_ERROR;
  }
}

/*
** Clear the error message in pParse, if any
*/
SQLITE_PRIVATE void sqlite3ErrorClear(Parse *pParse){
  sqliteFree(pParse->zErrMsg);
  pParse->zErrMsg = 0;
  pParse->nErr = 0;
}

/*
** Convert an SQL-style quoted string into a normal string by removing
** the quote characters.  The conversion is done in-place.  If the
** input does not begin with a quote character, then this routine
** is a no-op.
**
** 2002-Feb-14: This routine is extended to remove MS-Access style
** brackets from around identifers.  For example:  "[a-b-c]" becomes
** "a-b-c".
*/
SQLITE_PRIVATE void sqlite3Dequote(char *z){
  int quote;
  int i, j;
  if( z==0 ) return;
  quote = z[0];
  switch( quote ){
    case '\'':  break;
    case '"':   break;
    case '`':   break;                /* For MySQL compatibility */
    case '[':   quote = ']';  break;  /* For MS SqlServer compatibility */
    default:    return;
  }
  for(i=1, j=0; z[i]; i++){
    if( z[i]==quote ){
      if( z[i+1]==quote ){
        z[j++] = quote;
        i++;
      }else{
        z[j++] = 0;
        break;
      }
    }else{
      z[j++] = z[i];
    }
  }
}

/* An array to map all upper-case characters into their corresponding
** lower-case character. 
*/
const unsigned char sqlite3UpperToLower[] = {
#ifdef SQLITE_ASCII
      0,  1,  2,  3,  4,  5,  6,  7,  8,  9, 10, 11, 12, 13, 14, 15, 16, 17,
     18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35,
     36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53,
     54, 55, 56, 57, 58, 59, 60, 61, 62, 63, 64, 97, 98, 99,100,101,102,103,
    104,105,106,107,108,109,110,111,112,113,114,115,116,117,118,119,120,121,
    122, 91, 92, 93, 94, 95, 96, 97, 98, 99,100,101,102,103,104,105,106,107,
    108,109,110,111,112,113,114,115,116,117,118,119,120,121,122,123,124,125,
    126,127,128,129,130,131,132,133,134,135,136,137,138,139,140,141,142,143,
    144,145,146,147,148,149,150,151,152,153,154,155,156,157,158,159,160,161,
    162,163,164,165,166,167,168,169,170,171,172,173,174,175,176,177,178,179,
    180,181,182,183,184,185,186,187,188,189,190,191,192,193,194,195,196,197,
    198,199,200,201,202,203,204,205,206,207,208,209,210,211,212,213,214,215,
    216,217,218,219,220,221,222,223,224,225,226,227,228,229,230,231,232,233,
    234,235,236,237,238,239,240,241,242,243,244,245,246,247,248,249,250,251,
    252,253,254,255
#endif
#ifdef SQLITE_EBCDIC
      0,  1,  2,  3,  4,  5,  6,  7,  8,  9, 10, 11, 12, 13, 14, 15, /* 0x */
     16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, /* 1x */
     32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, /* 2x */
     48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, /* 3x */
     64, 65, 66, 67, 68, 69, 70, 71, 72, 73, 74, 75, 76, 77, 78, 79, /* 4x */
     80, 81, 82, 83, 84, 85, 86, 87, 88, 89, 90, 91, 92, 93, 94, 95, /* 5x */
     96, 97, 66, 67, 68, 69, 70, 71, 72, 73,106,107,108,109,110,111, /* 6x */
    112, 81, 82, 83, 84, 85, 86, 87, 88, 89,122,123,124,125,126,127, /* 7x */
    128,129,130,131,132,133,134,135,136,137,138,139,140,141,142,143, /* 8x */
    144,145,146,147,148,149,150,151,152,153,154,155,156,157,156,159, /* 9x */
    160,161,162,163,164,165,166,167,168,169,170,171,140,141,142,175, /* Ax */
    176,177,178,179,180,181,182,183,184,185,186,187,188,189,190,191, /* Bx */
    192,129,130,131,132,133,134,135,136,137,202,203,204,205,206,207, /* Cx */
    208,145,146,147,148,149,150,151,152,153,218,219,220,221,222,223, /* Dx */
    224,225,162,163,164,165,166,167,168,169,232,203,204,205,206,207, /* Ex */
    239,240,241,242,243,244,245,246,247,248,249,219,220,221,222,255, /* Fx */
#endif
};
#define UpperToLower sqlite3UpperToLower

/*
** Some systems have stricmp().  Others have strcasecmp().  Because
** there is no consistency, we will define our own.
*/
SQLITE_PRIVATE int sqlite3StrICmp(const char *zLeft, const char *zRight){
  register unsigned char *a, *b;
  a = (unsigned char *)zLeft;
  b = (unsigned char *)zRight;
  while( *a!=0 && UpperToLower[*a]==UpperToLower[*b]){ a++; b++; }
  return UpperToLower[*a] - UpperToLower[*b];
}
SQLITE_PRIVATE int sqlite3StrNICmp(const char *zLeft, const char *zRight, int N){
  register unsigned char *a, *b;
  a = (unsigned char *)zLeft;
  b = (unsigned char *)zRight;
  while( N-- > 0 && *a!=0 && UpperToLower[*a]==UpperToLower[*b]){ a++; b++; }
  return N<0 ? 0 : UpperToLower[*a] - UpperToLower[*b];
}

/*
** Return TRUE if z is a pure numeric string.  Return FALSE if the
** string contains any character which is not part of a number. If
** the string is numeric and contains the '.' character, set *realnum
** to TRUE (otherwise FALSE).
**
** An empty string is considered non-numeric.
*/
SQLITE_PRIVATE int sqlite3IsNumber(const char *z, int *realnum, u8 enc){
  int incr = (enc==SQLITE_UTF8?1:2);
  if( enc==SQLITE_UTF16BE ) z++;
  if( *z=='-' || *z=='+' ) z += incr;
  if( !isdigit(*(u8*)z) ){
    return 0;
  }
  z += incr;
  if( realnum ) *realnum = 0;
  while( isdigit(*(u8*)z) ){ z += incr; }
  if( *z=='.' ){
    z += incr;
    if( !isdigit(*(u8*)z) ) return 0;
    while( isdigit(*(u8*)z) ){ z += incr; }
    if( realnum ) *realnum = 1;
  }
  if( *z=='e' || *z=='E' ){
    z += incr;
    if( *z=='+' || *z=='-' ) z += incr;
    if( !isdigit(*(u8*)z) ) return 0;
    while( isdigit(*(u8*)z) ){ z += incr; }
    if( realnum ) *realnum = 1;
  }
  return *z==0;
}

/*
** The string z[] is an ascii representation of a real number.
** Convert this string to a double.
**
** This routine assumes that z[] really is a valid number.  If it
** is not, the result is undefined.
**
** This routine is used instead of the library atof() function because
** the library atof() might want to use "," as the decimal point instead
** of "." depending on how locale is set.  But that would cause problems
** for SQL.  So this routine always uses "." regardless of locale.
*/
SQLITE_PRIVATE int sqlite3AtoF(const char *z, double *pResult){
#ifndef SQLITE_OMIT_FLOATING_POINT
  int sign = 1;
  const char *zBegin = z;
  LONGDOUBLE_TYPE v1 = 0.0;
  while( isspace(*(u8*)z) ) z++;
  if( *z=='-' ){
    sign = -1;
    z++;
  }else if( *z=='+' ){
    z++;
  }
  while( isdigit(*(u8*)z) ){
    v1 = v1*10.0 + (*z - '0');
    z++;
  }
  if( *z=='.' ){
    LONGDOUBLE_TYPE divisor = 1.0;
    z++;
    while( isdigit(*(u8*)z) ){
      v1 = v1*10.0 + (*z - '0');
      divisor *= 10.0;
      z++;
    }
    v1 /= divisor;
  }
  if( *z=='e' || *z=='E' ){
    int esign = 1;
    int eval = 0;
    LONGDOUBLE_TYPE scale = 1.0;
    z++;
    if( *z=='-' ){
      esign = -1;
      z++;
    }else if( *z=='+' ){
      z++;
    }
    while( isdigit(*(u8*)z) ){
      eval = eval*10 + *z - '0';
      z++;
    }
    while( eval>=64 ){ scale *= 1.0e+64; eval -= 64; }
    while( eval>=16 ){ scale *= 1.0e+16; eval -= 16; }
    while( eval>=4 ){ scale *= 1.0e+4; eval -= 4; }
    while( eval>=1 ){ scale *= 1.0e+1; eval -= 1; }
    if( esign<0 ){
      v1 /= scale;
    }else{
      v1 *= scale;
    }
  }
  *pResult = sign<0 ? -v1 : v1;
  return z - zBegin;
#else
  return sqlite3Atoi64(z, pResult);
#endif /* SQLITE_OMIT_FLOATING_POINT */
}

/*
** Compare the 19-character string zNum against the text representation
** value 2^63:  9223372036854775808.  Return negative, zero, or positive
** if zNum is less than, equal to, or greater than the string.
**
** Unlike memcmp() this routine is guaranteed to return the difference
** in the values of the last digit if the only difference is in the
** last digit.  So, for example,
**
**      compare2pow63("9223372036854775800")
**
** will return -8.
*/
static int compare2pow63(const char *zNum){
  int c;
  c = memcmp(zNum,"922337203685477580",18);
  if( c==0 ){
    c = zNum[18] - '8';
  }
  return c;
}


/*
** Return TRUE if zNum is a 64-bit signed integer and write
** the value of the integer into *pNum.  If zNum is not an integer
** or is an integer that is too large to be expressed with 64 bits,
** then return false.
**
** When this routine was originally written it dealt with only
** 32-bit numbers.  At that time, it was much faster than the
** atoi() library routine in RedHat 7.2.
*/
SQLITE_PRIVATE int sqlite3Atoi64(const char *zNum, i64 *pNum){
  i64 v = 0;
  int neg;
  int i, c;
  while( isspace(*(u8*)zNum) ) zNum++;
  if( *zNum=='-' ){
    neg = 1;
    zNum++;
  }else if( *zNum=='+' ){
    neg = 0;
    zNum++;
  }else{
    neg = 0;
  }
  while( zNum[0]=='0' ){ zNum++; } /* Skip over leading zeros. Ticket #2454 */
  for(i=0; (c=zNum[i])>='0' && c<='9'; i++){
    v = v*10 + c - '0';
  }
  *pNum = neg ? -v : v;
  if( c!=0 || i==0 || i>19 ){
    /* zNum is empty or contains non-numeric text or is longer
    ** than 19 digits (thus guaranting that it is too large) */
    return 0;
  }else if( i<19 ){
    /* Less than 19 digits, so we know that it fits in 64 bits */
    return 1;
  }else{
    /* 19-digit numbers must be no larger than 9223372036854775807 if positive
    ** or 9223372036854775808 if negative.  Note that 9223372036854665808
    ** is 2^63. */
    return compare2pow63(zNum)<neg;
  }
}

/*
** The string zNum represents an integer.  There might be some other
** information following the integer too, but that part is ignored.
** If the integer that the prefix of zNum represents will fit in a
** 64-bit signed integer, return TRUE.  Otherwise return FALSE.
**
** This routine returns FALSE for the string -9223372036854775808 even that
** that number will, in theory fit in a 64-bit integer.  Positive
** 9223373036854775808 will not fit in 64 bits.  So it seems safer to return
** false.
*/
SQLITE_PRIVATE int sqlite3FitsIn64Bits(const char *zNum){
  int i, c;
  int neg = 0;
  if( *zNum=='-' ){
    neg = 1;
    zNum++;
  }else if( *zNum=='+' ){
    zNum++;
  }
  while( *zNum=='0' ){
    zNum++;   /* Skip leading zeros.  Ticket #2454 */
  }
  for(i=0; (c=zNum[i])>='0' && c<='9'; i++){}
  if( i<19 ){
    /* Guaranteed to fit if less than 19 digits */
    return 1;
  }else if( i>19 ){
    /* Guaranteed to be too big if greater than 19 digits */
    return 0;
  }else{
    /* Compare against 2^63. */
    return compare2pow63(zNum)<neg;
  }
}

/*
** If zNum represents an integer that will fit in 32-bits, then set
** *pValue to that integer and return true.  Otherwise return false.
**
** Any non-numeric characters that following zNum are ignored.
** This is different from sqlite3Atoi64() which requires the
** input number to be zero-terminated.
*/
SQLITE_PRIVATE int sqlite3GetInt32(const char *zNum, int *pValue){
  sqlite_int64 v = 0;
  int i, c;
  int neg = 0;
  if( zNum[0]=='-' ){
    neg = 1;
    zNum++;
  }else if( zNum[0]=='+' ){
    zNum++;
  }
  while( zNum[0]=='0' ) zNum++;
  for(i=0; i<10 && (c = zNum[i] - '0')>=0 && c<=9; i++){
    v = v*10 + c;
  }
  if( i>9 ){
    return 0;
  }
  if( v-neg>2147483647 ){
    return 0;
  }
  if( neg ){
    v = -v;
  }
  *pValue = (int)v;
  return 1;
}

/*
** Check to make sure we have a valid db pointer.  This test is not
** foolproof but it does provide some measure of protection against
** misuse of the interface such as passing in db pointers that are
** NULL or which have been previously closed.  If this routine returns
** TRUE it means that the db pointer is invalid and should not be
** dereferenced for any reason.  The calling function should invoke
** SQLITE_MISUSE immediately.
*/
SQLITE_PRIVATE int sqlite3SafetyCheck(sqlite3 *db){
  int magic;
  if( db==0 ) return 1;
  magic = db->magic;
  if( magic!=SQLITE_MAGIC_CLOSED &&
         magic!=SQLITE_MAGIC_OPEN &&
         magic!=SQLITE_MAGIC_BUSY ) return 1;
  return 0;
}

/*
** The variable-length integer encoding is as follows:
**
** KEY:
**         A = 0xxxxxxx    7 bits of data and one flag bit
**         B = 1xxxxxxx    7 bits of data and one flag bit
**         C = xxxxxxxx    8 bits of data
**
**  7 bits - A
** 14 bits - BA
** 21 bits - BBA
** 28 bits - BBBA
** 35 bits - BBBBA
** 42 bits - BBBBBA
** 49 bits - BBBBBBA
** 56 bits - BBBBBBBA
** 64 bits - BBBBBBBBC
*/

/*
** Write a 64-bit variable-length integer to memory starting at p[0].
** The length of data write will be between 1 and 9 bytes.  The number
** of bytes written is returned.
**
** A variable-length integer consists of the lower 7 bits of each byte
** for all bytes that have the 8th bit set and one byte with the 8th
** bit clear.  Except, if we get to the 9th byte, it stores the full
** 8 bits and is the last byte.
*/
SQLITE_PRIVATE int sqlite3PutVarint(unsigned char *p, u64 v){
  int i, j, n;
  u8 buf[10];
  if( v & (((u64)0xff000000)<<32) ){
    p[8] = v;
    v >>= 8;
    for(i=7; i>=0; i--){
      p[i] = (v & 0x7f) | 0x80;
      v >>= 7;
    }
    return 9;
  }    
  n = 0;
  do{
    buf[n++] = (v & 0x7f) | 0x80;
    v >>= 7;
  }while( v!=0 );
  buf[0] &= 0x7f;
  assert( n<=9 );
  for(i=0, j=n-1; j>=0; j--, i++){
    p[i] = buf[j];
  }
  return n;
}

/*
** Read a 64-bit variable-length integer from memory starting at p[0].
** Return the number of bytes read.  The value is stored in *v.
*/
SQLITE_PRIVATE int sqlite3GetVarint(const unsigned char *p, u64 *v){
  u32 x;
  u64 x64;
  int n;
  unsigned char c;
  if( ((c = p[0]) & 0x80)==0 ){
    *v = c;
    return 1;
  }
  x = c & 0x7f;
  if( ((c = p[1]) & 0x80)==0 ){
    *v = (x<<7) | c;
    return 2;
  }
  x = (x<<7) | (c&0x7f);
  if( ((c = p[2]) & 0x80)==0 ){
    *v = (x<<7) | c;
    return 3;
  }
  x = (x<<7) | (c&0x7f);
  if( ((c = p[3]) & 0x80)==0 ){
    *v = (x<<7) | c;
    return 4;
  }
  x64 = (x<<7) | (c&0x7f);
  n = 4;
  do{
    c = p[n++];
    if( n==9 ){
      x64 = (x64<<8) | c;
      break;
    }
    x64 = (x64<<7) | (c&0x7f);
  }while( (c & 0x80)!=0 );
  *v = x64;
  return n;
}

/*
** Read a 32-bit variable-length integer from memory starting at p[0].
** Return the number of bytes read.  The value is stored in *v.
*/
SQLITE_PRIVATE int sqlite3GetVarint32(const unsigned char *p, u32 *v){
  u32 x;
  int n;
  unsigned char c;
  if( ((signed char*)p)[0]>=0 ){
    *v = p[0];
    return 1;
  }
  x = p[0] & 0x7f;
  if( ((signed char*)p)[1]>=0 ){
    *v = (x<<7) | p[1];
    return 2;
  }
  x = (x<<7) | (p[1] & 0x7f);
  n = 2;
  do{
    x = (x<<7) | ((c = p[n++])&0x7f);
  }while( (c & 0x80)!=0 && n<9 );
  *v = x;
  return n;
}

/*
** Return the number of bytes that will be needed to store the given
** 64-bit integer.
*/
SQLITE_PRIVATE int sqlite3VarintLen(u64 v){
  int i = 0;
  do{
    i++;
    v >>= 7;
  }while( v!=0 && i<9 );
  return i;
}


/*
** Read or write a four-byte big-endian integer value.
*/
SQLITE_PRIVATE u32 sqlite3Get4byte(const u8 *p){
  return (p[0]<<24) | (p[1]<<16) | (p[2]<<8) | p[3];
}
SQLITE_PRIVATE void sqlite3Put4byte(unsigned char *p, u32 v){
  p[0] = v>>24;
  p[1] = v>>16;
  p[2] = v>>8;
  p[3] = v;
}



#if !defined(SQLITE_OMIT_BLOB_LITERAL) || defined(SQLITE_HAS_CODEC) \
    || defined(SQLITE_TEST)
/*
** Translate a single byte of Hex into an integer.
*/
static int hexToInt(int h){
  if( h>='0' && h<='9' ){
    return h - '0';
  }else if( h>='a' && h<='f' ){
    return h - 'a' + 10;
  }else{
    assert( h>='A' && h<='F' );
    return h - 'A' + 10;
  }
}
#endif /* !SQLITE_OMIT_BLOB_LITERAL || SQLITE_HAS_CODEC || SQLITE_TEST */

#if !defined(SQLITE_OMIT_BLOB_LITERAL) || defined(SQLITE_HAS_CODEC)
/*
** Convert a BLOB literal of the form "x'hhhhhh'" into its binary
** value.  Return a pointer to its binary value.  Space to hold the
** binary value has been obtained from malloc and must be freed by
** the calling routine.
*/
SQLITE_PRIVATE void *sqlite3HexToBlob(const char *z){
  char *zBlob;
  int i;
  int n = strlen(z);
  if( n%2 ) return 0;

  zBlob = (char *)sqliteMalloc(n/2);
  if( zBlob ){
    for(i=0; i<n; i+=2){
      zBlob[i/2] = (hexToInt(z[i])<<4) | hexToInt(z[i+1]);
    }
  }
  return zBlob;
}
#endif /* !SQLITE_OMIT_BLOB_LITERAL || SQLITE_HAS_CODEC */


/*
** Change the sqlite.magic from SQLITE_MAGIC_OPEN to SQLITE_MAGIC_BUSY.
** Return an error (non-zero) if the magic was not SQLITE_MAGIC_OPEN
** when this routine is called.
**
** This routine is called when entering an SQLite API.  The SQLITE_MAGIC_OPEN
** value indicates that the database connection passed into the API is
** open and is not being used by another thread.  By changing the value
** to SQLITE_MAGIC_BUSY we indicate that the connection is in use.
** sqlite3SafetyOff() below will change the value back to SQLITE_MAGIC_OPEN
** when the API exits. 
**
** This routine is a attempt to detect if two threads use the
** same sqlite* pointer at the same time.  There is a race 
** condition so it is possible that the error is not detected.
** But usually the problem will be seen.  The result will be an
** error which can be used to debug the application that is
** using SQLite incorrectly.
**
** Ticket #202:  If db->magic is not a valid open value, take care not
** to modify the db structure at all.  It could be that db is a stale
** pointer.  In other words, it could be that there has been a prior
** call to sqlite3_close(db) and db has been deallocated.  And we do
** not want to write into deallocated memory.
*/
SQLITE_PRIVATE int sqlite3SafetyOn(sqlite3 *db){
  if( db->magic==SQLITE_MAGIC_OPEN ){
    db->magic = SQLITE_MAGIC_BUSY;
    return 0;
  }else if( db->magic==SQLITE_MAGIC_BUSY ){
    db->magic = SQLITE_MAGIC_ERROR;
    db->u1.isInterrupted = 1;
  }
  return 1;
}

/*
** Change the magic from SQLITE_MAGIC_BUSY to SQLITE_MAGIC_OPEN.
** Return an error (non-zero) if the magic was not SQLITE_MAGIC_BUSY
** when this routine is called.
*/
SQLITE_PRIVATE int sqlite3SafetyOff(sqlite3 *db){
  if( db->magic==SQLITE_MAGIC_BUSY ){
    db->magic = SQLITE_MAGIC_OPEN;
    return 0;
  }else {
    db->magic = SQLITE_MAGIC_ERROR;
    db->u1.isInterrupted = 1;
    return 1;
  }
}

/*
** Return a pointer to the ThreadData associated with the calling thread.
*/
SQLITE_PRIVATE ThreadData *sqlite3ThreadData(){
  ThreadData *p = (ThreadData*)sqlite3OsThreadSpecificData(1);
  if( !p ){
    sqlite3FailedMalloc();
  }
  return p;
}

/*
** Return a pointer to the ThreadData associated with the calling thread.
** If no ThreadData has been allocated to this thread yet, return a pointer
** to a substitute ThreadData structure that is all zeros. 
*/
SQLITE_PRIVATE const ThreadData *sqlite3ThreadDataReadOnly(){
  static const ThreadData zeroData = {0};  /* Initializer to silence warnings
                                           ** from broken compilers */
  const ThreadData *pTd = sqlite3OsThreadSpecificData(0);
  return pTd ? pTd : &zeroData;
}

/*
** Check to see if the ThreadData for this thread is all zero.  If it
** is, then deallocate it. 
*/
SQLITE_PRIVATE void sqlite3ReleaseThreadData(){
  sqlite3OsThreadSpecificData(-1);
}

/************** End of util.c ************************************************/
/************** Begin file hash.c ********************************************/
/*
** 2001 September 22
**
** The author disclaims copyright to this source code.  In place of
** a legal notice, here is a blessing:
**
**    May you do good and not evil.
**    May you find forgiveness for yourself and forgive others.
**    May you share freely, never taking more than you give.
**
*************************************************************************
** This is the implementation of generic hash-tables
** used in SQLite.
**
** $Id: sqlite3.c,v 1.4 2007/07/21 08:51:43 rmsimpson Exp $
*/

/* Turn bulk memory into a hash table object by initializing the
** fields of the Hash structure.
**
** "pNew" is a pointer to the hash table that is to be initialized.
** keyClass is one of the constants SQLITE_HASH_INT, SQLITE_HASH_POINTER,
** SQLITE_HASH_BINARY, or SQLITE_HASH_STRING.  The value of keyClass 
** determines what kind of key the hash table will use.  "copyKey" is
** true if the hash table should make its own private copy of keys and
** false if it should just use the supplied pointer.  CopyKey only makes
** sense for SQLITE_HASH_STRING and SQLITE_HASH_BINARY and is ignored
** for other key classes.
*/
SQLITE_PRIVATE void sqlite3HashInit(Hash *pNew, int keyClass, int copyKey){
  assert( pNew!=0 );
  assert( keyClass>=SQLITE_HASH_STRING && keyClass<=SQLITE_HASH_BINARY );
  pNew->keyClass = keyClass;
#if 0
  if( keyClass==SQLITE_HASH_POINTER || keyClass==SQLITE_HASH_INT ) copyKey = 0;
#endif
  pNew->copyKey = copyKey;
  pNew->first = 0;
  pNew->count = 0;
  pNew->htsize = 0;
  pNew->ht = 0;
  pNew->xMalloc = sqlite3MallocX;
  pNew->xFree = sqlite3FreeX;
}

/* Remove all entries from a hash table.  Reclaim all memory.
** Call this routine to delete a hash table or to reset a hash table
** to the empty state.
*/
SQLITE_PRIVATE void sqlite3HashClear(Hash *pH){
  HashElem *elem;         /* For looping over all elements of the table */

  assert( pH!=0 );
  elem = pH->first;
  pH->first = 0;
  if( pH->ht ) pH->xFree(pH->ht);
  pH->ht = 0;
  pH->htsize = 0;
  while( elem ){
    HashElem *next_elem = elem->next;
    if( pH->copyKey && elem->pKey ){
      pH->xFree(elem->pKey);
    }
    pH->xFree(elem);
    elem = next_elem;
  }
  pH->count = 0;
}

#if 0 /* NOT USED */
/*
** Hash and comparison functions when the mode is SQLITE_HASH_INT
*/
static int intHash(const void *pKey, int nKey){
  return nKey ^ (nKey<<8) ^ (nKey>>8);
}
static int intCompare(const void *pKey1, int n1, const void *pKey2, int n2){
  return n2 - n1;
}
#endif

#if 0 /* NOT USED */
/*
** Hash and comparison functions when the mode is SQLITE_HASH_POINTER
*/
static int ptrHash(const void *pKey, int nKey){
  uptr x = Addr(pKey);
  return x ^ (x<<8) ^ (x>>8);
}
static int ptrCompare(const void *pKey1, int n1, const void *pKey2, int n2){
  if( pKey1==pKey2 ) return 0;
  if( pKey1<pKey2 ) return -1;
  return 1;
}
#endif

/*
** Hash and comparison functions when the mode is SQLITE_HASH_STRING
*/
static int strHash(const void *pKey, int nKey){
  const char *z = (const char *)pKey;
  int h = 0;
  if( nKey<=0 ) nKey = strlen(z);
  while( nKey > 0  ){
    h = (h<<3) ^ h ^ sqlite3UpperToLower[(unsigned char)*z++];
    nKey--;
  }
  return h & 0x7fffffff;
}
static int strCompare(const void *pKey1, int n1, const void *pKey2, int n2){
  if( n1!=n2 ) return 1;
  return sqlite3StrNICmp((const char*)pKey1,(const char*)pKey2,n1);
}

/*
** Hash and comparison functions when the mode is SQLITE_HASH_BINARY
*/
static int binHash(const void *pKey, int nKey){
  int h = 0;
  const char *z = (const char *)pKey;
  while( nKey-- > 0 ){
    h = (h<<3) ^ h ^ *(z++);
  }
  return h & 0x7fffffff;
}
static int binCompare(const void *pKey1, int n1, const void *pKey2, int n2){
  if( n1!=n2 ) return 1;
  return memcmp(pKey1,pKey2,n1);
}

/*
** Return a pointer to the appropriate hash function given the key class.
**
** The C syntax in this function definition may be unfamilar to some 
** programmers, so we provide the following additional explanation:
**
** The name of the function is "hashFunction".  The function takes a
** single parameter "keyClass".  The return value of hashFunction()
** is a pointer to another function.  Specifically, the return value
** of hashFunction() is a pointer to a function that takes two parameters
** with types "const void*" and "int" and returns an "int".
*/
static int (*hashFunction(int keyClass))(const void*,int){
#if 0  /* HASH_INT and HASH_POINTER are never used */
  switch( keyClass ){
    case SQLITE_HASH_INT:     return &intHash;
    case SQLITE_HASH_POINTER: return &ptrHash;
    case SQLITE_HASH_STRING:  return &strHash;
    case SQLITE_HASH_BINARY:  return &binHash;;
    default: break;
  }
  return 0;
#else
  if( keyClass==SQLITE_HASH_STRING ){
    return &strHash;
  }else{
    assert( keyClass==SQLITE_HASH_BINARY );
    return &binHash;
  }
#endif
}

/*
** Return a pointer to the appropriate hash function given the key class.
**
** For help in interpreted the obscure C code in the function definition,
** see the header comment on the previous function.
*/
static int (*compareFunction(int keyClass))(const void*,int,const void*,int){
#if 0 /* HASH_INT and HASH_POINTER are never used */
  switch( keyClass ){
    case SQLITE_HASH_INT:     return &intCompare;
    case SQLITE_HASH_POINTER: return &ptrCompare;
    case SQLITE_HASH_STRING:  return &strCompare;
    case SQLITE_HASH_BINARY:  return &binCompare;
    default: break;
  }
  return 0;
#else
  if( keyClass==SQLITE_HASH_STRING ){
    return &strCompare;
  }else{
    assert( keyClass==SQLITE_HASH_BINARY );
    return &binCompare;
  }
#endif
}

/* Link an element into the hash table
*/
static void insertElement(
  Hash *pH,              /* The complete hash table */
  struct _ht *pEntry,    /* The entry into which pNew is inserted */
  HashElem *pNew         /* The element to be inserted */
){
  HashElem *pHead;       /* First element already in pEntry */
  pHead = pEntry->chain;
  if( pHead ){
    pNew->next = pHead;
    pNew->prev = pHead->prev;
    if( pHead->prev ){ pHead->prev->next = pNew; }
    else             { pH->first = pNew; }
    pHead->prev = pNew;
  }else{
    pNew->next = pH->first;
    if( pH->first ){ pH->first->prev = pNew; }
    pNew->prev = 0;
    pH->first = pNew;
  }
  pEntry->count++;
  pEntry->chain = pNew;
}


/* Resize the hash table so that it cantains "new_size" buckets.
** "new_size" must be a power of 2.  The hash table might fail 
** to resize if sqliteMalloc() fails.
*/
static void rehash(Hash *pH, int new_size){
  struct _ht *new_ht;            /* The new hash table */
  HashElem *elem, *next_elem;    /* For looping over existing elements */
  int (*xHash)(const void*,int); /* The hash function */

  assert( (new_size & (new_size-1))==0 );
  new_ht = (struct _ht *)pH->xMalloc( new_size*sizeof(struct _ht) );
  if( new_ht==0 ) return;
  if( pH->ht ) pH->xFree(pH->ht);
  pH->ht = new_ht;
  pH->htsize = new_size;
  xHash = hashFunction(pH->keyClass);
  for(elem=pH->first, pH->first=0; elem; elem = next_elem){
    int h = (*xHash)(elem->pKey, elem->nKey) & (new_size-1);
    next_elem = elem->next;
    insertElement(pH, &new_ht[h], elem);
  }
}

/* This function (for internal use only) locates an element in an
** hash table that matches the given key.  The hash for this key has
** already been computed and is passed as the 4th parameter.
*/
static HashElem *findElementGivenHash(
  const Hash *pH,     /* The pH to be searched */
  const void *pKey,   /* The key we are searching for */
  int nKey,
  int h               /* The hash for this key. */
){
  HashElem *elem;                /* Used to loop thru the element list */
  int count;                     /* Number of elements left to test */
  int (*xCompare)(const void*,int,const void*,int);  /* comparison function */

  if( pH->ht ){
    struct _ht *pEntry = &pH->ht[h];
    elem = pEntry->chain;
    count = pEntry->count;
    xCompare = compareFunction(pH->keyClass);
    while( count-- && elem ){
      if( (*xCompare)(elem->pKey,elem->nKey,pKey,nKey)==0 ){ 
        return elem;
      }
      elem = elem->next;
    }
  }
  return 0;
}

/* Remove a single entry from the hash table given a pointer to that
** element and a hash on the element's key.
*/
static void removeElementGivenHash(
  Hash *pH,         /* The pH containing "elem" */
  HashElem* elem,   /* The element to be removed from the pH */
  int h             /* Hash value for the element */
){
  struct _ht *pEntry;
  if( elem->prev ){
    elem->prev->next = elem->next; 
  }else{
    pH->first = elem->next;
  }
  if( elem->next ){
    elem->next->prev = elem->prev;
  }
  pEntry = &pH->ht[h];
  if( pEntry->chain==elem ){
    pEntry->chain = elem->next;
  }
  pEntry->count--;
  if( pEntry->count<=0 ){
    pEntry->chain = 0;
  }
  if( pH->copyKey ){
    pH->xFree(elem->pKey);
  }
  pH->xFree( elem );
  pH->count--;
  if( pH->count<=0 ){
    assert( pH->first==0 );
    assert( pH->count==0 );
    sqlite3HashClear(pH);
  }
}

/* Attempt to locate an element of the hash table pH with a key
** that matches pKey,nKey.  Return the data for this element if it is
** found, or NULL if there is no match.
*/
SQLITE_PRIVATE void *sqlite3HashFind(const Hash *pH, const void *pKey, int nKey){
  int h;             /* A hash on key */
  HashElem *elem;    /* The element that matches key */
  int (*xHash)(const void*,int);  /* The hash function */

  if( pH==0 || pH->ht==0 ) return 0;
  xHash = hashFunction(pH->keyClass);
  assert( xHash!=0 );
  h = (*xHash)(pKey,nKey);
  assert( (pH->htsize & (pH->htsize-1))==0 );
  elem = findElementGivenHash(pH,pKey,nKey, h & (pH->htsize-1));
  return elem ? elem->data : 0;
}

/* Insert an element into the hash table pH.  The key is pKey,nKey
** and the data is "data".
**
** If no element exists with a matching key, then a new
** element is created.  A copy of the key is made if the copyKey
** flag is set.  NULL is returned.
**
** If another element already exists with the same key, then the
** new data replaces the old data and the old data is returned.
** The key is not copied in this instance.  If a malloc fails, then
** the new data is returned and the hash table is unchanged.
**
** If the "data" parameter to this function is NULL, then the
** element corresponding to "key" is removed from the hash table.
*/
SQLITE_PRIVATE void *sqlite3HashInsert(Hash *pH, const void *pKey, int nKey, void *data){
  int hraw;             /* Raw hash value of the key */
  int h;                /* the hash of the key modulo hash table size */
  HashElem *elem;       /* Used to loop thru the element list */
  HashElem *new_elem;   /* New element added to the pH */
  int (*xHash)(const void*,int);  /* The hash function */

  assert( pH!=0 );
  xHash = hashFunction(pH->keyClass);
  assert( xHash!=0 );
  hraw = (*xHash)(pKey, nKey);
  assert( (pH->htsize & (pH->htsize-1))==0 );
  h = hraw & (pH->htsize-1);
  elem = findElementGivenHash(pH,pKey,nKey,h);
  if( elem ){
    void *old_data = elem->data;
    if( data==0 ){
      removeElementGivenHash(pH,elem,h);
    }else{
      elem->data = data;
    }
    return old_data;
  }
  if( data==0 ) return 0;
  new_elem = (HashElem*)pH->xMalloc( sizeof(HashElem) );
  if( new_elem==0 ) return data;
  if( pH->copyKey && pKey!=0 ){
    new_elem->pKey = pH->xMalloc( nKey );
    if( new_elem->pKey==0 ){
      pH->xFree(new_elem);
      return data;
    }
    memcpy((void*)new_elem->pKey, pKey, nKey);
  }else{
    new_elem->pKey = (void*)pKey;
  }
  new_elem->nKey = nKey;
  pH->count++;
  if( pH->htsize==0 ){
    rehash(pH,8);
    if( pH->htsize==0 ){
      pH->count = 0;
      if( pH->copyKey ){
        pH->xFree(new_elem->pKey);
      }
      pH->xFree(new_elem);
      return data;
    }
  }
  if( pH->count > pH->htsize ){
    rehash(pH,pH->htsize*2);
  }
  assert( pH->htsize>0 );
  assert( (pH->htsize & (pH->htsize-1))==0 );
  h = hraw & (pH->htsize-1);
  insertElement(pH, &pH->ht[h], new_elem);
  new_elem->data = data;
  return 0;
}

/************** End of hash.c ************************************************/
/************** Begin file opcodes.c *****************************************/
/* Automatically generated.  Do not edit */
/* See the mkopcodec.awk script for details. */
#if !defined(SQLITE_OMIT_EXPLAIN) || !defined(NDEBUG) || defined(VDBE_PROFILE) || defined(SQLITE_DEBUG)
const char *const sqlite3OpcodeNames[] = { "?",
 /*   1 */ "MemLoad",
 /*   2 */ "VNext",
 /*   3 */ "Column",
 /*   4 */ "SetCookie",
 /*   5 */ "IfMemPos",
 /*   6 */ "Sequence",
 /*   7 */ "MoveGt",
 /*   8 */ "RowKey",
 /*   9 */ "OpenWrite",
 /*  10 */ "If",
 /*  11 */ "Pop",
 /*  12 */ "VRowid",
 /*  13 */ "CollSeq",
 /*  14 */ "OpenRead",
 /*  15 */ "Expire",
 /*  16 */ "Not",
 /*  17 */ "AutoCommit",
 /*  18 */ "IntegrityCk",
 /*  19 */ "Sort",
 /*  20 */ "Function",
 /*  21 */ "Noop",
 /*  22 */ "Return",
 /*  23 */ "NewRowid",
 /*  24 */ "IfMemNeg",
 /*  25 */ "Variable",
 /*  26 */ "String",
 /*  27 */ "RealAffinity",
 /*  28 */ "VRename",
 /*  29 */ "ParseSchema",
 /*  30 */ "VOpen",
 /*  31 */ "Close",
 /*  32 */ "CreateIndex",
 /*  33 */ "IsUnique",
 /*  34 */ "NotFound",
 /*  35 */ "Int64",
 /*  36 */ "MustBeInt",
 /*  37 */ "Halt",
 /*  38 */ "Rowid",
 /*  39 */ "IdxLT",
 /*  40 */ "AddImm",
 /*  41 */ "Statement",
 /*  42 */ "RowData",
 /*  43 */ "MemMax",
 /*  44 */ "Push",
 /*  45 */ "NotExists",
 /*  46 */ "MemIncr",
 /*  47 */ "Gosub",
 /*  48 */ "Integer",
 /*  49 */ "MemInt",
 /*  50 */ "Prev",
 /*  51 */ "VColumn",
 /*  52 */ "CreateTable",
 /*  53 */ "Last",
 /*  54 */ "IncrVacuum",
 /*  55 */ "IdxRowid",
 /*  56 */ "MakeIdxRec",
 /*  57 */ "ResetCount",
 /*  58 */ "FifoWrite",
 /*  59 */ "Callback",
 /*  60 */ "Or",
 /*  61 */ "And",
 /*  62 */ "ContextPush",
 /*  63 */ "DropTrigger",
 /*  64 */ "DropIndex",
 /*  65 */ "IsNull",
 /*  66 */ "NotNull",
 /*  67 */ "Ne",
 /*  68 */ "Eq",
 /*  69 */ "Gt",
 /*  70 */ "Le",
 /*  71 */ "Lt",
 /*  72 */ "Ge",
 /*  73 */ "IdxGE",
 /*  74 */ "BitAnd",
 /*  75 */ "BitOr",
 /*  76 */ "ShiftLeft",
 /*  77 */ "ShiftRight",
 /*  78 */ "Add",
 /*  79 */ "Subtract",
 /*  80 */ "Multiply",
 /*  81 */ "Divide",
 /*  82 */ "Remainder",
 /*  83 */ "Concat",
 /*  84 */ "IdxDelete",
 /*  85 */ "Negative",
 /*  86 */ "Vacuum",
 /*  87 */ "BitNot",
 /*  88 */ "String8",
 /*  89 */ "MoveLe",
 /*  90 */ "IfNot",
 /*  91 */ "DropTable",
 /*  92 */ "MakeRecord",
 /*  93 */ "Delete",
 /*  94 */ "AggFinal",
 /*  95 */ "Dup",
 /*  96 */ "Goto",
 /*  97 */ "TableLock",
 /*  98 */ "FifoRead",
 /*  99 */ "Clear",
 /* 100 */ "IdxGT",
 /* 101 */ "MoveLt",
 /* 102 */ "VerifyCookie",
 /* 103 */ "AggStep",
 /* 104 */ "Pull",
 /* 105 */ "SetNumColumns",
 /* 106 */ "AbsValue",
 /* 107 */ "Transaction",
 /* 108 */ "VFilter",
 /* 109 */ "VDestroy",
 /* 110 */ "ContextPop",
 /* 111 */ "Next",
 /* 112 */ "IdxInsert",
 /* 113 */ "Distinct",
 /* 114 */ "Insert",
 /* 115 */ "Destroy",
 /* 116 */ "ReadCookie",
 /* 117 */ "ForceInt",
 /* 118 */ "LoadAnalysis",
 /* 119 */ "Explain",
 /* 120 */ "IfMemZero",
 /* 121 */ "OpenPseudo",
 /* 122 */ "OpenEphemeral",
 /* 123 */ "Null",
 /* 124 */ "Blob",
 /* 125 */ "Real",
 /* 126 */ "HexBlob",
 /* 127 */ "MemStore",
 /* 128 */ "Rewind",
 /* 129 */ "MoveGe",
 /* 130 */ "VBegin",
 /* 131 */ "VUpdate",
 /* 132 */ "VCreate",
 /* 133 */ "MemMove",
 /* 134 */ "MemNull",
 /* 135 */ "Found",
 /* 136 */ "NullRow",
 /* 137 */ "NotUsed_137",
 /* 138 */ "ToText",
 /* 139 */ "ToBlob",
 /* 140 */ "ToNumeric",
 /* 141 */ "ToInt",
 /* 142 */ "ToReal",
};
#endif

/************** End of opcodes.c *********************************************/
/************** Begin file os_os2.c ******************************************/
/*
** 2006 Feb 14
**
** The author disclaims copyright to this source code.  In place of
** a legal notice, here is a blessing:
**
**    May you do good and not evil.
**    May you find forgiveness for yourself and forgive others.
**    May you share freely, never taking more than you give.
**
******************************************************************************
**
** This file contains code that is specific to OS/2.
*/


#if OS_OS2

/*
** Macros used to determine whether or not to use threads.
*/
#if defined(THREADSAFE) && THREADSAFE
# define SQLITE_OS2_THREADS 1
#endif

/*
** Include code that is common to all os_*.c files
*/
/************** Include os_common.h in the middle of os_os2.c ****************/
/************** Begin file os_common.h ***************************************/
/*
** 2004 May 22
**
** The author disclaims copyright to this source code.  In place of
** a legal notice, here is a blessing:
**
**    May you do good and not evil.
**    May you find forgiveness for yourself and forgive others.
**    May you share freely, never taking more than you give.
**
******************************************************************************
**
** This file contains macros and a little bit of code that is common to
** all of the platform-specific files (os_*.c) and is #included into those
** files.
**
** This file should be #included by the os_*.c files only.  It is not a
** general purpose header file.
*/

/*
** At least two bugs have slipped in because we changed the MEMORY_DEBUG
** macro to SQLITE_DEBUG and some older makefiles have not yet made the
** switch.  The following code should catch this problem at compile-time.
*/
#ifdef MEMORY_DEBUG
# error "The MEMORY_DEBUG macro is obsolete.  Use SQLITE_DEBUG instead."
#endif


/*
 * When testing, this global variable stores the location of the
 * pending-byte in the database file.
 */
#ifdef SQLITE_TEST
unsigned int sqlite3_pending_byte = 0x40000000;
#endif

int sqlite3_os_trace = 0;
#ifdef SQLITE_DEBUG
#define OSTRACE1(X)         if( sqlite3_os_trace ) sqlite3DebugPrintf(X)
#define OSTRACE2(X,Y)       if( sqlite3_os_trace ) sqlite3DebugPrintf(X,Y)
#define OSTRACE3(X,Y,Z)     if( sqlite3_os_trace ) sqlite3DebugPrintf(X,Y,Z)
#define OSTRACE4(X,Y,Z,A)   if( sqlite3_os_trace ) sqlite3DebugPrintf(X,Y,Z,A)
#define OSTRACE5(X,Y,Z,A,B) if( sqlite3_os_trace ) sqlite3DebugPrintf(X,Y,Z,A,B)
#define OSTRACE6(X,Y,Z,A,B,C) \
    if(sqlite3_os_trace) sqlite3DebugPrintf(X,Y,Z,A,B,C)
#define OSTRACE7(X,Y,Z,A,B,C,D) \
    if(sqlite3_os_trace) sqlite3DebugPrintf(X,Y,Z,A,B,C,D)
#else
#define OSTRACE1(X)
#define OSTRACE2(X,Y)
#define OSTRACE3(X,Y,Z)
#define OSTRACE4(X,Y,Z,A)
#define OSTRACE5(X,Y,Z,A,B)
#define OSTRACE6(X,Y,Z,A,B,C)
#define OSTRACE7(X,Y,Z,A,B,C,D)
#endif

/*
** Macros for performance tracing.  Normally turned off.  Only works
** on i486 hardware.
*/
#ifdef SQLITE_PERFORMANCE_TRACE
__inline__ unsigned long long int hwtime(void){
  unsigned long long int x;
  __asm__("rdtsc\n\t"
          "mov %%edx, %%ecx\n\t"
          :"=A" (x));
  return x;
}
static unsigned long long int g_start;
static unsigned int elapse;
#define TIMER_START       g_start=hwtime()
#define TIMER_END         elapse=hwtime()-g_start
#define TIMER_ELAPSED     elapse
#else
#define TIMER_START
#define TIMER_END
#define TIMER_ELAPSED     0
#endif

/*
** If we compile with the SQLITE_TEST macro set, then the following block
** of code will give us the ability to simulate a disk I/O error.  This
** is used for testing the I/O recovery logic.
*/
#ifdef SQLITE_TEST
int sqlite3_io_error_hit = 0;
int sqlite3_io_error_pending = 0;
int sqlite3_io_error_persist = 0;
int sqlite3_diskfull_pending = 0;
int sqlite3_diskfull = 0;
#define SimulateIOError(CODE)  \
  if( sqlite3_io_error_pending || sqlite3_io_error_hit ) \
     if( sqlite3_io_error_pending-- == 1 \
         || (sqlite3_io_error_persist && sqlite3_io_error_hit) ) \
                { local_ioerr(); CODE; }
static void local_ioerr(){
  IOTRACE(("IOERR\n"));
  sqlite3_io_error_hit = 1;
}
#define SimulateDiskfullError(CODE) \
   if( sqlite3_diskfull_pending ){ \
     if( sqlite3_diskfull_pending == 1 ){ \
       local_ioerr(); \
       sqlite3_diskfull = 1; \
       sqlite3_io_error_hit = 1; \
       CODE; \
     }else{ \
       sqlite3_diskfull_pending--; \
     } \
   }
#else
#define SimulateIOError(A)
#define SimulateDiskfullError(A)
#endif

/*
** When testing, keep a count of the number of open files.
*/
#ifdef SQLITE_TEST
int sqlite3_open_file_count = 0;
#define OpenCounter(X)  sqlite3_open_file_count+=(X)
#else
#define OpenCounter(X)
#endif

/*
** sqlite3GenericMalloc
** sqlite3GenericRealloc
** sqlite3GenericOsFree
** sqlite3GenericAllocationSize
**
** Implementation of the os level dynamic memory allocation interface in terms
** of the standard malloc(), realloc() and free() found in many operating
** systems. No rocket science here.
**
** There are two versions of these four functions here. The version
** implemented here is only used if memory-management or memory-debugging is
** enabled. This version allocates an extra 8-bytes at the beginning of each
** block and stores the size of the allocation there.
**
** If neither memory-management or debugging is enabled, the second
** set of implementations is used instead.
*/
#if defined(SQLITE_ENABLE_MEMORY_MANAGEMENT) || defined (SQLITE_MEMDEBUG)
SQLITE_PRIVATE void *sqlite3GenericMalloc(int n){
  char *p = (char *)malloc(n+8);
  assert(n>0);
  assert(sizeof(int)<=8);
  if( p ){
    *(int *)p = n;
    p += 8;
  }
  return (void *)p;
}
SQLITE_PRIVATE void *sqlite3GenericRealloc(void *p, int n){
  char *p2 = ((char *)p - 8);
  assert(n>0);
  p2 = (char*)realloc(p2, n+8);
  if( p2 ){
    *(int *)p2 = n;
    p2 += 8;
  }
  return (void *)p2;
}
SQLITE_PRIVATE void sqlite3GenericFree(void *p){
  assert(p);
  free((void *)((char *)p - 8));
}
SQLITE_PRIVATE int sqlite3GenericAllocationSize(void *p){
  return p ? *(int *)((char *)p - 8) : 0;
}
#else
SQLITE_PRIVATE void *sqlite3GenericMalloc(int n){
  char *p = (char *)malloc(n);
  return (void *)p;
}
SQLITE_PRIVATE void *sqlite3GenericRealloc(void *p, int n){
  assert(n>0);
  p = realloc(p, n);
  return p;
}
SQLITE_PRIVATE void sqlite3GenericFree(void *p){
  assert(p);
  free(p);
}
/* Never actually used, but needed for the linker */
SQLITE_PRIVATE int sqlite3GenericAllocationSize(void *p){ return 0; }
#endif

/*
** The default size of a disk sector
*/
#ifndef PAGER_SECTOR_SIZE
# define PAGER_SECTOR_SIZE 512
#endif

/************** End of os_common.h *******************************************/
/************** Continuing where we left off in os_os2.c *********************/

/*
** The os2File structure is subclass of OsFile specific for the OS/2
** protability layer.
*/
typedef struct os2File os2File;
struct os2File {
  IoMethod const *pMethod;  /* Always the first entry */
  HFILE h;                  /* Handle for accessing the file */
  int delOnClose;           /* True if file is to be deleted on close */
  char* pathToDel;          /* Name of file to delete on close */
  unsigned char locktype;   /* Type of lock currently held on this file */
};

/*
** Do not include any of the File I/O interface procedures if the
** SQLITE_OMIT_DISKIO macro is defined (indicating that there database
** will be in-memory only)
*/
#ifndef SQLITE_OMIT_DISKIO

/*
** Delete the named file
*/
SQLITE_PRIVATE int sqlite3Os2Delete( const char *zFilename ){
  APIRET rc = NO_ERROR;

  rc = DosDelete( (PSZ)zFilename );
  OSTRACE2( "DELETE \"%s\"\n", zFilename );
  return rc == NO_ERROR ? SQLITE_OK : SQLITE_IOERR;
}

/*
** Return TRUE if the named file exists.
*/
SQLITE_PRIVATE int sqlite3Os2FileExists( const char *zFilename ){
  FILESTATUS3 fsts3ConfigInfo;
  memset(&fsts3ConfigInfo, 0, sizeof(fsts3ConfigInfo));
  return DosQueryPathInfo( (PSZ)zFilename, FIL_STANDARD,
        &fsts3ConfigInfo, sizeof(FILESTATUS3) ) == NO_ERROR;
}

/* Forward declaration */
int allocateOs2File( os2File *pInit, OsFile **pld );

/*
** Attempt to open a file for both reading and writing.  If that
** fails, try opening it read-only.  If the file does not exist,
** try to create it.
**
** On success, a handle for the open file is written to *id
** and *pReadonly is set to 0 if the file was opened for reading and
** writing or 1 if the file was opened read-only.  The function returns
** SQLITE_OK.
**
** On failure, the function returns SQLITE_CANTOPEN and leaves
** *id and *pReadonly unchanged.
*/
SQLITE_PRIVATE int sqlite3Os2OpenReadWrite(
  const char *zFilename,
  OsFile **pld,
  int *pReadonly
){
  os2File  f;
  HFILE    hf;
  ULONG    ulAction;
  APIRET   rc = NO_ERROR;

  assert( *pld == 0 );
  rc = DosOpen( (PSZ)zFilename, &hf, &ulAction, 0L,
            FILE_ARCHIVED | FILE_NORMAL,
                OPEN_ACTION_CREATE_IF_NEW | OPEN_ACTION_OPEN_IF_EXISTS,
                OPEN_FLAGS_FAIL_ON_ERROR | OPEN_FLAGS_RANDOM |
                    OPEN_SHARE_DENYNONE | OPEN_ACCESS_READWRITE, (PEAOP2)NULL );
  if( rc != NO_ERROR ){
    rc = DosOpen( (PSZ)zFilename, &hf, &ulAction, 0L,
            FILE_ARCHIVED | FILE_NORMAL,
                OPEN_ACTION_CREATE_IF_NEW | OPEN_ACTION_OPEN_IF_EXISTS,
                OPEN_FLAGS_FAIL_ON_ERROR | OPEN_FLAGS_RANDOM |
                        OPEN_SHARE_DENYWRITE | OPEN_ACCESS_READONLY, (PEAOP2)NULL );
    if( rc != NO_ERROR ){
        return SQLITE_CANTOPEN;
    }
    *pReadonly = 1;
  }
  else{
    *pReadonly = 0;
  }
  f.h = hf;
  f.locktype = NO_LOCK;
  f.delOnClose = 0;
  f.pathToDel = NULL;
  OpenCounter(+1);
  OSTRACE3( "OPEN R/W %d \"%s\"\n", hf, zFilename );
  return allocateOs2File( &f, pld );
}


/*
** Attempt to open a new file for exclusive access by this process.
** The file will be opened for both reading and writing.  To avoid
** a potential security problem, we do not allow the file to have
** previously existed.  Nor do we allow the file to be a symbolic
** link.
**
** If delFlag is true, then make arrangements to automatically delete
** the file when it is closed.
**
** On success, write the file handle into *id and return SQLITE_OK.
**
** On failure, return SQLITE_CANTOPEN.
*/
SQLITE_PRIVATE int sqlite3Os2OpenExclusive( const char *zFilename, OsFile **pld, int delFlag ){
  os2File  f;
  HFILE    hf;
  ULONG    ulAction;
  APIRET   rc = NO_ERROR;

  assert( *pld == 0 );
  rc = DosOpen( (PSZ)zFilename, &hf, &ulAction, 0L, FILE_NORMAL,
            OPEN_ACTION_CREATE_IF_NEW | OPEN_ACTION_REPLACE_IF_EXISTS,
            OPEN_FLAGS_FAIL_ON_ERROR | OPEN_FLAGS_RANDOM |
                OPEN_SHARE_DENYREADWRITE | OPEN_ACCESS_READWRITE, (PEAOP2)NULL );
  if( rc != NO_ERROR ){
    return SQLITE_CANTOPEN;
  }

  f.h = hf;
  f.locktype = NO_LOCK;
  f.delOnClose = delFlag ? 1 : 0;
  f.pathToDel = delFlag ? sqlite3OsFullPathname( zFilename ) : NULL;
  OpenCounter( +1 );
  if( delFlag ) DosForceDelete( (PSZ)sqlite3OsFullPathname( zFilename ) );
  OSTRACE3( "OPEN EX %d \"%s\"\n", hf, sqlite3OsFullPathname ( zFilename ) );
  return allocateOs2File( &f, pld );
}

/*
** Attempt to open a new file for read-only access.
**
** On success, write the file handle into *id and return SQLITE_OK.
**
** On failure, return SQLITE_CANTOPEN.
*/
SQLITE_PRIVATE int sqlite3Os2OpenReadOnly( const char *zFilename, OsFile **pld ){
  os2File  f;
  HFILE    hf;
  ULONG    ulAction;
  APIRET   rc = NO_ERROR;

  assert( *pld == 0 );
  rc = DosOpen( (PSZ)zFilename, &hf, &ulAction, 0L,
            FILE_NORMAL, OPEN_ACTION_OPEN_IF_EXISTS,
            OPEN_FLAGS_FAIL_ON_ERROR | OPEN_FLAGS_RANDOM |
                OPEN_SHARE_DENYWRITE | OPEN_ACCESS_READONLY, (PEAOP2)NULL );
  if( rc != NO_ERROR ){
    return SQLITE_CANTOPEN;
  }
  f.h = hf;
  f.locktype = NO_LOCK;
  f.delOnClose = 0;
  f.pathToDel = NULL;
  OpenCounter( +1 );
  OSTRACE3( "OPEN RO %d \"%s\"\n", hf, zFilename );
  return allocateOs2File( &f, pld );
}

/*
** Attempt to open a file descriptor for the directory that contains a
** file.  This file descriptor can be used to fsync() the directory
** in order to make sure the creation of a new file is actually written
** to disk.
**
** This routine is only meaningful for Unix.  It is a no-op under
** OS/2 since OS/2 does not support hard links.
**
** On success, a handle for a previously open file is at *id is
** updated with the new directory file descriptor and SQLITE_OK is
** returned.
**
** On failure, the function returns SQLITE_CANTOPEN and leaves
** *id unchanged.
*/
int os2OpenDirectory(
  OsFile *id,
  const char *zDirname
){
  return SQLITE_OK;
}

/*
** Create a temporary file name in zBuf.  zBuf must be big enough to
** hold at least SQLITE_TEMPNAME_SIZE characters.
*/
SQLITE_PRIVATE int sqlite3Os2TempFileName( char *zBuf ){
  static const unsigned char zChars[] =
    "abcdefghijklmnopqrstuvwxyz"
    "ABCDEFGHIJKLMNOPQRSTUVWXYZ"
    "0123456789";
  int i, j;
  PSZ zTempPath = 0;
  if( DosScanEnv( (PSZ)"TEMP", &zTempPath ) ){
    if( DosScanEnv( (PSZ)"TMP", &zTempPath ) ){
      if( DosScanEnv( (PSZ)"TMPDIR", &zTempPath ) ){
           ULONG ulDriveNum = 0, ulDriveMap = 0;
           DosQueryCurrentDisk( &ulDriveNum, &ulDriveMap );
           sprintf( (char*)zTempPath, "%c:", (char)( 'A' + ulDriveNum - 1 ) );
      }
    }
  }
  for(;;){
      sprintf( zBuf, "%s\\"TEMP_FILE_PREFIX, zTempPath );
      j = strlen( zBuf );
      sqlite3Randomness( 15, &zBuf[j] );
      for( i = 0; i < 15; i++, j++ ){
        zBuf[j] = (char)zChars[ ((unsigned char)zBuf[j])%(sizeof(zChars)-1) ];
      }
      zBuf[j] = 0;
      if( !sqlite3OsFileExists( zBuf ) ) break;
  }
  OSTRACE2( "TEMP FILENAME: %s\n", zBuf );
  return SQLITE_OK;
}

/*
** Close a file.
*/
int os2Close( OsFile **pld ){
  os2File *pFile;
  APIRET rc = NO_ERROR;
  if( pld && (pFile = (os2File*)*pld) != 0 ){
    OSTRACE2( "CLOSE %d\n", pFile->h );
    rc = DosClose( pFile->h );
    pFile->locktype = NO_LOCK;
    if( pFile->delOnClose != 0 ){
        rc = DosForceDelete( (PSZ)pFile->pathToDel );
    }
    *pld = 0;
    OpenCounter( -1 );
  }

  return rc == NO_ERROR ? SQLITE_OK : SQLITE_IOERR;
}

/*
** Read data from a file into a buffer.  Return SQLITE_OK if all
** bytes were read successfully and SQLITE_IOERR if anything goes
** wrong.
*/
int os2Read( OsFile *id, void *pBuf, int amt ){
  ULONG got;
  assert( id!=0 );
  SimulateIOError( return SQLITE_IOERR );
  OSTRACE3( "READ %d lock=%d\n", ((os2File*)id)->h, ((os2File*)id)->locktype );
  DosRead( ((os2File*)id)->h, pBuf, amt, &got );
  if (got == (ULONG)amt)
    return SQLITE_OK;
  else if (got == 0)
    return SQLITE_IOERR_READ;
  else {
    memset(&((char*)pBuf)[got], 0, amt-got);
    return SQLITE_IOERR_SHORT_READ;
  }
}

/*
** Write data from a buffer into a file.  Return SQLITE_OK on success
** or some other error code on failure.
*/
int os2Write( OsFile *id, const void *pBuf, int amt ){
  APIRET rc = NO_ERROR;
  ULONG wrote;
  assert( id!=0 );
  SimulateIOError( return SQLITE_IOERR );
  SimulateDiskfullError( return SQLITE_FULL );
  OSTRACE3( "WRITE %d lock=%d\n", ((os2File*)id)->h, ((os2File*)id)->locktype );
  while( amt > 0 &&
      (rc = DosWrite( ((os2File*)id)->h, (PVOID)pBuf, amt, &wrote )) && wrote > 0 ){
      amt -= wrote;
      pBuf = &((char*)pBuf)[wrote];
  }

  return ( rc != NO_ERROR || amt > (int)wrote ) ? SQLITE_FULL : SQLITE_OK;
}

/*
** Move the read/write pointer in a file.
*/
int os2Seek( OsFile *id, i64 offset ){
  APIRET rc = NO_ERROR;
  ULONG filePointer = 0L;
  assert( id!=0 );
  rc = DosSetFilePtr( ((os2File*)id)->h, offset, FILE_BEGIN, &filePointer );
  OSTRACE3( "SEEK %d %lld\n", ((os2File*)id)->h, offset );
  return rc == NO_ERROR ? SQLITE_OK : SQLITE_IOERR;
}

/*
** Make sure all writes to a particular file are committed to disk.
*/
int os2Sync( OsFile *id, int dataOnly ){
  assert( id!=0 );
  OSTRACE3( "SYNC %d lock=%d\n", ((os2File*)id)->h, ((os2File*)id)->locktype );
  return DosResetBuffer( ((os2File*)id)->h ) == NO_ERROR ? SQLITE_OK : SQLITE_IOERR;
}

/*
** Sync the directory zDirname. This is a no-op on operating systems other
** than UNIX.
*/
SQLITE_PRIVATE int sqlite3Os2SyncDirectory( const char *zDirname ){
  SimulateIOError( return SQLITE_IOERR );
  return SQLITE_OK;
}

/*
** Truncate an open file to a specified size
*/
int os2Truncate( OsFile *id, i64 nByte ){
  APIRET rc = NO_ERROR;
  ULONG upperBits = nByte>>32;
  assert( id!=0 );
  OSTRACE3( "TRUNCATE %d %lld\n", ((os2File*)id)->h, nByte );
  SimulateIOError( return SQLITE_IOERR );
  rc = DosSetFilePtr( ((os2File*)id)->h, nByte, FILE_BEGIN, &upperBits );
  if( rc != NO_ERROR ){
    return SQLITE_IOERR;
  }
  rc = DosSetFilePtr( ((os2File*)id)->h, 0L, FILE_END, &upperBits );
  return rc == NO_ERROR ? SQLITE_OK : SQLITE_IOERR;
}

/*
** Determine the current size of a file in bytes
*/
int os2FileSize( OsFile *id, i64 *pSize ){
  APIRET rc = NO_ERROR;
  FILESTATUS3 fsts3FileInfo;
  memset(&fsts3FileInfo, 0, sizeof(fsts3FileInfo));
  assert( id!=0 );
  SimulateIOError( return SQLITE_IOERR );
  rc = DosQueryFileInfo( ((os2File*)id)->h, FIL_STANDARD, &fsts3FileInfo, sizeof(FILESTATUS3) );
  if( rc == NO_ERROR ){
    *pSize = fsts3FileInfo.cbFile;
    return SQLITE_OK;
  }
  else{
    return SQLITE_IOERR;
  }
}

/*
** Acquire a reader lock.
*/
static int getReadLock( os2File *id ){
  FILELOCK  LockArea,
            UnlockArea;
  memset(&LockArea, 0, sizeof(LockArea));
  memset(&UnlockArea, 0, sizeof(UnlockArea));
  LockArea.lOffset = SHARED_FIRST;
  LockArea.lRange = SHARED_SIZE;
  UnlockArea.lOffset = 0L;
  UnlockArea.lRange = 0L;
  return DosSetFileLocks( id->h, &UnlockArea, &LockArea, 2000L, 1L );
}

/*
** Undo a readlock
*/
static int unlockReadLock( os2File *id ){
  FILELOCK  LockArea,
            UnlockArea;
  memset(&LockArea, 0, sizeof(LockArea));
  memset(&UnlockArea, 0, sizeof(UnlockArea));
  LockArea.lOffset = 0L;
  LockArea.lRange = 0L;
  UnlockArea.lOffset = SHARED_FIRST;
  UnlockArea.lRange = SHARED_SIZE;
  return DosSetFileLocks( id->h, &UnlockArea, &LockArea, 2000L, 1L );
}

#ifndef SQLITE_OMIT_PAGER_PRAGMAS
/*
** Check that a given pathname is a directory and is writable
**
*/
SQLITE_PRIVATE int sqlite3Os2IsDirWritable( char *zDirname ){
  FILESTATUS3 fsts3ConfigInfo;
  APIRET rc = NO_ERROR;
  memset(&fsts3ConfigInfo, 0, sizeof(fsts3ConfigInfo));
  if( zDirname==0 ) return 0;
  if( strlen(zDirname)>CCHMAXPATH ) return 0;
  rc = DosQueryPathInfo( (PSZ)zDirname, FIL_STANDARD, &fsts3ConfigInfo, sizeof(FILESTATUS3) );
  if( rc != NO_ERROR ) return 0;
  if( (fsts3ConfigInfo.attrFile & FILE_DIRECTORY) != FILE_DIRECTORY ) return 0;

  return 1;
}
#endif /* SQLITE_OMIT_PAGER_PRAGMAS */

/*
** Lock the file with the lock specified by parameter locktype - one
** of the following:
**
**     (1) SHARED_LOCK
**     (2) RESERVED_LOCK
**     (3) PENDING_LOCK
**     (4) EXCLUSIVE_LOCK
**
** Sometimes when requesting one lock state, additional lock states
** are inserted in between.  The locking might fail on one of the later
** transitions leaving the lock state different from what it started but
** still short of its goal.  The following chart shows the allowed
** transitions and the inserted intermediate states:
**
**    UNLOCKED -> SHARED
**    SHARED -> RESERVED
**    SHARED -> (PENDING) -> EXCLUSIVE
**    RESERVED -> (PENDING) -> EXCLUSIVE
**    PENDING -> EXCLUSIVE
**
** This routine will only increase a lock.  The os2Unlock() routine
** erases all locks at once and returns us immediately to locking level 0.
** It is not possible to lower the locking level one step at a time.  You
** must go straight to locking level 0.
*/
int os2Lock( OsFile *id, int locktype ){
  APIRET rc = SQLITE_OK;    /* Return code from subroutines */
  APIRET res = NO_ERROR;    /* Result of an OS/2 lock call */
  int newLocktype;       /* Set id->locktype to this value before exiting */
  int gotPendingLock = 0;/* True if we acquired a PENDING lock this time */
  FILELOCK  LockArea,
            UnlockArea;
  os2File *pFile = (os2File*)id;
  memset(&LockArea, 0, sizeof(LockArea));
  memset(&UnlockArea, 0, sizeof(UnlockArea));
  assert( pFile!=0 );
  OSTRACE4( "LOCK %d %d was %d\n", pFile->h, locktype, pFile->locktype );

  /* If there is already a lock of this type or more restrictive on the
  ** OsFile, do nothing. Don't use the end_lock: exit path, as
  ** sqlite3OsEnterMutex() hasn't been called yet.
  */
  if( pFile->locktype>=locktype ){
    return SQLITE_OK;
  }

  /* Make sure the locking sequence is correct
  */
  assert( pFile->locktype!=NO_LOCK || locktype==SHARED_LOCK );
  assert( locktype!=PENDING_LOCK );
  assert( locktype!=RESERVED_LOCK || pFile->locktype==SHARED_LOCK );

  /* Lock the PENDING_LOCK byte if we need to acquire a PENDING lock or
  ** a SHARED lock.  If we are acquiring a SHARED lock, the acquisition of
  ** the PENDING_LOCK byte is temporary.
  */
  newLocktype = pFile->locktype;
  if( pFile->locktype==NO_LOCK
   || (locktype==EXCLUSIVE_LOCK && pFile->locktype==RESERVED_LOCK)
  ){
    int cnt = 3;

    LockArea.lOffset = PENDING_BYTE;
    LockArea.lRange = 1L;
    UnlockArea.lOffset = 0L;
    UnlockArea.lRange = 0L;

    while( cnt-->0 && (res = DosSetFileLocks( pFile->h, &UnlockArea, &LockArea, 2000L, 1L) )!=NO_ERROR ){
      /* Try 3 times to get the pending lock.  The pending lock might be
      ** held by another reader process who will release it momentarily.
      */
      OSTRACE2( "could not get a PENDING lock. cnt=%d\n", cnt );
      DosSleep(1);
    }
    gotPendingLock = res;
  }

  /* Acquire a shared lock
  */
  if( locktype==SHARED_LOCK && res ){
    assert( pFile->locktype==NO_LOCK );
    res = getReadLock(pFile);
    if( res == NO_ERROR ){
      newLocktype = SHARED_LOCK;
    }
  }

  /* Acquire a RESERVED lock
  */
  if( locktype==RESERVED_LOCK && res ){
    assert( pFile->locktype==SHARED_LOCK );
    LockArea.lOffset = RESERVED_BYTE;
    LockArea.lRange = 1L;
    UnlockArea.lOffset = 0L;
    UnlockArea.lRange = 0L;
    res = DosSetFileLocks( pFile->h, &UnlockArea, &LockArea, 2000L, 1L );
    if( res == NO_ERROR ){
      newLocktype = RESERVED_LOCK;
    }
  }

  /* Acquire a PENDING lock
  */
  if( locktype==EXCLUSIVE_LOCK && res ){
    newLocktype = PENDING_LOCK;
    gotPendingLock = 0;
  }

  /* Acquire an EXCLUSIVE lock
  */
  if( locktype==EXCLUSIVE_LOCK && res ){
    assert( pFile->locktype>=SHARED_LOCK );
    res = unlockReadLock(pFile);
    OSTRACE2( "unreadlock = %d\n", res );
    LockArea.lOffset = SHARED_FIRST;
    LockArea.lRange = SHARED_SIZE;
    UnlockArea.lOffset = 0L;
    UnlockArea.lRange = 0L;
    res = DosSetFileLocks( pFile->h, &UnlockArea, &LockArea, 2000L, 1L );
    if( res == NO_ERROR ){
      newLocktype = EXCLUSIVE_LOCK;
    }else{
      OSTRACE2( "error-code = %d\n", res );
    }
  }

  /* If we are holding a PENDING lock that ought to be released, then
  ** release it now.
  */
  if( gotPendingLock && locktype==SHARED_LOCK ){
    LockArea.lOffset = 0L;
    LockArea.lRange = 0L;
    UnlockArea.lOffset = PENDING_BYTE;
    UnlockArea.lRange = 1L;
    DosSetFileLocks( pFile->h, &UnlockArea, &LockArea, 2000L, 1L );
  }

  /* Update the state of the lock has held in the file descriptor then
  ** return the appropriate result code.
  */
  if( res == NO_ERROR ){
    rc = SQLITE_OK;
  }else{
    OSTRACE4( "LOCK FAILED %d trying for %d but got %d\n", pFile->h,
           locktype, newLocktype );
    rc = SQLITE_BUSY;
  }
  pFile->locktype = newLocktype;
  return rc;
}

/*
** This routine checks if there is a RESERVED lock held on the specified
** file by this or any other process. If such a lock is held, return
** non-zero, otherwise zero.
*/
int os2CheckReservedLock( OsFile *id ){
  APIRET rc = NO_ERROR;
  os2File *pFile = (os2File*)id;
  assert( pFile!=0 );
  if( pFile->locktype>=RESERVED_LOCK ){
    rc = 1;
    OSTRACE3( "TEST WR-LOCK %d %d (local)\n", pFile->h, rc );
  }else{
    FILELOCK  LockArea,
              UnlockArea;
    memset(&LockArea, 0, sizeof(LockArea));
    memset(&UnlockArea, 0, sizeof(UnlockArea));
    LockArea.lOffset = RESERVED_BYTE;
    LockArea.lRange = 1L;
    UnlockArea.lOffset = 0L;
    UnlockArea.lRange = 0L;
    rc = DosSetFileLocks( pFile->h, &UnlockArea, &LockArea, 2000L, 1L );
    if( rc == NO_ERROR ){
      LockArea.lOffset = 0L;
      LockArea.lRange = 0L;
      UnlockArea.lOffset = RESERVED_BYTE;
      UnlockArea.lRange = 1L;
      rc = DosSetFileLocks( pFile->h, &UnlockArea, &LockArea, 2000L, 1L );
    }
    OSTRACE3( "TEST WR-LOCK %d %d (remote)\n", pFile->h, rc );
  }
  return rc;
}

/*
** Lower the locking level on file descriptor id to locktype.  locktype
** must be either NO_LOCK or SHARED_LOCK.
**
** If the locking level of the file descriptor is already at or below
** the requested locking level, this routine is a no-op.
**
** It is not possible for this routine to fail if the second argument
** is NO_LOCK.  If the second argument is SHARED_LOCK then this routine
** might return SQLITE_IOERR;
*/
int os2Unlock( OsFile *id, int locktype ){
  int type;
  APIRET rc = SQLITE_OK;
  os2File *pFile = (os2File*)id;
  FILELOCK  LockArea,
            UnlockArea;
  memset(&LockArea, 0, sizeof(LockArea));
  memset(&UnlockArea, 0, sizeof(UnlockArea));
  assert( pFile!=0 );
  assert( locktype<=SHARED_LOCK );
  OSTRACE4( "UNLOCK %d to %d was %d\n", pFile->h, locktype, pFile->locktype );
  type = pFile->locktype;
  if( type>=EXCLUSIVE_LOCK ){
    LockArea.lOffset = 0L;
    LockArea.lRange = 0L;
    UnlockArea.lOffset = SHARED_FIRST;
    UnlockArea.lRange = SHARED_SIZE;
    DosSetFileLocks( pFile->h, &UnlockArea, &LockArea, 2000L, 1L );
    if( locktype==SHARED_LOCK && getReadLock(pFile) != NO_ERROR ){
      /* This should never happen.  We should always be able to
      ** reacquire the read lock */
      rc = SQLITE_IOERR;
    }
  }
  if( type>=RESERVED_LOCK ){
    LockArea.lOffset = 0L;
    LockArea.lRange = 0L;
    UnlockArea.lOffset = RESERVED_BYTE;
    UnlockArea.lRange = 1L;
    DosSetFileLocks( pFile->h, &UnlockArea, &LockArea, 2000L, 1L );
  }
  if( locktype==NO_LOCK && type>=SHARED_LOCK ){
    unlockReadLock(pFile);
  }
  if( type>=PENDING_LOCK ){
    LockArea.lOffset = 0L;
    LockArea.lRange = 0L;
    UnlockArea.lOffset = PENDING_BYTE;
    UnlockArea.lRange = 1L;
    DosSetFileLocks( pFile->h, &UnlockArea, &LockArea, 2000L, 1L );
  }
  pFile->locktype = locktype;
  return rc;
}

/*
** Turn a relative pathname into a full pathname.  Return a pointer
** to the full pathname stored in space obtained from sqliteMalloc().
** The calling function is responsible for freeing this space once it
** is no longer needed.
*/
SQLITE_PRIVATE char *sqlite3Os2FullPathname( const char *zRelative ){
  char *zFull = 0;
  if( strchr(zRelative, ':') ){
    sqlite3SetString( &zFull, zRelative, (char*)0 );
  }else{
    ULONG ulDriveNum = 0;
    ULONG ulDriveMap = 0;
    ULONG cbzBufLen = SQLITE_TEMPNAME_SIZE;
    char zDrive[2];
    char *zBuff;

    zBuff = sqliteMalloc( cbzBufLen );
    if( zBuff != 0 ){
      DosQueryCurrentDisk( &ulDriveNum, &ulDriveMap );
      if( DosQueryCurrentDir( ulDriveNum, (PBYTE)zBuff, &cbzBufLen ) == NO_ERROR ){
        sprintf( zDrive, "%c", (char)('A' + ulDriveNum - 1) );
        sqlite3SetString( &zFull, zDrive, ":\\", zBuff,
                          "\\", zRelative, (char*)0 );
      }
      sqliteFree( zBuff );
    }
  }
  return zFull;
}

/*
** The fullSync option is meaningless on os2, or correct me if I'm wrong.  This is a no-op.
** From os_unix.c: Change the value of the fullsync flag in the given file descriptor.
** From os_unix.c: ((unixFile*)id)->fullSync = v;
*/
static void os2SetFullSync( OsFile *id, int v ){
  return;
}

/*
** Return the underlying file handle for an OsFile
*/
static int os2FileHandle( OsFile *id ){
  return (int)((os2File*)id)->h;
}

/*
** Return an integer that indices the type of lock currently held
** by this handle.  (Used for testing and analysis only.)
*/
static int os2LockState( OsFile *id ){
  return ((os2File*)id)->locktype;
}

/*
** Return the sector size in bytes of the underlying block device for
** the specified file. This is almost always 512 bytes, but may be
** larger for some devices.
**
** SQLite code assumes this function cannot fail. It also assumes that
** if two files are created in the same file-system directory (i.e.
** a database and it's journal file) that the sector size will be the
** same for both.
*/
static int os2SectorSize(OsFile *id){
  return SQLITE_DEFAULT_SECTOR_SIZE;
}

/*
** This vector defines all the methods that can operate on an OsFile
** for os2.
*/
static const IoMethod sqlite3Os2IoMethod = {
  os2Close,
  os2OpenDirectory,
  os2Read,
  os2Write,
  os2Seek,
  os2Truncate,
  os2Sync,
  os2SetFullSync,
  os2FileHandle,
  os2FileSize,
  os2Lock,
  os2Unlock,
  os2LockState,
  os2CheckReservedLock,
  os2SectorSize,
};

/*
** Allocate memory for an OsFile.  Initialize the new OsFile
** to the value given in pInit and return a pointer to the new
** OsFile.  If we run out of memory, close the file and return NULL.
*/
int allocateOs2File( os2File *pInit, OsFile **pld ){
  os2File *pNew;
  pNew = sqliteMalloc( sizeof(*pNew) );
  if( pNew==0 ){
    DosClose( pInit->h );
    *pld = 0;
    return SQLITE_NOMEM;
  }else{
    *pNew = *pInit;
    pNew->pMethod = &sqlite3Os2IoMethod;
    pNew->locktype = NO_LOCK;
    *pld = (OsFile*)pNew;
    OpenCounter(+1);
    return SQLITE_OK;
  }
}

#endif /* SQLITE_OMIT_DISKIO */
/***************************************************************************
** Everything above deals with file I/O.  Everything that follows deals
** with other miscellanous aspects of the operating system interface
****************************************************************************/

#ifndef SQLITE_OMIT_LOAD_EXTENSION
/*
** Interfaces for opening a shared library, finding entry points
** within the shared library, and closing the shared library.
*/
SQLITE_PRIVATE void *sqlite3Os2Dlopen(const char *zFilename){
  UCHAR loadErr[256];
  HMODULE hmod;
  APIRET rc;
  rc = DosLoadModule((PSZ)loadErr, sizeof(loadErr), zFilename, &hmod);
  if (rc != NO_ERROR) return 0;
  return (void*)hmod;
}
SQLITE_PRIVATE void *sqlite3Os2Dlsym(void *pHandle, const char *zSymbol){
  PFN pfn;
  APIRET rc;
  rc = DosQueryProcAddr((HMODULE)pHandle, 0L, zSymbol, &pfn);
  if (rc != NO_ERROR) {
    /* if the symbol itself was not found, search again for the same
     * symbol with an extra underscore, that might be needed depending
     * on the calling convention */
    char _zSymbol[256] = "_";
    strncat(_zSymbol, zSymbol, 255);
    rc = DosQueryProcAddr((HMODULE)pHandle, 0L, _zSymbol, &pfn);
  }
  if (rc != NO_ERROR) return 0;
  return (void *)pfn;
}
SQLITE_PRIVATE int sqlite3Os2Dlclose(void *pHandle){
  return DosFreeModule((HMODULE)pHandle);
}
#endif /* SQLITE_OMIT_LOAD_EXTENSION */


/*
** Get information to seed the random number generator.  The seed
** is written into the buffer zBuf[256].  The calling function must
** supply a sufficiently large buffer.
*/
SQLITE_PRIVATE int sqlite3Os2RandomSeed( char *zBuf ){
  /* We have to initialize zBuf to prevent valgrind from reporting
  ** errors.  The reports issued by valgrind are incorrect - we would
  ** prefer that the randomness be increased by making use of the
  ** uninitialized space in zBuf - but valgrind errors tend to worry
  ** some users.  Rather than argue, it seems easier just to initialize
  ** the whole array and silence valgrind, even if that means less randomness
  ** in the random seed.
  **
  ** When testing, initializing zBuf[] to zero is all we do.  That means
  ** that we always use the same random number sequence. This makes the
  ** tests repeatable.
  */
  memset( zBuf, 0, 256 );
  DosGetDateTime( (PDATETIME)zBuf );
  return SQLITE_OK;
}

/*
** Sleep for a little while.  Return the amount of time slept.
*/
SQLITE_PRIVATE int sqlite3Os2Sleep( int ms ){
  DosSleep( ms );
  return ms;
}

/*
** Static variables used for thread synchronization
*/
static int inMutex = 0;
#ifdef SQLITE_OS2_THREADS
static ULONG mutexOwner;
#endif

/*
** The following pair of routines implement mutual exclusion for
** multi-threaded processes.  Only a single thread is allowed to
** executed code that is surrounded by EnterMutex() and LeaveMutex().
**
** SQLite uses only a single Mutex.  There is not much critical
** code and what little there is executes quickly and without blocking.
*/
SQLITE_PRIVATE void sqlite3Os2EnterMutex(){
#ifdef SQLITE_OS2_THREADS
  PTIB ptib;
  DosEnterCritSec();
  DosGetInfoBlocks( &ptib, NULL );
  mutexOwner = ptib->tib_ptib2->tib2_ultid;
#endif
  assert( !inMutex );
  inMutex = 1;
}
SQLITE_PRIVATE void sqlite3Os2LeaveMutex(){
#ifdef SQLITE_OS2_THREADS
  PTIB ptib;
#endif
  assert( inMutex );
  inMutex = 0;
#ifdef SQLITE_OS2_THREADS
  DosGetInfoBlocks( &ptib, NULL );
  assert( mutexOwner == ptib->tib_ptib2->tib2_ultid );
  DosExitCritSec();
#endif
}

/*
** Return TRUE if the mutex is currently held.
**
** If the thisThreadOnly parameter is true, return true if and only if the
** calling thread holds the mutex.  If the parameter is false, return
** true if any thread holds the mutex.
*/
SQLITE_PRIVATE int sqlite3Os2InMutex( int thisThreadOnly ){
#ifdef SQLITE_OS2_THREADS
  PTIB ptib;
  DosGetInfoBlocks( &ptib, NULL );
  return inMutex>0 && (thisThreadOnly==0 || mutexOwner==ptib->tib_ptib2->tib2_ultid);
#else
  return inMutex>0;
#endif
}

/*
** The following variable, if set to a non-zero value, becomes the result
** returned from sqlite3OsCurrentTime().  This is used for testing.
*/
#ifdef SQLITE_TEST
int sqlite3_current_time = 0;
#endif

/*
** Find the current time (in Universal Coordinated Time).  Write the
** current time and date as a Julian Day number into *prNow and
** return 0.  Return 1 if the time and date cannot be found.
*/
SQLITE_PRIVATE int sqlite3Os2CurrentTime( double *prNow ){
  double now;
  USHORT second, minute, hour,
         day, month, year;
  DATETIME dt;
  DosGetDateTime( &dt );
  second = (USHORT)dt.seconds;
  minute = (USHORT)dt.minutes + dt.timezone;
  hour = (USHORT)dt.hours;
  day = (USHORT)dt.day;
  month = (USHORT)dt.month;
  year = (USHORT)dt.year;

  /* Calculations from http://www.astro.keele.ac.uk/~rno/Astronomy/hjd.html
     http://www.astro.keele.ac.uk/~rno/Astronomy/hjd-0.1.c */
  /* Calculate the Julian days */
  now = day - 32076 +
    1461*(year + 4800 + (month - 14)/12)/4 +
    367*(month - 2 - (month - 14)/12*12)/12 -
    3*((year + 4900 + (month - 14)/12)/100)/4;

  /* Add the fractional hours, mins and seconds */
  now += (hour + 12.0)/24.0;
  now += minute/1440.0;
  now += second/86400.0;
  *prNow = now;
#ifdef SQLITE_TEST
  if( sqlite3_current_time ){
    *prNow = sqlite3_current_time/86400.0 + 2440587.5;
  }
#endif
  return 0;
}

/*
** Remember the number of thread-specific-data blocks allocated.
** Use this to verify that we are not leaking thread-specific-data.
** Ticket #1601
*/
#ifdef SQLITE_TEST
int sqlite3_tsd_count = 0;
# define TSD_COUNTER_INCR InterlockedIncrement( &sqlite3_tsd_count )
# define TSD_COUNTER_DECR InterlockedDecrement( &sqlite3_tsd_count )
#else
# define TSD_COUNTER_INCR  /* no-op */
# define TSD_COUNTER_DECR  /* no-op */
#endif

/*
** If called with allocateFlag>1, then return a pointer to thread
** specific data for the current thread.  Allocate and zero the
** thread-specific data if it does not already exist necessary.
**
** If called with allocateFlag==0, then check the current thread
** specific data.  Return it if it exists.  If it does not exist,
** then return NULL.
**
** If called with allocateFlag<0, check to see if the thread specific
** data is allocated and is all zero.  If it is then deallocate it.
** Return a pointer to the thread specific data or NULL if it is
** unallocated or gets deallocated.
*/
SQLITE_PRIVATE ThreadData *sqlite3Os2ThreadSpecificData( int allocateFlag ){
  static ThreadData **s_ppTsd = NULL;
  static const ThreadData zeroData = {0, 0, 0};
  ThreadData *pTsd;

  if( !s_ppTsd ){
    sqlite3OsEnterMutex();
    if( !s_ppTsd ){
      PULONG pul;
      APIRET rc = DosAllocThreadLocalMemory(1, &pul);
      if( rc != NO_ERROR ){
        sqlite3OsLeaveMutex();
        return 0;
      }
      s_ppTsd = (ThreadData **)pul;
    }
    sqlite3OsLeaveMutex();
  }
  pTsd = *s_ppTsd;
  if( allocateFlag>0 ){
    if( !pTsd ){
      pTsd = sqlite3OsMalloc( sizeof(zeroData) );
      if( pTsd ){
        *pTsd = zeroData;
        *s_ppTsd = pTsd;
        TSD_COUNTER_INCR;
      }
    }
  }else if( pTsd!=0 && allocateFlag<0
              && memcmp( pTsd, &zeroData, sizeof(ThreadData) )==0 ){
    sqlite3OsFree(pTsd);
    *s_ppTsd = NULL;
    TSD_COUNTER_DECR;
    pTsd = 0;
  }
  return pTsd;
}
#endif /* OS_OS2 */

/************** End of os_os2.c **********************************************/
/************** Begin file os_unix.c *****************************************/
/*
** 2004 May 22
**
** The author disclaims copyright to this source code.  In place of
** a legal notice, here is a blessing:
**
**    May you do good and not evil.
**    May you find forgiveness for yourself and forgive others.
**    May you share freely, never taking more than you give.
**
******************************************************************************
**
** This file contains code that is specific to Unix systems.
*/
#if OS_UNIX              /* This file is used on unix only */

/* #define SQLITE_ENABLE_LOCKING_STYLE 0 */

/*
** These #defines should enable >2GB file support on Posix if the
** underlying operating system supports it.  If the OS lacks
** large file support, these should be no-ops.
**
** Large file support can be disabled using the -DSQLITE_DISABLE_LFS switch
** on the compiler command line.  This is necessary if you are compiling
** on a recent machine (ex: RedHat 7.2) but you want your code to work
** on an older machine (ex: RedHat 6.0).  If you compile on RedHat 7.2
** without this option, LFS is enable.  But LFS does not exist in the kernel
** in RedHat 6.0, so the code won't work.  Hence, for maximum binary
** portability you should omit LFS.
*/
#ifndef SQLITE_DISABLE_LFS
# define _LARGE_FILE       1
# ifndef _FILE_OFFSET_BITS
#   define _FILE_OFFSET_BITS 64
# endif
# define _LARGEFILE_SOURCE 1
#endif

/*
** standard include files.
*/
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <unistd.h>
#include <sys/time.h>
#include <errno.h>
#ifdef SQLITE_ENABLE_LOCKING_STYLE
#include <sys/ioctl.h>
#include <sys/param.h>
#include <sys/mount.h>
#endif /* SQLITE_ENABLE_LOCKING_STYLE */

/*
** If we are to be thread-safe, include the pthreads header and define
** the SQLITE_UNIX_THREADS macro.
*/
#ifndef THREADSAFE
# define THREADSAFE 1
#endif
#if THREADSAFE
# include <pthread.h>
# define SQLITE_UNIX_THREADS 1
#endif

/*
** Default permissions when creating a new file
*/
#ifndef SQLITE_DEFAULT_FILE_PERMISSIONS
# define SQLITE_DEFAULT_FILE_PERMISSIONS 0644
#endif



/*
** The unixFile structure is subclass of OsFile specific for the unix
** protability layer.
*/
typedef struct unixFile unixFile;
struct unixFile {
  IoMethod const *pMethod;  /* Always the first entry */
  struct openCnt *pOpen;    /* Info about all open fd's on this inode */
  struct lockInfo *pLock;   /* Info about locks on this inode */
#ifdef SQLITE_ENABLE_LOCKING_STYLE
  void *lockingContext;     /* Locking style specific state */
#endif /* SQLITE_ENABLE_LOCKING_STYLE */
  int h;                    /* The file descriptor */
  unsigned char locktype;   /* The type of lock held on this fd */
  unsigned char isOpen;     /* True if needs to be closed */
  unsigned char fullSync;   /* Use F_FULLSYNC if available */
  int dirfd;                /* File descriptor for the directory */
  i64 offset;               /* Seek offset */
#ifdef SQLITE_UNIX_THREADS
  pthread_t tid;            /* The thread that "owns" this OsFile */
#endif
};

/*
** Provide the ability to override some OS-layer functions during
** testing.  This is used to simulate OS crashes to verify that 
** commits are atomic even in the event of an OS crash.
*/
#ifdef SQLITE_CRASH_TEST
  extern int sqlite3CrashTestEnable;
  extern int sqlite3CrashOpenReadWrite(const char*, OsFile**, int*);
  extern int sqlite3CrashOpenExclusive(const char*, OsFile**, int);
  extern int sqlite3CrashOpenReadOnly(const char*, OsFile**, int);
# define CRASH_TEST_OVERRIDE(X,A,B,C) \
    if(sqlite3CrashTestEnable){ return X(A,B,C); }
#else
# define CRASH_TEST_OVERRIDE(X,A,B,C)  /* no-op */
#endif


/*
** Include code that is common to all os_*.c files
*/
/************** Include os_common.h in the middle of os_unix.c ***************/
/************** Begin file os_common.h ***************************************/
/*
** 2004 May 22
**
** The author disclaims copyright to this source code.  In place of
** a legal notice, here is a blessing:
**
**    May you do good and not evil.
**    May you find forgiveness for yourself and forgive others.
**    May you share freely, never taking more than you give.
**
******************************************************************************
**
** This file contains macros and a little bit of code that is common to
** all of the platform-specific files (os_*.c) and is #included into those
** files.
**
** This file should be #included by the os_*.c files only.  It is not a
** general purpose header file.
*/

/*
** At least two bugs have slipped in because we changed the MEMORY_DEBUG
** macro to SQLITE_DEBUG and some older makefiles have not yet made the
** switch.  The following code should catch this problem at compile-time.
*/
#ifdef MEMORY_DEBUG
# error "The MEMORY_DEBUG macro is obsolete.  Use SQLITE_DEBUG instead."
#endif


/*
 * When testing, this global variable stores the location of the
 * pending-byte in the database file.
 */
#ifdef SQLITE_TEST
unsigned int sqlite3_pending_byte = 0x40000000;
#endif

int sqlite3_os_trace = 0;
#ifdef SQLITE_DEBUG
#define OSTRACE1(X)         if( sqlite3_os_trace ) sqlite3DebugPrintf(X)
#define OSTRACE2(X,Y)       if( sqlite3_os_trace ) sqlite3DebugPrintf(X,Y)
#define OSTRACE3(X,Y,Z)     if( sqlite3_os_trace ) sqlite3DebugPrintf(X,Y,Z)
#define OSTRACE4(X,Y,Z,A)   if( sqlite3_os_trace ) sqlite3DebugPrintf(X,Y,Z,A)
#define OSTRACE5(X,Y,Z,A,B) if( sqlite3_os_trace ) sqlite3DebugPrintf(X,Y,Z,A,B)
#define OSTRACE6(X,Y,Z,A,B,C) \
    if(sqlite3_os_trace) sqlite3DebugPrintf(X,Y,Z,A,B,C)
#define OSTRACE7(X,Y,Z,A,B,C,D) \
    if(sqlite3_os_trace) sqlite3DebugPrintf(X,Y,Z,A,B,C,D)
#else
#define OSTRACE1(X)
#define OSTRACE2(X,Y)
#define OSTRACE3(X,Y,Z)
#define OSTRACE4(X,Y,Z,A)
#define OSTRACE5(X,Y,Z,A,B)
#define OSTRACE6(X,Y,Z,A,B,C)
#define OSTRACE7(X,Y,Z,A,B,C,D)
#endif

/*
** Macros for performance tracing.  Normally turned off.  Only works
** on i486 hardware.
*/
#ifdef SQLITE_PERFORMANCE_TRACE
__inline__ unsigned long long int hwtime(void){
  unsigned long long int x;
  __asm__("rdtsc\n\t"
          "mov %%edx, %%ecx\n\t"
          :"=A" (x));
  return x;
}
static unsigned long long int g_start;
static unsigned int elapse;
#define TIMER_START       g_start=hwtime()
#define TIMER_END         elapse=hwtime()-g_start
#define TIMER_ELAPSED     elapse
#else
#define TIMER_START
#define TIMER_END
#define TIMER_ELAPSED     0
#endif

/*
** If we compile with the SQLITE_TEST macro set, then the following block
** of code will give us the ability to simulate a disk I/O error.  This
** is used for testing the I/O recovery logic.
*/
#ifdef SQLITE_TEST
int sqlite3_io_error_hit = 0;
int sqlite3_io_error_pending = 0;
int sqlite3_io_error_persist = 0;
int sqlite3_diskfull_pending = 0;
int sqlite3_diskfull = 0;
#define SimulateIOError(CODE)  \
  if( sqlite3_io_error_pending || sqlite3_io_error_hit ) \
     if( sqlite3_io_error_pending-- == 1 \
         || (sqlite3_io_error_persist && sqlite3_io_error_hit) ) \
                { local_ioerr(); CODE; }
static void local_ioerr(){
  IOTRACE(("IOERR\n"));
  sqlite3_io_error_hit = 1;
}
#define SimulateDiskfullError(CODE) \
   if( sqlite3_diskfull_pending ){ \
     if( sqlite3_diskfull_pending == 1 ){ \
       local_ioerr(); \
       sqlite3_diskfull = 1; \
       sqlite3_io_error_hit = 1; \
       CODE; \
     }else{ \
       sqlite3_diskfull_pending--; \
     } \
   }
#else
#define SimulateIOError(A)
#define SimulateDiskfullError(A)
#endif

/*
** When testing, keep a count of the number of open files.
*/
#ifdef SQLITE_TEST
int sqlite3_open_file_count = 0;
#define OpenCounter(X)  sqlite3_open_file_count+=(X)
#else
#define OpenCounter(X)
#endif

/*
** sqlite3GenericMalloc
** sqlite3GenericRealloc
** sqlite3GenericOsFree
** sqlite3GenericAllocationSize
**
** Implementation of the os level dynamic memory allocation interface in terms
** of the standard malloc(), realloc() and free() found in many operating
** systems. No rocket science here.
**
** There are two versions of these four functions here. The version
** implemented here is only used if memory-management or memory-debugging is
** enabled. This version allocates an extra 8-bytes at the beginning of each
** block and stores the size of the allocation there.
**
** If neither memory-management or debugging is enabled, the second
** set of implementations is used instead.
*/
#if defined(SQLITE_ENABLE_MEMORY_MANAGEMENT) || defined (SQLITE_MEMDEBUG)
SQLITE_PRIVATE void *sqlite3GenericMalloc(int n){
  char *p = (char *)malloc(n+8);
  assert(n>0);
  assert(sizeof(int)<=8);
  if( p ){
    *(int *)p = n;
    p += 8;
  }
  return (void *)p;
}
SQLITE_PRIVATE void *sqlite3GenericRealloc(void *p, int n){
  char *p2 = ((char *)p - 8);
  assert(n>0);
  p2 = (char*)realloc(p2, n+8);
  if( p2 ){
    *(int *)p2 = n;
    p2 += 8;
  }
  return (void *)p2;
}
SQLITE_PRIVATE void sqlite3GenericFree(void *p){
  assert(p);
  free((void *)((char *)p - 8));
}
SQLITE_PRIVATE int sqlite3GenericAllocationSize(void *p){
  return p ? *(int *)((char *)p - 8) : 0;
}
#else
SQLITE_PRIVATE void *sqlite3GenericMalloc(int n){
  char *p = (char *)malloc(n);
  return (void *)p;
}
SQLITE_PRIVATE void *sqlite3GenericRealloc(void *p, int n){
  assert(n>0);
  p = realloc(p, n);
  return p;
}
SQLITE_PRIVATE void sqlite3GenericFree(void *p){
  assert(p);
  free(p);
}
/* Never actually used, but needed for the linker */
SQLITE_PRIVATE int sqlite3GenericAllocationSize(void *p){ return 0; }
#endif

/*
** The default size of a disk sector
*/
#ifndef PAGER_SECTOR_SIZE
# define PAGER_SECTOR_SIZE 512
#endif

/************** End of os_common.h *******************************************/
/************** Continuing where we left off in os_unix.c ********************/

/*
** Do not include any of the File I/O interface procedures if the
** SQLITE_OMIT_DISKIO macro is defined (indicating that the database
** will be in-memory only)
*/
#ifndef SQLITE_OMIT_DISKIO


/*
** Define various macros that are missing from some systems.
*/
#ifndef O_LARGEFILE
# define O_LARGEFILE 0
#endif
#ifdef SQLITE_DISABLE_LFS
# undef O_LARGEFILE
# define O_LARGEFILE 0
#endif
#ifndef O_NOFOLLOW
# define O_NOFOLLOW 0
#endif
#ifndef O_BINARY
# define O_BINARY 0
#endif

/*
** The DJGPP compiler environment looks mostly like Unix, but it
** lacks the fcntl() system call.  So redefine fcntl() to be something
** that always succeeds.  This means that locking does not occur under
** DJGPP.  But it's DOS - what did you expect?
*/
#ifdef __DJGPP__
# define fcntl(A,B,C) 0
#endif

/*
** The threadid macro resolves to the thread-id or to 0.  Used for
** testing and debugging only.
*/
#ifdef SQLITE_UNIX_THREADS
#define threadid pthread_self()
#else
#define threadid 0
#endif

/*
** Set or check the OsFile.tid field.  This field is set when an OsFile
** is first opened.  All subsequent uses of the OsFile verify that the
** same thread is operating on the OsFile.  Some operating systems do
** not allow locks to be overridden by other threads and that restriction
** means that sqlite3* database handles cannot be moved from one thread
** to another.  This logic makes sure a user does not try to do that
** by mistake.
**
** Version 3.3.1 (2006-01-15):  OsFiles can be moved from one thread to
** another as long as we are running on a system that supports threads
** overriding each others locks (which now the most common behavior)
** or if no locks are held.  But the OsFile.pLock field needs to be
** recomputed because its key includes the thread-id.  See the 
** transferOwnership() function below for additional information
*/
#if defined(SQLITE_UNIX_THREADS)
# define SET_THREADID(X)   (X)->tid = pthread_self()
# define CHECK_THREADID(X) (threadsOverrideEachOthersLocks==0 && \
                            !pthread_equal((X)->tid, pthread_self()))
#else
# define SET_THREADID(X)
# define CHECK_THREADID(X) 0
#endif

/*
** Here is the dirt on POSIX advisory locks:  ANSI STD 1003.1 (1996)
** section 6.5.2.2 lines 483 through 490 specify that when a process
** sets or clears a lock, that operation overrides any prior locks set
** by the same process.  It does not explicitly say so, but this implies
** that it overrides locks set by the same process using a different
** file descriptor.  Consider this test case:
**
**       int fd1 = open("./file1", O_RDWR|O_CREAT, 0644);
**       int fd2 = open("./file2", O_RDWR|O_CREAT, 0644);
**
** Suppose ./file1 and ./file2 are really the same file (because
** one is a hard or symbolic link to the other) then if you set
** an exclusive lock on fd1, then try to get an exclusive lock
** on fd2, it works.  I would have expected the second lock to
** fail since there was already a lock on the file due to fd1.
** But not so.  Since both locks came from the same process, the
** second overrides the first, even though they were on different
** file descriptors opened on different file names.
**
** Bummer.  If you ask me, this is broken.  Badly broken.  It means
** that we cannot use POSIX locks to synchronize file access among
** competing threads of the same process.  POSIX locks will work fine
** to synchronize access for threads in separate processes, but not
** threads within the same process.
**
** To work around the problem, SQLite has to manage file locks internally
** on its own.  Whenever a new database is opened, we have to find the
** specific inode of the database file (the inode is determined by the
** st_dev and st_ino fields of the stat structure that fstat() fills in)
** and check for locks already existing on that inode.  When locks are
** created or removed, we have to look at our own internal record of the
** locks to see if another thread has previously set a lock on that same
** inode.
**
** The OsFile structure for POSIX is no longer just an integer file
** descriptor.  It is now a structure that holds the integer file
** descriptor and a pointer to a structure that describes the internal
** locks on the corresponding inode.  There is one locking structure
** per inode, so if the same inode is opened twice, both OsFile structures
** point to the same locking structure.  The locking structure keeps
** a reference count (so we will know when to delete it) and a "cnt"
** field that tells us its internal lock status.  cnt==0 means the
** file is unlocked.  cnt==-1 means the file has an exclusive lock.
** cnt>0 means there are cnt shared locks on the file.
**
** Any attempt to lock or unlock a file first checks the locking
** structure.  The fcntl() system call is only invoked to set a 
** POSIX lock if the internal lock structure transitions between
** a locked and an unlocked state.
**
** 2004-Jan-11:
** More recent discoveries about POSIX advisory locks.  (The more
** I discover, the more I realize the a POSIX advisory locks are
** an abomination.)
**
** If you close a file descriptor that points to a file that has locks,
** all locks on that file that are owned by the current process are
** released.  To work around this problem, each OsFile structure contains
** a pointer to an openCnt structure.  There is one openCnt structure
** per open inode, which means that multiple OsFiles can point to a single
** openCnt.  When an attempt is made to close an OsFile, if there are
** other OsFiles open on the same inode that are holding locks, the call
** to close() the file descriptor is deferred until all of the locks clear.
** The openCnt structure keeps a list of file descriptors that need to
** be closed and that list is walked (and cleared) when the last lock
** clears.
**
** First, under Linux threads, because each thread has a separate
** process ID, lock operations in one thread do not override locks
** to the same file in other threads.  Linux threads behave like
** separate processes in this respect.  But, if you close a file
** descriptor in linux threads, all locks are cleared, even locks
** on other threads and even though the other threads have different
** process IDs.  Linux threads is inconsistent in this respect.
** (I'm beginning to think that linux threads is an abomination too.)
** The consequence of this all is that the hash table for the lockInfo
** structure has to include the process id as part of its key because
** locks in different threads are treated as distinct.  But the 
** openCnt structure should not include the process id in its
** key because close() clears lock on all threads, not just the current
** thread.  Were it not for this goofiness in linux threads, we could
** combine the lockInfo and openCnt structures into a single structure.
**
** 2004-Jun-28:
** On some versions of linux, threads can override each others locks.
** On others not.  Sometimes you can change the behavior on the same
** system by setting the LD_ASSUME_KERNEL environment variable.  The
** POSIX standard is silent as to which behavior is correct, as far
** as I can tell, so other versions of unix might show the same
** inconsistency.  There is no little doubt in my mind that posix
** advisory locks and linux threads are profoundly broken.
**
** To work around the inconsistencies, we have to test at runtime 
** whether or not threads can override each others locks.  This test
** is run once, the first time any lock is attempted.  A static 
** variable is set to record the results of this test for future
** use.
*/

/*
** An instance of the following structure serves as the key used
** to locate a particular lockInfo structure given its inode.
**
** If threads cannot override each others locks, then we set the
** lockKey.tid field to the thread ID.  If threads can override
** each others locks then tid is always set to zero.  tid is omitted
** if we compile without threading support.
*/
struct lockKey {
  dev_t dev;       /* Device number */
  ino_t ino;       /* Inode number */
#ifdef SQLITE_UNIX_THREADS
  pthread_t tid;   /* Thread ID or zero if threads can override each other */
#endif
};

/*
** An instance of the following structure is allocated for each open
** inode on each thread with a different process ID.  (Threads have
** different process IDs on linux, but not on most other unixes.)
**
** A single inode can have multiple file descriptors, so each OsFile
** structure contains a pointer to an instance of this object and this
** object keeps a count of the number of OsFiles pointing to it.
*/
struct lockInfo {
  struct lockKey key;  /* The lookup key */
  int cnt;             /* Number of SHARED locks held */
  int locktype;        /* One of SHARED_LOCK, RESERVED_LOCK etc. */
  int nRef;            /* Number of pointers to this structure */
};

/*
** An instance of the following structure serves as the key used
** to locate a particular openCnt structure given its inode.  This
** is the same as the lockKey except that the thread ID is omitted.
*/
struct openKey {
  dev_t dev;   /* Device number */
  ino_t ino;   /* Inode number */
};

/*
** An instance of the following structure is allocated for each open
** inode.  This structure keeps track of the number of locks on that
** inode.  If a close is attempted against an inode that is holding
** locks, the close is deferred until all locks clear by adding the
** file descriptor to be closed to the pending list.
*/
struct openCnt {
  struct openKey key;   /* The lookup key */
  int nRef;             /* Number of pointers to this structure */
  int nLock;            /* Number of outstanding locks */
  int nPending;         /* Number of pending close() operations */
  int *aPending;        /* Malloced space holding fd's awaiting a close() */
};

/* 
** These hash tables map inodes and file descriptors (really, lockKey and
** openKey structures) into lockInfo and openCnt structures.  Access to 
** these hash tables must be protected by a mutex.
*/
static Hash lockHash = {SQLITE_HASH_BINARY, 0, 0, 0, 
    sqlite3ThreadSafeMalloc, sqlite3ThreadSafeFree, 0, 0};
static Hash openHash = {SQLITE_HASH_BINARY, 0, 0, 0, 
    sqlite3ThreadSafeMalloc, sqlite3ThreadSafeFree, 0, 0};

#ifdef SQLITE_ENABLE_LOCKING_STYLE
/*
** The locking styles are associated with the different file locking
** capabilities supported by different file systems.  
**
** POSIX locking style fully supports shared and exclusive byte-range locks 
** ADP locking only supports exclusive byte-range locks
** FLOCK only supports a single file-global exclusive lock
** DOTLOCK isn't a true locking style, it refers to the use of a special
**   file named the same as the database file with a '.lock' extension, this
**   can be used on file systems that do not offer any reliable file locking
** NO locking means that no locking will be attempted, this is only used for
**   read-only file systems currently
** UNSUPPORTED means that no locking will be attempted, this is only used for
**   file systems that are known to be unsupported
*/
typedef enum {
	posixLockingStyle = 0,       /* standard posix-advisory locks */
	afpLockingStyle,             /* use afp locks */
	flockLockingStyle,           /* use flock() */
	dotlockLockingStyle,         /* use <file>.lock files */
	noLockingStyle,              /* useful for read-only file system */
	unsupportedLockingStyle      /* indicates unsupported file system */
} sqlite3LockingStyle;
#endif /* SQLITE_ENABLE_LOCKING_STYLE */

#ifdef SQLITE_UNIX_THREADS
/*
** This variable records whether or not threads can override each others
** locks.
**
**    0:  No.  Threads cannot override each others locks.
**    1:  Yes.  Threads can override each others locks.
**   -1:  We don't know yet.
**
** On some systems, we know at compile-time if threads can override each
** others locks.  On those systems, the SQLITE_THREAD_OVERRIDE_LOCK macro
** will be set appropriately.  On other systems, we have to check at
** runtime.  On these latter systems, SQLTIE_THREAD_OVERRIDE_LOCK is
** undefined.
**
** This variable normally has file scope only.  But during testing, we make
** it a global so that the test code can change its value in order to verify
** that the right stuff happens in either case.
*/
#ifndef SQLITE_THREAD_OVERRIDE_LOCK
# define SQLITE_THREAD_OVERRIDE_LOCK -1
#endif
#ifdef SQLITE_TEST
int threadsOverrideEachOthersLocks = SQLITE_THREAD_OVERRIDE_LOCK;
#else
static int threadsOverrideEachOthersLocks = SQLITE_THREAD_OVERRIDE_LOCK;
#endif

/*
** This structure holds information passed into individual test
** threads by the testThreadLockingBehavior() routine.
*/
struct threadTestData {
  int fd;                /* File to be locked */
  struct flock lock;     /* The locking operation */
  int result;            /* Result of the locking operation */
};

#ifdef SQLITE_LOCK_TRACE
/*
** Print out information about all locking operations.
**
** This routine is used for troubleshooting locks on multithreaded
** platforms.  Enable by compiling with the -DSQLITE_LOCK_TRACE
** command-line option on the compiler.  This code is normally
** turned off.
*/
static int lockTrace(int fd, int op, struct flock *p){
  char *zOpName, *zType;
  int s;
  int savedErrno;
  if( op==F_GETLK ){
    zOpName = "GETLK";
  }else if( op==F_SETLK ){
    zOpName = "SETLK";
  }else{
    s = fcntl(fd, op, p);
    sqlite3DebugPrintf("fcntl unknown %d %d %d\n", fd, op, s);
    return s;
  }
  if( p->l_type==F_RDLCK ){
    zType = "RDLCK";
  }else if( p->l_type==F_WRLCK ){
    zType = "WRLCK";
  }else if( p->l_type==F_UNLCK ){
    zType = "UNLCK";
  }else{
    assert( 0 );
  }
  assert( p->l_whence==SEEK_SET );
  s = fcntl(fd, op, p);
  savedErrno = errno;
  sqlite3DebugPrintf("fcntl %d %d %s %s %d %d %d %d\n",
     threadid, fd, zOpName, zType, (int)p->l_start, (int)p->l_len,
     (int)p->l_pid, s);
  if( s==(-1) && op==F_SETLK && (p->l_type==F_RDLCK || p->l_type==F_WRLCK) ){
    struct flock l2;
    l2 = *p;
    fcntl(fd, F_GETLK, &l2);
    if( l2.l_type==F_RDLCK ){
      zType = "RDLCK";
    }else if( l2.l_type==F_WRLCK ){
      zType = "WRLCK";
    }else if( l2.l_type==F_UNLCK ){
      zType = "UNLCK";
    }else{
      assert( 0 );
    }
    sqlite3DebugPrintf("fcntl-failure-reason: %s %d %d %d\n",
       zType, (int)l2.l_start, (int)l2.l_len, (int)l2.l_pid);
  }
  errno = savedErrno;
  return s;
}
#define fcntl lockTrace
#endif /* SQLITE_LOCK_TRACE */

/*
** The testThreadLockingBehavior() routine launches two separate
** threads on this routine.  This routine attempts to lock a file
** descriptor then returns.  The success or failure of that attempt
** allows the testThreadLockingBehavior() procedure to determine
** whether or not threads can override each others locks.
*/
static void *threadLockingTest(void *pArg){
  struct threadTestData *pData = (struct threadTestData*)pArg;
  pData->result = fcntl(pData->fd, F_SETLK, &pData->lock);
  return pArg;
}

/*
** This procedure attempts to determine whether or not threads
** can override each others locks then sets the 
** threadsOverrideEachOthersLocks variable appropriately.
*/
static void testThreadLockingBehavior(int fd_orig){
  int fd;
  struct threadTestData d[2];
  pthread_t t[2];

  fd = dup(fd_orig);
  if( fd<0 ) return;
  memset(d, 0, sizeof(d));
  d[0].fd = fd;
  d[0].lock.l_type = F_RDLCK;
  d[0].lock.l_len = 1;
  d[0].lock.l_start = 0;
  d[0].lock.l_whence = SEEK_SET;
  d[1] = d[0];
  d[1].lock.l_type = F_WRLCK;
  pthread_create(&t[0], 0, threadLockingTest, &d[0]);
  pthread_create(&t[1], 0, threadLockingTest, &d[1]);
  pthread_join(t[0], 0);
  pthread_join(t[1], 0);
  close(fd);
  threadsOverrideEachOthersLocks =  d[0].result==0 && d[1].result==0;
}
#endif /* SQLITE_UNIX_THREADS */

/*
** Release a lockInfo structure previously allocated by findLockInfo().
*/
static void releaseLockInfo(struct lockInfo *pLock){
  assert( sqlite3OsInMutex(1) );
  if (pLock == NULL)
    return;
  pLock->nRef--;
  if( pLock->nRef==0 ){
    sqlite3HashInsert(&lockHash, &pLock->key, sizeof(pLock->key), 0);
    sqlite3ThreadSafeFree(pLock);
  }
}

/*
** Release a openCnt structure previously allocated by findLockInfo().
*/
static void releaseOpenCnt(struct openCnt *pOpen){
  assert( sqlite3OsInMutex(1) );
  if (pOpen == NULL)
    return;
  pOpen->nRef--;
  if( pOpen->nRef==0 ){
    sqlite3HashInsert(&openHash, &pOpen->key, sizeof(pOpen->key), 0);
    free(pOpen->aPending);
    sqlite3ThreadSafeFree(pOpen);
  }
}

#ifdef SQLITE_ENABLE_LOCKING_STYLE
/*
** Tests a byte-range locking query to see if byte range locks are 
** supported, if not we fall back to dotlockLockingStyle.
*/
static sqlite3LockingStyle sqlite3TestLockingStyle(const char *filePath, 
  int fd) {
  /* test byte-range lock using fcntl */
  struct flock lockInfo;
  
  lockInfo.l_len = 1;
  lockInfo.l_start = 0;
  lockInfo.l_whence = SEEK_SET;
  lockInfo.l_type = F_RDLCK;
  
  if (fcntl(fd, F_GETLK, &lockInfo) != -1) {
    return posixLockingStyle;
  } 
  
  /* testing for flock can give false positives.  So if if the above test
  ** fails, then we fall back to using dot-lock style locking.
  */  
  return dotlockLockingStyle;
}

/* 
** Examines the f_fstypename entry in the statfs structure as returned by 
** stat() for the file system hosting the database file, assigns the 
** appropriate locking style based on it's value.  These values and 
** assignments are based on Darwin/OSX behavior and have not been tested on 
** other systems.
*/
static sqlite3LockingStyle sqlite3DetectLockingStyle(const char *filePath, 
  int fd) {

#ifdef SQLITE_FIXED_LOCKING_STYLE
  return (sqlite3LockingStyle)SQLITE_FIXED_LOCKING_STYLE;
#else
  struct statfs fsInfo;

  if (statfs(filePath, &fsInfo) == -1)
    return sqlite3TestLockingStyle(filePath, fd);
  
  if (fsInfo.f_flags & MNT_RDONLY)
    return noLockingStyle;
  
  if( (!strcmp(fsInfo.f_fstypename, "hfs")) ||
    (!strcmp(fsInfo.f_fstypename, "ufs")) )
		return posixLockingStyle;
  
  if(!strcmp(fsInfo.f_fstypename, "afpfs"))
    return afpLockingStyle;
  
  if(!strcmp(fsInfo.f_fstypename, "nfs")) 
    return sqlite3TestLockingStyle(filePath, fd);
  
  if(!strcmp(fsInfo.f_fstypename, "smbfs"))
    return flockLockingStyle;
  
  if(!strcmp(fsInfo.f_fstypename, "msdos"))
    return dotlockLockingStyle;
  
  if(!strcmp(fsInfo.f_fstypename, "webdav"))
    return unsupportedLockingStyle;
  
  return sqlite3TestLockingStyle(filePath, fd);  
#endif /* SQLITE_FIXED_LOCKING_STYLE */
}

#endif /* SQLITE_ENABLE_LOCKING_STYLE */

/*
** Given a file descriptor, locate lockInfo and openCnt structures that
** describes that file descriptor.  Create new ones if necessary.  The
** return values might be uninitialized if an error occurs.
**
** Return the number of errors.
*/
static int findLockInfo(
  int fd,                      /* The file descriptor used in the key */
  struct lockInfo **ppLock,    /* Return the lockInfo structure here */
  struct openCnt **ppOpen      /* Return the openCnt structure here */
){
  int rc;
  struct lockKey key1;
  struct openKey key2;
  struct stat statbuf;
  struct lockInfo *pLock;
  struct openCnt *pOpen;
  rc = fstat(fd, &statbuf);
  if( rc!=0 ) return 1;

  assert( sqlite3OsInMutex(1) );
  memset(&key1, 0, sizeof(key1));
  key1.dev = statbuf.st_dev;
  key1.ino = statbuf.st_ino;
#ifdef SQLITE_UNIX_THREADS
  if( threadsOverrideEachOthersLocks<0 ){
    testThreadLockingBehavior(fd);
  }
  key1.tid = threadsOverrideEachOthersLocks ? 0 : pthread_self();
#endif
  memset(&key2, 0, sizeof(key2));
  key2.dev = statbuf.st_dev;
  key2.ino = statbuf.st_ino;
  pLock = (struct lockInfo*)sqlite3HashFind(&lockHash, &key1, sizeof(key1));
  if( pLock==0 ){
    struct lockInfo *pOld;
    pLock = sqlite3ThreadSafeMalloc( sizeof(*pLock) );
    if( pLock==0 ){
      rc = 1;
      goto exit_findlockinfo;
    }
    pLock->key = key1;
    pLock->nRef = 1;
    pLock->cnt = 0;
    pLock->locktype = 0;
    pOld = sqlite3HashInsert(&lockHash, &pLock->key, sizeof(key1), pLock);
    if( pOld!=0 ){
      assert( pOld==pLock );
      sqlite3ThreadSafeFree(pLock);
      rc = 1;
      goto exit_findlockinfo;
    }
  }else{
    pLock->nRef++;
  }
  *ppLock = pLock;
  if( ppOpen!=0 ){
    pOpen = (struct openCnt*)sqlite3HashFind(&openHash, &key2, sizeof(key2));
    if( pOpen==0 ){
      struct openCnt *pOld;
      pOpen = sqlite3ThreadSafeMalloc( sizeof(*pOpen) );
      if( pOpen==0 ){
        releaseLockInfo(pLock);
        rc = 1;
        goto exit_findlockinfo;
      }
      pOpen->key = key2;
      pOpen->nRef = 1;
      pOpen->nLock = 0;
      pOpen->nPending = 0;
      pOpen->aPending = 0;
      pOld = sqlite3HashInsert(&openHash, &pOpen->key, sizeof(key2), pOpen);
      if( pOld!=0 ){
        assert( pOld==pOpen );
        sqlite3ThreadSafeFree(pOpen);
        releaseLockInfo(pLock);
        rc = 1;
        goto exit_findlockinfo;
      }
    }else{
      pOpen->nRef++;
    }
    *ppOpen = pOpen;
  }

exit_findlockinfo:
  return rc;
}

#ifdef SQLITE_DEBUG
/*
** Helper function for printing out trace information from debugging
** binaries. This returns the string represetation of the supplied
** integer lock-type.
*/
static const char *locktypeName(int locktype){
  switch( locktype ){
  case NO_LOCK: return "NONE";
  case SHARED_LOCK: return "SHARED";
  case RESERVED_LOCK: return "RESERVED";
  case PENDING_LOCK: return "PENDING";
  case EXCLUSIVE_LOCK: return "EXCLUSIVE";
  }
  return "ERROR";
}
#endif

/*
** If we are currently in a different thread than the thread that the
** unixFile argument belongs to, then transfer ownership of the unixFile
** over to the current thread.
**
** A unixFile is only owned by a thread on systems where one thread is
** unable to override locks created by a different thread.  RedHat9 is
** an example of such a system.
**
** Ownership transfer is only allowed if the unixFile is currently unlocked.
** If the unixFile is locked and an ownership is wrong, then return
** SQLITE_MISUSE.  SQLITE_OK is returned if everything works.
*/
#ifdef SQLITE_UNIX_THREADS
static int transferOwnership(unixFile *pFile){
  int rc;
  pthread_t hSelf;
  if( threadsOverrideEachOthersLocks ){
    /* Ownership transfers not needed on this system */
    return SQLITE_OK;
  }
  hSelf = pthread_self();
  if( pthread_equal(pFile->tid, hSelf) ){
    /* We are still in the same thread */
    OSTRACE1("No-transfer, same thread\n");
    return SQLITE_OK;
  }
  if( pFile->locktype!=NO_LOCK ){
    /* We cannot change ownership while we are holding a lock! */
    return SQLITE_MISUSE;
  }
  OSTRACE4("Transfer ownership of %d from %d to %d\n",
            pFile->h, pFile->tid, hSelf);
  pFile->tid = hSelf;
  if (pFile->pLock != NULL) {
    releaseLockInfo(pFile->pLock);
    rc = findLockInfo(pFile->h, &pFile->pLock, 0);
    OSTRACE5("LOCK    %d is now %s(%s,%d)\n", pFile->h,
           locktypeName(pFile->locktype),
           locktypeName(pFile->pLock->locktype), pFile->pLock->cnt);
    return rc;
  } else {
    return SQLITE_OK;
  }
}
#else
  /* On single-threaded builds, ownership transfer is a no-op */
# define transferOwnership(X) SQLITE_OK
#endif

/*
** Delete the named file
*/
SQLITE_PRIVATE int sqlite3UnixDelete(const char *zFilename){
  SimulateIOError(return SQLITE_IOERR_DELETE);
  unlink(zFilename);
  return SQLITE_OK;
}

/*
** Return TRUE if the named file exists.
*/
SQLITE_PRIVATE int sqlite3UnixFileExists(const char *zFilename){
  return access(zFilename, 0)==0;
}

/* Forward declaration */
static int allocateUnixFile(
  int h,                    /* File descriptor of the open file */
  OsFile **pId,             /* Write the real file descriptor here */
  const char *zFilename,    /* Name of the file being opened */
  int delFlag               /* If true, make sure the file deletes on close */
);

/*
** Attempt to open a file for both reading and writing.  If that
** fails, try opening it read-only.  If the file does not exist,
** try to create it.
**
** On success, a handle for the open file is written to *id
** and *pReadonly is set to 0 if the file was opened for reading and
** writing or 1 if the file was opened read-only.  The function returns
** SQLITE_OK.
**
** On failure, the function returns SQLITE_CANTOPEN and leaves
** *id and *pReadonly unchanged.
*/
SQLITE_PRIVATE int sqlite3UnixOpenReadWrite(
  const char *zFilename,
  OsFile **pId,
  int *pReadonly
){
  int h;
  
  CRASH_TEST_OVERRIDE(sqlite3CrashOpenReadWrite, zFilename, pId, pReadonly);
  assert( 0==*pId );
  h = open(zFilename, O_RDWR|O_CREAT|O_LARGEFILE|O_BINARY,
                        SQLITE_DEFAULT_FILE_PERMISSIONS);
  if( h<0 ){
#ifdef EISDIR
    if( errno==EISDIR ){
      return SQLITE_CANTOPEN;
    }
#endif
    h = open(zFilename, O_RDONLY|O_LARGEFILE|O_BINARY);
    if( h<0 ){
      return SQLITE_CANTOPEN; 
    }
    *pReadonly = 1;
  }else{
    *pReadonly = 0;
  }
  return allocateUnixFile(h, pId, zFilename, 0);
}


/*
** Attempt to open a new file for exclusive access by this process.
** The file will be opened for both reading and writing.  To avoid
** a potential security problem, we do not allow the file to have
** previously existed.  Nor do we allow the file to be a symbolic
** link.
**
** If delFlag is true, then make arrangements to automatically delete
** the file when it is closed.
**
** On success, write the file handle into *id and return SQLITE_OK.
**
** On failure, return SQLITE_CANTOPEN.
*/
SQLITE_PRIVATE int sqlite3UnixOpenExclusive(const char *zFilename, OsFile **pId, int delFlag){
  int h;

  CRASH_TEST_OVERRIDE(sqlite3CrashOpenExclusive, zFilename, pId, delFlag);
  assert( 0==*pId );
  h = open(zFilename,
                O_RDWR|O_CREAT|O_EXCL|O_NOFOLLOW|O_LARGEFILE|O_BINARY,
                delFlag ? 0600 : SQLITE_DEFAULT_FILE_PERMISSIONS);
  if( h<0 ){
    return SQLITE_CANTOPEN;
  }
  return allocateUnixFile(h, pId, zFilename, delFlag);
}

/*
** Attempt to open a new file for read-only access.
**
** On success, write the file handle into *id and return SQLITE_OK.
**
** On failure, return SQLITE_CANTOPEN.
*/
SQLITE_PRIVATE int sqlite3UnixOpenReadOnly(const char *zFilename, OsFile **pId){
  int h;
  
  CRASH_TEST_OVERRIDE(sqlite3CrashOpenReadOnly, zFilename, pId, 0);
  assert( 0==*pId );
  h = open(zFilename, O_RDONLY|O_LARGEFILE|O_BINARY);
  if( h<0 ){
    return SQLITE_CANTOPEN;
  }
  return allocateUnixFile(h, pId, zFilename, 0);
}

/*
** Attempt to open a file descriptor for the directory that contains a
** file.  This file descriptor can be used to fsync() the directory
** in order to make sure the creation of a new file is actually written
** to disk.
**
** This routine is only meaningful for Unix.  It is a no-op under
** windows since windows does not support hard links.
**
** If FULL_FSYNC is enabled, this function is not longer useful, 
** a FULL_FSYNC sync applies to all pending disk operations.
**
** On success, a handle for a previously open file at *id is
** updated with the new directory file descriptor and SQLITE_OK is
** returned.
**
** On failure, the function returns SQLITE_CANTOPEN and leaves
** *id unchanged.
*/
static int unixOpenDirectory(
  OsFile *id,
  const char *zDirname
){
  int h;
  unixFile *pFile = (unixFile*)id;
  assert( pFile!=0 );
  SET_THREADID(pFile);
  assert( pFile->dirfd<0 );
  pFile->dirfd = h = open(zDirname, O_RDONLY|O_BINARY, 0);
  if( h<0 ){
    return SQLITE_CANTOPEN; 
  }
#ifdef FD_CLOEXEC
  fcntl(h, F_SETFD, fcntl(h, F_GETFD, 0) | FD_CLOEXEC);
#endif
  OSTRACE3("OPENDIR %-3d %s\n", h, zDirname);
  return SQLITE_OK;
}

/*
** Create a temporary file name in zBuf.  zBuf must be big enough to
** hold at least SQLITE_TEMPNAME_SIZE characters.
*/
SQLITE_PRIVATE int sqlite3UnixTempFileName(char *zBuf){
  static const char *azDirs[] = {
     0,
     "/var/tmp",
     "/usr/tmp",
     "/tmp",
     ".",
  };
  static const unsigned char zChars[] =
    "abcdefghijklmnopqrstuvwxyz"
    "ABCDEFGHIJKLMNOPQRSTUVWXYZ"
    "0123456789";
  int i, j;
  struct stat buf;
  const char *zDir = ".";
  azDirs[0] = sqlite3_temp_directory;
  for(i=0; i<sizeof(azDirs)/sizeof(azDirs[0]); i++){
    if( azDirs[i]==0 ) continue;
    if( stat(azDirs[i], &buf) ) continue;
    if( !S_ISDIR(buf.st_mode) ) continue;
    if( access(azDirs[i], 07) ) continue;
    zDir = azDirs[i];
    break;
  }
  do{
    sqlite3_snprintf(SQLITE_TEMPNAME_SIZE, zBuf, "%s/"TEMP_FILE_PREFIX, zDir);
    j = strlen(zBuf);
    sqlite3Randomness(15, &zBuf[j]);
    for(i=0; i<15; i++, j++){
      zBuf[j] = (char)zChars[ ((unsigned char)zBuf[j])%(sizeof(zChars)-1) ];
    }
    zBuf[j] = 0;
  }while( access(zBuf,0)==0 );
  return SQLITE_OK; 
}

/*
** Check that a given pathname is a directory and is writable 
**
*/
SQLITE_PRIVATE int sqlite3UnixIsDirWritable(char *zBuf){
#ifndef SQLITE_OMIT_PAGER_PRAGMAS
  struct stat buf;
  if( zBuf==0 ) return 0;
  if( zBuf[0]==0 ) return 0;
  if( stat(zBuf, &buf) ) return 0;
  if( !S_ISDIR(buf.st_mode) ) return 0;
  if( access(zBuf, 07) ) return 0;
#endif /* SQLITE_OMIT_PAGER_PRAGMAS */
  return 1;
}

/*
** Seek to the offset in id->offset then read cnt bytes into pBuf.
** Return the number of bytes actually read.  Update the offset.
*/
static int seekAndRead(unixFile *id, void *pBuf, int cnt){
  int got;
  i64 newOffset;
  TIMER_START;
#if defined(USE_PREAD)
  got = pread(id->h, pBuf, cnt, id->offset);
  SimulateIOError( got = -1 );
#elif defined(USE_PREAD64)
  got = pread64(id->h, pBuf, cnt, id->offset);
  SimulateIOError( got = -1 );
#else
  newOffset = lseek(id->h, id->offset, SEEK_SET);
  SimulateIOError( newOffset-- );
  if( newOffset!=id->offset ){
    return -1;
  }
  got = read(id->h, pBuf, cnt);
#endif
  TIMER_END;
  OSTRACE5("READ    %-3d %5d %7lld %d\n", id->h, got, id->offset, TIMER_ELAPSED);
  if( got>0 ){
    id->offset += got;
  }
  return got;
}

/*
** Read data from a file into a buffer.  Return SQLITE_OK if all
** bytes were read successfully and SQLITE_IOERR if anything goes
** wrong.
*/
static int unixRead(OsFile *id, void *pBuf, int amt){
  int got;
  assert( id );
  got = seekAndRead((unixFile*)id, pBuf, amt);
  if( got==amt ){
    return SQLITE_OK;
  }else if( got<0 ){
    return SQLITE_IOERR_READ;
  }else{
    memset(&((char*)pBuf)[got], 0, amt-got);
    return SQLITE_IOERR_SHORT_READ;
  }
}

/*
** Seek to the offset in id->offset then read cnt bytes into pBuf.
** Return the number of bytes actually read.  Update the offset.
*/
static int seekAndWrite(unixFile *id, const void *pBuf, int cnt){
  int got;
  i64 newOffset;
  TIMER_START;
#if defined(USE_PREAD)
  got = pwrite(id->h, pBuf, cnt, id->offset);
#elif defined(USE_PREAD64)
  got = pwrite64(id->h, pBuf, cnt, id->offset);
#else
  newOffset = lseek(id->h, id->offset, SEEK_SET);
  if( newOffset!=id->offset ){
    return -1;
  }
  got = write(id->h, pBuf, cnt);
#endif
  TIMER_END;
  OSTRACE5("WRITE   %-3d %5d %7lld %d\n", id->h, got, id->offset, TIMER_ELAPSED);
  if( got>0 ){
    id->offset += got;
  }
  return got;
}


/*
** Write data from a buffer into a file.  Return SQLITE_OK on success
** or some other error code on failure.
*/
static int unixWrite(OsFile *id, const void *pBuf, int amt){
  int wrote = 0;
  assert( id );
  assert( amt>0 );
  while( amt>0 && (wrote = seekAndWrite((unixFile*)id, pBuf, amt))>0 ){
    amt -= wrote;
    pBuf = &((char*)pBuf)[wrote];
  }
  SimulateIOError(( wrote=(-1), amt=1 ));
  SimulateDiskfullError(( wrote=0, amt=1 ));
  if( amt>0 ){
    if( wrote<0 ){
      return SQLITE_IOERR_WRITE;
    }else{
      return SQLITE_FULL;
    }
  }
  return SQLITE_OK;
}

/*
** Move the read/write pointer in a file.
*/
static int unixSeek(OsFile *id, i64 offset){
  assert( id );
#ifdef SQLITE_TEST
  if( offset ) SimulateDiskfullError(return SQLITE_FULL);
#endif
  ((unixFile*)id)->offset = offset;
  return SQLITE_OK;
}

#ifdef SQLITE_TEST
/*
** Count the number of fullsyncs and normal syncs.  This is used to test
** that syncs and fullsyncs are occuring at the right times.
*/
int sqlite3_sync_count = 0;
int sqlite3_fullsync_count = 0;
#endif

/*
** Use the fdatasync() API only if the HAVE_FDATASYNC macro is defined.
** Otherwise use fsync() in its place.
*/
#ifndef HAVE_FDATASYNC
# define fdatasync fsync
#endif

/*
** Define HAVE_FULLFSYNC to 0 or 1 depending on whether or not
** the F_FULLFSYNC macro is defined.  F_FULLFSYNC is currently
** only available on Mac OS X.  But that could change.
*/
#ifdef F_FULLFSYNC
# define HAVE_FULLFSYNC 1
#else
# define HAVE_FULLFSYNC 0
#endif


/*
** The fsync() system call does not work as advertised on many
** unix systems.  The following procedure is an attempt to make
** it work better.
**
** The SQLITE_NO_SYNC macro disables all fsync()s.  This is useful
** for testing when we want to run through the test suite quickly.
** You are strongly advised *not* to deploy with SQLITE_NO_SYNC
** enabled, however, since with SQLITE_NO_SYNC enabled, an OS crash
** or power failure will likely corrupt the database file.
*/
static int full_fsync(int fd, int fullSync, int dataOnly){
  int rc;

  /* Record the number of times that we do a normal fsync() and 
  ** FULLSYNC.  This is used during testing to verify that this procedure
  ** gets called with the correct arguments.
  */
#ifdef SQLITE_TEST
  if( fullSync ) sqlite3_fullsync_count++;
  sqlite3_sync_count++;
#endif

  /* If we compiled with the SQLITE_NO_SYNC flag, then syncing is a
  ** no-op
  */
#ifdef SQLITE_NO_SYNC
  rc = SQLITE_OK;
#else

#if HAVE_FULLFSYNC
  if( fullSync ){
    rc = fcntl(fd, F_FULLFSYNC, 0);
  }else{
    rc = 1;
  }
  /* If the FULLFSYNC failed, fall back to attempting an fsync().
   * It shouldn't be possible for fullfsync to fail on the local 
   * file system (on OSX), so failure indicates that FULLFSYNC
   * isn't supported for this file system. So, attempt an fsync 
   * and (for now) ignore the overhead of a superfluous fcntl call.  
   * It'd be better to detect fullfsync support once and avoid 
   * the fcntl call every time sync is called.
   */
  if( rc ) rc = fsync(fd);

#else 
  if( dataOnly ){
    rc = fdatasync(fd);
  }else{
    rc = fsync(fd);
  }
#endif /* HAVE_FULLFSYNC */
#endif /* defined(SQLITE_NO_SYNC) */

  return rc;
}

/*
** Make sure all writes to a particular file are committed to disk.
**
** If dataOnly==0 then both the file itself and its metadata (file
** size, access time, etc) are synced.  If dataOnly!=0 then only the
** file data is synced.
**
** Under Unix, also make sure that the directory entry for the file
** has been created by fsync-ing the directory that contains the file.
** If we do not do this and we encounter a power failure, the directory
** entry for the journal might not exist after we reboot.  The next
** SQLite to access the file will not know that the journal exists (because
** the directory entry for the journal was never created) and the transaction
** will not roll back - possibly leading to database corruption.
*/
static int unixSync(OsFile *id, int dataOnly){
  int rc;
  unixFile *pFile = (unixFile*)id;
  assert( pFile );
  OSTRACE2("SYNC    %-3d\n", pFile->h);
  rc = full_fsync(pFile->h, pFile->fullSync, dataOnly);
  SimulateIOError( rc=1 );
  if( rc ){
    return SQLITE_IOERR_FSYNC;
  }
  if( pFile->dirfd>=0 ){
    OSTRACE4("DIRSYNC %-3d (have_fullfsync=%d fullsync=%d)\n", pFile->dirfd,
            HAVE_FULLFSYNC, pFile->fullSync);
#ifndef SQLITE_DISABLE_DIRSYNC
    /* The directory sync is only attempted if full_fsync is
    ** turned off or unavailable.  If a full_fsync occurred above,
    ** then the directory sync is superfluous.
    */
    if( (!HAVE_FULLFSYNC || !pFile->fullSync) && full_fsync(pFile->dirfd,0,0) ){
       /*
       ** We have received multiple reports of fsync() returning
       ** errors when applied to directories on certain file systems.
       ** A failed directory sync is not a big deal.  So it seems
       ** better to ignore the error.  Ticket #1657
       */
       /* return SQLITE_IOERR; */
    }
#endif
    close(pFile->dirfd);  /* Only need to sync once, so close the directory */
    pFile->dirfd = -1;    /* when we are done. */
  }
  return SQLITE_OK;
}

/*
** Sync the directory zDirname. This is a no-op on operating systems other
** than UNIX.
**
** This is used to make sure the master journal file has truely been deleted
** before making changes to individual journals on a multi-database commit.
** The F_FULLFSYNC option is not needed here.
*/
SQLITE_PRIVATE int sqlite3UnixSyncDirectory(const char *zDirname){
#ifdef SQLITE_DISABLE_DIRSYNC
  return SQLITE_OK;
#else
  int fd;
  int r;
  fd = open(zDirname, O_RDONLY|O_BINARY, 0);
  OSTRACE3("DIRSYNC %-3d (%s)\n", fd, zDirname);
  if( fd<0 ){
    return SQLITE_CANTOPEN; 
  }
  r = fsync(fd);
  close(fd);
  SimulateIOError( r=1 );
  if( r ){
    return SQLITE_IOERR_DIR_FSYNC;
  }else{
    return SQLITE_OK;
  }
#endif
}

/*
** Truncate an open file to a specified size
*/
static int unixTruncate(OsFile *id, i64 nByte){
  int rc;
  assert( id );
  rc = ftruncate(((unixFile*)id)->h, (off_t)nByte);
  SimulateIOError( rc=1 );
  if( rc ){
    return SQLITE_IOERR_TRUNCATE;
  }else{
    return SQLITE_OK;
  }
}

/*
** Determine the current size of a file in bytes
*/
static int unixFileSize(OsFile *id, i64 *pSize){
  int rc;
  struct stat buf;
  assert( id );
  rc = fstat(((unixFile*)id)->h, &buf);
  SimulateIOError( rc=1 );
  if( rc!=0 ){
    return SQLITE_IOERR_FSTAT;
  }
  *pSize = buf.st_size;
  return SQLITE_OK;
}

/*
** This routine checks if there is a RESERVED lock held on the specified
** file by this or any other process. If such a lock is held, return
** non-zero.  If the file is unlocked or holds only SHARED locks, then
** return zero.
*/
static int unixCheckReservedLock(OsFile *id){
  int r = 0;
  unixFile *pFile = (unixFile*)id;

  assert( pFile );
  sqlite3OsEnterMutex(); /* Because pFile->pLock is shared across threads */

  /* Check if a thread in this process holds such a lock */
  if( pFile->pLock->locktype>SHARED_LOCK ){
    r = 1;
  }

  /* Otherwise see if some other process holds it.
  */
  if( !r ){
    struct flock lock;
    lock.l_whence = SEEK_SET;
    lock.l_start = RESERVED_BYTE;
    lock.l_len = 1;
    lock.l_type = F_WRLCK;
    fcntl(pFile->h, F_GETLK, &lock);
    if( lock.l_type!=F_UNLCK ){
      r = 1;
    }
  }
  
  sqlite3OsLeaveMutex();
  OSTRACE3("TEST WR-LOCK %d %d\n", pFile->h, r);

  return r;
}

/*
** Lock the file with the lock specified by parameter locktype - one
** of the following:
**
**     (1) SHARED_LOCK
**     (2) RESERVED_LOCK
**     (3) PENDING_LOCK
**     (4) EXCLUSIVE_LOCK
**
** Sometimes when requesting one lock state, additional lock states
** are inserted in between.  The locking might fail on one of the later
** transitions leaving the lock state different from what it started but
** still short of its goal.  The following chart shows the allowed
** transitions and the inserted intermediate states:
**
**    UNLOCKED -> SHARED
**    SHARED -> RESERVED
**    SHARED -> (PENDING) -> EXCLUSIVE
**    RESERVED -> (PENDING) -> EXCLUSIVE
**    PENDING -> EXCLUSIVE
**
** This routine will only increase a lock.  Use the sqlite3OsUnlock()
** routine to lower a locking level.
*/
static int unixLock(OsFile *id, int locktype){
  /* The following describes the implementation of the various locks and
  ** lock transitions in terms of the POSIX advisory shared and exclusive
  ** lock primitives (called read-locks and write-locks below, to avoid
  ** confusion with SQLite lock names). The algorithms are complicated
  ** slightly in order to be compatible with windows systems simultaneously
  ** accessing the same database file, in case that is ever required.
  **
  ** Symbols defined in os.h indentify the 'pending byte' and the 'reserved
  ** byte', each single bytes at well known offsets, and the 'shared byte
  ** range', a range of 510 bytes at a well known offset.
  **
  ** To obtain a SHARED lock, a read-lock is obtained on the 'pending
  ** byte'.  If this is successful, a random byte from the 'shared byte
  ** range' is read-locked and the lock on the 'pending byte' released.
  **
  ** A process may only obtain a RESERVED lock after it has a SHARED lock.
  ** A RESERVED lock is implemented by grabbing a write-lock on the
  ** 'reserved byte'. 
  **
  ** A process may only obtain a PENDING lock after it has obtained a
  ** SHARED lock. A PENDING lock is implemented by obtaining a write-lock
  ** on the 'pending byte'. This ensures that no new SHARED locks can be
  ** obtained, but existing SHARED locks are allowed to persist. A process
  ** does not have to obtain a RESERVED lock on the way to a PENDING lock.
  ** This property is used by the algorithm for rolling back a journal file
  ** after a crash.
  **
  ** An EXCLUSIVE lock, obtained after a PENDING lock is held, is
  ** implemented by obtaining a write-lock on the entire 'shared byte
  ** range'. Since all other locks require a read-lock on one of the bytes
  ** within this range, this ensures that no other locks are held on the
  ** database. 
  **
  ** The reason a single byte cannot be used instead of the 'shared byte
  ** range' is that some versions of windows do not support read-locks. By
  ** locking a random byte from a range, concurrent SHARED locks may exist
  ** even if the locking primitive used is always a write-lock.
  */
  int rc = SQLITE_OK;
  unixFile *pFile = (unixFile*)id;
  struct lockInfo *pLock = pFile->pLock;
  struct flock lock;
  int s;

  assert( pFile );
  OSTRACE7("LOCK    %d %s was %s(%s,%d) pid=%d\n", pFile->h,
      locktypeName(locktype), locktypeName(pFile->locktype),
      locktypeName(pLock->locktype), pLock->cnt , getpid());

  /* If there is already a lock of this type or more restrictive on the
  ** OsFile, do nothing. Don't use the end_lock: exit path, as
  ** sqlite3OsEnterMutex() hasn't been called yet.
  */
  if( pFile->locktype>=locktype ){
    OSTRACE3("LOCK    %d %s ok (already held)\n", pFile->h,
            locktypeName(locktype));
    return SQLITE_OK;
  }

  /* Make sure the locking sequence is correct
  */
  assert( pFile->locktype!=NO_LOCK || locktype==SHARED_LOCK );
  assert( locktype!=PENDING_LOCK );
  assert( locktype!=RESERVED_LOCK || pFile->locktype==SHARED_LOCK );

  /* This mutex is needed because pFile->pLock is shared across threads
  */
  sqlite3OsEnterMutex();

  /* Make sure the current thread owns the pFile.
  */
  rc = transferOwnership(pFile);
  if( rc!=SQLITE_OK ){
    sqlite3OsLeaveMutex();
    return rc;
  }
  pLock = pFile->pLock;

  /* If some thread using this PID has a lock via a different OsFile*
  ** handle that precludes the requested lock, return BUSY.
  */
  if( (pFile->locktype!=pLock->locktype && 
          (pLock->locktype>=PENDING_LOCK || locktype>SHARED_LOCK))
  ){
    rc = SQLITE_BUSY;
    goto end_lock;
  }

  /* If a SHARED lock is requested, and some thread using this PID already
  ** has a SHARED or RESERVED lock, then increment reference counts and
  ** return SQLITE_OK.
  */
  if( locktype==SHARED_LOCK && 
      (pLock->locktype==SHARED_LOCK || pLock->locktype==RESERVED_LOCK) ){
    assert( locktype==SHARED_LOCK );
    assert( pFile->locktype==0 );
    assert( pLock->cnt>0 );
    pFile->locktype = SHARED_LOCK;
    pLock->cnt++;
    pFile->pOpen->nLock++;
    goto end_lock;
  }

  lock.l_len = 1L;

  lock.l_whence = SEEK_SET;

  /* A PENDING lock is needed before acquiring a SHARED lock and before
  ** acquiring an EXCLUSIVE lock.  For the SHARED lock, the PENDING will
  ** be released.
  */
  if( locktype==SHARED_LOCK 
      || (locktype==EXCLUSIVE_LOCK && pFile->locktype<PENDING_LOCK)
  ){
    lock.l_type = (locktype==SHARED_LOCK?F_RDLCK:F_WRLCK);
    lock.l_start = PENDING_BYTE;
    s = fcntl(pFile->h, F_SETLK, &lock);
    if( s==(-1) ){
      rc = (errno==EINVAL) ? SQLITE_NOLFS : SQLITE_BUSY;
      goto end_lock;
    }
  }


  /* If control gets to this point, then actually go ahead and make
  ** operating system calls for the specified lock.
  */
  if( locktype==SHARED_LOCK ){
    assert( pLock->cnt==0 );
    assert( pLock->locktype==0 );

    /* Now get the read-lock */
    lock.l_start = SHARED_FIRST;
    lock.l_len = SHARED_SIZE;
    s = fcntl(pFile->h, F_SETLK, &lock);

    /* Drop the temporary PENDING lock */
    lock.l_start = PENDING_BYTE;
    lock.l_len = 1L;
    lock.l_type = F_UNLCK;
    if( fcntl(pFile->h, F_SETLK, &lock)!=0 ){
      rc = SQLITE_IOERR_UNLOCK;  /* This should never happen */
      goto end_lock;
    }
    if( s==(-1) ){
      rc = (errno==EINVAL) ? SQLITE_NOLFS : SQLITE_BUSY;
    }else{
      pFile->locktype = SHARED_LOCK;
      pFile->pOpen->nLock++;
      pLock->cnt = 1;
    }
  }else if( locktype==EXCLUSIVE_LOCK && pLock->cnt>1 ){
    /* We are trying for an exclusive lock but another thread in this
    ** same process is still holding a shared lock. */
    rc = SQLITE_BUSY;
  }else{
    /* The request was for a RESERVED or EXCLUSIVE lock.  It is
    ** assumed that there is a SHARED or greater lock on the file
    ** already.
    */
    assert( 0!=pFile->locktype );
    lock.l_type = F_WRLCK;
    switch( locktype ){
      case RESERVED_LOCK:
        lock.l_start = RESERVED_BYTE;
        break;
      case EXCLUSIVE_LOCK:
        lock.l_start = SHARED_FIRST;
        lock.l_len = SHARED_SIZE;
        break;
      default:
        assert(0);
    }
    s = fcntl(pFile->h, F_SETLK, &lock);
    if( s==(-1) ){
      rc = (errno==EINVAL) ? SQLITE_NOLFS : SQLITE_BUSY;
    }
  }
  
  if( rc==SQLITE_OK ){
    pFile->locktype = locktype;
    pLock->locktype = locktype;
  }else if( locktype==EXCLUSIVE_LOCK ){
    pFile->locktype = PENDING_LOCK;
    pLock->locktype = PENDING_LOCK;
  }

end_lock:
  sqlite3OsLeaveMutex();
  OSTRACE4("LOCK    %d %s %s\n", pFile->h, locktypeName(locktype), 
      rc==SQLITE_OK ? "ok" : "failed");
  return rc;
}

/*
** Lower the locking level on file descriptor pFile to locktype.  locktype
** must be either NO_LOCK or SHARED_LOCK.
**
** If the locking level of the file descriptor is already at or below
** the requested locking level, this routine is a no-op.
*/
static int unixUnlock(OsFile *id, int locktype){
  struct lockInfo *pLock;
  struct flock lock;
  int rc = SQLITE_OK;
  unixFile *pFile = (unixFile*)id;

  assert( pFile );
  OSTRACE7("UNLOCK  %d %d was %d(%d,%d) pid=%d\n", pFile->h, locktype,
      pFile->locktype, pFile->pLock->locktype, pFile->pLock->cnt, getpid());

  assert( locktype<=SHARED_LOCK );
  if( pFile->locktype<=locktype ){
    return SQLITE_OK;
  }
  if( CHECK_THREADID(pFile) ){
    return SQLITE_MISUSE;
  }
  sqlite3OsEnterMutex();
  pLock = pFile->pLock;
  assert( pLock->cnt!=0 );
  if( pFile->locktype>SHARED_LOCK ){
    assert( pLock->locktype==pFile->locktype );
    if( locktype==SHARED_LOCK ){
      lock.l_type = F_RDLCK;
      lock.l_whence = SEEK_SET;
      lock.l_start = SHARED_FIRST;
      lock.l_len = SHARED_SIZE;
      if( fcntl(pFile->h, F_SETLK, &lock)==(-1) ){
        /* This should never happen */
        rc = SQLITE_IOERR_RDLOCK;
      }
    }
    lock.l_type = F_UNLCK;
    lock.l_whence = SEEK_SET;
    lock.l_start = PENDING_BYTE;
    lock.l_len = 2L;  assert( PENDING_BYTE+1==RESERVED_BYTE );
    if( fcntl(pFile->h, F_SETLK, &lock)!=(-1) ){
      pLock->locktype = SHARED_LOCK;
    }else{
      rc = SQLITE_IOERR_UNLOCK;  /* This should never happen */
    }
  }
  if( locktype==NO_LOCK ){
    struct openCnt *pOpen;

    /* Decrement the shared lock counter.  Release the lock using an
    ** OS call only when all threads in this same process have released
    ** the lock.
    */
    pLock->cnt--;
    if( pLock->cnt==0 ){
      lock.l_type = F_UNLCK;
      lock.l_whence = SEEK_SET;
      lock.l_start = lock.l_len = 0L;
      if( fcntl(pFile->h, F_SETLK, &lock)!=(-1) ){
        pLock->locktype = NO_LOCK;
      }else{
        rc = SQLITE_IOERR_UNLOCK;  /* This should never happen */
      }
    }

    /* Decrement the count of locks against this same file.  When the
    ** count reaches zero, close any other file descriptors whose close
    ** was deferred because of outstanding locks.
    */
    pOpen = pFile->pOpen;
    pOpen->nLock--;
    assert( pOpen->nLock>=0 );
    if( pOpen->nLock==0 && pOpen->nPending>0 ){
      int i;
      for(i=0; i<pOpen->nPending; i++){
        close(pOpen->aPending[i]);
      }
      free(pOpen->aPending);
      pOpen->nPending = 0;
      pOpen->aPending = 0;
    }
  }
  sqlite3OsLeaveMutex();
  pFile->locktype = locktype;
  return rc;
}

/*
** Close a file.
*/
static int unixClose(OsFile **pId){
  unixFile *id = (unixFile*)*pId;

  if( !id ) return SQLITE_OK;
  unixUnlock(*pId, NO_LOCK);
  if( id->dirfd>=0 ) close(id->dirfd);
  id->dirfd = -1;
  sqlite3OsEnterMutex();

  if( id->pOpen->nLock ){
    /* If there are outstanding locks, do not actually close the file just
    ** yet because that would clear those locks.  Instead, add the file
    ** descriptor to pOpen->aPending.  It will be automatically closed when
    ** the last lock is cleared.
    */
    int *aNew;
    struct openCnt *pOpen = id->pOpen;
    aNew = realloc( pOpen->aPending, (pOpen->nPending+1)*sizeof(int) );
    if( aNew==0 ){
      /* If a malloc fails, just leak the file descriptor */
    }else{
      pOpen->aPending = aNew;
      pOpen->aPending[pOpen->nPending] = id->h;
      pOpen->nPending++;
    }
  }else{
    /* There are no outstanding locks so we can close the file immediately */
    close(id->h);
  }
  releaseLockInfo(id->pLock);
  releaseOpenCnt(id->pOpen);

  sqlite3OsLeaveMutex();
  id->isOpen = 0;
  OSTRACE2("CLOSE   %-3d\n", id->h);
  OpenCounter(-1);
  sqlite3ThreadSafeFree(id);
  *pId = 0;
  return SQLITE_OK;
}


#ifdef SQLITE_ENABLE_LOCKING_STYLE
#pragma mark AFP Support

/*
 ** The afpLockingContext structure contains all afp lock specific state
 */
typedef struct afpLockingContext afpLockingContext;
struct afpLockingContext {
  unsigned long long sharedLockByte;
  char *filePath;
};

struct ByteRangeLockPB2
{
  unsigned long long offset;        /* offset to first byte to lock */
  unsigned long long length;        /* nbr of bytes to lock */
  unsigned long long retRangeStart; /* nbr of 1st byte locked if successful */
  unsigned char unLockFlag;         /* 1 = unlock, 0 = lock */
  unsigned char startEndFlag;       /* 1=rel to end of fork, 0=rel to start */
  int fd;                           /* file desc to assoc this lock with */
};

#define afpfsByteRangeLock2FSCTL	_IOWR('z', 23, struct ByteRangeLockPB2)

/* return 0 on success, 1 on failure.  To match the behavior of the 
  normal posix file locking (used in unixLock for example), we should 
  provide 'richer' return codes - specifically to differentiate between
  'file busy' and 'file system error' results */
static int _AFPFSSetLock(const char *path, int fd, unsigned long long offset, 
                         unsigned long long length, int setLockFlag)
{
  struct ByteRangeLockPB2	pb;
  int                     err;
  
  pb.unLockFlag = setLockFlag ? 0 : 1;
  pb.startEndFlag = 0;
  pb.offset = offset;
  pb.length = length; 
  pb.fd = fd;
  OSTRACE5("AFPLOCK setting lock %s for %d in range %llx:%llx\n", 
    (setLockFlag?"ON":"OFF"), fd, offset, length);
  err = fsctl(path, afpfsByteRangeLock2FSCTL, &pb, 0);
  if ( err==-1 ) {
    OSTRACE4("AFPLOCK failed to fsctl() '%s' %d %s\n", path, errno, 
      strerror(errno));
    return 1; /* error */
  } else {
    return 0;
  }
}

/*
 ** This routine checks if there is a RESERVED lock held on the specified
 ** file by this or any other process. If such a lock is held, return
 ** non-zero.  If the file is unlocked or holds only SHARED locks, then
 ** return zero.
 */
static int afpUnixCheckReservedLock(OsFile *id){
  int r = 0;
  unixFile *pFile = (unixFile*)id;
  
  assert( pFile ); 
  afpLockingContext *context = (afpLockingContext *) pFile->lockingContext;
  
  /* Check if a thread in this process holds such a lock */
  if( pFile->locktype>SHARED_LOCK ){
    r = 1;
  }
  
  /* Otherwise see if some other process holds it.
   */
  if ( !r ) {
    /* lock the byte */
    int failed = _AFPFSSetLock(context->filePath, pFile->h, RESERVED_BYTE, 1,1);  
    if (failed) {
      /* if we failed to get the lock then someone else must have it */
      r = 1;
    } else {
      /* if we succeeded in taking the reserved lock, unlock it to restore
      ** the original state */
      _AFPFSSetLock(context->filePath, pFile->h, RESERVED_BYTE, 1, 0);
    }
  }
  OSTRACE3("TEST WR-LOCK %d %d\n", pFile->h, r);
  
  return r;
}

/* AFP-style locking following the behavior of unixLock, see the unixLock 
** function comments for details of lock management. */
static int afpUnixLock(OsFile *id, int locktype)
{
  int rc = SQLITE_OK;
  unixFile *pFile = (unixFile*)id;
  afpLockingContext *context = (afpLockingContext *) pFile->lockingContext;
  int gotPendingLock = 0;
  
  assert( pFile );
  OSTRACE5("LOCK    %d %s was %s pid=%d\n", pFile->h,
         locktypeName(locktype), locktypeName(pFile->locktype), getpid());  
  /* If there is already a lock of this type or more restrictive on the
    ** OsFile, do nothing. Don't use the afp_end_lock: exit path, as
    ** sqlite3OsEnterMutex() hasn't been called yet.
    */
  if( pFile->locktype>=locktype ){
    OSTRACE3("LOCK    %d %s ok (already held)\n", pFile->h,
           locktypeName(locktype));
    return SQLITE_OK;
  }

  /* Make sure the locking sequence is correct
    */
  assert( pFile->locktype!=NO_LOCK || locktype==SHARED_LOCK );
  assert( locktype!=PENDING_LOCK );
  assert( locktype!=RESERVED_LOCK || pFile->locktype==SHARED_LOCK );
  
  /* This mutex is needed because pFile->pLock is shared across threads
    */
  sqlite3OsEnterMutex();

  /* Make sure the current thread owns the pFile.
    */
  rc = transferOwnership(pFile);
  if( rc!=SQLITE_OK ){
    sqlite3OsLeaveMutex();
    return rc;
  }
    
  /* A PENDING lock is needed before acquiring a SHARED lock and before
    ** acquiring an EXCLUSIVE lock.  For the SHARED lock, the PENDING will
    ** be released.
    */
  if( locktype==SHARED_LOCK 
      || (locktype==EXCLUSIVE_LOCK && pFile->locktype<PENDING_LOCK)
      ){
    int failed = _AFPFSSetLock(context->filePath, pFile->h, 
      PENDING_BYTE, 1, 1);
    if (failed) {
      rc = SQLITE_BUSY;
      goto afp_end_lock;
    }
  }
  
  /* If control gets to this point, then actually go ahead and make
    ** operating system calls for the specified lock.
    */
  if( locktype==SHARED_LOCK ){
    int lk, failed;
    int tries = 0;
    
    /* Now get the read-lock */
    /* note that the quality of the randomness doesn't matter that much */
    lk = random(); 
    context->sharedLockByte = (lk & 0x7fffffff)%(SHARED_SIZE - 1);
    failed = _AFPFSSetLock(context->filePath, pFile->h, 
      SHARED_FIRST+context->sharedLockByte, 1, 1);
    
    /* Drop the temporary PENDING lock */
    if (_AFPFSSetLock(context->filePath, pFile->h, PENDING_BYTE, 1, 0)) {
      rc = SQLITE_IOERR_UNLOCK;  /* This should never happen */
      goto afp_end_lock;
    }
    
    if( failed ){
      rc = SQLITE_BUSY;
    } else {
      pFile->locktype = SHARED_LOCK;
    }
  }else{
    /* The request was for a RESERVED or EXCLUSIVE lock.  It is
    ** assumed that there is a SHARED or greater lock on the file
    ** already.
    */
    int failed = 0;
    assert( 0!=pFile->locktype );
    if (locktype >= RESERVED_LOCK && pFile->locktype < RESERVED_LOCK) {
        /* Acquire a RESERVED lock */
        failed = _AFPFSSetLock(context->filePath, pFile->h, RESERVED_BYTE, 1,1);
    }
    if (!failed && locktype == EXCLUSIVE_LOCK) {
      /* Acquire an EXCLUSIVE lock */
        
      /* Remove the shared lock before trying the range.  we'll need to 
      ** reestablish the shared lock if we can't get the  afpUnixUnlock
      */
      if (!_AFPFSSetLock(context->filePath, pFile->h, SHARED_FIRST +
                         context->sharedLockByte, 1, 0)) {
        /* now attemmpt to get the exclusive lock range */
        failed = _AFPFSSetLock(context->filePath, pFile->h, SHARED_FIRST, 
                               SHARED_SIZE, 1);
        if (failed && _AFPFSSetLock(context->filePath, pFile->h, SHARED_FIRST +
                                    context->sharedLockByte, 1, 1)) {
          rc = SQLITE_IOERR_RDLOCK; /* this should never happen */
        }
      } else {
        /* */
        rc = SQLITE_IOERR_UNLOCK; /* this should never happen */
      }
    }
    if( failed && rc == SQLITE_OK){
      rc = SQLITE_BUSY;
    }
  }
  
  if( rc==SQLITE_OK ){
    pFile->locktype = locktype;
  }else if( locktype==EXCLUSIVE_LOCK ){
    pFile->locktype = PENDING_LOCK;
  }
  
afp_end_lock:
    sqlite3OsLeaveMutex();
  OSTRACE4("LOCK    %d %s %s\n", pFile->h, locktypeName(locktype), 
         rc==SQLITE_OK ? "ok" : "failed");
  return rc;
}

/*
 ** Lower the locking level on file descriptor pFile to locktype.  locktype
 ** must be either NO_LOCK or SHARED_LOCK.
 **
 ** If the locking level of the file descriptor is already at or below
 ** the requested locking level, this routine is a no-op.
 */
static int afpUnixUnlock(OsFile *id, int locktype) {
  struct flock lock;
  int rc = SQLITE_OK;
  unixFile *pFile = (unixFile*)id;
  afpLockingContext *context = (afpLockingContext *) pFile->lockingContext;

  assert( pFile );
  OSTRACE5("UNLOCK  %d %d was %d pid=%d\n", pFile->h, locktype,
         pFile->locktype, getpid());
  
  assert( locktype<=SHARED_LOCK );
  if( pFile->locktype<=locktype ){
    return SQLITE_OK;
  }
  if( CHECK_THREADID(pFile) ){
    return SQLITE_MISUSE;
  }
  sqlite3OsEnterMutex();
  if( pFile->locktype>SHARED_LOCK ){
    if( locktype==SHARED_LOCK ){
      int failed = 0;

      /* unlock the exclusive range - then re-establish the shared lock */
      if (pFile->locktype==EXCLUSIVE_LOCK) {
        failed = _AFPFSSetLock(context->filePath, pFile->h, SHARED_FIRST, 
                                 SHARED_SIZE, 0);
        if (!failed) {
          /* successfully removed the exclusive lock */
          if (_AFPFSSetLock(context->filePath, pFile->h, SHARED_FIRST+
                            context->sharedLockByte, 1, 1)) {
            /* failed to re-establish our shared lock */
            rc = SQLITE_IOERR_RDLOCK; /* This should never happen */
          }
        } else {
          /* This should never happen - failed to unlock the exclusive range */
          rc = SQLITE_IOERR_UNLOCK;
        } 
      }
    }
    if (rc == SQLITE_OK && pFile->locktype>=PENDING_LOCK) {
      if (_AFPFSSetLock(context->filePath, pFile->h, PENDING_BYTE, 1, 0)){
        /* failed to release the pending lock */
        rc = SQLITE_IOERR_UNLOCK; /* This should never happen */
      }
    } 
    if (rc == SQLITE_OK && pFile->locktype>=RESERVED_LOCK) {
      if (_AFPFSSetLock(context->filePath, pFile->h, RESERVED_BYTE, 1, 0)) {
        /* failed to release the reserved lock */
        rc = SQLITE_IOERR_UNLOCK;  /* This should never happen */
      }
    } 
  }
  if( locktype==NO_LOCK ){
    int failed = _AFPFSSetLock(context->filePath, pFile->h, 
                               SHARED_FIRST + context->sharedLockByte, 1, 0);
    if (failed) {
      rc = SQLITE_IOERR_UNLOCK;  /* This should never happen */
    }
  }
  if (rc == SQLITE_OK)
    pFile->locktype = locktype;
  sqlite3OsLeaveMutex();
  return rc;
}

/*
 ** Close a file & cleanup AFP specific locking context 
 */
static int afpUnixClose(OsFile **pId) {
  unixFile *id = (unixFile*)*pId;
  
  if( !id ) return SQLITE_OK;
  afpUnixUnlock(*pId, NO_LOCK);
  /* free the AFP locking structure */
  if (id->lockingContext != NULL) {
    if (((afpLockingContext *)id->lockingContext)->filePath != NULL)
      sqlite3ThreadSafeFree(((afpLockingContext*)id->lockingContext)->filePath);
    sqlite3ThreadSafeFree(id->lockingContext);
  }
  
  if( id->dirfd>=0 ) close(id->dirfd);
  id->dirfd = -1;
  close(id->h);
  id->isOpen = 0;
  OSTRACE2("CLOSE   %-3d\n", id->h);
  OpenCounter(-1);
  sqlite3ThreadSafeFree(id);
  *pId = 0;
  return SQLITE_OK;
}


#pragma mark flock() style locking

/*
 ** The flockLockingContext is not used
 */
typedef void flockLockingContext;

static int flockUnixCheckReservedLock(OsFile *id) {
  unixFile *pFile = (unixFile*)id;
  
  if (pFile->locktype == RESERVED_LOCK) {
    return 1; /* already have a reserved lock */
  } else {
    /* attempt to get the lock */
    int rc = flock(pFile->h, LOCK_EX | LOCK_NB);
    if (!rc) {
      /* got the lock, unlock it */
      flock(pFile->h, LOCK_UN);
      return 0;  /* no one has it reserved */
    }
    return 1; /* someone else might have it reserved */
  }
}

static int flockUnixLock(OsFile *id, int locktype) {
  unixFile *pFile = (unixFile*)id;
  
  /* if we already have a lock, it is exclusive.  
  ** Just adjust level and punt on outta here. */
  if (pFile->locktype > NO_LOCK) {
    pFile->locktype = locktype;
    return SQLITE_OK;
  }
  
  /* grab an exclusive lock */
  int rc = flock(pFile->h, LOCK_EX | LOCK_NB);
  if (rc) {
    /* didn't get, must be busy */
    return SQLITE_BUSY;
  } else {
    /* got it, set the type and return ok */
    pFile->locktype = locktype;
    return SQLITE_OK;
  }
}

static int flockUnixUnlock(OsFile *id, int locktype) {
  unixFile *pFile = (unixFile*)id;
  
  assert( locktype<=SHARED_LOCK );
  
  /* no-op if possible */
  if( pFile->locktype==locktype ){
    return SQLITE_OK;
  }
  
  /* shared can just be set because we always have an exclusive */
  if (locktype==SHARED_LOCK) {
    pFile->locktype = locktype;
    return SQLITE_OK;
  }
  
  /* no, really, unlock. */
  int rc = flock(pFile->h, LOCK_UN);
  if (rc)
    return SQLITE_IOERR_UNLOCK;
  else {
    pFile->locktype = NO_LOCK;
    return SQLITE_OK;
  }
}

/*
 ** Close a file.
 */
static int flockUnixClose(OsFile **pId) {
  unixFile *id = (unixFile*)*pId;
  
  if( !id ) return SQLITE_OK;
  flockUnixUnlock(*pId, NO_LOCK);
  
  if( id->dirfd>=0 ) close(id->dirfd);
  id->dirfd = -1;
  sqlite3OsEnterMutex();
  
  close(id->h);  
  sqlite3OsLeaveMutex();
  id->isOpen = 0;
  OSTRACE2("CLOSE   %-3d\n", id->h);
  OpenCounter(-1);
  sqlite3ThreadSafeFree(id);
  *pId = 0;
  return SQLITE_OK;
}

#pragma mark Old-School .lock file based locking

/*
 ** The dotlockLockingContext structure contains all dotlock (.lock) lock
 ** specific state
 */
typedef struct dotlockLockingContext dotlockLockingContext;
struct dotlockLockingContext {
  char *lockPath;
};


static int dotlockUnixCheckReservedLock(OsFile *id) {
  unixFile *pFile = (unixFile*)id;
  dotlockLockingContext *context = 
    (dotlockLockingContext *) pFile->lockingContext;
  
  if (pFile->locktype == RESERVED_LOCK) {
    return 1; /* already have a reserved lock */
  } else {
    struct stat statBuf;
    if (lstat(context->lockPath,&statBuf) == 0)
      /* file exists, someone else has the lock */
      return 1;
    else
      /* file does not exist, we could have it if we want it */
      return 0;
  }
}

static int dotlockUnixLock(OsFile *id, int locktype) {
  unixFile *pFile = (unixFile*)id;
  dotlockLockingContext *context = 
    (dotlockLockingContext *) pFile->lockingContext;
  
  /* if we already have a lock, it is exclusive.  
  ** Just adjust level and punt on outta here. */
  if (pFile->locktype > NO_LOCK) {
    pFile->locktype = locktype;
    
    /* Always update the timestamp on the old file */
    utimes(context->lockPath,NULL);
    return SQLITE_OK;
  }
  
  /* check to see if lock file already exists */
  struct stat statBuf;
  if (lstat(context->lockPath,&statBuf) == 0){
    return SQLITE_BUSY; /* it does, busy */
  }
  
  /* grab an exclusive lock */
  int fd = open(context->lockPath,O_RDONLY|O_CREAT|O_EXCL,0600);
  if (fd < 0) {
    /* failed to open/create the file, someone else may have stolen the lock */
    return SQLITE_BUSY; 
  }
  close(fd);
  
  /* got it, set the type and return ok */
  pFile->locktype = locktype;
  return SQLITE_OK;
}

static int dotlockUnixUnlock(OsFile *id, int locktype) {
  unixFile *pFile = (unixFile*)id;
  dotlockLockingContext *context = 
    (dotlockLockingContext *) pFile->lockingContext;
  
  assert( locktype<=SHARED_LOCK );
  
  /* no-op if possible */
  if( pFile->locktype==locktype ){
    return SQLITE_OK;
  }
  
  /* shared can just be set because we always have an exclusive */
  if (locktype==SHARED_LOCK) {
    pFile->locktype = locktype;
    return SQLITE_OK;
  }
  
  /* no, really, unlock. */
  unlink(context->lockPath);
  pFile->locktype = NO_LOCK;
  return SQLITE_OK;
}

/*
 ** Close a file.
 */
static int dotlockUnixClose(OsFile **pId) {
  unixFile *id = (unixFile*)*pId;
  
  if( !id ) return SQLITE_OK;
  dotlockUnixUnlock(*pId, NO_LOCK);
  /* free the dotlock locking structure */
  if (id->lockingContext != NULL) {
    if (((dotlockLockingContext *)id->lockingContext)->lockPath != NULL)
      sqlite3ThreadSafeFree( ( (dotlockLockingContext *)
        id->lockingContext)->lockPath);
    sqlite3ThreadSafeFree(id->lockingContext);
  }
  
  if( id->dirfd>=0 ) close(id->dirfd);
  id->dirfd = -1;
  sqlite3OsEnterMutex();
  
  close(id->h);
  
  sqlite3OsLeaveMutex();
  id->isOpen = 0;
  OSTRACE2("CLOSE   %-3d\n", id->h);
  OpenCounter(-1);
  sqlite3ThreadSafeFree(id);
  *pId = 0;
  return SQLITE_OK;
}


#pragma mark No locking

/*
 ** The nolockLockingContext is void
 */
typedef void nolockLockingContext;

static int nolockUnixCheckReservedLock(OsFile *id) {
  return 0;
}

static int nolockUnixLock(OsFile *id, int locktype) {
  return SQLITE_OK;
}

static int nolockUnixUnlock(OsFile *id, int locktype) {
  return SQLITE_OK;
}

/*
 ** Close a file.
 */
static int nolockUnixClose(OsFile **pId) {
  unixFile *id = (unixFile*)*pId;
  
  if( !id ) return SQLITE_OK;
  if( id->dirfd>=0 ) close(id->dirfd);
  id->dirfd = -1;
  sqlite3OsEnterMutex();
  
  close(id->h);
  
  sqlite3OsLeaveMutex();
  id->isOpen = 0;
  OSTRACE2("CLOSE   %-3d\n", id->h);
  OpenCounter(-1);
  sqlite3ThreadSafeFree(id);
  *pId = 0;
  return SQLITE_OK;
}

#endif /* SQLITE_ENABLE_LOCKING_STYLE */

/*
** Turn a relative pathname into a full pathname.  Return a pointer
** to the full pathname stored in space obtained from sqliteMalloc().
** The calling function is responsible for freeing this space once it
** is no longer needed.
*/
SQLITE_PRIVATE char *sqlite3UnixFullPathname(const char *zRelative){
  char *zFull = 0;
  if( zRelative[0]=='/' ){
    sqlite3SetString(&zFull, zRelative, (char*)0);
  }else{
    char *zBuf = sqliteMalloc(5000);
    if( zBuf==0 ){
      return 0;
    }
    zBuf[0] = 0;
    sqlite3SetString(&zFull, getcwd(zBuf, 5000), "/", zRelative,
                    (char*)0);
    sqliteFree(zBuf);
  }

#if 0
  /*
  ** Remove "/./" path elements and convert "/A/./" path elements
  ** to just "/".
  */
  if( zFull ){
    int i, j;
    for(i=j=0; zFull[i]; i++){
      if( zFull[i]=='/' ){
        if( zFull[i+1]=='/' ) continue;
        if( zFull[i+1]=='.' && zFull[i+2]=='/' ){
          i += 1;
          continue;
        }
        if( zFull[i+1]=='.' && zFull[i+2]=='.' && zFull[i+3]=='/' ){
          while( j>0 && zFull[j-1]!='/' ){ j--; }
          i += 3;
          continue;
        }
      }
      zFull[j++] = zFull[i];
    }
    zFull[j] = 0;
  }
#endif

  return zFull;
}

/*
** Change the value of the fullsync flag in the given file descriptor.
*/
static void unixSetFullSync(OsFile *id, int v){
  ((unixFile*)id)->fullSync = v;
}

/*
** Return the underlying file handle for an OsFile
*/
static int unixFileHandle(OsFile *id){
  return ((unixFile*)id)->h;
}

/*
** Return an integer that indices the type of lock currently held
** by this handle.  (Used for testing and analysis only.)
*/
static int unixLockState(OsFile *id){
  return ((unixFile*)id)->locktype;
}

/*
** Return the sector size in bytes of the underlying block device for
** the specified file. This is almost always 512 bytes, but may be
** larger for some devices.
**
** SQLite code assumes this function cannot fail. It also assumes that
** if two files are created in the same file-system directory (i.e.
** a database and it's journal file) that the sector size will be the
** same for both.
*/
static int unixSectorSize(OsFile *id){
  return SQLITE_DEFAULT_SECTOR_SIZE;
}

/*
** This vector defines all the methods that can operate on an OsFile
** for unix.
*/
static const IoMethod sqlite3UnixIoMethod = {
  unixClose,
  unixOpenDirectory,
  unixRead,
  unixWrite,
  unixSeek,
  unixTruncate,
  unixSync,
  unixSetFullSync,
  unixFileHandle,
  unixFileSize,
  unixLock,
  unixUnlock,
  unixLockState,
  unixCheckReservedLock,
  unixSectorSize,
};

#ifdef SQLITE_ENABLE_LOCKING_STYLE
/*
 ** This vector defines all the methods that can operate on an OsFile
 ** for unix with AFP style file locking.
 */
static const IoMethod sqlite3AFPLockingUnixIoMethod = {
    afpUnixClose,
    unixOpenDirectory,
    unixRead,
    unixWrite,
    unixSeek,
    unixTruncate,
    unixSync,
    unixSetFullSync,
    unixFileHandle,
    unixFileSize,
    afpUnixLock,
    afpUnixUnlock,
    unixLockState,
    afpUnixCheckReservedLock,
    unixSectorSize,
};

/*
 ** This vector defines all the methods that can operate on an OsFile
 ** for unix with flock() style file locking.
 */
static const IoMethod sqlite3FlockLockingUnixIoMethod = {
    flockUnixClose,
    unixOpenDirectory,
    unixRead,
    unixWrite,
    unixSeek,
    unixTruncate,
    unixSync,
    unixSetFullSync,
    unixFileHandle,
    unixFileSize,
    flockUnixLock,
    flockUnixUnlock,
    unixLockState,
    flockUnixCheckReservedLock,
    unixSectorSize,
};

/*
 ** This vector defines all the methods that can operate on an OsFile
 ** for unix with dotlock style file locking.
 */
static const IoMethod sqlite3DotlockLockingUnixIoMethod = {
    dotlockUnixClose,
    unixOpenDirectory,
    unixRead,
    unixWrite,
    unixSeek,
    unixTruncate,
    unixSync,
    unixSetFullSync,
    unixFileHandle,
    unixFileSize,
    dotlockUnixLock,
    dotlockUnixUnlock,
    unixLockState,
    dotlockUnixCheckReservedLock,
    unixSectorSize,
};

/*
 ** This vector defines all the methods that can operate on an OsFile
 ** for unix with dotlock style file locking.
 */
static const IoMethod sqlite3NolockLockingUnixIoMethod = {
  nolockUnixClose,
  unixOpenDirectory,
  unixRead,
  unixWrite,
  unixSeek,
  unixTruncate,
  unixSync,
  unixSetFullSync,
  unixFileHandle,
  unixFileSize,
  nolockUnixLock,
  nolockUnixUnlock,
  unixLockState,
  nolockUnixCheckReservedLock,
  unixSectorSize,
};

#endif /* SQLITE_ENABLE_LOCKING_STYLE */

/*
** Allocate memory for a new unixFile and initialize that unixFile.
** Write a pointer to the new unixFile into *pId.
** If we run out of memory, close the file and return an error.
*/
#ifdef SQLITE_ENABLE_LOCKING_STYLE
/* 
 ** When locking extensions are enabled, the filepath and locking style 
 ** are needed to determine the unixFile pMethod to use for locking operations.
 ** The locking-style specific lockingContext data structure is created 
 ** and assigned here also.
 */
static int allocateUnixFile(
  int h,                  /* Open file descriptor of file being opened */
  OsFile **pId,           /* Write completed initialization here */
  const char *zFilename,  /* Name of the file being opened */
  int delFlag             /* Delete-on-or-before-close flag */
){
  sqlite3LockingStyle lockingStyle;
  unixFile *pNew;
  unixFile f;
  int rc;

  memset(&f, 0, sizeof(f));
  lockingStyle = sqlite3DetectLockingStyle(zFilename, h);
  if ( lockingStyle == posixLockingStyle ) {
    sqlite3OsEnterMutex();
    rc = findLockInfo(h, &f.pLock, &f.pOpen);
    sqlite3OsLeaveMutex();
    if( rc ){
      close(h);
      unlink(zFilename);
      return SQLITE_NOMEM;
    }
  } else {
    /*  pLock and pOpen are only used for posix advisory locking */
    f.pLock = NULL;
    f.pOpen = NULL;
  }
  if( delFlag ){
    unlink(zFilename);
  }
  f.dirfd = -1;
  f.h = h;
  SET_THREADID(&f);
  pNew = sqlite3ThreadSafeMalloc( sizeof(unixFile) );
  if( pNew==0 ){
    close(h);
    sqlite3OsEnterMutex();
    releaseLockInfo(f.pLock);
    releaseOpenCnt(f.pOpen);
    sqlite3OsLeaveMutex();
    *pId = 0;
    return SQLITE_NOMEM;
  }else{
    *pNew = f;
    switch(lockingStyle) {
      case afpLockingStyle: {
        /* afp locking uses the file path so it needs to be included in
        ** the afpLockingContext */
        int nFilename;
        pNew->pMethod = &sqlite3AFPLockingUnixIoMethod;
        pNew->lockingContext = 
          sqlite3ThreadSafeMalloc(sizeof(afpLockingContext));
        nFilename = strlen(zFilename)+1;
        ((afpLockingContext *)pNew->lockingContext)->filePath = 
          sqlite3ThreadSafeMalloc(nFilename);
        memcpy(((afpLockingContext *)pNew->lockingContext)->filePath, 
               zFilename, nFilename);
        srandomdev();
        break;
      }
      case flockLockingStyle:
        /* flock locking doesn't need additional lockingContext information */
        pNew->pMethod = &sqlite3FlockLockingUnixIoMethod;
        break;
      case dotlockLockingStyle: {
        /* dotlock locking uses the file path so it needs to be included in
         ** the dotlockLockingContext */
        int nFilename;
        pNew->pMethod = &sqlite3DotlockLockingUnixIoMethod;
        pNew->lockingContext = sqlite3ThreadSafeMalloc(
          sizeof(dotlockLockingContext));
        nFilename = strlen(zFilename) + 6;
        ((dotlockLockingContext *)pNew->lockingContext)->lockPath = 
            sqlite3ThreadSafeMalloc( nFilename );
        sqlite3_snprintf(nFilename, 
                ((dotlockLockingContext *)pNew->lockingContext)->lockPath, 
                "%s.lock", zFilename);
        break;
      }
      case posixLockingStyle:
        /* posix locking doesn't need additional lockingContext information */
        pNew->pMethod = &sqlite3UnixIoMethod;
        break;
      case noLockingStyle:
      case unsupportedLockingStyle:
      default: 
        pNew->pMethod = &sqlite3NolockLockingUnixIoMethod;
    }
    *pId = (OsFile*)pNew;
    OpenCounter(+1);
    return SQLITE_OK;
  }
}
#else /* SQLITE_ENABLE_LOCKING_STYLE */
static int allocateUnixFile(
  int h,                 /* Open file descriptor on file being opened */
  OsFile **pId,          /* Write the resul unixFile structure here */
  const char *zFilename, /* Name of the file being opened */
  int delFlag            /* If true, delete the file on or before closing */
){
  unixFile *pNew;
  unixFile f;
  int rc;

#ifdef FD_CLOEXEC
  fcntl(h, F_SETFD, fcntl(h, F_GETFD, 0) | FD_CLOEXEC);
#endif
  memset(&f, 0, sizeof(f));
  sqlite3OsEnterMutex();
  rc = findLockInfo(h, &f.pLock, &f.pOpen);
  sqlite3OsLeaveMutex();
  if( delFlag ){
    unlink(zFilename);
  }
  if( rc ){
    close(h);
    return SQLITE_NOMEM;
  }
  OSTRACE3("OPEN    %-3d %s\n", h, zFilename);
  f.dirfd = -1;
  f.h = h;
  SET_THREADID(&f);
  pNew = sqlite3ThreadSafeMalloc( sizeof(unixFile) );
  if( pNew==0 ){
    close(h);
    sqlite3OsEnterMutex();
    releaseLockInfo(f.pLock);
    releaseOpenCnt(f.pOpen);
    sqlite3OsLeaveMutex();
    *pId = 0;
    return SQLITE_NOMEM;
  }else{
    *pNew = f;
    pNew->pMethod = &sqlite3UnixIoMethod;
    *pId = (OsFile*)pNew;
    OpenCounter(+1);
    return SQLITE_OK;
  }
}
#endif /* SQLITE_ENABLE_LOCKING_STYLE */

#endif /* SQLITE_OMIT_DISKIO */
/***************************************************************************
** Everything above deals with file I/O.  Everything that follows deals
** with other miscellanous aspects of the operating system interface
****************************************************************************/


#ifndef SQLITE_OMIT_LOAD_EXTENSION
/*
** Interfaces for opening a shared library, finding entry points
** within the shared library, and closing the shared library.
*/
#include <dlfcn.h>
SQLITE_PRIVATE void *sqlite3UnixDlopen(const char *zFilename){
  return dlopen(zFilename, RTLD_NOW | RTLD_GLOBAL);
}
SQLITE_PRIVATE void *sqlite3UnixDlsym(void *pHandle, const char *zSymbol){
  return dlsym(pHandle, zSymbol);
}
SQLITE_PRIVATE int sqlite3UnixDlclose(void *pHandle){
  return dlclose(pHandle);
}
#endif /* SQLITE_OMIT_LOAD_EXTENSION */

/*
** Get information to seed the random number generator.  The seed
** is written into the buffer zBuf[256].  The calling function must
** supply a sufficiently large buffer.
*/
SQLITE_PRIVATE int sqlite3UnixRandomSeed(char *zBuf){
  /* We have to initialize zBuf to prevent valgrind from reporting
  ** errors.  The reports issued by valgrind are incorrect - we would
  ** prefer that the randomness be increased by making use of the
  ** uninitialized space in zBuf - but valgrind errors tend to worry
  ** some users.  Rather than argue, it seems easier just to initialize
  ** the whole array and silence valgrind, even if that means less randomness
  ** in the random seed.
  **
  ** When testing, initializing zBuf[] to zero is all we do.  That means
  ** that we always use the same random number sequence.  This makes the
  ** tests repeatable.
  */
  memset(zBuf, 0, 256);
#if !defined(SQLITE_TEST)
  {
    int pid, fd;
    fd = open("/dev/urandom", O_RDONLY);
    if( fd<0 ){
      time_t t;
      time(&t);
      memcpy(zBuf, &t, sizeof(t));
      pid = getpid();
      memcpy(&zBuf[sizeof(time_t)], &pid, sizeof(pid));
    }else{
      read(fd, zBuf, 256);
      close(fd);
    }
  }
#endif
  return SQLITE_OK;
}

/*
** Sleep for a little while.  Return the amount of time slept.
** The argument is the number of milliseconds we want to sleep.
*/
SQLITE_PRIVATE int sqlite3UnixSleep(int ms){
#if defined(HAVE_USLEEP) && HAVE_USLEEP
  usleep(ms*1000);
  return ms;
#else
  sleep((ms+999)/1000);
  return 1000*((ms+999)/1000);
#endif
}

/*
** Static variables used for thread synchronization.
**
** inMutex      the nesting depth of the recursive mutex.  The thread
**              holding mutexMain can read this variable at any time.
**              But is must hold mutexAux to change this variable.  Other
**              threads must hold mutexAux to read the variable and can
**              never write.
**
** mutexOwner   The thread id of the thread holding mutexMain.  Same
**              access rules as for inMutex.
**
** mutexOwnerValid   True if the value in mutexOwner is valid.  The same
**                   access rules apply as for inMutex.
**
** mutexMain    The main mutex.  Hold this mutex in order to get exclusive
**              access to SQLite data structures.
**
** mutexAux     An auxiliary mutex needed to access variables defined above.
**
** Mutexes are always acquired in this order: mutexMain mutexAux.   It
** is not necessary to acquire mutexMain in order to get mutexAux - just
** do not attempt to acquire them in the reverse order: mutexAux mutexMain.
** Either get the mutexes with mutexMain first or get mutexAux only.
**
** When running on a platform where the three variables inMutex, mutexOwner,
** and mutexOwnerValid can be set atomically, the mutexAux is not required.
** On many systems, all three are 32-bit integers and writing to a 32-bit
** integer is atomic.  I think.  But there are no guarantees.  So it seems
** safer to protect them using mutexAux.
*/
static int inMutex = 0;
#ifdef SQLITE_UNIX_THREADS
static pthread_t mutexOwner;          /* Thread holding mutexMain */
static int mutexOwnerValid = 0;       /* True if mutexOwner is valid */
static pthread_mutex_t mutexMain = PTHREAD_MUTEX_INITIALIZER; /* The mutex */
static pthread_mutex_t mutexAux = PTHREAD_MUTEX_INITIALIZER;  /* Aux mutex */
#endif

/*
** The following pair of routine implement mutual exclusion for
** multi-threaded processes.  Only a single thread is allowed to
** executed code that is surrounded by EnterMutex() and LeaveMutex().
**
** SQLite uses only a single Mutex.  There is not much critical
** code and what little there is executes quickly and without blocking.
**
** As of version 3.3.2, this mutex must be recursive.
*/
SQLITE_PRIVATE void sqlite3UnixEnterMutex(){
#ifdef SQLITE_UNIX_THREADS
  pthread_mutex_lock(&mutexAux);
  if( !mutexOwnerValid || !pthread_equal(mutexOwner, pthread_self()) ){
    pthread_mutex_unlock(&mutexAux);
    pthread_mutex_lock(&mutexMain);
    assert( inMutex==0 );
    assert( !mutexOwnerValid );
    pthread_mutex_lock(&mutexAux);
    mutexOwner = pthread_self();
    mutexOwnerValid = 1;
  }
  inMutex++;
  pthread_mutex_unlock(&mutexAux);
#else
  inMutex++;
#endif
}
SQLITE_PRIVATE void sqlite3UnixLeaveMutex(){
  assert( inMutex>0 );
#ifdef SQLITE_UNIX_THREADS
  pthread_mutex_lock(&mutexAux);
  inMutex--;
  assert( pthread_equal(mutexOwner, pthread_self()) );
  if( inMutex==0 ){
    assert( mutexOwnerValid );
    mutexOwnerValid = 0;
    pthread_mutex_unlock(&mutexMain);
  }
  pthread_mutex_unlock(&mutexAux);
#else
  inMutex--;
#endif
}

/*
** Return TRUE if the mutex is currently held.
**
** If the thisThrd parameter is true, return true only if the
** calling thread holds the mutex.  If the parameter is false, return
** true if any thread holds the mutex.
*/
SQLITE_PRIVATE int sqlite3UnixInMutex(int thisThrd){
#ifdef SQLITE_UNIX_THREADS
  int rc;
  pthread_mutex_lock(&mutexAux);
  rc = inMutex>0 && (thisThrd==0 || pthread_equal(mutexOwner,pthread_self()));
  pthread_mutex_unlock(&mutexAux);
  return rc;
#else
  return inMutex>0;
#endif
}

/*
** Remember the number of thread-specific-data blocks allocated.
** Use this to verify that we are not leaking thread-specific-data.
** Ticket #1601
*/
#ifdef SQLITE_TEST
int sqlite3_tsd_count = 0;
# ifdef SQLITE_UNIX_THREADS
    static pthread_mutex_t tsd_counter_mutex = PTHREAD_MUTEX_INITIALIZER;
#   define TSD_COUNTER(N) \
             pthread_mutex_lock(&tsd_counter_mutex); \
             sqlite3_tsd_count += N; \
             pthread_mutex_unlock(&tsd_counter_mutex);
# else
#   define TSD_COUNTER(N)  sqlite3_tsd_count += N
# endif
#else
# define TSD_COUNTER(N)  /* no-op */
#endif

/*
** If called with allocateFlag>0, then return a pointer to thread
** specific data for the current thread.  Allocate and zero the
** thread-specific data if it does not already exist.
**
** If called with allocateFlag==0, then check the current thread
** specific data.  Return it if it exists.  If it does not exist,
** then return NULL.
**
** If called with allocateFlag<0, check to see if the thread specific
** data is allocated and is all zero.  If it is then deallocate it.
** Return a pointer to the thread specific data or NULL if it is
** unallocated or gets deallocated.
*/
SQLITE_PRIVATE ThreadData *sqlite3UnixThreadSpecificData(int allocateFlag){
  static const ThreadData zeroData = {0};  /* Initializer to silence warnings
                                           ** from broken compilers */
#ifdef SQLITE_UNIX_THREADS
  static pthread_key_t key;
  static int keyInit = 0;
  ThreadData *pTsd;

  if( !keyInit ){
    sqlite3OsEnterMutex();
    if( !keyInit ){
      int rc;
      rc = pthread_key_create(&key, 0);
      if( rc ){
        sqlite3OsLeaveMutex();
        return 0;
      }
      keyInit = 1;
    }
    sqlite3OsLeaveMutex();
  }

  pTsd = pthread_getspecific(key);
  if( allocateFlag>0 ){
    if( pTsd==0 ){
      if( !sqlite3TestMallocFail() ){
        pTsd = sqlite3OsMalloc(sizeof(zeroData));
      }
#ifdef SQLITE_MEMDEBUG
      sqlite3_isFail = 0;
#endif
      if( pTsd ){
        *pTsd = zeroData;
        pthread_setspecific(key, pTsd);
        TSD_COUNTER(+1);
      }
    }
  }else if( pTsd!=0 && allocateFlag<0 
            && memcmp(pTsd, &zeroData, sizeof(ThreadData))==0 ){
    sqlite3OsFree(pTsd);
    pthread_setspecific(key, 0);
    TSD_COUNTER(-1);
    pTsd = 0;
  }
  return pTsd;
#else
  static ThreadData *pTsd = 0;
  if( allocateFlag>0 ){
    if( pTsd==0 ){
      if( !sqlite3TestMallocFail() ){
        pTsd = sqlite3OsMalloc( sizeof(zeroData) );
      }
#ifdef SQLITE_MEMDEBUG
      sqlite3_isFail = 0;
#endif
      if( pTsd ){
        *pTsd = zeroData;
        TSD_COUNTER(+1);
      }
    }
  }else if( pTsd!=0 && allocateFlag<0
            && memcmp(pTsd, &zeroData, sizeof(ThreadData))==0 ){
    sqlite3OsFree(pTsd);
    TSD_COUNTER(-1);
    pTsd = 0;
  }
  return pTsd;
#endif
}

/*
** The following variable, if set to a non-zero value, becomes the result
** returned from sqlite3OsCurrentTime().  This is used for testing.
*/
#ifdef SQLITE_TEST
int sqlite3_current_time = 0;
#endif

/*
** Find the current time (in Universal Coordinated Time).  Write the
** current time and date as a Julian Day number into *prNow and
** return 0.  Return 1 if the time and date cannot be found.
*/
SQLITE_PRIVATE int sqlite3UnixCurrentTime(double *prNow){
#ifdef NO_GETTOD
  time_t t;
  time(&t);
  *prNow = t/86400.0 + 2440587.5;
#else
  struct timeval sNow;
  gettimeofday(&sNow, 0);
  *prNow = 2440587.5 + sNow.tv_sec/86400.0 + sNow.tv_usec/86400000000.0;
#endif
#ifdef SQLITE_TEST
  if( sqlite3_current_time ){
    *prNow = sqlite3_current_time/86400.0 + 2440587.5;
  }
#endif
  return 0;
}

#endif /* OS_UNIX */

/************** End of os_unix.c *********************************************/
/************** Begin file os_win.c ******************************************/
/*
** 2004 May 22
**
** The author disclaims copyright to this source code.  In place of
** a legal notice, here is a blessing:
**
**    May you do good and not evil.
**    May you find forgiveness for yourself and forgive others.
**    May you share freely, never taking more than you give.
**
******************************************************************************
**
** This file contains code that is specific to windows.
*/
#if OS_WIN               /* This file is used for windows only */

#include <winbase.h>

#ifdef __CYGWIN__
# include <sys/cygwin.h>
#endif

/*
** Macros used to determine whether or not to use threads.
*/
#if defined(THREADSAFE) && THREADSAFE
# define SQLITE_W32_THREADS 1
#endif

/*
** Include code that is common to all os_*.c files
*/
/************** Include os_common.h in the middle of os_win.c ****************/
/************** Begin file os_common.h ***************************************/
/*
** 2004 May 22
**
** The author disclaims copyright to this source code.  In place of
** a legal notice, here is a blessing:
**
**    May you do good and not evil.
**    May you find forgiveness for yourself and forgive others.
**    May you share freely, never taking more than you give.
**
******************************************************************************
**
** This file contains macros and a little bit of code that is common to
** all of the platform-specific files (os_*.c) and is #included into those
** files.
**
** This file should be #included by the os_*.c files only.  It is not a
** general purpose header file.
*/

/*
** At least two bugs have slipped in because we changed the MEMORY_DEBUG
** macro to SQLITE_DEBUG and some older makefiles have not yet made the
** switch.  The following code should catch this problem at compile-time.
*/
#ifdef MEMORY_DEBUG
# error "The MEMORY_DEBUG macro is obsolete.  Use SQLITE_DEBUG instead."
#endif


/*
 * When testing, this global variable stores the location of the
 * pending-byte in the database file.
 */
#ifdef SQLITE_TEST
unsigned int sqlite3_pending_byte = 0x40000000;
#endif

int sqlite3_os_trace = 0;
#ifdef SQLITE_DEBUG
#define OSTRACE1(X)         if( sqlite3_os_trace ) sqlite3DebugPrintf(X)
#define OSTRACE2(X,Y)       if( sqlite3_os_trace ) sqlite3DebugPrintf(X,Y)
#define OSTRACE3(X,Y,Z)     if( sqlite3_os_trace ) sqlite3DebugPrintf(X,Y,Z)
#define OSTRACE4(X,Y,Z,A)   if( sqlite3_os_trace ) sqlite3DebugPrintf(X,Y,Z,A)
#define OSTRACE5(X,Y,Z,A,B) if( sqlite3_os_trace ) sqlite3DebugPrintf(X,Y,Z,A,B)
#define OSTRACE6(X,Y,Z,A,B,C) \
    if(sqlite3_os_trace) sqlite3DebugPrintf(X,Y,Z,A,B,C)
#define OSTRACE7(X,Y,Z,A,B,C,D) \
    if(sqlite3_os_trace) sqlite3DebugPrintf(X,Y,Z,A,B,C,D)
#else
#define OSTRACE1(X)
#define OSTRACE2(X,Y)
#define OSTRACE3(X,Y,Z)
#define OSTRACE4(X,Y,Z,A)
#define OSTRACE5(X,Y,Z,A,B)
#define OSTRACE6(X,Y,Z,A,B,C)
#define OSTRACE7(X,Y,Z,A,B,C,D)
#endif

/*
** Macros for performance tracing.  Normally turned off.  Only works
** on i486 hardware.
*/
#ifdef SQLITE_PERFORMANCE_TRACE
__inline__ unsigned long long int hwtime(void){
  unsigned long long int x;
  __asm__("rdtsc\n\t"
          "mov %%edx, %%ecx\n\t"
          :"=A" (x));
  return x;
}
static unsigned long long int g_start;
static unsigned int elapse;
#define TIMER_START       g_start=hwtime()
#define TIMER_END         elapse=hwtime()-g_start
#define TIMER_ELAPSED     elapse
#else
#define TIMER_START
#define TIMER_END
#define TIMER_ELAPSED     0
#endif

/*
** If we compile with the SQLITE_TEST macro set, then the following block
** of code will give us the ability to simulate a disk I/O error.  This
** is used for testing the I/O recovery logic.
*/
#ifdef SQLITE_TEST
int sqlite3_io_error_hit = 0;
int sqlite3_io_error_pending = 0;
int sqlite3_io_error_persist = 0;
int sqlite3_diskfull_pending = 0;
int sqlite3_diskfull = 0;
#define SimulateIOError(CODE)  \
  if( sqlite3_io_error_pending || sqlite3_io_error_hit ) \
     if( sqlite3_io_error_pending-- == 1 \
         || (sqlite3_io_error_persist && sqlite3_io_error_hit) ) \
                { local_ioerr(); CODE; }
static void local_ioerr(){
  IOTRACE(("IOERR\n"));
  sqlite3_io_error_hit = 1;
}
#define SimulateDiskfullError(CODE) \
   if( sqlite3_diskfull_pending ){ \
     if( sqlite3_diskfull_pending == 1 ){ \
       local_ioerr(); \
       sqlite3_diskfull = 1; \
       sqlite3_io_error_hit = 1; \
       CODE; \
     }else{ \
       sqlite3_diskfull_pending--; \
     } \
   }
#else
#define SimulateIOError(A)
#define SimulateDiskfullError(A)
#endif

/*
** When testing, keep a count of the number of open files.
*/
#ifdef SQLITE_TEST
int sqlite3_open_file_count = 0;
#define OpenCounter(X)  sqlite3_open_file_count+=(X)
#else
#define OpenCounter(X)
#endif

/*
** sqlite3GenericMalloc
** sqlite3GenericRealloc
** sqlite3GenericOsFree
** sqlite3GenericAllocationSize
**
** Implementation of the os level dynamic memory allocation interface in terms
** of the standard malloc(), realloc() and free() found in many operating
** systems. No rocket science here.
**
** There are two versions of these four functions here. The version
** implemented here is only used if memory-management or memory-debugging is
** enabled. This version allocates an extra 8-bytes at the beginning of each
** block and stores the size of the allocation there.
**
** If neither memory-management or debugging is enabled, the second
** set of implementations is used instead.
*/
#if defined(SQLITE_ENABLE_MEMORY_MANAGEMENT) || defined (SQLITE_MEMDEBUG)
SQLITE_PRIVATE void *sqlite3GenericMalloc(int n){
  char *p = (char *)malloc(n+8);
  assert(n>0);
  assert(sizeof(int)<=8);
  if( p ){
    *(int *)p = n;
    p += 8;
  }
  return (void *)p;
}
SQLITE_PRIVATE void *sqlite3GenericRealloc(void *p, int n){
  char *p2 = ((char *)p - 8);
  assert(n>0);
  p2 = (char*)realloc(p2, n+8);
  if( p2 ){
    *(int *)p2 = n;
    p2 += 8;
  }
  return (void *)p2;
}
SQLITE_PRIVATE void sqlite3GenericFree(void *p){
  assert(p);
  free((void *)((char *)p - 8));
}
SQLITE_PRIVATE int sqlite3GenericAllocationSize(void *p){
  return p ? *(int *)((char *)p - 8) : 0;
}
#else
SQLITE_PRIVATE void *sqlite3GenericMalloc(int n){
  char *p = (char *)malloc(n);
  return (void *)p;
}
SQLITE_PRIVATE void *sqlite3GenericRealloc(void *p, int n){
  assert(n>0);
  p = realloc(p, n);
  return p;
}
SQLITE_PRIVATE void sqlite3GenericFree(void *p){
  assert(p);
  free(p);
}
/* Never actually used, but needed for the linker */
SQLITE_PRIVATE int sqlite3GenericAllocationSize(void *p){ return 0; }
#endif

/*
** The default size of a disk sector
*/
#ifndef PAGER_SECTOR_SIZE
# define PAGER_SECTOR_SIZE 512
#endif

/************** End of os_common.h *******************************************/
/************** Continuing where we left off in os_win.c *********************/

/*
** Determine if we are dealing with WindowsCE - which has a much
** reduced API.
*/
#if defined(_WIN32_WCE)
# define OS_WINCE 1
# define AreFileApisANSI() 1
#else
# define OS_WINCE 0
#endif

/*
** WinCE lacks native support for file locking so we have to fake it
** with some code of our own.
*/
#if OS_WINCE
typedef struct winceLock {
  int nReaders;       /* Number of reader locks obtained */
  BOOL bPending;      /* Indicates a pending lock has been obtained */
  BOOL bReserved;     /* Indicates a reserved lock has been obtained */
  BOOL bExclusive;    /* Indicates an exclusive lock has been obtained */
} winceLock;
#endif

/*
** The winFile structure is a subclass of OsFile specific to the win32
** portability layer.
*/
typedef struct winFile winFile;
struct winFile {
  IoMethod const *pMethod;/* Must be first */
  HANDLE h;               /* Handle for accessing the file */
  unsigned char locktype; /* Type of lock currently held on this file */
  short sharedLockByte;   /* Randomly chosen byte used as a shared lock */
#if OS_WINCE
  WCHAR *zDeleteOnClose;  /* Name of file to delete when closing */
  HANDLE hMutex;          /* Mutex used to control access to shared lock */  
  HANDLE hShared;         /* Shared memory segment used for locking */
  winceLock local;        /* Locks obtained by this instance of winFile */
  winceLock *shared;      /* Global shared lock memory for the file  */
#endif
};


/*
** Do not include any of the File I/O interface procedures if the
** SQLITE_OMIT_DISKIO macro is defined (indicating that there database
** will be in-memory only)
*/
#ifndef SQLITE_OMIT_DISKIO

/*
** The following variable is (normally) set once and never changes
** thereafter.  It records whether the operating system is Win95
** or WinNT.
**
** 0:   Operating system unknown.
** 1:   Operating system is Win95.
** 2:   Operating system is WinNT.
**
** In order to facilitate testing on a WinNT system, the test fixture
** can manually set this value to 1 to emulate Win98 behavior.
*/
int sqlite3_os_type = 0;

/*
** Return true (non-zero) if we are running under WinNT, Win2K, WinXP,
** or WinCE.  Return false (zero) for Win95, Win98, or WinME.
**
** Here is an interesting observation:  Win95, Win98, and WinME lack
** the LockFileEx() API.  But we can still statically link against that
** API as long as we don't call it win running Win95/98/ME.  A call to
** this routine is used to determine if the host is Win95/98/ME or
** WinNT/2K/XP so that we will know whether or not we can safely call
** the LockFileEx() API.
*/
#if OS_WINCE
# define isNT()  (1)
#else
  static int isNT(void){
    if( sqlite3_os_type==0 ){
      OSVERSIONINFO sInfo;
      sInfo.dwOSVersionInfoSize = sizeof(sInfo);
      GetVersionEx(&sInfo);
      sqlite3_os_type = sInfo.dwPlatformId==VER_PLATFORM_WIN32_NT ? 2 : 1;
    }
    return sqlite3_os_type==2;
  }
#endif /* OS_WINCE */

/*
** Convert a UTF-8 string to microsoft unicode (UTF-16?). 
**
** Space to hold the returned string is obtained from sqliteMalloc.
*/
static WCHAR *utf8ToUnicode(const char *zFilename){
  int nChar;
  WCHAR *zWideFilename;

  nChar = MultiByteToWideChar(CP_UTF8, 0, zFilename, -1, NULL, 0);
  zWideFilename = sqliteMalloc( nChar*sizeof(zWideFilename[0]) );
  if( zWideFilename==0 ){
    return 0;
  }
  nChar = MultiByteToWideChar(CP_UTF8, 0, zFilename, -1, zWideFilename, nChar);
  if( nChar==0 ){
    sqliteFree(zWideFilename);
    zWideFilename = 0;
  }
  return zWideFilename;
}

/*
** Convert microsoft unicode to UTF-8.  Space to hold the returned string is
** obtained from sqliteMalloc().
*/
static char *unicodeToUtf8(const WCHAR *zWideFilename){
  int nByte;
  char *zFilename;

  nByte = WideCharToMultiByte(CP_UTF8, 0, zWideFilename, -1, 0, 0, 0, 0);
  zFilename = sqliteMalloc( nByte );
  if( zFilename==0 ){
    return 0;
  }
  nByte = WideCharToMultiByte(CP_UTF8, 0, zWideFilename, -1, zFilename, nByte,
                              0, 0);
  if( nByte == 0 ){
    sqliteFree(zFilename);
    zFilename = 0;
  }
  return zFilename;
}

/*
** Convert an ansi string to microsoft unicode, based on the
** current codepage settings for file apis.
** 
** Space to hold the returned string is obtained
** from sqliteMalloc.
*/
static WCHAR *mbcsToUnicode(const char *zFilename){
  int nByte;
  WCHAR *zMbcsFilename;
  int codepage = AreFileApisANSI() ? CP_ACP : CP_OEMCP;

  nByte = MultiByteToWideChar(codepage, 0, zFilename, -1, NULL,0)*sizeof(WCHAR);
  zMbcsFilename = sqliteMalloc( nByte*sizeof(zMbcsFilename[0]) );
  if( zMbcsFilename==0 ){
    return 0;
  }
  nByte = MultiByteToWideChar(codepage, 0, zFilename, -1, zMbcsFilename, nByte);
  if( nByte==0 ){
    sqliteFree(zMbcsFilename);
    zMbcsFilename = 0;
  }
  return zMbcsFilename;
}

/*
** Convert microsoft unicode to multibyte character string, based on the
** user's Ansi codepage.
**
** Space to hold the returned string is obtained from
** sqliteMalloc().
*/
static char *unicodeToMbcs(const WCHAR *zWideFilename){
  int nByte;
  char *zFilename;
  int codepage = AreFileApisANSI() ? CP_ACP : CP_OEMCP;

  nByte = WideCharToMultiByte(codepage, 0, zWideFilename, -1, 0, 0, 0, 0);
  zFilename = sqliteMalloc( nByte );
  if( zFilename==0 ){
    return 0;
  }
  nByte = WideCharToMultiByte(codepage, 0, zWideFilename, -1, zFilename, nByte,
                              0, 0);
  if( nByte == 0 ){
    sqliteFree(zFilename);
    zFilename = 0;
  }
  return zFilename;
}

/*
** Convert multibyte character string to UTF-8.  Space to hold the
** returned string is obtained from sqliteMalloc().
*/
static char *mbcsToUtf8(const char *zFilename){
  char *zFilenameUtf8;
  WCHAR *zTmpWide;

  zTmpWide = mbcsToUnicode(zFilename);
  if( zTmpWide==0 ){
    return 0;
  }
  zFilenameUtf8 = unicodeToUtf8(zTmpWide);
  sqliteFree(zTmpWide);
  return zFilenameUtf8;
}

/*
** Convert UTF-8 to multibyte character string.  Space to hold the 
** returned string is obtained from sqliteMalloc().
*/
static char *utf8ToMbcs(const char *zFilename){
  char *zFilenameMbcs;
  WCHAR *zTmpWide;

  zTmpWide = utf8ToUnicode(zFilename);
  if( zTmpWide==0 ){
    return 0;
  }
  zFilenameMbcs = unicodeToMbcs(zTmpWide);
  sqliteFree(zTmpWide);
  return zFilenameMbcs;
}

#if OS_WINCE
/*************************************************************************
** This section contains code for WinCE only.
*/
/*
** WindowsCE does not have a localtime() function.  So create a
** substitute.
*/
struct tm *__cdecl localtime(const time_t *t)
{
  static struct tm y;
  FILETIME uTm, lTm;
  SYSTEMTIME pTm;
  i64 t64;
  t64 = *t;
  t64 = (t64 + 11644473600)*10000000;
  uTm.dwLowDateTime = t64 & 0xFFFFFFFF;
  uTm.dwHighDateTime= t64 >> 32;
  FileTimeToLocalFileTime(&uTm,&lTm);
  FileTimeToSystemTime(&lTm,&pTm);
  y.tm_year = pTm.wYear - 1900;
  y.tm_mon = pTm.wMonth - 1;
  y.tm_wday = pTm.wDayOfWeek;
  y.tm_mday = pTm.wDay;
  y.tm_hour = pTm.wHour;
  y.tm_min = pTm.wMinute;
  y.tm_sec = pTm.wSecond;
  return &y;
}

/* This will never be called, but defined to make the code compile */
#define GetTempPathA(a,b)

#define LockFile(a,b,c,d,e)       winceLockFile(&a, b, c, d, e)
#define UnlockFile(a,b,c,d,e)     winceUnlockFile(&a, b, c, d, e)
#define LockFileEx(a,b,c,d,e,f)   winceLockFileEx(&a, b, c, d, e, f)

#define HANDLE_TO_WINFILE(a) (winFile*)&((char*)a)[-offsetof(winFile,h)]

/*
** Acquire a lock on the handle h
*/
static void winceMutexAcquire(HANDLE h){
   DWORD dwErr;
   do {
     dwErr = WaitForSingleObject(h, INFINITE);
   } while (dwErr != WAIT_OBJECT_0 && dwErr != WAIT_ABANDONED);
}
/*
** Release a lock acquired by winceMutexAcquire()
*/
#define winceMutexRelease(h) ReleaseMutex(h)

/*
** Create the mutex and shared memory used for locking in the file
** descriptor pFile
*/
static BOOL winceCreateLock(const char *zFilename, winFile *pFile){
  WCHAR *zTok;
  WCHAR *zName = utf8ToUnicode(zFilename);
  BOOL bInit = TRUE;

  /* Initialize the local lockdata */
  ZeroMemory(&pFile->local, sizeof(pFile->local));

  /* Replace the backslashes from the filename and lowercase it
  ** to derive a mutex name. */
  zTok = CharLowerW(zName);
  for (;*zTok;zTok++){
    if (*zTok == '\\') *zTok = '_';
  }

  /* Create/open the named mutex */
  pFile->hMutex = CreateMutexW(NULL, FALSE, zName);
  if (!pFile->hMutex){
    sqliteFree(zName);
    return FALSE;
  }

  /* Acquire the mutex before continuing */
  winceMutexAcquire(pFile->hMutex);
  
  /* Since the names of named mutexes, semaphores, file mappings etc are 
  ** case-sensitive, take advantage of that by uppercasing the mutex name
  ** and using that as the shared filemapping name.
  */
  CharUpperW(zName);
  pFile->hShared = CreateFileMappingW(INVALID_HANDLE_VALUE, NULL,
                                       PAGE_READWRITE, 0, sizeof(winceLock),
                                       zName);  

  /* Set a flag that indicates we're the first to create the memory so it 
  ** must be zero-initialized */
  if (GetLastError() == ERROR_ALREADY_EXISTS){
    bInit = FALSE;
  }

  sqliteFree(zName);

  /* If we succeeded in making the shared memory handle, map it. */
  if (pFile->hShared){
    pFile->shared = (winceLock*)MapViewOfFile(pFile->hShared, 
             FILE_MAP_READ|FILE_MAP_WRITE, 0, 0, sizeof(winceLock));
    /* If mapping failed, close the shared memory handle and erase it */
    if (!pFile->shared){
      CloseHandle(pFile->hShared);
      pFile->hShared = NULL;
    }
  }

  /* If shared memory could not be created, then close the mutex and fail */
  if (pFile->hShared == NULL){
    winceMutexRelease(pFile->hMutex);
    CloseHandle(pFile->hMutex);
    pFile->hMutex = NULL;
    return FALSE;
  }
  
  /* Initialize the shared memory if we're supposed to */
  if (bInit) {
    ZeroMemory(pFile->shared, sizeof(winceLock));
  }

  winceMutexRelease(pFile->hMutex);
  return TRUE;
}

/*
** Destroy the part of winFile that deals with wince locks
*/
static void winceDestroyLock(winFile *pFile){
  if (pFile->hMutex){
    /* Acquire the mutex */
    winceMutexAcquire(pFile->hMutex);

    /* The following blocks should probably assert in debug mode, but they
       are to cleanup in case any locks remained open */
    if (pFile->local.nReaders){
      pFile->shared->nReaders --;
    }
    if (pFile->local.bReserved){
      pFile->shared->bReserved = FALSE;
    }
    if (pFile->local.bPending){
      pFile->shared->bPending = FALSE;
    }
    if (pFile->local.bExclusive){
      pFile->shared->bExclusive = FALSE;
    }

    /* De-reference and close our copy of the shared memory handle */
    UnmapViewOfFile(pFile->shared);
    CloseHandle(pFile->hShared);

    if( pFile->zDeleteOnClose ){
      DeleteFileW(pFile->zDeleteOnClose);
      sqliteFree(pFile->zDeleteOnClose);
      pFile->zDeleteOnClose = 0;
    }

    /* Done with the mutex */
    winceMutexRelease(pFile->hMutex);    
    CloseHandle(pFile->hMutex);
    pFile->hMutex = NULL;
  }
}

/* 
** An implementation of the LockFile() API of windows for wince
*/
static BOOL winceLockFile(
  HANDLE *phFile,
  DWORD dwFileOffsetLow,
  DWORD dwFileOffsetHigh,
  DWORD nNumberOfBytesToLockLow,
  DWORD nNumberOfBytesToLockHigh
){
  winFile *pFile = HANDLE_TO_WINFILE(phFile);
  BOOL bReturn = FALSE;

  if (!pFile->hMutex) return TRUE;
  winceMutexAcquire(pFile->hMutex);

  /* Wanting an exclusive lock? */
  if (dwFileOffsetLow == SHARED_FIRST
       && nNumberOfBytesToLockLow == SHARED_SIZE){
    if (pFile->shared->nReaders == 0 && pFile->shared->bExclusive == 0){
       pFile->shared->bExclusive = TRUE;
       pFile->local.bExclusive = TRUE;
       bReturn = TRUE;
    }
  }

  /* Want a read-only lock? */
  else if ((dwFileOffsetLow >= SHARED_FIRST &&
            dwFileOffsetLow < SHARED_FIRST + SHARED_SIZE) &&
            nNumberOfBytesToLockLow == 1){
    if (pFile->shared->bExclusive == 0){
      pFile->local.nReaders ++;
      if (pFile->local.nReaders == 1){
        pFile->shared->nReaders ++;
      }
      bReturn = TRUE;
    }
  }

  /* Want a pending lock? */
  else if (dwFileOffsetLow == PENDING_BYTE && nNumberOfBytesToLockLow == 1){
    /* If no pending lock has been acquired, then acquire it */
    if (pFile->shared->bPending == 0) {
      pFile->shared->bPending = TRUE;
      pFile->local.bPending = TRUE;
      bReturn = TRUE;
    }
  }
  /* Want a reserved lock? */
  else if (dwFileOffsetLow == RESERVED_BYTE && nNumberOfBytesToLockLow == 1){
    if (pFile->shared->bReserved == 0) {
      pFile->shared->bReserved = TRUE;
      pFile->local.bReserved = TRUE;
      bReturn = TRUE;
    }
  }

  winceMutexRelease(pFile->hMutex);
  return bReturn;
}

/*
** An implementation of the UnlockFile API of windows for wince
*/
static BOOL winceUnlockFile(
  HANDLE *phFile,
  DWORD dwFileOffsetLow,
  DWORD dwFileOffsetHigh,
  DWORD nNumberOfBytesToUnlockLow,
  DWORD nNumberOfBytesToUnlockHigh
){
  winFile *pFile = HANDLE_TO_WINFILE(phFile);
  BOOL bReturn = FALSE;

  if (!pFile->hMutex) return TRUE;
  winceMutexAcquire(pFile->hMutex);

  /* Releasing a reader lock or an exclusive lock */
  if (dwFileOffsetLow >= SHARED_FIRST &&
       dwFileOffsetLow < SHARED_FIRST + SHARED_SIZE){
    /* Did we have an exclusive lock? */
    if (pFile->local.bExclusive){
      pFile->local.bExclusive = FALSE;
      pFile->shared->bExclusive = FALSE;
      bReturn = TRUE;
    }

    /* Did we just have a reader lock? */
    else if (pFile->local.nReaders){
      pFile->local.nReaders --;
      if (pFile->local.nReaders == 0)
      {
        pFile->shared->nReaders --;
      }
      bReturn = TRUE;
    }
  }

  /* Releasing a pending lock */
  else if (dwFileOffsetLow == PENDING_BYTE && nNumberOfBytesToUnlockLow == 1){
    if (pFile->local.bPending){
      pFile->local.bPending = FALSE;
      pFile->shared->bPending = FALSE;
      bReturn = TRUE;
    }
  }
  /* Releasing a reserved lock */
  else if (dwFileOffsetLow == RESERVED_BYTE && nNumberOfBytesToUnlockLow == 1){
    if (pFile->local.bReserved) {
      pFile->local.bReserved = FALSE;
      pFile->shared->bReserved = FALSE;
      bReturn = TRUE;
    }
  }

  winceMutexRelease(pFile->hMutex);
  return bReturn;
}

/*
** An implementation of the LockFileEx() API of windows for wince
*/
static BOOL winceLockFileEx(
  HANDLE *phFile,
  DWORD dwFlags,
  DWORD dwReserved,
  DWORD nNumberOfBytesToLockLow,
  DWORD nNumberOfBytesToLockHigh,
  LPOVERLAPPED lpOverlapped
){
  /* If the caller wants a shared read lock, forward this call
  ** to winceLockFile */
  if (lpOverlapped->Offset == SHARED_FIRST &&
      dwFlags == 1 &&
      nNumberOfBytesToLockLow == SHARED_SIZE){
    return winceLockFile(phFile, SHARED_FIRST, 0, 1, 0);
  }
  return FALSE;
}
/*
** End of the special code for wince
*****************************************************************************/
#endif /* OS_WINCE */

/*
** Convert a UTF-8 filename into whatever form the underlying
** operating system wants filenames in.  Space to hold the result
** is obtained from sqliteMalloc and must be freed by the calling
** function.
*/
static void *convertUtf8Filename(const char *zFilename){
  void *zConverted = 0;
  if( isNT() ){
    zConverted = utf8ToUnicode(zFilename);
  }else{
    zConverted = utf8ToMbcs(zFilename);
  }
  /* caller will handle out of memory */
  return zConverted;
}

/*
** Delete the named file.
**
** Note that windows does not allow a file to be deleted if some other
** process has it open.  Sometimes a virus scanner or indexing program
** will open a journal file shortly after it is created in order to do
** whatever it is it does.  While this other process is holding the
** file open, we will be unable to delete it.  To work around this
** problem, we delay 100 milliseconds and try to delete again.  Up
** to MX_DELETION_ATTEMPTs deletion attempts are run before giving
** up and returning an error.
*/
#define MX_DELETION_ATTEMPTS 3
SQLITE_PRIVATE int sqlite3WinDelete(const char *zFilename){
  int cnt = 0;
  int rc;
  void *zConverted = convertUtf8Filename(zFilename);
  if( zConverted==0 ){
    return SQLITE_NOMEM;
  }
  SimulateIOError(return SQLITE_IOERR_DELETE);
  if( isNT() ){
    do{
      rc = DeleteFileW(zConverted);
    }while( rc==0 && GetFileAttributesW(zConverted)!=0xffffffff 
            && cnt++ < MX_DELETION_ATTEMPTS && (Sleep(100), 1) );
  }else{
#if OS_WINCE
    return SQLITE_NOMEM;
#else
    do{
      rc = DeleteFileA(zConverted);
    }while( rc==0 && GetFileAttributesA(zConverted)!=0xffffffff
            && cnt++ < MX_DELETION_ATTEMPTS && (Sleep(100), 1) );
#endif
  }
  sqliteFree(zConverted);
  OSTRACE2("DELETE \"%s\"\n", zFilename);
  return rc!=0 ? SQLITE_OK : SQLITE_IOERR;
}

/*
** Return TRUE if the named file exists.
*/
SQLITE_PRIVATE int sqlite3WinFileExists(const char *zFilename){
  int exists = 0;
  void *zConverted = convertUtf8Filename(zFilename);
  if( zConverted==0 ){
    return SQLITE_NOMEM;
  }
  if( isNT() ){
    exists = GetFileAttributesW((WCHAR*)zConverted) != 0xffffffff;
  }else{
#if OS_WINCE
    return SQLITE_NOMEM;
#else
    exists = GetFileAttributesA((char*)zConverted) != 0xffffffff;
#endif
  }
  sqliteFree(zConverted);
  return exists;
}

/* Forward declaration */
static int allocateWinFile(winFile *pInit, OsFile **pId);

/*
** Attempt to open a file for both reading and writing.  If that
** fails, try opening it read-only.  If the file does not exist,
** try to create it.
**
** On success, a handle for the open file is written to *id
** and *pReadonly is set to 0 if the file was opened for reading and
** writing or 1 if the file was opened read-only.  The function returns
** SQLITE_OK.
**
** On failure, the function returns SQLITE_CANTOPEN and leaves
** *id and *pReadonly unchanged.
*/
SQLITE_PRIVATE int sqlite3WinOpenReadWrite(
  const char *zFilename,
  OsFile **pId,
  int *pReadonly
){
  winFile f;
  HANDLE h;
  void *zConverted = convertUtf8Filename(zFilename);
  if( zConverted==0 ){
    return SQLITE_NOMEM;
  }
  assert( *pId==0 );

  if( isNT() ){
    h = CreateFileW((WCHAR*)zConverted,
       GENERIC_READ | GENERIC_WRITE,
       FILE_SHARE_READ | FILE_SHARE_WRITE,
       NULL,
       OPEN_ALWAYS,
       FILE_ATTRIBUTE_NORMAL | FILE_FLAG_RANDOM_ACCESS,
       NULL
    );
    if( h==INVALID_HANDLE_VALUE ){
      h = CreateFileW((WCHAR*)zConverted,
         GENERIC_READ,
         FILE_SHARE_READ | FILE_SHARE_WRITE,
         NULL,
         OPEN_ALWAYS,
         FILE_ATTRIBUTE_NORMAL | FILE_FLAG_RANDOM_ACCESS,
         NULL
      );
      if( h==INVALID_HANDLE_VALUE ){
        sqliteFree(zConverted);
        return SQLITE_CANTOPEN;
      }
      *pReadonly = 1;
    }else{
      *pReadonly = 0;
    }
#if OS_WINCE
    if (!winceCreateLock(zFilename, &f)){
      CloseHandle(h);
      sqliteFree(zConverted);
      return SQLITE_CANTOPEN;
    }
#endif
  }else{
#if OS_WINCE
    return SQLITE_NOMEM;
#else
    h = CreateFileA((char*)zConverted,
       GENERIC_READ | GENERIC_WRITE,
       FILE_SHARE_READ | FILE_SHARE_WRITE,
       NULL,
       OPEN_ALWAYS,
       FILE_ATTRIBUTE_NORMAL | FILE_FLAG_RANDOM_ACCESS,
       NULL
    );
    if( h==INVALID_HANDLE_VALUE ){
      h = CreateFileA((char*)zConverted,
         GENERIC_READ,
         FILE_SHARE_READ | FILE_SHARE_WRITE,
         NULL,
         OPEN_ALWAYS,
         FILE_ATTRIBUTE_NORMAL | FILE_FLAG_RANDOM_ACCESS,
         NULL
      );
      if( h==INVALID_HANDLE_VALUE ){
        sqliteFree(zConverted);
        return SQLITE_CANTOPEN;
      }
      *pReadonly = 1;
    }else{
      *pReadonly = 0;
    }
#endif /* OS_WINCE */
  }

  sqliteFree(zConverted);

  f.h = h;
#if OS_WINCE
  f.zDeleteOnClose = 0;
#endif
  OSTRACE3("OPEN R/W %d \"%s\"\n", h, zFilename);
  return allocateWinFile(&f, pId);
}


/*
** Attempt to open a new file for exclusive access by this process.
** The file will be opened for both reading and writing.  To avoid
** a potential security problem, we do not allow the file to have
** previously existed.  Nor do we allow the file to be a symbolic
** link.
**
** If delFlag is true, then make arrangements to automatically delete
** the file when it is closed.
**
** On success, write the file handle into *id and return SQLITE_OK.
**
** On failure, return SQLITE_CANTOPEN.
**
** Sometimes if we have just deleted a prior journal file, windows
** will fail to open a new one because there is a "pending delete".
** To work around this bug, we pause for 100 milliseconds and attempt
** a second open after the first one fails.  The whole operation only
** fails if both open attempts are unsuccessful.
*/
SQLITE_PRIVATE int sqlite3WinOpenExclusive(const char *zFilename, OsFile **pId, int delFlag){
  winFile f;
  HANDLE h;
  DWORD fileflags;
  void *zConverted = convertUtf8Filename(zFilename);
  if( zConverted==0 ){
    return SQLITE_NOMEM;
  }
  assert( *pId == 0 );
  fileflags = FILE_FLAG_RANDOM_ACCESS;
#if !OS_WINCE
  if( delFlag ){
    fileflags |= FILE_ATTRIBUTE_TEMPORARY | FILE_FLAG_DELETE_ON_CLOSE;
  }
#endif
  if( isNT() ){
    int cnt = 0;
    do{
      h = CreateFileW((WCHAR*)zConverted,
         GENERIC_READ | GENERIC_WRITE,
         0,
         NULL,
         CREATE_ALWAYS,
         fileflags,
         NULL
      );
    }while( h==INVALID_HANDLE_VALUE && cnt++ < 2 && (Sleep(100), 1) );
  }else{
#if OS_WINCE
    return SQLITE_NOMEM;
#else
    int cnt = 0;
    do{
      h = CreateFileA((char*)zConverted,
        GENERIC_READ | GENERIC_WRITE,
        0,
        NULL,
        CREATE_ALWAYS,
        fileflags,
        NULL
      );
    }while( h==INVALID_HANDLE_VALUE && cnt++ < 2 && (Sleep(100), 1) );
#endif /* OS_WINCE */
  }
#if OS_WINCE
  if( delFlag && h!=INVALID_HANDLE_VALUE ){
    f.zDeleteOnClose = zConverted;
    zConverted = 0;
  }
  f.hMutex = NULL;
#endif
  sqliteFree(zConverted);
  if( h==INVALID_HANDLE_VALUE ){
    return SQLITE_CANTOPEN;
  }
  f.h = h;
  OSTRACE3("OPEN EX %d \"%s\"\n", h, zFilename);
  return allocateWinFile(&f, pId);
}

/*
** Attempt to open a new file for read-only access.
**
** On success, write the file handle into *id and return SQLITE_OK.
**
** On failure, return SQLITE_CANTOPEN.
*/
SQLITE_PRIVATE int sqlite3WinOpenReadOnly(const char *zFilename, OsFile **pId){
  winFile f;
  HANDLE h;
  void *zConverted = convertUtf8Filename(zFilename);
  if( zConverted==0 ){
    return SQLITE_NOMEM;
  }
  assert( *pId==0 );
  if( isNT() ){
    h = CreateFileW((WCHAR*)zConverted,
       GENERIC_READ,
       0,
       NULL,
       OPEN_EXISTING,
       FILE_ATTRIBUTE_NORMAL | FILE_FLAG_RANDOM_ACCESS,
       NULL
    );
  }else{
#if OS_WINCE
    return SQLITE_NOMEM;
#else
    h = CreateFileA((char*)zConverted,
       GENERIC_READ,
       0,
       NULL,
       OPEN_EXISTING,
       FILE_ATTRIBUTE_NORMAL | FILE_FLAG_RANDOM_ACCESS,
       NULL
    );
#endif
  }
  sqliteFree(zConverted);
  if( h==INVALID_HANDLE_VALUE ){
    return SQLITE_CANTOPEN;
  }
  f.h = h;
#if OS_WINCE
  f.zDeleteOnClose = 0;
  f.hMutex = NULL;
#endif
  OSTRACE3("OPEN RO %d \"%s\"\n", h, zFilename);
  return allocateWinFile(&f, pId);
}

/*
** Attempt to open a file descriptor for the directory that contains a
** file.  This file descriptor can be used to fsync() the directory
** in order to make sure the creation of a new file is actually written
** to disk.
**
** This routine is only meaningful for Unix.  It is a no-op under
** windows since windows does not support hard links.
**
** On success, a handle for a previously open file is at *id is
** updated with the new directory file descriptor and SQLITE_OK is
** returned.
**
** On failure, the function returns SQLITE_CANTOPEN and leaves
** *id unchanged.
*/
static int winOpenDirectory(
  OsFile *id,
  const char *zDirname
){
  return SQLITE_OK;
}

/*
** Create a temporary file name in zBuf.  zBuf must be big enough to
** hold at least SQLITE_TEMPNAME_SIZE characters.
*/
SQLITE_PRIVATE int sqlite3WinTempFileName(char *zBuf){
  static char zChars[] =
    "abcdefghijklmnopqrstuvwxyz"
    "ABCDEFGHIJKLMNOPQRSTUVWXYZ"
    "0123456789";
  int i, j;
  char zTempPath[SQLITE_TEMPNAME_SIZE];
  if( sqlite3_temp_directory ){
    strncpy(zTempPath, sqlite3_temp_directory, SQLITE_TEMPNAME_SIZE-30);
    zTempPath[SQLITE_TEMPNAME_SIZE-30] = 0;
  }else if( isNT() ){
    char *zMulti;
    WCHAR zWidePath[SQLITE_TEMPNAME_SIZE];
    GetTempPathW(SQLITE_TEMPNAME_SIZE-30, zWidePath);
    zMulti = unicodeToUtf8(zWidePath);
    if( zMulti ){
      strncpy(zTempPath, zMulti, SQLITE_TEMPNAME_SIZE-30);
      zTempPath[SQLITE_TEMPNAME_SIZE-30] = 0;
      sqliteFree(zMulti);
    }else{
      return SQLITE_NOMEM;
    }
  }else{
    char *zUtf8;
    char zMbcsPath[SQLITE_TEMPNAME_SIZE];
    GetTempPathA(SQLITE_TEMPNAME_SIZE-30, zMbcsPath);
    zUtf8 = mbcsToUtf8(zMbcsPath);
    if( zUtf8 ){
      strncpy(zTempPath, zUtf8, SQLITE_TEMPNAME_SIZE-30);
      zTempPath[SQLITE_TEMPNAME_SIZE-30] = 0;
      sqliteFree(zUtf8);
    }else{
      return SQLITE_NOMEM;
    }
  }
  for(i=strlen(zTempPath); i>0 && zTempPath[i-1]=='\\'; i--){}
  zTempPath[i] = 0;
  for(;;){
    sqlite3_snprintf(SQLITE_TEMPNAME_SIZE, zBuf,
                     "%s\\"TEMP_FILE_PREFIX, zTempPath);
    j = strlen(zBuf);
    sqlite3Randomness(15, &zBuf[j]);
    for(i=0; i<15; i++, j++){
      zBuf[j] = (char)zChars[ ((unsigned char)zBuf[j])%(sizeof(zChars)-1) ];
    }
    zBuf[j] = 0;
    if( !sqlite3OsFileExists(zBuf) ) break;
  }
  OSTRACE2("TEMP FILENAME: %s\n", zBuf);
  return SQLITE_OK; 
}

/*
** Close a file.
**
** It is reported that an attempt to close a handle might sometimes
** fail.  This is a very unreasonable result, but windows is notorious
** for being unreasonable so I do not doubt that it might happen.  If
** the close fails, we pause for 100 milliseconds and try again.  As
** many as MX_CLOSE_ATTEMPT attempts to close the handle are made before
** giving up and returning an error.
*/
#define MX_CLOSE_ATTEMPT 3
static int winClose(OsFile **pId){
  winFile *pFile;
  int rc = 1;
  if( pId && (pFile = (winFile*)*pId)!=0 ){
    int rc, cnt = 0;
    OSTRACE2("CLOSE %d\n", pFile->h);
    do{
      rc = CloseHandle(pFile->h);
    }while( rc==0 && cnt++ < MX_CLOSE_ATTEMPT && (Sleep(100), 1) );
#if OS_WINCE
    winceDestroyLock(pFile);
#endif
    OpenCounter(-1);
    sqliteFree(pFile);
    *pId = 0;
  }
  return rc ? SQLITE_OK : SQLITE_IOERR;
}

/*
** Read data from a file into a buffer.  Return SQLITE_OK if all
** bytes were read successfully and SQLITE_IOERR if anything goes
** wrong.
*/
static int winRead(OsFile *id, void *pBuf, int amt){
  DWORD got;
  assert( id!=0 );
  SimulateIOError(return SQLITE_IOERR_READ);
  OSTRACE3("READ %d lock=%d\n", ((winFile*)id)->h, ((winFile*)id)->locktype);
  if( !ReadFile(((winFile*)id)->h, pBuf, amt, &got, 0) ){
    return SQLITE_IOERR_READ;
  }
  if( got==(DWORD)amt ){
    return SQLITE_OK;
  }else{
    memset(&((char*)pBuf)[got], 0, amt-got);
    return SQLITE_IOERR_SHORT_READ;
  }
}

/*
** Write data from a buffer into a file.  Return SQLITE_OK on success
** or some other error code on failure.
*/
static int winWrite(OsFile *id, const void *pBuf, int amt){
  int rc = 0;
  DWORD wrote;
  assert( id!=0 );
  SimulateIOError(return SQLITE_IOERR_READ);
  SimulateDiskfullError(return SQLITE_FULL);
  OSTRACE3("WRITE %d lock=%d\n", ((winFile*)id)->h, ((winFile*)id)->locktype);
  assert( amt>0 );
  while( amt>0 && (rc = WriteFile(((winFile*)id)->h, pBuf, amt, &wrote, 0))!=0
         && wrote>0 ){
    amt -= wrote;
    pBuf = &((char*)pBuf)[wrote];
  }
  if( !rc || amt>(int)wrote ){
    return SQLITE_FULL;
  }
  return SQLITE_OK;
}

/*
** Some microsoft compilers lack this definition.
*/
#ifndef INVALID_SET_FILE_POINTER
# define INVALID_SET_FILE_POINTER ((DWORD)-1)
#endif

/*
** Move the read/write pointer in a file.
*/
static int winSeek(OsFile *id, i64 offset){
  LONG upperBits = offset>>32;
  LONG lowerBits = offset & 0xffffffff;
  DWORD rc;
  assert( id!=0 );
#ifdef SQLITE_TEST
  if( offset ) SimulateDiskfullError(return SQLITE_FULL);
#endif
  rc = SetFilePointer(((winFile*)id)->h, lowerBits, &upperBits, FILE_BEGIN);
  OSTRACE3("SEEK %d %lld\n", ((winFile*)id)->h, offset);
  if( rc==INVALID_SET_FILE_POINTER && GetLastError()!=NO_ERROR ){
    return SQLITE_FULL;
  }
  return SQLITE_OK;
}

/*
** Make sure all writes to a particular file are committed to disk.
*/
static int winSync(OsFile *id, int dataOnly){
  assert( id!=0 );
  OSTRACE3("SYNC %d lock=%d\n", ((winFile*)id)->h, ((winFile*)id)->locktype);
  if( FlushFileBuffers(((winFile*)id)->h) ){
    return SQLITE_OK;
  }else{
    return SQLITE_IOERR;
  }
}

/*
** Sync the directory zDirname. This is a no-op on operating systems other
** than UNIX.
*/
SQLITE_PRIVATE int sqlite3WinSyncDirectory(const char *zDirname){
  SimulateIOError(return SQLITE_IOERR_READ);
  return SQLITE_OK;
}

/*
** Truncate an open file to a specified size
*/
static int winTruncate(OsFile *id, i64 nByte){
  LONG upperBits = nByte>>32;
  assert( id!=0 );
  OSTRACE3("TRUNCATE %d %lld\n", ((winFile*)id)->h, nByte);
  SimulateIOError(return SQLITE_IOERR_TRUNCATE);
  SetFilePointer(((winFile*)id)->h, nByte, &upperBits, FILE_BEGIN);
  SetEndOfFile(((winFile*)id)->h);
  return SQLITE_OK;
}

/*
** Determine the current size of a file in bytes
*/
static int winFileSize(OsFile *id, i64 *pSize){
  DWORD upperBits, lowerBits;
  assert( id!=0 );
  SimulateIOError(return SQLITE_IOERR_FSTAT);
  lowerBits = GetFileSize(((winFile*)id)->h, &upperBits);
  *pSize = (((i64)upperBits)<<32) + lowerBits;
  return SQLITE_OK;
}

/*
** LOCKFILE_FAIL_IMMEDIATELY is undefined on some Windows systems.
*/
#ifndef LOCKFILE_FAIL_IMMEDIATELY
# define LOCKFILE_FAIL_IMMEDIATELY 1
#endif

/*
** Acquire a reader lock.
** Different API routines are called depending on whether or not this
** is Win95 or WinNT.
*/
static int getReadLock(winFile *id){
  int res;
  if( isNT() ){
    OVERLAPPED ovlp;
    ovlp.Offset = SHARED_FIRST;
    ovlp.OffsetHigh = 0;
    ovlp.hEvent = 0;
    res = LockFileEx(id->h, LOCKFILE_FAIL_IMMEDIATELY, 0, SHARED_SIZE,0,&ovlp);
  }else{
    int lk;
    sqlite3Randomness(sizeof(lk), &lk);
    id->sharedLockByte = (lk & 0x7fffffff)%(SHARED_SIZE - 1);
    res = LockFile(id->h, SHARED_FIRST+id->sharedLockByte, 0, 1, 0);
  }
  return res;
}

/*
** Undo a readlock
*/
static int unlockReadLock(winFile *pFile){
  int res;
  if( isNT() ){
    res = UnlockFile(pFile->h, SHARED_FIRST, 0, SHARED_SIZE, 0);
  }else{
    res = UnlockFile(pFile->h, SHARED_FIRST + pFile->sharedLockByte, 0, 1, 0);
  }
  return res;
}

#ifndef SQLITE_OMIT_PAGER_PRAGMAS
/*
** Check that a given pathname is a directory and is writable 
**
*/
SQLITE_PRIVATE int sqlite3WinIsDirWritable(char *zDirname){
  int fileAttr;
  void *zConverted;
  if( zDirname==0 ) return 0;
  if( !isNT() && strlen(zDirname)>MAX_PATH ) return 0;

  zConverted = convertUtf8Filename(zDirname);
  if( zConverted==0 ){
    return SQLITE_NOMEM;
  }
  if( isNT() ){
    fileAttr = GetFileAttributesW((WCHAR*)zConverted);
  }else{
#if OS_WINCE
    return 0;
#else
    fileAttr = GetFileAttributesA((char*)zConverted);
#endif
  }
  sqliteFree(zConverted);
  if( fileAttr == 0xffffffff ) return 0;
  if( (fileAttr & FILE_ATTRIBUTE_DIRECTORY) != FILE_ATTRIBUTE_DIRECTORY ){
    return 0;
  }
  return 1;
}
#endif /* SQLITE_OMIT_PAGER_PRAGMAS */

/*
** Lock the file with the lock specified by parameter locktype - one
** of the following:
**
**     (1) SHARED_LOCK
**     (2) RESERVED_LOCK
**     (3) PENDING_LOCK
**     (4) EXCLUSIVE_LOCK
**
** Sometimes when requesting one lock state, additional lock states
** are inserted in between.  The locking might fail on one of the later
** transitions leaving the lock state different from what it started but
** still short of its goal.  The following chart shows the allowed
** transitions and the inserted intermediate states:
**
**    UNLOCKED -> SHARED
**    SHARED -> RESERVED
**    SHARED -> (PENDING) -> EXCLUSIVE
**    RESERVED -> (PENDING) -> EXCLUSIVE
**    PENDING -> EXCLUSIVE
**
** This routine will only increase a lock.  The winUnlock() routine
** erases all locks at once and returns us immediately to locking level 0.
** It is not possible to lower the locking level one step at a time.  You
** must go straight to locking level 0.
*/
static int winLock(OsFile *id, int locktype){
  int rc = SQLITE_OK;    /* Return code from subroutines */
  int res = 1;           /* Result of a windows lock call */
  int newLocktype;       /* Set id->locktype to this value before exiting */
  int gotPendingLock = 0;/* True if we acquired a PENDING lock this time */
  winFile *pFile = (winFile*)id;

  assert( pFile!=0 );
  OSTRACE5("LOCK %d %d was %d(%d)\n",
          pFile->h, locktype, pFile->locktype, pFile->sharedLockByte);

  /* If there is already a lock of this type or more restrictive on the
  ** OsFile, do nothing. Don't use the end_lock: exit path, as
  ** sqlite3OsEnterMutex() hasn't been called yet.
  */
  if( pFile->locktype>=locktype ){
    return SQLITE_OK;
  }

  /* Make sure the locking sequence is correct
  */
  assert( pFile->locktype!=NO_LOCK || locktype==SHARED_LOCK );
  assert( locktype!=PENDING_LOCK );
  assert( locktype!=RESERVED_LOCK || pFile->locktype==SHARED_LOCK );

  /* Lock the PENDING_LOCK byte if we need to acquire a PENDING lock or
  ** a SHARED lock.  If we are acquiring a SHARED lock, the acquisition of
  ** the PENDING_LOCK byte is temporary.
  */
  newLocktype = pFile->locktype;
  if( pFile->locktype==NO_LOCK
   || (locktype==EXCLUSIVE_LOCK && pFile->locktype==RESERVED_LOCK)
  ){
    int cnt = 3;
    while( cnt-->0 && (res = LockFile(pFile->h, PENDING_BYTE, 0, 1, 0))==0 ){
      /* Try 3 times to get the pending lock.  The pending lock might be
      ** held by another reader process who will release it momentarily.
      */
      OSTRACE2("could not get a PENDING lock. cnt=%d\n", cnt);
      Sleep(1);
    }
    gotPendingLock = res;
  }

  /* Acquire a shared lock
  */
  if( locktype==SHARED_LOCK && res ){
    assert( pFile->locktype==NO_LOCK );
    res = getReadLock(pFile);
    if( res ){
      newLocktype = SHARED_LOCK;
    }
  }

  /* Acquire a RESERVED lock
  */
  if( locktype==RESERVED_LOCK && res ){
    assert( pFile->locktype==SHARED_LOCK );
    res = LockFile(pFile->h, RESERVED_BYTE, 0, 1, 0);
    if( res ){
      newLocktype = RESERVED_LOCK;
    }
  }

  /* Acquire a PENDING lock
  */
  if( locktype==EXCLUSIVE_LOCK && res ){
    newLocktype = PENDING_LOCK;
    gotPendingLock = 0;
  }

  /* Acquire an EXCLUSIVE lock
  */
  if( locktype==EXCLUSIVE_LOCK && res ){
    assert( pFile->locktype>=SHARED_LOCK );
    res = unlockReadLock(pFile);
    OSTRACE2("unreadlock = %d\n", res);
    res = LockFile(pFile->h, SHARED_FIRST, 0, SHARED_SIZE, 0);
    if( res ){
      newLocktype = EXCLUSIVE_LOCK;
    }else{
      OSTRACE2("error-code = %d\n", GetLastError());
      getReadLock(pFile);
    }
  }

  /* If we are holding a PENDING lock that ought to be released, then
  ** release it now.
  */
  if( gotPendingLock && locktype==SHARED_LOCK ){
    UnlockFile(pFile->h, PENDING_BYTE, 0, 1, 0);
  }

  /* Update the state of the lock has held in the file descriptor then
  ** return the appropriate result code.
  */
  if( res ){
    rc = SQLITE_OK;
  }else{
    OSTRACE4("LOCK FAILED %d trying for %d but got %d\n", pFile->h,
           locktype, newLocktype);
    rc = SQLITE_BUSY;
  }
  pFile->locktype = newLocktype;
  return rc;
}

/*
** This routine checks if there is a RESERVED lock held on the specified
** file by this or any other process. If such a lock is held, return
** non-zero, otherwise zero.
*/
static int winCheckReservedLock(OsFile *id){
  int rc;
  winFile *pFile = (winFile*)id;
  assert( pFile!=0 );
  if( pFile->locktype>=RESERVED_LOCK ){
    rc = 1;
    OSTRACE3("TEST WR-LOCK %d %d (local)\n", pFile->h, rc);
  }else{
    rc = LockFile(pFile->h, RESERVED_BYTE, 0, 1, 0);
    if( rc ){
      UnlockFile(pFile->h, RESERVED_BYTE, 0, 1, 0);
    }
    rc = !rc;
    OSTRACE3("TEST WR-LOCK %d %d (remote)\n", pFile->h, rc);
  }
  return rc;
}

/*
** Lower the locking level on file descriptor id to locktype.  locktype
** must be either NO_LOCK or SHARED_LOCK.
**
** If the locking level of the file descriptor is already at or below
** the requested locking level, this routine is a no-op.
**
** It is not possible for this routine to fail if the second argument
** is NO_LOCK.  If the second argument is SHARED_LOCK then this routine
** might return SQLITE_IOERR;
*/
static int winUnlock(OsFile *id, int locktype){
  int type;
  int rc = SQLITE_OK;
  winFile *pFile = (winFile*)id;
  assert( pFile!=0 );
  assert( locktype<=SHARED_LOCK );
  OSTRACE5("UNLOCK %d to %d was %d(%d)\n", pFile->h, locktype,
          pFile->locktype, pFile->sharedLockByte);
  type = pFile->locktype;
  if( type>=EXCLUSIVE_LOCK ){
    UnlockFile(pFile->h, SHARED_FIRST, 0, SHARED_SIZE, 0);
    if( locktype==SHARED_LOCK && !getReadLock(pFile) ){
      /* This should never happen.  We should always be able to
      ** reacquire the read lock */
      rc = SQLITE_IOERR_UNLOCK;
    }
  }
  if( type>=RESERVED_LOCK ){
    UnlockFile(pFile->h, RESERVED_BYTE, 0, 1, 0);
  }
  if( locktype==NO_LOCK && type>=SHARED_LOCK ){
    unlockReadLock(pFile);
  }
  if( type>=PENDING_LOCK ){
    UnlockFile(pFile->h, PENDING_BYTE, 0, 1, 0);
  }
  pFile->locktype = locktype;
  return rc;
}

/*
** Turn a relative pathname into a full pathname.  Return a pointer
** to the full pathname stored in space obtained from sqliteMalloc().
** The calling function is responsible for freeing this space once it
** is no longer needed.
*/
SQLITE_PRIVATE char *sqlite3WinFullPathname(const char *zRelative){
  char *zFull;
#if defined(__CYGWIN__)
  int nByte;
  nByte = strlen(zRelative) + MAX_PATH + 1001;
  zFull = sqliteMalloc( nByte );
  if( zFull==0 ) return 0;
  if( cygwin_conv_to_full_win32_path(zRelative, zFull) ) return 0;
#elif OS_WINCE
  /* WinCE has no concept of a relative pathname, or so I am told. */
  zFull = sqliteStrDup(zRelative);
#else
  int nByte;
  void *zConverted;
  zConverted = convertUtf8Filename(zRelative);
  if( isNT() ){
    WCHAR *zTemp;
    nByte = GetFullPathNameW((WCHAR*)zConverted, 0, 0, 0) + 3;
    zTemp = sqliteMalloc( nByte*sizeof(zTemp[0]) );
    if( zTemp==0 ){
      sqliteFree(zConverted);
      return 0;
    }
    GetFullPathNameW((WCHAR*)zConverted, nByte, zTemp, 0);
    sqliteFree(zConverted);
    zFull = unicodeToUtf8(zTemp);
    sqliteFree(zTemp);
  }else{
    char *zTemp;
    nByte = GetFullPathNameA((char*)zConverted, 0, 0, 0) + 3;
    zTemp = sqliteMalloc( nByte*sizeof(zTemp[0]) );
    if( zTemp==0 ){
      sqliteFree(zConverted);
      return 0;
    }
    GetFullPathNameA((char*)zConverted, nByte, zTemp, 0);
    sqliteFree(zConverted);
    zFull = mbcsToUtf8(zTemp);
    sqliteFree(zTemp);
  }
#endif
  return zFull;
}

/*
** The fullSync option is meaningless on windows.   This is a no-op.
*/
static void winSetFullSync(OsFile *id, int v){
  return;
}

/*
** Return the underlying file handle for an OsFile
*/
static int winFileHandle(OsFile *id){
  return (int)((winFile*)id)->h;
}

/*
** Return an integer that indices the type of lock currently held
** by this handle.  (Used for testing and analysis only.)
*/
static int winLockState(OsFile *id){
  return ((winFile*)id)->locktype;
}

/*
** Return the sector size in bytes of the underlying block device for
** the specified file. This is almost always 512 bytes, but may be
** larger for some devices.
**
** SQLite code assumes this function cannot fail. It also assumes that
** if two files are created in the same file-system directory (i.e.
** a database and it's journal file) that the sector size will be the
** same for both.
*/
static int winSectorSize(OsFile *id){
  return SQLITE_DEFAULT_SECTOR_SIZE;
}

/*
** This vector defines all the methods that can operate on an OsFile
** for win32.
*/
static const IoMethod sqlite3WinIoMethod = {
  winClose,
  winOpenDirectory,
  winRead,
  winWrite,
  winSeek,
  winTruncate,
  winSync,
  winSetFullSync,
  winFileHandle,
  winFileSize,
  winLock,
  winUnlock,
  winLockState,
  winCheckReservedLock,
  winSectorSize,
};

/*
** Allocate memory for an OsFile.  Initialize the new OsFile
** to the value given in pInit and return a pointer to the new
** OsFile.  If we run out of memory, close the file and return NULL.
*/
static int allocateWinFile(winFile *pInit, OsFile **pId){
  winFile *pNew;
  pNew = sqliteMalloc( sizeof(*pNew) );
  if( pNew==0 ){
    CloseHandle(pInit->h);
#if OS_WINCE
    sqliteFree(pInit->zDeleteOnClose);
#endif
    *pId = 0;
    return SQLITE_NOMEM;
  }else{
    *pNew = *pInit;
    pNew->pMethod = &sqlite3WinIoMethod;
    pNew->locktype = NO_LOCK;
    pNew->sharedLockByte = 0;
    *pId = (OsFile*)pNew;
    OpenCounter(+1);
    return SQLITE_OK;
  }
}


#endif /* SQLITE_OMIT_DISKIO */
/***************************************************************************
** Everything above deals with file I/O.  Everything that follows deals
** with other miscellanous aspects of the operating system interface
****************************************************************************/

#if !defined(SQLITE_OMIT_LOAD_EXTENSION)
/*
** Interfaces for opening a shared library, finding entry points
** within the shared library, and closing the shared library.
*/
SQLITE_PRIVATE void *sqlite3WinDlopen(const char *zFilename){
  HANDLE h;
  void *zConverted = convertUtf8Filename(zFilename);
  if( zConverted==0 ){
    return 0;
  }
  if( isNT() ){
    h = LoadLibraryW((WCHAR*)zConverted);
  }else{
#if OS_WINCE
    return 0;
#else
    h = LoadLibraryA((char*)zConverted);
#endif
  }
  sqliteFree(zConverted);
  return (void*)h;
  
}
SQLITE_PRIVATE void *sqlite3WinDlsym(void *pHandle, const char *zSymbol){
#if OS_WINCE
  /* The GetProcAddressA() routine is only available on wince. */
  return GetProcAddressA((HANDLE)pHandle, zSymbol);
#else
  /* All other windows platforms expect GetProcAddress() to take
  ** an Ansi string regardless of the _UNICODE setting */
  return GetProcAddress((HANDLE)pHandle, zSymbol);
#endif
}
SQLITE_PRIVATE int sqlite3WinDlclose(void *pHandle){
  return FreeLibrary((HANDLE)pHandle);
}
#endif /* !SQLITE_OMIT_LOAD_EXTENSION */

/*
** Get information to seed the random number generator.  The seed
** is written into the buffer zBuf[256].  The calling function must
** supply a sufficiently large buffer.
*/
SQLITE_PRIVATE int sqlite3WinRandomSeed(char *zBuf){
  /* We have to initialize zBuf to prevent valgrind from reporting
  ** errors.  The reports issued by valgrind are incorrect - we would
  ** prefer that the randomness be increased by making use of the
  ** uninitialized space in zBuf - but valgrind errors tend to worry
  ** some users.  Rather than argue, it seems easier just to initialize
  ** the whole array and silence valgrind, even if that means less randomness
  ** in the random seed.
  **
  ** When testing, initializing zBuf[] to zero is all we do.  That means
  ** that we always use the same random number sequence.* This makes the
  ** tests repeatable.
  */
  memset(zBuf, 0, 256);
  GetSystemTime((LPSYSTEMTIME)zBuf);
  return SQLITE_OK;
}

/*
** Sleep for a little while.  Return the amount of time slept.
*/
SQLITE_PRIVATE int sqlite3WinSleep(int ms){
  Sleep(ms);
  return ms;
}

/*
** Static variables used for thread synchronization
*/
static int inMutex = 0;
#ifdef SQLITE_W32_THREADS
  static DWORD mutexOwner;
  static CRITICAL_SECTION cs;
#endif

/*
** The following pair of routines implement mutual exclusion for
** multi-threaded processes.  Only a single thread is allowed to
** executed code that is surrounded by EnterMutex() and LeaveMutex().
**
** SQLite uses only a single Mutex.  There is not much critical
** code and what little there is executes quickly and without blocking.
**
** Version 3.3.1 and earlier used a simple mutex.  Beginning with
** version 3.3.2, a recursive mutex is required.
*/
SQLITE_PRIVATE void sqlite3WinEnterMutex(){
#ifdef SQLITE_W32_THREADS
  static int isInit = 0;
  while( !isInit ){
    static long lock = 0;
    if( InterlockedIncrement(&lock)==1 ){
      InitializeCriticalSection(&cs);
      isInit = 1;
    }else{
      Sleep(1);
    }
  }
  EnterCriticalSection(&cs);
  mutexOwner = GetCurrentThreadId();
#endif
  inMutex++;
}
SQLITE_PRIVATE void sqlite3WinLeaveMutex(){
  assert( inMutex );
  inMutex--;
#ifdef SQLITE_W32_THREADS
  assert( mutexOwner==GetCurrentThreadId() );
  LeaveCriticalSection(&cs);
#endif
}

/*
** Return TRUE if the mutex is currently held.
**
** If the thisThreadOnly parameter is true, return true if and only if the
** calling thread holds the mutex.  If the parameter is false, return
** true if any thread holds the mutex.
*/
SQLITE_PRIVATE int sqlite3WinInMutex(int thisThreadOnly){
#ifdef SQLITE_W32_THREADS
  return inMutex>0 && (thisThreadOnly==0 || mutexOwner==GetCurrentThreadId());
#else
  return inMutex>0;
#endif
}


/*
** The following variable, if set to a non-zero value, becomes the result
** returned from sqlite3OsCurrentTime().  This is used for testing.
*/
#ifdef SQLITE_TEST
int sqlite3_current_time = 0;
#endif

/*
** Find the current time (in Universal Coordinated Time).  Write the
** current time and date as a Julian Day number into *prNow and
** return 0.  Return 1 if the time and date cannot be found.
*/
SQLITE_PRIVATE int sqlite3WinCurrentTime(double *prNow){
  FILETIME ft;
  /* FILETIME structure is a 64-bit value representing the number of 
     100-nanosecond intervals since January 1, 1601 (= JD 2305813.5). 
  */
  double now;
#if OS_WINCE
  SYSTEMTIME time;
  GetSystemTime(&time);
  SystemTimeToFileTime(&time,&ft);
#else
  GetSystemTimeAsFileTime( &ft );
#endif
  now = ((double)ft.dwHighDateTime) * 4294967296.0; 
  *prNow = (now + ft.dwLowDateTime)/864000000000.0 + 2305813.5;
#ifdef SQLITE_TEST
  if( sqlite3_current_time ){
    *prNow = sqlite3_current_time/86400.0 + 2440587.5;
  }
#endif
  return 0;
}

/*
** Remember the number of thread-specific-data blocks allocated.
** Use this to verify that we are not leaking thread-specific-data.
** Ticket #1601
*/
#ifdef SQLITE_TEST
int sqlite3_tsd_count = 0;
# define TSD_COUNTER_INCR InterlockedIncrement(&sqlite3_tsd_count)
# define TSD_COUNTER_DECR InterlockedDecrement(&sqlite3_tsd_count)
#else
# define TSD_COUNTER_INCR  /* no-op */
# define TSD_COUNTER_DECR  /* no-op */
#endif



/*
** If called with allocateFlag>1, then return a pointer to thread
** specific data for the current thread.  Allocate and zero the
** thread-specific data if it does not already exist necessary.
**
** If called with allocateFlag==0, then check the current thread
** specific data.  Return it if it exists.  If it does not exist,
** then return NULL.
**
** If called with allocateFlag<0, check to see if the thread specific
** data is allocated and is all zero.  If it is then deallocate it.
** Return a pointer to the thread specific data or NULL if it is
** unallocated or gets deallocated.
*/
SQLITE_PRIVATE ThreadData *sqlite3WinThreadSpecificData(int allocateFlag){
  static int key;
  static int keyInit = 0;
  static const ThreadData zeroData = {0};
  ThreadData *pTsd;

  if( !keyInit ){
    sqlite3OsEnterMutex();
    if( !keyInit ){
      key = TlsAlloc();
      if( key==0xffffffff ){
        sqlite3OsLeaveMutex();
        return 0;
      }
      keyInit = 1;
    }
    sqlite3OsLeaveMutex();
  }
  pTsd = TlsGetValue(key);
  if( allocateFlag>0 ){
    if( !pTsd ){
      pTsd = sqlite3OsMalloc( sizeof(zeroData) );
      if( pTsd ){
        *pTsd = zeroData;
        TlsSetValue(key, pTsd);
        TSD_COUNTER_INCR;
      }
    }
  }else if( pTsd!=0 && allocateFlag<0 
              && memcmp(pTsd, &zeroData, sizeof(ThreadData))==0 ){
    sqlite3OsFree(pTsd);
    TlsSetValue(key, 0);
    TSD_COUNTER_DECR;
    pTsd = 0;
  }
  return pTsd;
}
#endif /* OS_WIN */

/************** End of os_win.c **********************************************/
/************** Begin file pager.c *******************************************/
/*
** 2001 September 15
**
** The author disclaims copyright to this source code.  In place of
** a legal notice, here is a blessing:
**
**    May you do good and not evil.
**    May you find forgiveness for yourself and forgive others.
**    May you share freely, never taking more than you give.
**
*************************************************************************
** This is the implementation of the page cache subsystem or "pager".
** 
** The pager is used to access a database disk file.  It implements
** atomic commit and rollback through the use of a journal file that
** is separate from the database file.  The pager also implements file
** locking to prevent two processes from writing the same database
** file simultaneously, or one process from reading the database while
** another is writing.
**
** @(#) $Id: sqlite3.c,v 1.4 2007/07/21 08:51:43 rmsimpson Exp $
*/
#ifndef SQLITE_OMIT_DISKIO

/*
** Macros for troubleshooting.  Normally turned off
*/
#if 0
#define sqlite3DebugPrintf printf
#define PAGERTRACE1(X)       sqlite3DebugPrintf(X)
#define PAGERTRACE2(X,Y)     sqlite3DebugPrintf(X,Y)
#define PAGERTRACE3(X,Y,Z)   sqlite3DebugPrintf(X,Y,Z)
#define PAGERTRACE4(X,Y,Z,W) sqlite3DebugPrintf(X,Y,Z,W)
#define PAGERTRACE5(X,Y,Z,W,V) sqlite3DebugPrintf(X,Y,Z,W,V)
#else
#define PAGERTRACE1(X)
#define PAGERTRACE2(X,Y)
#define PAGERTRACE3(X,Y,Z)
#define PAGERTRACE4(X,Y,Z,W)
#define PAGERTRACE5(X,Y,Z,W,V)
#endif

/*
** The following two macros are used within the PAGERTRACEX() macros above
** to print out file-descriptors. 
**
** PAGERID() takes a pointer to a Pager struct as it's argument. The
** associated file-descriptor is returned. FILEHANDLEID() takes an OsFile
** struct as it's argument.
*/
#define PAGERID(p) ((int)(p->fd))
#define FILEHANDLEID(fd) ((int)fd)

/*
** The page cache as a whole is always in one of the following
** states:
**
**   PAGER_UNLOCK        The page cache is not currently reading or 
**                       writing the database file.  There is no
**                       data held in memory.  This is the initial
**                       state.
**
**   PAGER_SHARED        The page cache is reading the database.
**                       Writing is not permitted.  There can be
**                       multiple readers accessing the same database
**                       file at the same time.
**
**   PAGER_RESERVED      This process has reserved the database for writing
**                       but has not yet made any changes.  Only one process
**                       at a time can reserve the database.  The original
**                       database file has not been modified so other
**                       processes may still be reading the on-disk
**                       database file.
**
**   PAGER_EXCLUSIVE     The page cache is writing the database.
**                       Access is exclusive.  No other processes or
**                       threads can be reading or writing while one
**                       process is writing.
**
**   PAGER_SYNCED        The pager moves to this state from PAGER_EXCLUSIVE
**                       after all dirty pages have been written to the
**                       database file and the file has been synced to
**                       disk. All that remains to do is to remove or
**                       truncate the journal file and the transaction 
**                       will be committed.
**
** The page cache comes up in PAGER_UNLOCK.  The first time a
** sqlite3PagerGet() occurs, the state transitions to PAGER_SHARED.
** After all pages have been released using sqlite_page_unref(),
** the state transitions back to PAGER_UNLOCK.  The first time
** that sqlite3PagerWrite() is called, the state transitions to
** PAGER_RESERVED.  (Note that sqlite3PagerWrite() can only be
** called on an outstanding page which means that the pager must
** be in PAGER_SHARED before it transitions to PAGER_RESERVED.)
** PAGER_RESERVED means that there is an open rollback journal.
** The transition to PAGER_EXCLUSIVE occurs before any changes
** are made to the database file, though writes to the rollback
** journal occurs with just PAGER_RESERVED.  After an sqlite3PagerRollback()
** or sqlite3PagerCommitPhaseTwo(), the state can go back to PAGER_SHARED,
** or it can stay at PAGER_EXCLUSIVE if we are in exclusive access mode.
*/
#define PAGER_UNLOCK      0
#define PAGER_SHARED      1   /* same as SHARED_LOCK */
#define PAGER_RESERVED    2   /* same as RESERVED_LOCK */
#define PAGER_EXCLUSIVE   4   /* same as EXCLUSIVE_LOCK */
#define PAGER_SYNCED      5

/*
** If the SQLITE_BUSY_RESERVED_LOCK macro is set to true at compile-time,
** then failed attempts to get a reserved lock will invoke the busy callback.
** This is off by default.  To see why, consider the following scenario:
** 
** Suppose thread A already has a shared lock and wants a reserved lock.
** Thread B already has a reserved lock and wants an exclusive lock.  If
** both threads are using their busy callbacks, it might be a long time
** be for one of the threads give up and allows the other to proceed.
** But if the thread trying to get the reserved lock gives up quickly
** (if it never invokes its busy callback) then the contention will be
** resolved quickly.
*/
#ifndef SQLITE_BUSY_RESERVED_LOCK
# define SQLITE_BUSY_RESERVED_LOCK 0
#endif

/*
** This macro rounds values up so that if the value is an address it
** is guaranteed to be an address that is aligned to an 8-byte boundary.
*/
#define FORCE_ALIGNMENT(X)   (((X)+7)&~7)

/*
** Each in-memory image of a page begins with the following header.
** This header is only visible to this pager module.  The client
** code that calls pager sees only the data that follows the header.
**
** Client code should call sqlite3PagerWrite() on a page prior to making
** any modifications to that page.  The first time sqlite3PagerWrite()
** is called, the original page contents are written into the rollback
** journal and PgHdr.inJournal and PgHdr.needSync are set.  Later, once
** the journal page has made it onto the disk surface, PgHdr.needSync
** is cleared.  The modified page cannot be written back into the original
** database file until the journal pages has been synced to disk and the
** PgHdr.needSync has been cleared.
**
** The PgHdr.dirty flag is set when sqlite3PagerWrite() is called and
** is cleared again when the page content is written back to the original
** database file.
**
** Details of important structure elements:
**
** needSync
**
**     If this is true, this means that it is not safe to write the page
**     content to the database because the original content needed
**     for rollback has not by synced to the main rollback journal.
**     The original content may have been written to the rollback journal
**     but it has not yet been synced.  So we cannot write to the database
**     file because power failure might cause the page in the journal file
**     to never reach the disk.  It is as if the write to the journal file
**     does not occur until the journal file is synced.
**     
**     This flag is false if the page content exactly matches what
**     currently exists in the database file.  The needSync flag is also
**     false if the original content has been written to the main rollback
**     journal and synced.  If the page represents a new page that has
**     been added onto the end of the database during the current
**     transaction, the needSync flag is true until the original database
**     size in the journal header has been synced to disk.
**
** inJournal
**
**     This is true if the original page has been written into the main
**     rollback journal.  This is always false for new pages added to
**     the end of the database file during the current transaction.
**     And this flag says nothing about whether or not the journal
**     has been synced to disk.  For pages that are in the original
**     database file, the following expression should always be true:
**
**       inJournal = (pPager->aInJournal[(pgno-1)/8] & (1<<((pgno-1)%8))!=0
**
**     The pPager->aInJournal[] array is only valid for the original
**     pages of the database, not new pages that are added to the end
**     of the database, so obviously the above expression cannot be
**     valid for new pages.  For new pages inJournal is always 0.
**
** dirty
**
**     When true, this means that the content of the page has been
**     modified and needs to be written back to the database file.
**     If false, it means that either the content of the page is
**     unchanged or else the content is unimportant and we do not
**     care whether or not it is preserved.
**
** alwaysRollback
**
**     This means that the sqlite3PagerDontRollback() API should be
**     ignored for this page.  The DontRollback() API attempts to say
**     that the content of the page on disk is unimportant (it is an
**     unused page on the freelist) so that it is unnecessary to 
**     rollback changes to this page because the content of the page
**     can change without changing the meaning of the database.  This
**     flag overrides any DontRollback() attempt.  This flag is set
**     when a page that originally contained valid data is added to
**     the freelist.  Later in the same transaction, this page might
**     be pulled from the freelist and reused for something different
**     and at that point the DontRollback() API will be called because
**     pages taken from the freelist do not need to be protected by
**     the rollback journal.  But this flag says that the page was
**     not originally part of the freelist so that it still needs to
**     be rolled back in spite of any subsequent DontRollback() calls.
**
** needRead 
**
**     This flag means (when true) that the content of the page has
**     not yet been loaded from disk.  The in-memory content is just
**     garbage.  (Actually, we zero the content, but you should not
**     make any assumptions about the content nevertheless.)  If the
**     content is needed in the future, it should be read from the
**     original database file.
*/
typedef struct PgHdr PgHdr;
struct PgHdr {
  Pager *pPager;                 /* The pager to which this page belongs */
  Pgno pgno;                     /* The page number for this page */
  PgHdr *pNextHash, *pPrevHash;  /* Hash collision chain for PgHdr.pgno */
  PgHdr *pNextFree, *pPrevFree;  /* Freelist of pages where nRef==0 */
  PgHdr *pNextAll;               /* A list of all pages */
  u8 inJournal;                  /* TRUE if has been written to journal */
  u8 dirty;                      /* TRUE if we need to write back changes */
  u8 needSync;                   /* Sync journal before writing this page */
  u8 alwaysRollback;             /* Disable DontRollback() for this page */
  u8 needRead;                   /* Read content if PagerWrite() is called */
  short int nRef;                /* Number of users of this page */
  PgHdr *pDirty, *pPrevDirty;    /* Dirty pages */
  u32 notUsed;                   /* Buffer space */
#ifdef SQLITE_CHECK_PAGES
  u32 pageHash;
#endif
  /* pPager->pageSize bytes of page data follow this header */
  /* Pager.nExtra bytes of local data follow the page data */
};

/*
** For an in-memory only database, some extra information is recorded about
** each page so that changes can be rolled back.  (Journal files are not
** used for in-memory databases.)  The following information is added to
** the end of every EXTRA block for in-memory databases.
**
** This information could have been added directly to the PgHdr structure.
** But then it would take up an extra 8 bytes of storage on every PgHdr
** even for disk-based databases.  Splitting it out saves 8 bytes.  This
** is only a savings of 0.8% but those percentages add up.
*/
typedef struct PgHistory PgHistory;
struct PgHistory {
  u8 *pOrig;     /* Original page text.  Restore to this on a full rollback */
  u8 *pStmt;     /* Text as it was at the beginning of the current statement */
  PgHdr *pNextStmt, *pPrevStmt;  /* List of pages in the statement journal */
  u8 inStmt;                     /* TRUE if in the statement subjournal */
};

/*
** A macro used for invoking the codec if there is one
*/
#ifdef SQLITE_HAS_CODEC
# define CODEC1(P,D,N,X) if( P->xCodec!=0 ){ P->xCodec(P->pCodecArg,D,N,X); }
# define CODEC2(P,D,N,X) ((char*)(P->xCodec!=0?P->xCodec(P->pCodecArg,D,N,X):D))
#else
# define CODEC1(P,D,N,X) /* NO-OP */
# define CODEC2(P,D,N,X) ((char*)D)
#endif

/*
** Convert a pointer to a PgHdr into a pointer to its data
** and back again.
*/
#define PGHDR_TO_DATA(P)  ((void*)(&(P)[1]))
#define DATA_TO_PGHDR(D)  (&((PgHdr*)(D))[-1])
#define PGHDR_TO_EXTRA(G,P) ((void*)&((char*)(&(G)[1]))[(P)->pageSize])
#define PGHDR_TO_HIST(P,PGR)  \
            ((PgHistory*)&((char*)(&(P)[1]))[(PGR)->pageSize+(PGR)->nExtra])

/*
** A open page cache is an instance of the following structure.
**
** Pager.errCode may be set to SQLITE_IOERR, SQLITE_CORRUPT, or
** or SQLITE_FULL. Once one of the first three errors occurs, it persists
** and is returned as the result of every major pager API call.  The
** SQLITE_FULL return code is slightly different. It persists only until the
** next successful rollback is performed on the pager cache. Also,
** SQLITE_FULL does not affect the sqlite3PagerGet() and sqlite3PagerLookup()
** APIs, they may still be used successfully.
*/
struct Pager {
  u8 journalOpen;             /* True if journal file descriptors is valid */
  u8 journalStarted;          /* True if header of journal is synced */
  u8 useJournal;              /* Use a rollback journal on this file */
  u8 noReadlock;              /* Do not bother to obtain readlocks */
  u8 stmtOpen;                /* True if the statement subjournal is open */
  u8 stmtInUse;               /* True we are in a statement subtransaction */
  u8 stmtAutoopen;            /* Open stmt journal when main journal is opened*/
  u8 noSync;                  /* Do not sync the journal if true */
  u8 fullSync;                /* Do extra syncs of the journal for robustness */
  u8 full_fsync;              /* Use F_FULLFSYNC when available */
  u8 state;                   /* PAGER_UNLOCK, _SHARED, _RESERVED, etc. */
  u8 tempFile;                /* zFilename is a temporary file */
  u8 readOnly;                /* True for a read-only database */
  u8 needSync;                /* True if an fsync() is needed on the journal */
  u8 dirtyCache;              /* True if cached pages have changed */
  u8 alwaysRollback;          /* Disable DontRollback() for all pages */
  u8 memDb;                   /* True to inhibit all file I/O */
  u8 setMaster;               /* True if a m-j name has been written to jrnl */
  u8 doNotSync;               /* Boolean. While true, do not spill the cache */
  u8 exclusiveMode;           /* Boolean. True if locking_mode==EXCLUSIVE */
  u8 changeCountDone;         /* Set after incrementing the change-counter */
  int errCode;                /* One of several kinds of errors */
  int dbSize;                 /* Number of pages in the file */
  int origDbSize;             /* dbSize before the current change */
  int stmtSize;               /* Size of database (in pages) at stmt_begin() */
  int nRec;                   /* Number of pages written to the journal */
  u32 cksumInit;              /* Quasi-random value added to every checksum */
  int stmtNRec;               /* Number of records in stmt subjournal */
  int nExtra;                 /* Add this many bytes to each in-memory page */
  int pageSize;               /* Number of bytes in a page */
  int nPage;                  /* Total number of in-memory pages */
  int nRef;                   /* Number of in-memory pages with PgHdr.nRef>0 */
  int mxPage;                 /* Maximum number of pages to hold in cache */
  Pgno mxPgno;                /* Maximum allowed size of the database */
  u8 *aInJournal;             /* One bit for each page in the database file */
  u8 *aInStmt;                /* One bit for each page in the database */
  char *zFilename;            /* Name of the database file */
  char *zJournal;             /* Name of the journal file */
  char *zDirectory;           /* Directory hold database and journal files */
  OsFile *fd, *jfd;           /* File descriptors for database and journal */
  OsFile *stfd;               /* File descriptor for the statement subjournal*/
  BusyHandler *pBusyHandler;  /* Pointer to sqlite.busyHandler */
  PgHdr *pFirst, *pLast;      /* List of free pages */
  PgHdr *pFirstSynced;        /* First free page with PgHdr.needSync==0 */
  PgHdr *pAll;                /* List of all pages */
  PgHdr *pStmt;               /* List of pages in the statement subjournal */
  PgHdr *pDirty;              /* List of all dirty pages */
  i64 journalOff;             /* Current byte offset in the journal file */
  i64 journalHdr;             /* Byte offset to previous journal header */
  i64 stmtHdrOff;             /* First journal header written this statement */
  i64 stmtCksum;              /* cksumInit when statement was started */
  i64 stmtJSize;              /* Size of journal at stmt_begin() */
  int sectorSize;             /* Assumed sector size during rollback */
#ifdef SQLITE_TEST
  int nHit, nMiss;            /* Cache hits and missing */
  int nRead, nWrite;          /* Database pages read/written */
#endif
  void (*xDestructor)(DbPage*,int); /* Call this routine when freeing pages */
  void (*xReiniter)(DbPage*,int);   /* Call this routine when reloading pages */
#ifdef SQLITE_HAS_CODEC
  void *(*xCodec)(void*,void*,Pgno,int); /* Routine for en/decoding data */
  void *pCodecArg;            /* First argument to xCodec() */
#endif
  int nHash;                  /* Size of the pager hash table */
  PgHdr **aHash;              /* Hash table to map page number to PgHdr */
#ifdef SQLITE_ENABLE_MEMORY_MANAGEMENT
  Pager *pNext;               /* Linked list of pagers in this thread */
#endif
  char *pTmpSpace;            /* Pager.pageSize bytes of space for tmp use */
  char dbFileVers[16];        /* Changes whenever database file changes */
};

/*
** The following global variables hold counters used for
** testing purposes only.  These variables do not exist in
** a non-testing build.  These variables are not thread-safe.
*/
#ifdef SQLITE_TEST
int sqlite3_pager_readdb_count = 0;    /* Number of full pages read from DB */
int sqlite3_pager_writedb_count = 0;   /* Number of full pages written to DB */
int sqlite3_pager_writej_count = 0;    /* Number of pages written to journal */
int sqlite3_pager_pgfree_count = 0;    /* Number of cache pages freed */
# define PAGER_INCR(v)  v++
#else
# define PAGER_INCR(v)
#endif



/*
** Journal files begin with the following magic string.  The data
** was obtained from /dev/random.  It is used only as a sanity check.
**
** Since version 2.8.0, the journal format contains additional sanity
** checking information.  If the power fails while the journal is begin
** written, semi-random garbage data might appear in the journal
** file after power is restored.  If an attempt is then made
** to roll the journal back, the database could be corrupted.  The additional
** sanity checking data is an attempt to discover the garbage in the
** journal and ignore it.
**
** The sanity checking information for the new journal format consists
** of a 32-bit checksum on each page of data.  The checksum covers both
** the page number and the pPager->pageSize bytes of data for the page.
** This cksum is initialized to a 32-bit random value that appears in the
** journal file right after the header.  The random initializer is important,
** because garbage data that appears at the end of a journal is likely
** data that was once in other files that have now been deleted.  If the
** garbage data came from an obsolete journal file, the checksums might
** be correct.  But by initializing the checksum to random value which
** is different for every journal, we minimize that risk.
*/
static const unsigned char aJournalMagic[] = {
  0xd9, 0xd5, 0x05, 0xf9, 0x20, 0xa1, 0x63, 0xd7,
};

/*
** The size of the header and of each page in the journal is determined
** by the following macros.
*/
#define JOURNAL_PG_SZ(pPager)  ((pPager->pageSize) + 8)

/*
** The journal header size for this pager. In the future, this could be
** set to some value read from the disk controller. The important
** characteristic is that it is the same size as a disk sector.
*/
#define JOURNAL_HDR_SZ(pPager) (pPager->sectorSize)

/*
** The macro MEMDB is true if we are dealing with an in-memory database.
** We do this as a macro so that if the SQLITE_OMIT_MEMORYDB macro is set,
** the value of MEMDB will be a constant and the compiler will optimize
** out code that would never execute.
*/
#ifdef SQLITE_OMIT_MEMORYDB
# define MEMDB 0
#else
# define MEMDB pPager->memDb
#endif

/*
** Page number PAGER_MJ_PGNO is never used in an SQLite database (it is
** reserved for working around a windows/posix incompatibility). It is
** used in the journal to signify that the remainder of the journal file 
** is devoted to storing a master journal name - there are no more pages to
** roll back. See comments for function writeMasterJournal() for details.
*/
/* #define PAGER_MJ_PGNO(x) (PENDING_BYTE/((x)->pageSize)) */
#define PAGER_MJ_PGNO(x) ((PENDING_BYTE/((x)->pageSize))+1)

/*
** The maximum legal page number is (2^31 - 1).
*/
#define PAGER_MAX_PGNO 2147483647

/*
** Enable reference count tracking (for debugging) here:
*/
#ifdef SQLITE_DEBUG
  int pager3_refinfo_enable = 0;
  static void pager_refinfo(PgHdr *p){
    static int cnt = 0;
    if( !pager3_refinfo_enable ) return;
    sqlite3DebugPrintf(
       "REFCNT: %4d addr=%p nRef=%-3d total=%d\n",
       p->pgno, PGHDR_TO_DATA(p), p->nRef, p->pPager->nRef
    );
    cnt++;   /* Something to set a breakpoint on */
  }
# define REFINFO(X)  pager_refinfo(X)
#else
# define REFINFO(X)
#endif

/*
** Return true if page *pPg has already been written to the statement
** journal (or statement snapshot has been created, if *pPg is part
** of an in-memory database).
*/
static int pageInStatement(PgHdr *pPg){
  Pager *pPager = pPg->pPager;
  if( MEMDB ){
    return PGHDR_TO_HIST(pPg, pPager)->inStmt;
  }else{
    Pgno pgno = pPg->pgno;
    u8 *a = pPager->aInStmt;
    return (a && (int)pgno<=pPager->stmtSize && (a[pgno/8] & (1<<(pgno&7))));
  }
}

/*
** Change the size of the pager hash table to N.  N must be a power
** of two.
*/
static void pager_resize_hash_table(Pager *pPager, int N){
  PgHdr **aHash, *pPg;
  assert( N>0 && (N&(N-1))==0 );
  aHash = sqliteMalloc( sizeof(aHash[0])*N );
  if( aHash==0 ){
    /* Failure to rehash is not an error.  It is only a performance hit. */
    return;
  }
  sqliteFree(pPager->aHash);
  pPager->nHash = N;
  pPager->aHash = aHash;
  for(pPg=pPager->pAll; pPg; pPg=pPg->pNextAll){
    int h;
    if( pPg->pgno==0 ){
      assert( pPg->pNextHash==0 && pPg->pPrevHash==0 );
      continue;
    }
    h = pPg->pgno & (N-1);
    pPg->pNextHash = aHash[h];
    if( aHash[h] ){
      aHash[h]->pPrevHash = pPg;
    }
    aHash[h] = pPg;
    pPg->pPrevHash = 0;
  }
}

/*
** Read a 32-bit integer from the given file descriptor.  Store the integer
** that is read in *pRes.  Return SQLITE_OK if everything worked, or an
** error code is something goes wrong.
**
** All values are stored on disk as big-endian.
*/
static int read32bits(OsFile *fd, u32 *pRes){
  unsigned char ac[4];
  int rc = sqlite3OsRead(fd, ac, sizeof(ac));
  if( rc==SQLITE_OK ){
    *pRes = sqlite3Get4byte(ac);
  }
  return rc;
}

/*
** Write a 32-bit integer into a string buffer in big-endian byte order.
*/
#define put32bits(A,B)  sqlite3Put4byte((u8*)A,B)

/*
** Write a 32-bit integer into the given file descriptor.  Return SQLITE_OK
** on success or an error code is something goes wrong.
*/
static int write32bits(OsFile *fd, u32 val){
  char ac[4];
  put32bits(ac, val);
  return sqlite3OsWrite(fd, ac, 4);
}

/*
** This function should be called when an error occurs within the pager
** code. The first argument is a pointer to the pager structure, the
** second the error-code about to be returned by a pager API function. 
** The value returned is a copy of the second argument to this function. 
**
** If the second argument is SQLITE_IOERR, SQLITE_CORRUPT, or SQLITE_FULL
** the error becomes persistent. All subsequent API calls on this Pager
** will immediately return the same error code.
*/
static int pager_error(Pager *pPager, int rc){
  int rc2 = rc & 0xff;
  assert( pPager->errCode==SQLITE_FULL || pPager->errCode==SQLITE_OK );
  if(
    rc2==SQLITE_FULL ||
    rc2==SQLITE_IOERR ||
    rc2==SQLITE_CORRUPT
  ){
    pPager->errCode = rc;
  }
  return rc;
}

/*
** If SQLITE_CHECK_PAGES is defined then we do some sanity checking
** on the cache using a hash function.  This is used for testing
** and debugging only.
*/
#ifdef SQLITE_CHECK_PAGES
/*
** Return a 32-bit hash of the page data for pPage.
*/
static u32 pager_datahash(int nByte, unsigned char *pData){
  u32 hash = 0;
  int i;
  for(i=0; i<nByte; i++){
    hash = (hash*1039) + pData[i];
  }
  return hash;
}
static u32 pager_pagehash(PgHdr *pPage){
  return pager_datahash(pPage->pPager->pageSize, 
                        (unsigned char *)PGHDR_TO_DATA(pPage));
}

/*
** The CHECK_PAGE macro takes a PgHdr* as an argument. If SQLITE_CHECK_PAGES
** is defined, and NDEBUG is not defined, an assert() statement checks
** that the page is either dirty or still matches the calculated page-hash.
*/
#define CHECK_PAGE(x) checkPage(x)
static void checkPage(PgHdr *pPg){
  Pager *pPager = pPg->pPager;
  assert( !pPg->pageHash || pPager->errCode || MEMDB || pPg->dirty || 
      pPg->pageHash==pager_pagehash(pPg) );
}

#else
#define pager_datahash(X,Y)  0
#define pager_pagehash(X)  0
#define CHECK_PAGE(x)
#endif

/*
** When this is called the journal file for pager pPager must be open.
** The master journal file name is read from the end of the file and 
** written into memory obtained from sqliteMalloc(). *pzMaster is
** set to point at the memory and SQLITE_OK returned. The caller must
** sqliteFree() *pzMaster.
**
** If no master journal file name is present *pzMaster is set to 0 and
** SQLITE_OK returned.
*/
static int readMasterJournal(OsFile *pJrnl, char **pzMaster){
  int rc;
  u32 len;
  i64 szJ;
  u32 cksum;
  int i;
  unsigned char aMagic[8]; /* A buffer to hold the magic header */

  *pzMaster = 0;

  rc = sqlite3OsFileSize(pJrnl, &szJ);
  if( rc!=SQLITE_OK || szJ<16 ) return rc;

  rc = sqlite3OsSeek(pJrnl, szJ-16);
  if( rc!=SQLITE_OK ) return rc;
 
  rc = read32bits(pJrnl, &len);
  if( rc!=SQLITE_OK ) return rc;

  rc = read32bits(pJrnl, &cksum);
  if( rc!=SQLITE_OK ) return rc;

  rc = sqlite3OsRead(pJrnl, aMagic, 8);
  if( rc!=SQLITE_OK || memcmp(aMagic, aJournalMagic, 8) ) return rc;

  rc = sqlite3OsSeek(pJrnl, szJ-16-len);
  if( rc!=SQLITE_OK ) return rc;

  *pzMaster = (char *)sqliteMalloc(len+1);
  if( !*pzMaster ){
    return SQLITE_NOMEM;
  }
  rc = sqlite3OsRead(pJrnl, *pzMaster, len);
  if( rc!=SQLITE_OK ){
    sqliteFree(*pzMaster);
    *pzMaster = 0;
    return rc;
  }

  /* See if the checksum matches the master journal name */
  for(i=0; i<len; i++){
    cksum -= (*pzMaster)[i];
  }
  if( cksum ){
    /* If the checksum doesn't add up, then one or more of the disk sectors
    ** containing the master journal filename is corrupted. This means
    ** definitely roll back, so just return SQLITE_OK and report a (nul)
    ** master-journal filename.
    */
    sqliteFree(*pzMaster);
    *pzMaster = 0;
  }else{
    (*pzMaster)[len] = '\0';
  }
   
  return SQLITE_OK;
}

/*
** Seek the journal file descriptor to the next sector boundary where a
** journal header may be read or written. Pager.journalOff is updated with
** the new seek offset.
**
** i.e for a sector size of 512:
**
** Input Offset              Output Offset
** ---------------------------------------
** 0                         0
** 512                       512
** 100                       512
** 2000                      2048
** 
*/
static int seekJournalHdr(Pager *pPager){
  i64 offset = 0;
  i64 c = pPager->journalOff;
  if( c ){
    offset = ((c-1)/JOURNAL_HDR_SZ(pPager) + 1) * JOURNAL_HDR_SZ(pPager);
  }
  assert( offset%JOURNAL_HDR_SZ(pPager)==0 );
  assert( offset>=c );
  assert( (offset-c)<JOURNAL_HDR_SZ(pPager) );
  pPager->journalOff = offset;
  return sqlite3OsSeek(pPager->jfd, pPager->journalOff);
}

/*
** The journal file must be open when this routine is called. A journal
** header (JOURNAL_HDR_SZ bytes) is written into the journal file at the
** current location.
**
** The format for the journal header is as follows:
** - 8 bytes: Magic identifying journal format.
** - 4 bytes: Number of records in journal, or -1 no-sync mode is on.
** - 4 bytes: Random number used for page hash.
** - 4 bytes: Initial database page count.
** - 4 bytes: Sector size used by the process that wrote this journal.
** 
** Followed by (JOURNAL_HDR_SZ - 24) bytes of unused space.
*/
static int writeJournalHdr(Pager *pPager){
  char zHeader[sizeof(aJournalMagic)+16];
  int rc;

  if( pPager->stmtHdrOff==0 ){
    pPager->stmtHdrOff = pPager->journalOff;
  }

  rc = seekJournalHdr(pPager);
  if( rc ) return rc;

  pPager->journalHdr = pPager->journalOff;
  pPager->journalOff += JOURNAL_HDR_SZ(pPager);

  /* FIX ME: 
  **
  ** Possibly for a pager not in no-sync mode, the journal magic should not
  ** be written until nRec is filled in as part of next syncJournal(). 
  **
  ** Actually maybe the whole journal header should be delayed until that
  ** point. Think about this.
  */
  memcpy(zHeader, aJournalMagic, sizeof(aJournalMagic));
  /* The nRec Field. 0xFFFFFFFF for no-sync journals. */
  put32bits(&zHeader[sizeof(aJournalMagic)], pPager->noSync ? 0xffffffff : 0);
  /* The random check-hash initialiser */ 
  sqlite3Randomness(sizeof(pPager->cksumInit), &pPager->cksumInit);
  put32bits(&zHeader[sizeof(aJournalMagic)+4], pPager->cksumInit);
  /* The initial database size */
  put32bits(&zHeader[sizeof(aJournalMagic)+8], pPager->dbSize);
  /* The assumed sector size for this process */
  put32bits(&zHeader[sizeof(aJournalMagic)+12], pPager->sectorSize);
  IOTRACE(("JHDR %p %lld %d\n", pPager, pPager->journalHdr, sizeof(zHeader)))
  rc = sqlite3OsWrite(pPager->jfd, zHeader, sizeof(zHeader));

  /* The journal header has been written successfully. Seek the journal
  ** file descriptor to the end of the journal header sector.
  */
  if( rc==SQLITE_OK ){
    IOTRACE(("JTAIL %p %lld\n", pPager, pPager->journalOff-1))
    rc = sqlite3OsSeek(pPager->jfd, pPager->journalOff-1);
    if( rc==SQLITE_OK ){
      rc = sqlite3OsWrite(pPager->jfd, "\000", 1);
    }
  }
  return rc;
}

/*
** The journal file must be open when this is called. A journal header file
** (JOURNAL_HDR_SZ bytes) is read from the current location in the journal
** file. See comments above function writeJournalHdr() for a description of
** the journal header format.
**
** If the header is read successfully, *nRec is set to the number of
** page records following this header and *dbSize is set to the size of the
** database before the transaction began, in pages. Also, pPager->cksumInit
** is set to the value read from the journal header. SQLITE_OK is returned
** in this case.
**
** If the journal header file appears to be corrupted, SQLITE_DONE is
** returned and *nRec and *dbSize are not set.  If JOURNAL_HDR_SZ bytes
** cannot be read from the journal file an error code is returned.
*/
static int readJournalHdr(
  Pager *pPager, 
  i64 journalSize,
  u32 *pNRec, 
  u32 *pDbSize
){
  int rc;
  unsigned char aMagic[8]; /* A buffer to hold the magic header */

  rc = seekJournalHdr(pPager);
  if( rc ) return rc;

  if( pPager->journalOff+JOURNAL_HDR_SZ(pPager) > journalSize ){
    return SQLITE_DONE;
  }

  rc = sqlite3OsRead(pPager->jfd, aMagic, sizeof(aMagic));
  if( rc ) return rc;

  if( memcmp(aMagic, aJournalMagic, sizeof(aMagic))!=0 ){
    return SQLITE_DONE;
  }

  rc = read32bits(pPager->jfd, pNRec);
  if( rc ) return rc;

  rc = read32bits(pPager->jfd, &pPager->cksumInit);
  if( rc ) return rc;

  rc = read32bits(pPager->jfd, pDbSize);
  if( rc ) return rc;

  /* Update the assumed sector-size to match the value used by 
  ** the process that created this journal. If this journal was
  ** created by a process other than this one, then this routine
  ** is being called from within pager_playback(). The local value
  ** of Pager.sectorSize is restored at the end of that routine.
  */
  rc = read32bits(pPager->jfd, (u32 *)&pPager->sectorSize);
  if( rc ) return rc;

  pPager->journalOff += JOURNAL_HDR_SZ(pPager);
  rc = sqlite3OsSeek(pPager->jfd, pPager->journalOff);
  return rc;
}


/*
** Write the supplied master journal name into the journal file for pager
** pPager at the current location. The master journal name must be the last
** thing written to a journal file. If the pager is in full-sync mode, the
** journal file descriptor is advanced to the next sector boundary before
** anything is written. The format is:
**
** + 4 bytes: PAGER_MJ_PGNO.
** + N bytes: length of master journal name.
** + 4 bytes: N
** + 4 bytes: Master journal name checksum.
** + 8 bytes: aJournalMagic[].
**
** The master journal page checksum is the sum of the bytes in the master
** journal name.
**
** If zMaster is a NULL pointer (occurs for a single database transaction), 
** this call is a no-op.
*/
static int writeMasterJournal(Pager *pPager, const char *zMaster){
  int rc;
  int len; 
  int i; 
  u32 cksum = 0;
  char zBuf[sizeof(aJournalMagic)+2*4];

  if( !zMaster || pPager->setMaster) return SQLITE_OK;
  pPager->setMaster = 1;

  len = strlen(zMaster);
  for(i=0; i<len; i++){
    cksum += zMaster[i];
  }

  /* If in full-sync mode, advance to the next disk sector before writing
  ** the master journal name. This is in case the previous page written to
  ** the journal has already been synced.
  */
  if( pPager->fullSync ){
    rc = seekJournalHdr(pPager);
    if( rc!=SQLITE_OK ) return rc;
  }
  pPager->journalOff += (len+20);

  rc = write32bits(pPager->jfd, PAGER_MJ_PGNO(pPager));
  if( rc!=SQLITE_OK ) return rc;

  rc = sqlite3OsWrite(pPager->jfd, zMaster, len);
  if( rc!=SQLITE_OK ) return rc;

  put32bits(zBuf, len);
  put32bits(&zBuf[4], cksum);
  memcpy(&zBuf[8], aJournalMagic, sizeof(aJournalMagic));
  rc = sqlite3OsWrite(pPager->jfd, zBuf, 8+sizeof(aJournalMagic));
  pPager->needSync = !pPager->noSync;
  return rc;
}

/*
** Add or remove a page from the list of all pages that are in the
** statement journal.
**
** The Pager keeps a separate list of pages that are currently in
** the statement journal.  This helps the sqlite3PagerStmtCommit()
** routine run MUCH faster for the common case where there are many
** pages in memory but only a few are in the statement journal.
*/
static void page_add_to_stmt_list(PgHdr *pPg){
  Pager *pPager = pPg->pPager;
  PgHistory *pHist = PGHDR_TO_HIST(pPg, pPager);
  assert( MEMDB );
  if( !pHist->inStmt ){
    assert( pHist->pPrevStmt==0 && pHist->pNextStmt==0 );
    if( pPager->pStmt ){
      PGHDR_TO_HIST(pPager->pStmt, pPager)->pPrevStmt = pPg;
    }
    pHist->pNextStmt = pPager->pStmt;
    pPager->pStmt = pPg;
    pHist->inStmt = 1;
  }
}

/*
** Find a page in the hash table given its page number.  Return
** a pointer to the page or NULL if not found.
*/
static PgHdr *pager_lookup(Pager *pPager, Pgno pgno){
  PgHdr *p;
  if( pPager->aHash==0 ) return 0;
  p = pPager->aHash[pgno & (pPager->nHash-1)];
  while( p && p->pgno!=pgno ){
    p = p->pNextHash;
  }
  return p;
}

/*
** Unlock the database file.
*/
static void pager_unlock(Pager *pPager){
  if( !pPager->exclusiveMode ){
    if( !MEMDB ){
      sqlite3OsUnlock(pPager->fd, NO_LOCK);
      pPager->dbSize = -1;
      IOTRACE(("UNLOCK %p\n", pPager))
    }
    pPager->state = PAGER_UNLOCK;
    pPager->changeCountDone = 0;
  }
}

/*
** Execute a rollback if a transaction is active and unlock the 
** database file. This is a no-op if the pager has already entered
** the error-state.
*/
static void pagerUnlockAndRollback(Pager *p){
  if( p->errCode ) return;
  assert( p->state>=PAGER_RESERVED || p->journalOpen==0 );
  if( p->state>=PAGER_RESERVED ){
    sqlite3PagerRollback(p);
  }
  pager_unlock(p);
  assert( p->errCode || !p->journalOpen || (p->exclusiveMode&&!p->journalOff) );
  assert( p->errCode || !p->stmtOpen || p->exclusiveMode );
}


/*
** Clear the in-memory cache.  This routine
** sets the state of the pager back to what it was when it was first
** opened.  Any outstanding pages are invalidated and subsequent attempts
** to access those pages will likely result in a coredump.
*/
static void pager_reset(Pager *pPager){
  PgHdr *pPg, *pNext;
  if( pPager->errCode ) return;
  for(pPg=pPager->pAll; pPg; pPg=pNext){
    IOTRACE(("PGFREE %p %d\n", pPager, pPg->pgno));
    PAGER_INCR(sqlite3_pager_pgfree_count);
    pNext = pPg->pNextAll;
    sqliteFree(pPg);
  }
  pPager->pStmt = 0;
  pPager->pFirst = 0;
  pPager->pFirstSynced = 0;
  pPager->pLast = 0;
  pPager->pAll = 0;
  pPager->nHash = 0;
  sqliteFree(pPager->aHash);
  pPager->nPage = 0;
  pPager->aHash = 0;
  pPager->nRef = 0;
}

/*
** This routine ends a transaction.  A transaction is ended by either
** a COMMIT or a ROLLBACK.
**
** When this routine is called, the pager has the journal file open and
** a RESERVED or EXCLUSIVE lock on the database.  This routine will release
** the database lock and acquires a SHARED lock in its place if that is
** the appropriate thing to do.  Release locks usually is appropriate,
** unless we are in exclusive access mode or unless this is a 
** COMMIT AND BEGIN or ROLLBACK AND BEGIN operation.
**
** The journal file is either deleted or truncated.
**
** TODO: Consider keeping the journal file open for temporary databases.
** This might give a performance improvement on windows where opening
** a file is an expensive operation.
*/
static int pager_end_transaction(Pager *pPager){
  PgHdr *pPg;
  int rc = SQLITE_OK;
  int rc2 = SQLITE_OK;
  assert( !MEMDB );
  if( pPager->state<PAGER_RESERVED ){
    return SQLITE_OK;
  }
  sqlite3PagerStmtCommit(pPager);
  if( pPager->stmtOpen && !pPager->exclusiveMode ){
    sqlite3OsClose(&pPager->stfd);
    pPager->stmtOpen = 0;
  }
  if( pPager->journalOpen ){
    if( pPager->exclusiveMode 
          && (rc = sqlite3OsTruncate(pPager->jfd, 0))==SQLITE_OK ){;
      sqlite3OsSeek(pPager->jfd, 0);
      pPager->journalOff = 0;
      pPager->journalStarted = 0;
    }else{
      sqlite3OsClose(&pPager->jfd);
      pPager->journalOpen = 0;
      if( rc==SQLITE_OK ){
        rc = sqlite3OsDelete(pPager->zJournal);
      }
    }
    sqliteFree( pPager->aInJournal );
    pPager->aInJournal = 0;
    for(pPg=pPager->pAll; pPg; pPg=pPg->pNextAll){
      pPg->inJournal = 0;
      pPg->dirty = 0;
      pPg->needSync = 0;
      pPg->alwaysRollback = 0;
#ifdef SQLITE_CHECK_PAGES
      pPg->pageHash = pager_pagehash(pPg);
#endif
    }
    pPager->pDirty = 0;
    pPager->dirtyCache = 0;
    pPager->nRec = 0;
  }else{
    assert( pPager->aInJournal==0 );
    assert( pPager->dirtyCache==0 || pPager->useJournal==0 );
  }

  if( !pPager->exclusiveMode ){
    rc2 = sqlite3OsUnlock(pPager->fd, SHARED_LOCK);
    pPager->state = PAGER_SHARED;
  }else if( pPager->state==PAGER_SYNCED ){
    pPager->state = PAGER_EXCLUSIVE;
  }
  pPager->origDbSize = 0;
  pPager->setMaster = 0;
  pPager->needSync = 0;
  pPager->pFirstSynced = pPager->pFirst;
  pPager->dbSize = -1;

  return (rc==SQLITE_OK?rc2:rc);
}

/*
** Compute and return a checksum for the page of data.
**
** This is not a real checksum.  It is really just the sum of the 
** random initial value and the page number.  We experimented with
** a checksum of the entire data, but that was found to be too slow.
**
** Note that the page number is stored at the beginning of data and
** the checksum is stored at the end.  This is important.  If journal
** corruption occurs due to a power failure, the most likely scenario
** is that one end or the other of the record will be changed.  It is
** much less likely that the two ends of the journal record will be
** correct and the middle be corrupt.  Thus, this "checksum" scheme,
** though fast and simple, catches the mostly likely kind of corruption.
**
** FIX ME:  Consider adding every 200th (or so) byte of the data to the
** checksum.  That way if a single page spans 3 or more disk sectors and
** only the middle sector is corrupt, we will still have a reasonable
** chance of failing the checksum and thus detecting the problem.
*/
static u32 pager_cksum(Pager *pPager, const u8 *aData){
  u32 cksum = pPager->cksumInit;
  int i = pPager->pageSize-200;
  while( i>0 ){
    cksum += aData[i];
    i -= 200;
  }
  return cksum;
}

/* Forward declaration */
static void makeClean(PgHdr*);

/*
** Read a single page from the journal file opened on file descriptor
** jfd.  Playback this one page.
**
** If useCksum==0 it means this journal does not use checksums.  Checksums
** are not used in statement journals because statement journals do not
** need to survive power failures.
*/
static int pager_playback_one_page(Pager *pPager, OsFile *jfd, int useCksum){
  int rc;
  PgHdr *pPg;                   /* An existing page in the cache */
  Pgno pgno;                    /* The page number of a page in journal */
  u32 cksum;                    /* Checksum used for sanity checking */
  u8 *aData = (u8 *)pPager->pTmpSpace;   /* Temp storage for a page */

  /* useCksum should be true for the main journal and false for
  ** statement journals.  Verify that this is always the case
  */
  assert( jfd == (useCksum ? pPager->jfd : pPager->stfd) );
  assert( aData );

  rc = read32bits(jfd, &pgno);
  if( rc!=SQLITE_OK ) return rc;
  rc = sqlite3OsRead(jfd, aData, pPager->pageSize);
  if( rc!=SQLITE_OK ) return rc;
  pPager->journalOff += pPager->pageSize + 4;

  /* Sanity checking on the page.  This is more important that I originally
  ** thought.  If a power failure occurs while the journal is being written,
  ** it could cause invalid data to be written into the journal.  We need to
  ** detect this invalid data (with high probability) and ignore it.
  */
  if( pgno==0 || pgno==PAGER_MJ_PGNO(pPager) ){
    return SQLITE_DONE;
  }
  if( pgno>(unsigned)pPager->dbSize ){
    return SQLITE_OK;
  }
  if( useCksum ){
    rc = read32bits(jfd, &cksum);
    if( rc ) return rc;
    pPager->journalOff += 4;
    if( pager_cksum(pPager, aData)!=cksum ){
      return SQLITE_DONE;
    }
  }

  assert( pPager->state==PAGER_RESERVED || pPager->state>=PAGER_EXCLUSIVE );

  /* If the pager is in RESERVED state, then there must be a copy of this
  ** page in the pager cache. In this case just update the pager cache,
  ** not the database file. The page is left marked dirty in this case.
  **
  ** An exception to the above rule: If the database is in no-sync mode
  ** and a page is moved during an incremental vacuum then the page may
  ** not be in the pager cache. Later: if a malloc() or IO error occurs
  ** during a Movepage() call, then the page may not be in the cache
  ** either. So the condition described in the above paragraph is not
  ** assert()able.
  **
  ** If in EXCLUSIVE state, then we update the pager cache if it exists
  ** and the main file. The page is then marked not dirty.
  **
  ** Ticket #1171:  The statement journal might contain page content that is
  ** different from the page content at the start of the transaction.
  ** This occurs when a page is changed prior to the start of a statement
  ** then changed again within the statement.  When rolling back such a
  ** statement we must not write to the original database unless we know
  ** for certain that original page contents are synced into the main rollback
  ** journal.  Otherwise, a power loss might leave modified data in the
  ** database file without an entry in the rollback journal that can
  ** restore the database to its original form.  Two conditions must be
  ** met before writing to the database files. (1) the database must be
  ** locked.  (2) we know that the original page content is fully synced
  ** in the main journal either because the page is not in cache or else
  ** the page is marked as needSync==0.
  */
  pPg = pager_lookup(pPager, pgno);
  PAGERTRACE4("PLAYBACK %d page %d hash(%08x)\n",
               PAGERID(pPager), pgno, pager_datahash(pPager->pageSize, aData));
  if( pPager->state>=PAGER_EXCLUSIVE && (pPg==0 || pPg->needSync==0) ){
    rc = sqlite3OsSeek(pPager->fd, (pgno-1)*(i64)pPager->pageSize);
    if( rc==SQLITE_OK ){
      rc = sqlite3OsWrite(pPager->fd, aData, pPager->pageSize);
    }
    if( pPg ){
      makeClean(pPg);
    }
  }
  if( pPg ){
    /* No page should ever be explicitly rolled back that is in use, except
    ** for page 1 which is held in use in order to keep the lock on the
    ** database active. However such a page may be rolled back as a result
    ** of an internal error resulting in an automatic call to
    ** sqlite3PagerRollback().
    */
    void *pData;
    /* assert( pPg->nRef==0 || pPg->pgno==1 ); */
    pData = PGHDR_TO_DATA(pPg);
    memcpy(pData, aData, pPager->pageSize);
    if( pPager->xReiniter ){
      pPager->xReiniter(pPg, pPager->pageSize);
    }
#ifdef SQLITE_CHECK_PAGES
    pPg->pageHash = pager_pagehash(pPg);
#endif
    /* If this was page 1, then restore the value of Pager.dbFileVers.
    ** Do this before any decoding. */
    if( pgno==1 ){
      memcpy(&pPager->dbFileVers, &((u8*)pData)[24],sizeof(pPager->dbFileVers));
    }

    /* Decode the page just read from disk */
    CODEC1(pPager, pData, pPg->pgno, 3);
  }
  return rc;
}

/*
** Parameter zMaster is the name of a master journal file. A single journal
** file that referred to the master journal file has just been rolled back.
** This routine checks if it is possible to delete the master journal file,
** and does so if it is.
**
** The master journal file contains the names of all child journals.
** To tell if a master journal can be deleted, check to each of the
** children.  If all children are either missing or do not refer to
** a different master journal, then this master journal can be deleted.
*/
static int pager_delmaster(const char *zMaster){
  int rc;
  int master_open = 0;
  OsFile *master = 0;
  char *zMasterJournal = 0; /* Contents of master journal file */
  i64 nMasterJournal;       /* Size of master journal file */

  /* Open the master journal file exclusively in case some other process
  ** is running this routine also. Not that it makes too much difference.
  */
  rc = sqlite3OsOpenReadOnly(zMaster, &master);
  assert( rc!=SQLITE_OK || master );
  if( rc!=SQLITE_OK ) goto delmaster_out;
  master_open = 1;
  rc = sqlite3OsFileSize(master, &nMasterJournal);
  if( rc!=SQLITE_OK ) goto delmaster_out;

  if( nMasterJournal>0 ){
    char *zJournal;
    char *zMasterPtr = 0;

    /* Load the entire master journal file into space obtained from
    ** sqliteMalloc() and pointed to by zMasterJournal. 
    */
    zMasterJournal = (char *)sqliteMalloc(nMasterJournal);
    if( !zMasterJournal ){
      rc = SQLITE_NOMEM;
      goto delmaster_out;
    }
    rc = sqlite3OsRead(master, zMasterJournal, nMasterJournal);
    if( rc!=SQLITE_OK ) goto delmaster_out;

    zJournal = zMasterJournal;
    while( (zJournal-zMasterJournal)<nMasterJournal ){
      if( sqlite3OsFileExists(zJournal) ){
        /* One of the journals pointed to by the master journal exists.
        ** Open it and check if it points at the master journal. If
        ** so, return without deleting the master journal file.
        */
        OsFile *journal = 0;
        int c;

        rc = sqlite3OsOpenReadOnly(zJournal, &journal);
        assert( rc!=SQLITE_OK || journal );
        if( rc!=SQLITE_OK ){
          goto delmaster_out;
        }

        rc = readMasterJournal(journal, &zMasterPtr);
        sqlite3OsClose(&journal);
        if( rc!=SQLITE_OK ){
          goto delmaster_out;
        }

        c = zMasterPtr!=0 && strcmp(zMasterPtr, zMaster)==0;
        sqliteFree(zMasterPtr);
        if( c ){
          /* We have a match. Do not delete the master journal file. */
          goto delmaster_out;
        }
      }
      zJournal += (strlen(zJournal)+1);
    }
  }
  
  rc = sqlite3OsDelete(zMaster);

delmaster_out:
  if( zMasterJournal ){
    sqliteFree(zMasterJournal);
  }  
  if( master_open ){
    sqlite3OsClose(&master);
  }
  return rc;
}


static void pager_truncate_cache(Pager *pPager);

/*
** Truncate the main file of the given pager to the number of pages
** indicated. Also truncate the cached representation of the file.
*/
static int pager_truncate(Pager *pPager, int nPage){
  int rc = SQLITE_OK;
  if( pPager->state>=PAGER_EXCLUSIVE ){
    rc = sqlite3OsTruncate(pPager->fd, pPager->pageSize*(i64)nPage);
  }
  if( rc==SQLITE_OK ){
    pPager->dbSize = nPage;
    pager_truncate_cache(pPager);
  }
  return rc;
}

/*
** Set the sectorSize for the given pager.
**
** The sector size is the larger of the sector size reported
** by sqlite3OsSectorSize() and the pageSize.
*/
static void setSectorSize(Pager *pPager){
  pPager->sectorSize = sqlite3OsSectorSize(pPager->fd);
  if( pPager->sectorSize<pPager->pageSize ){
    pPager->sectorSize = pPager->pageSize;
  }
}

/*
** Playback the journal and thus restore the database file to
** the state it was in before we started making changes.  
**
** The journal file format is as follows: 
**
**  (1)  8 byte prefix.  A copy of aJournalMagic[].
**  (2)  4 byte big-endian integer which is the number of valid page records
**       in the journal.  If this value is 0xffffffff, then compute the
**       number of page records from the journal size.
**  (3)  4 byte big-endian integer which is the initial value for the 
**       sanity checksum.
**  (4)  4 byte integer which is the number of pages to truncate the
**       database to during a rollback.
**  (5)  4 byte integer which is the number of bytes in the master journal
**       name.  The value may be zero (indicate that there is no master
**       journal.)
**  (6)  N bytes of the master journal name.  The name will be nul-terminated
**       and might be shorter than the value read from (5).  If the first byte
**       of the name is \000 then there is no master journal.  The master
**       journal name is stored in UTF-8.
**  (7)  Zero or more pages instances, each as follows:
**        +  4 byte page number.
**        +  pPager->pageSize bytes of data.
**        +  4 byte checksum
**
** When we speak of the journal header, we mean the first 6 items above.
** Each entry in the journal is an instance of the 7th item.
**
** Call the value from the second bullet "nRec".  nRec is the number of
** valid page entries in the journal.  In most cases, you can compute the
** value of nRec from the size of the journal file.  But if a power
** failure occurred while the journal was being written, it could be the
** case that the size of the journal file had already been increased but
** the extra entries had not yet made it safely to disk.  In such a case,
** the value of nRec computed from the file size would be too large.  For
** that reason, we always use the nRec value in the header.
**
** If the nRec value is 0xffffffff it means that nRec should be computed
** from the file size.  This value is used when the user selects the
** no-sync option for the journal.  A power failure could lead to corruption
** in this case.  But for things like temporary table (which will be
** deleted when the power is restored) we don't care.  
**
** If the file opened as the journal file is not a well-formed
** journal file then all pages up to the first corrupted page are rolled
** back (or no pages if the journal header is corrupted). The journal file
** is then deleted and SQLITE_OK returned, just as if no corruption had
** been encountered.
**
** If an I/O or malloc() error occurs, the journal-file is not deleted
** and an error code is returned.
*/
static int pager_playback(Pager *pPager, int isHot){
  i64 szJ;                 /* Size of the journal file in bytes */
  u32 nRec;                /* Number of Records in the journal */
  int i;                   /* Loop counter */
  Pgno mxPg = 0;           /* Size of the original file in pages */
  int rc;                  /* Result code of a subroutine */
  char *zMaster = 0;       /* Name of master journal file if any */

  /* Figure out how many records are in the journal.  Abort early if
  ** the journal is empty.
  */
  assert( pPager->journalOpen );
  rc = sqlite3OsFileSize(pPager->jfd, &szJ);
  if( rc!=SQLITE_OK || szJ==0 ){
    goto end_playback;
  }

  /* Read the master journal name from the journal, if it is present.
  ** If a master journal file name is specified, but the file is not
  ** present on disk, then the journal is not hot and does not need to be
  ** played back.
  */
  rc = readMasterJournal(pPager->jfd, &zMaster);
  assert( rc!=SQLITE_DONE );
  if( rc!=SQLITE_OK || (zMaster && !sqlite3OsFileExists(zMaster)) ){
    sqliteFree(zMaster);
    zMaster = 0;
    if( rc==SQLITE_DONE ) rc = SQLITE_OK;
    goto end_playback;
  }
  sqlite3OsSeek(pPager->jfd, 0);
  pPager->journalOff = 0;

  /* This loop terminates either when the readJournalHdr() call returns
  ** SQLITE_DONE or an IO error occurs. */
  while( 1 ){

    /* Read the next journal header from the journal file.  If there are
    ** not enough bytes left in the journal file for a complete header, or
    ** it is corrupted, then a process must of failed while writing it.
    ** This indicates nothing more needs to be rolled back.
    */
    rc = readJournalHdr(pPager, szJ, &nRec, &mxPg);
    if( rc!=SQLITE_OK ){ 
      if( rc==SQLITE_DONE ){
        rc = SQLITE_OK;
      }
      goto end_playback;
    }

    /* If nRec is 0xffffffff, then this journal was created by a process
    ** working in no-sync mode. This means that the rest of the journal
    ** file consists of pages, there are no more journal headers. Compute
    ** the value of nRec based on this assumption.
    */
    if( nRec==0xffffffff ){
      assert( pPager->journalOff==JOURNAL_HDR_SZ(pPager) );
      nRec = (szJ - JOURNAL_HDR_SZ(pPager))/JOURNAL_PG_SZ(pPager);
    }

    /* If nRec is 0 and this rollback is of a transaction created by this
    ** process. In this case the rest of the journal file consists of
    ** journalled copies of pages that need to be read back into the cache.
    */
    if( nRec==0 && !isHot ){
      nRec = (szJ - pPager->journalOff) / JOURNAL_PG_SZ(pPager);
    }

    /* If this is the first header read from the journal, truncate the
    ** database file back to it's original size.
    */
    if( pPager->journalOff==JOURNAL_HDR_SZ(pPager) ){
      rc = pager_truncate(pPager, mxPg);
      if( rc!=SQLITE_OK ){
        goto end_playback;
      }
    }

    /* Copy original pages out of the journal and back into the database file.
    */
    for(i=0; i<nRec; i++){
      rc = pager_playback_one_page(pPager, pPager->jfd, 1);
      if( rc!=SQLITE_OK ){
        if( rc==SQLITE_DONE ){
          rc = SQLITE_OK;
          pPager->journalOff = szJ;
          break;
        }else{
          goto end_playback;
        }
      }
    }
  }
  /*NOTREACHED*/
  assert( 0 );

end_playback:
  if( rc==SQLITE_OK ){
    rc = pager_end_transaction(pPager);
  }
  if( zMaster ){
    /* If there was a master journal and this routine will return success,
    ** see if it is possible to delete the master journal.
    */
    if( rc==SQLITE_OK ){
      rc = pager_delmaster(zMaster);
    }
    sqliteFree(zMaster);
  }

  /* The Pager.sectorSize variable may have been updated while rolling
  ** back a journal created by a process with a different sector size
  ** value. Reset it to the correct value for this process.
  */
  setSectorSize(pPager);
  return rc;
}

/*
** Playback the statement journal.
**
** This is similar to playing back the transaction journal but with
** a few extra twists.
**
**    (1)  The number of pages in the database file at the start of
**         the statement is stored in pPager->stmtSize, not in the
**         journal file itself.
**
**    (2)  In addition to playing back the statement journal, also
**         playback all pages of the transaction journal beginning
**         at offset pPager->stmtJSize.
*/
static int pager_stmt_playback(Pager *pPager){
  i64 szJ;                 /* Size of the full journal */
  i64 hdrOff;
  int nRec;                /* Number of Records */
  int i;                   /* Loop counter */
  int rc;

  szJ = pPager->journalOff;
#ifndef NDEBUG 
  {
    i64 os_szJ;
    rc = sqlite3OsFileSize(pPager->jfd, &os_szJ);
    if( rc!=SQLITE_OK ) return rc;
    assert( szJ==os_szJ );
  }
#endif

  /* Set hdrOff to be the offset just after the end of the last journal
  ** page written before the first journal-header for this statement
  ** transaction was written, or the end of the file if no journal
  ** header was written.
  */
  hdrOff = pPager->stmtHdrOff;
  assert( pPager->fullSync || !hdrOff );
  if( !hdrOff ){
    hdrOff = szJ;
  }
  
  /* Truncate the database back to its original size.
  */
  rc = pager_truncate(pPager, pPager->stmtSize);
  assert( pPager->state>=PAGER_SHARED );

  /* Figure out how many records are in the statement journal.
  */
  assert( pPager->stmtInUse && pPager->journalOpen );
  sqlite3OsSeek(pPager->stfd, 0);
  nRec = pPager->stmtNRec;
  
  /* Copy original pages out of the statement journal and back into the
  ** database file.  Note that the statement journal omits checksums from
  ** each record since power-failure recovery is not important to statement
  ** journals.
  */
  for(i=nRec-1; i>=0; i--){
    rc = pager_playback_one_page(pPager, pPager->stfd, 0);
    assert( rc!=SQLITE_DONE );
    if( rc!=SQLITE_OK ) goto end_stmt_playback;
  }

  /* Now roll some pages back from the transaction journal. Pager.stmtJSize
  ** was the size of the journal file when this statement was started, so
  ** everything after that needs to be rolled back, either into the
  ** database, the memory cache, or both.
  **
  ** If it is not zero, then Pager.stmtHdrOff is the offset to the start
  ** of the first journal header written during this statement transaction.
  */
  rc = sqlite3OsSeek(pPager->jfd, pPager->stmtJSize);
  if( rc!=SQLITE_OK ){
    goto end_stmt_playback;
  }
  pPager->journalOff = pPager->stmtJSize;
  pPager->cksumInit = pPager->stmtCksum;
  while( pPager->journalOff < hdrOff ){
    rc = pager_playback_one_page(pPager, pPager->jfd, 1);
    assert( rc!=SQLITE_DONE );
    if( rc!=SQLITE_OK ) goto end_stmt_playback;
  }

  while( pPager->journalOff < szJ ){
    u32 nJRec;         /* Number of Journal Records */
    u32 dummy;
    rc = readJournalHdr(pPager, szJ, &nJRec, &dummy);
    if( rc!=SQLITE_OK ){
      assert( rc!=SQLITE_DONE );
      goto end_stmt_playback;
    }
    if( nJRec==0 ){
      nJRec = (szJ - pPager->journalOff) / (pPager->pageSize+8);
    }
    for(i=nJRec-1; i>=0 && pPager->journalOff < szJ; i--){
      rc = pager_playback_one_page(pPager, pPager->jfd, 1);
      assert( rc!=SQLITE_DONE );
      if( rc!=SQLITE_OK ) goto end_stmt_playback;
    }
  }

  pPager->journalOff = szJ;
  
end_stmt_playback:
  if( rc==SQLITE_OK) {
    pPager->journalOff = szJ;
    /* pager_reload_cache(pPager); */
  }
  return rc;
}

/*
** Change the maximum number of in-memory pages that are allowed.
*/
SQLITE_PRIVATE void sqlite3PagerSetCachesize(Pager *pPager, int mxPage){
  if( mxPage>10 ){
    pPager->mxPage = mxPage;
  }else{
    pPager->mxPage = 10;
  }
}

/*
** Adjust the robustness of the database to damage due to OS crashes
** or power failures by changing the number of syncs()s when writing
** the rollback journal.  There are three levels:
**
**    OFF       sqlite3OsSync() is never called.  This is the default
**              for temporary and transient files.
**
**    NORMAL    The journal is synced once before writes begin on the
**              database.  This is normally adequate protection, but
**              it is theoretically possible, though very unlikely,
**              that an inopertune power failure could leave the journal
**              in a state which would cause damage to the database
**              when it is rolled back.
**
**    FULL      The journal is synced twice before writes begin on the
**              database (with some additional information - the nRec field
**              of the journal header - being written in between the two
**              syncs).  If we assume that writing a
**              single disk sector is atomic, then this mode provides
**              assurance that the journal will not be corrupted to the
**              point of causing damage to the database during rollback.
**
** Numeric values associated with these states are OFF==1, NORMAL=2,
** and FULL=3.
*/
#ifndef SQLITE_OMIT_PAGER_PRAGMAS
SQLITE_PRIVATE void sqlite3PagerSetSafetyLevel(Pager *pPager, int level, int full_fsync){
  pPager->noSync =  level==1 || pPager->tempFile;
  pPager->fullSync = level==3 && !pPager->tempFile;
  pPager->full_fsync = full_fsync;
  if( pPager->noSync ) pPager->needSync = 0;
}
#endif

/*
** The following global variable is incremented whenever the library
** attempts to open a temporary file.  This information is used for
** testing and analysis only.  
*/
#ifdef SQLITE_TEST
int sqlite3_opentemp_count = 0;
#endif

/*
** Open a temporary file. 
**
** Write the file descriptor into *fd.  Return SQLITE_OK on success or some
** other error code if we fail.
**
** The OS will automatically delete the temporary file when it is
** closed.
*/
static int sqlite3PagerOpentemp(OsFile **pFd){
  int cnt = 8;
  int rc;
  char zFile[SQLITE_TEMPNAME_SIZE];

#ifdef SQLITE_TEST
  sqlite3_opentemp_count++;  /* Used for testing and analysis only */
#endif
  do{
    cnt--;
    sqlite3OsTempFileName(zFile);
    rc = sqlite3OsOpenExclusive(zFile, pFd, 1);
    assert( rc!=SQLITE_OK || *pFd );
  }while( cnt>0 && rc!=SQLITE_OK && rc!=SQLITE_NOMEM );
  return rc;
}

/*
** Create a new page cache and put a pointer to the page cache in *ppPager.
** The file to be cached need not exist.  The file is not locked until
** the first call to sqlite3PagerGet() and is only held open until the
** last page is released using sqlite3PagerUnref().
**
** If zFilename is NULL then a randomly-named temporary file is created
** and used as the file to be cached.  The file will be deleted
** automatically when it is closed.
**
** If zFilename is ":memory:" then all information is held in cache.
** It is never written to disk.  This can be used to implement an
** in-memory database.
*/
SQLITE_PRIVATE int sqlite3PagerOpen(
  Pager **ppPager,         /* Return the Pager structure here */
  const char *zFilename,   /* Name of the database file to open */
  int nExtra,              /* Extra bytes append to each in-memory page */
  int flags                /* flags controlling this file */
){
  Pager *pPager = 0;
  char *zFullPathname = 0;
  int nameLen;  /* Compiler is wrong. This is always initialized before use */
  OsFile *fd = 0;
  int rc = SQLITE_OK;
  int i;
  int tempFile = 0;
  int memDb = 0;
  int readOnly = 0;
  int useJournal = (flags & PAGER_OMIT_JOURNAL)==0;
  int noReadlock = (flags & PAGER_NO_READLOCK)!=0;
  char zTemp[SQLITE_TEMPNAME_SIZE];
#ifdef SQLITE_ENABLE_MEMORY_MANAGEMENT
  /* A malloc() cannot fail in sqlite3ThreadData() as one or more calls to 
  ** malloc() must have already been made by this thread before it gets
  ** to this point. This means the ThreadData must have been allocated already
  ** so that ThreadData.nAlloc can be set. It would be nice to assert
  ** that ThreadData.nAlloc is non-zero, but alas this breaks test cases 
  ** written to invoke the pager directly.
  */
  ThreadData *pTsd = sqlite3ThreadData();
  assert( pTsd );
#endif

  /* We used to test if malloc() had already failed before proceeding. 
  ** But the way this function is used in SQLite means that can never
  ** happen. Furthermore, if the malloc-failed flag is already set, 
  ** either the call to sqliteStrDup() or sqliteMalloc() below will
  ** fail shortly and SQLITE_NOMEM returned anyway.
  */
  *ppPager = 0;

  /* Open the pager file and set zFullPathname to point at malloc()ed 
  ** memory containing the complete filename (i.e. including the directory).
  */
  if( zFilename && zFilename[0] ){
#ifndef SQLITE_OMIT_MEMORYDB
    if( strcmp(zFilename,":memory:")==0 ){
      memDb = 1;
      zFullPathname = sqliteStrDup("");
    }else
#endif
    {
      zFullPathname = sqlite3OsFullPathname(zFilename);
      if( zFullPathname ){
        rc = sqlite3OsOpenReadWrite(zFullPathname, &fd, &readOnly);
        assert( rc!=SQLITE_OK || fd );
      }
    }
  }else{
    rc = sqlite3PagerOpentemp(&fd);
    sqlite3OsTempFileName(zTemp);
    zFilename = zTemp;
    zFullPathname = sqlite3OsFullPathname(zFilename);
    if( rc==SQLITE_OK ){
      tempFile = 1;
    }
  }

  /* Allocate the Pager structure. As part of the same allocation, allocate
  ** space for the full paths of the file, directory and journal 
  ** (Pager.zFilename, Pager.zDirectory and Pager.zJournal).
  */
  if( zFullPathname ){
    nameLen = strlen(zFullPathname);
    pPager = sqliteMalloc( sizeof(*pPager) + nameLen*3 + 30 );
    if( pPager && rc==SQLITE_OK ){
      pPager->pTmpSpace = (char *)sqliteMallocRaw(SQLITE_DEFAULT_PAGE_SIZE);
    }
  }


  /* If an error occured in either of the blocks above, free the memory 
  ** pointed to by zFullPathname, free the Pager structure and close the 
  ** file. Since the pager is not allocated there is no need to set 
  ** any Pager.errMask variables.
  */
  if( !pPager || !zFullPathname || !pPager->pTmpSpace || rc!=SQLITE_OK ){
    sqlite3OsClose(&fd);
    sqliteFree(zFullPathname);
    sqliteFree(pPager);
    return ((rc==SQLITE_OK)?SQLITE_NOMEM:rc);
  }

  PAGERTRACE3("OPEN %d %s\n", FILEHANDLEID(fd), zFullPathname);
  IOTRACE(("OPEN %p %s\n", pPager, zFullPathname))
  pPager->zFilename = (char*)&pPager[1];
  pPager->zDirectory = &pPager->zFilename[nameLen+1];
  pPager->zJournal = &pPager->zDirectory[nameLen+1];
  memcpy(pPager->zFilename, zFullPathname, nameLen+1);
  memcpy(pPager->zDirectory, zFullPathname, nameLen+1);

  for(i=nameLen; i>0 && pPager->zDirectory[i-1]!='/'; i--){}
  if( i>0 ) pPager->zDirectory[i-1] = 0;
  memcpy(pPager->zJournal, zFullPathname,nameLen);
  sqliteFree(zFullPathname);
  memcpy(&pPager->zJournal[nameLen], "-journal",sizeof("-journal"));
  pPager->fd = fd;
  /* pPager->journalOpen = 0; */
  pPager->useJournal = useJournal && !memDb;
  pPager->noReadlock = noReadlock && readOnly;
  /* pPager->stmtOpen = 0; */
  /* pPager->stmtInUse = 0; */
  /* pPager->nRef = 0; */
  pPager->dbSize = memDb-1;
  pPager->pageSize = SQLITE_DEFAULT_PAGE_SIZE;
  /* pPager->stmtSize = 0; */
  /* pPager->stmtJSize = 0; */
  /* pPager->nPage = 0; */
  pPager->mxPage = 100;
  pPager->mxPgno = SQLITE_MAX_PAGE_COUNT;
  assert( PAGER_UNLOCK==0 );
  /* pPager->state = PAGER_UNLOCK; */
  /* pPager->errMask = 0; */
  pPager->tempFile = tempFile;
  assert( tempFile==PAGER_LOCKINGMODE_NORMAL 
          || tempFile==PAGER_LOCKINGMODE_EXCLUSIVE );
  assert( PAGER_LOCKINGMODE_EXCLUSIVE==1 );
  pPager->exclusiveMode = tempFile; 
  pPager->memDb = memDb;
  pPager->readOnly = readOnly;
  /* pPager->needSync = 0; */
  pPager->noSync = pPager->tempFile || !useJournal;
  pPager->fullSync = (pPager->noSync?0:1);
  /* pPager->pFirst = 0; */
  /* pPager->pFirstSynced = 0; */
  /* pPager->pLast = 0; */
  pPager->nExtra = FORCE_ALIGNMENT(nExtra);
  assert(fd||memDb);
  if( !memDb ){
    setSectorSize(pPager);
  }
  /* pPager->pBusyHandler = 0; */
  /* memset(pPager->aHash, 0, sizeof(pPager->aHash)); */
  *ppPager = pPager;
#ifdef SQLITE_ENABLE_MEMORY_MANAGEMENT
  pPager->pNext = pTsd->pPager;
  pTsd->pPager = pPager;
#endif
  return SQLITE_OK;
}

/*
** Set the busy handler function.
*/
SQLITE_PRIVATE void sqlite3PagerSetBusyhandler(Pager *pPager, BusyHandler *pBusyHandler){
  pPager->pBusyHandler = pBusyHandler;
}

/*
** Set the destructor for this pager.  If not NULL, the destructor is called
** when the reference count on each page reaches zero.  The destructor can
** be used to clean up information in the extra segment appended to each page.
**
** The destructor is not called as a result sqlite3PagerClose().  
** Destructors are only called by sqlite3PagerUnref().
*/
SQLITE_PRIVATE void sqlite3PagerSetDestructor(Pager *pPager, void (*xDesc)(DbPage*,int)){
  pPager->xDestructor = xDesc;
}

/*
** Set the reinitializer for this pager.  If not NULL, the reinitializer
** is called when the content of a page in cache is restored to its original
** value as a result of a rollback.  The callback gives higher-level code
** an opportunity to restore the EXTRA section to agree with the restored
** page data.
*/
SQLITE_PRIVATE void sqlite3PagerSetReiniter(Pager *pPager, void (*xReinit)(DbPage*,int)){
  pPager->xReiniter = xReinit;
}

/*
** Set the page size.  Return the new size.  If the suggest new page
** size is inappropriate, then an alternative page size is selected
** and returned.
*/
SQLITE_PRIVATE int sqlite3PagerSetPagesize(Pager *pPager, int pageSize){
  assert( pageSize>=512 && pageSize<=SQLITE_MAX_PAGE_SIZE );
  if( !pPager->memDb && pPager->nRef==0 ){
    pager_reset(pPager);
    pPager->pageSize = pageSize;
    pPager->pTmpSpace = sqlite3ReallocOrFree(pPager->pTmpSpace, pageSize);
  }
  return pPager->pageSize;
}

/*
** Attempt to set the maximum database page count if mxPage is positive. 
** Make no changes if mxPage is zero or negative.  And never reduce the
** maximum page count below the current size of the database.
**
** Regardless of mxPage, return the current maximum page count.
*/
SQLITE_PRIVATE int sqlite3PagerMaxPageCount(Pager *pPager, int mxPage){
  if( mxPage>0 ){
    pPager->mxPgno = mxPage;
  }
  sqlite3PagerPagecount(pPager);
  return pPager->mxPgno;
}

/*
** The following set of routines are used to disable the simulated
** I/O error mechanism.  These routines are used to avoid simulated
** errors in places where we do not care about errors.
**
** Unless -DSQLITE_TEST=1 is used, these routines are all no-ops
** and generate no code.
*/
#ifdef SQLITE_TEST
extern int sqlite3_io_error_pending;
extern int sqlite3_io_error_hit;
static int saved_cnt;
void disable_simulated_io_errors(void){
  saved_cnt = sqlite3_io_error_pending;
  sqlite3_io_error_pending = -1;
}
void enable_simulated_io_errors(void){
  sqlite3_io_error_pending = saved_cnt;
}
#else
# define disable_simulated_io_errors()
# define enable_simulated_io_errors()
#endif

/*
** Read the first N bytes from the beginning of the file into memory
** that pDest points to. 
**
** No error checking is done. The rational for this is that this function 
** may be called even if the file does not exist or contain a header. In 
** these cases sqlite3OsRead() will return an error, to which the correct 
** response is to zero the memory at pDest and continue.  A real IO error 
** will presumably recur and be picked up later (Todo: Think about this).
*/
SQLITE_PRIVATE int sqlite3PagerReadFileheader(Pager *pPager, int N, unsigned char *pDest){
  int rc = SQLITE_OK;
  memset(pDest, 0, N);
  if( MEMDB==0 ){
    disable_simulated_io_errors();
    sqlite3OsSeek(pPager->fd, 0);
    enable_simulated_io_errors();
    IOTRACE(("DBHDR %p 0 %d\n", pPager, N))
    rc = sqlite3OsRead(pPager->fd, pDest, N);
    if( rc==SQLITE_IOERR_SHORT_READ ){
      rc = SQLITE_OK;
    }
  }
  return rc;
}

/*
** Return the total number of pages in the disk file associated with
** pPager. 
**
** If the PENDING_BYTE lies on the page directly after the end of the
** file, then consider this page part of the file too. For example, if
** PENDING_BYTE is byte 4096 (the first byte of page 5) and the size of the
** file is 4096 bytes, 5 is returned instead of 4.
*/
SQLITE_PRIVATE int sqlite3PagerPagecount(Pager *pPager){
  i64 n;
  int rc;
  assert( pPager!=0 );
  if( pPager->errCode ){
    return 0;
  }
  if( pPager->dbSize>=0 ){
    n = pPager->dbSize;
  } else {
    if( (rc = sqlite3OsFileSize(pPager->fd, &n))!=SQLITE_OK ){
      pager_error(pPager, rc);
      return 0;
    }
    if( n>0 && n<pPager->pageSize ){
      n = 1;
    }else{
      n /= pPager->pageSize;
    }
    if( pPager->state!=PAGER_UNLOCK ){
      pPager->dbSize = n;
    }
  }
  if( n==(PENDING_BYTE/pPager->pageSize) ){
    n++;
  }
  if( n>pPager->mxPgno ){
    pPager->mxPgno = n;
  }
  return n;
}


#ifndef SQLITE_OMIT_MEMORYDB
/*
** Clear a PgHistory block
*/
static void clearHistory(PgHistory *pHist){
  sqliteFree(pHist->pOrig);
  sqliteFree(pHist->pStmt);
  pHist->pOrig = 0;
  pHist->pStmt = 0;
}
#else
#define clearHistory(x)
#endif

/*
** Forward declaration
*/
static int syncJournal(Pager*);

/*
** Unlink pPg from it's hash chain. Also set the page number to 0 to indicate
** that the page is not part of any hash chain. This is required because the
** sqlite3PagerMovepage() routine can leave a page in the 
** pNextFree/pPrevFree list that is not a part of any hash-chain.
*/
static void unlinkHashChain(Pager *pPager, PgHdr *pPg){
  if( pPg->pgno==0 ){
    assert( pPg->pNextHash==0 && pPg->pPrevHash==0 );
    return;
  }
  if( pPg->pNextHash ){
    pPg->pNextHash->pPrevHash = pPg->pPrevHash;
  }
  if( pPg->pPrevHash ){
    assert( pPager->aHash[pPg->pgno & (pPager->nHash-1)]!=pPg );
    pPg->pPrevHash->pNextHash = pPg->pNextHash;
  }else{
    int h = pPg->pgno & (pPager->nHash-1);
    pPager->aHash[h] = pPg->pNextHash;
  }
  if( MEMDB ){
    clearHistory(PGHDR_TO_HIST(pPg, pPager));
  }
  pPg->pgno = 0;
  pPg->pNextHash = pPg->pPrevHash = 0;
}

/*
** Unlink a page from the free list (the list of all pages where nRef==0)
** and from its hash collision chain.
*/
static void unlinkPage(PgHdr *pPg){
  Pager *pPager = pPg->pPager;

  /* Keep the pFirstSynced pointer pointing at the first synchronized page */
  if( pPg==pPager->pFirstSynced ){
    PgHdr *p = pPg->pNextFree;
    while( p && p->needSync ){ p = p->pNextFree; }
    pPager->pFirstSynced = p;
  }

  /* Unlink from the freelist */
  if( pPg->pPrevFree ){
    pPg->pPrevFree->pNextFree = pPg->pNextFree;
  }else{
    assert( pPager->pFirst==pPg );
    pPager->pFirst = pPg->pNextFree;
  }
  if( pPg->pNextFree ){
    pPg->pNextFree->pPrevFree = pPg->pPrevFree;
  }else{
    assert( pPager->pLast==pPg );
    pPager->pLast = pPg->pPrevFree;
  }
  pPg->pNextFree = pPg->pPrevFree = 0;

  /* Unlink from the pgno hash table */
  unlinkHashChain(pPager, pPg);
}

/*
** This routine is used to truncate the cache when a database
** is truncated.  Drop from the cache all pages whose pgno is
** larger than pPager->dbSize and is unreferenced.
**
** Referenced pages larger than pPager->dbSize are zeroed.
**
** Actually, at the point this routine is called, it would be
** an error to have a referenced page.  But rather than delete
** that page and guarantee a subsequent segfault, it seems better
** to zero it and hope that we error out sanely.
*/
static void pager_truncate_cache(Pager *pPager){
  PgHdr *pPg;
  PgHdr **ppPg;
  int dbSize = pPager->dbSize;

  ppPg = &pPager->pAll;
  while( (pPg = *ppPg)!=0 ){
    if( pPg->pgno<=dbSize ){
      ppPg = &pPg->pNextAll;
    }else if( pPg->nRef>0 ){
      memset(PGHDR_TO_DATA(pPg), 0, pPager->pageSize);
      ppPg = &pPg->pNextAll;
    }else{
      *ppPg = pPg->pNextAll;
      IOTRACE(("PGFREE %p %d\n", pPager, pPg->pgno));
      PAGER_INCR(sqlite3_pager_pgfree_count);
      unlinkPage(pPg);
      makeClean(pPg);
      sqliteFree(pPg);
      pPager->nPage--;
    }
  }
}

/*
** Try to obtain a lock on a file.  Invoke the busy callback if the lock
** is currently not available.  Repeat until the busy callback returns
** false or until the lock succeeds.
**
** Return SQLITE_OK on success and an error code if we cannot obtain
** the lock.
*/
static int pager_wait_on_lock(Pager *pPager, int locktype){
  int rc;

  /* The OS lock values must be the same as the Pager lock values */
  assert( PAGER_SHARED==SHARED_LOCK );
  assert( PAGER_RESERVED==RESERVED_LOCK );
  assert( PAGER_EXCLUSIVE==EXCLUSIVE_LOCK );

  /* If the file is currently unlocked then the size must be unknown */
  assert( pPager->state>=PAGER_SHARED || pPager->dbSize<0 || MEMDB );

  if( pPager->state>=locktype ){
    rc = SQLITE_OK;
  }else{
    do {
      rc = sqlite3OsLock(pPager->fd, locktype);
    }while( rc==SQLITE_BUSY && sqlite3InvokeBusyHandler(pPager->pBusyHandler) );
    if( rc==SQLITE_OK ){
      pPager->state = locktype;
      IOTRACE(("LOCK %p %d\n", pPager, locktype))
    }
  }
  return rc;
}

/*
** Truncate the file to the number of pages specified.
*/
SQLITE_PRIVATE int sqlite3PagerTruncate(Pager *pPager, Pgno nPage){
  int rc;
  assert( pPager->state>=PAGER_SHARED || MEMDB );
  sqlite3PagerPagecount(pPager);
  if( pPager->errCode ){
    rc = pPager->errCode;
    return rc;
  }
  if( nPage>=(unsigned)pPager->dbSize ){
    return SQLITE_OK;
  }
  if( MEMDB ){
    pPager->dbSize = nPage;
    pager_truncate_cache(pPager);
    return SQLITE_OK;
  }
  rc = syncJournal(pPager);
  if( rc!=SQLITE_OK ){
    return rc;
  }

  /* Get an exclusive lock on the database before truncating. */
  rc = pager_wait_on_lock(pPager, EXCLUSIVE_LOCK);
  if( rc!=SQLITE_OK ){
    return rc;
  }

  rc = pager_truncate(pPager, nPage);
  return rc;
}

/*
** Shutdown the page cache.  Free all memory and close all files.
**
** If a transaction was in progress when this routine is called, that
** transaction is rolled back.  All outstanding pages are invalidated
** and their memory is freed.  Any attempt to use a page associated
** with this page cache after this function returns will likely
** result in a coredump.
**
** This function always succeeds. If a transaction is active an attempt
** is made to roll it back. If an error occurs during the rollback 
** a hot journal may be left in the filesystem but no error is returned
** to the caller.
*/
SQLITE_PRIVATE int sqlite3PagerClose(Pager *pPager){
#ifdef SQLITE_ENABLE_MEMORY_MANAGEMENT
  /* A malloc() cannot fail in sqlite3ThreadData() as one or more calls to 
  ** malloc() must have already been made by this thread before it gets
  ** to this point. This means the ThreadData must have been allocated already
  ** so that ThreadData.nAlloc can be set.
  */
  ThreadData *pTsd = sqlite3ThreadData();
  assert( pPager );
  assert( pTsd && pTsd->nAlloc );
#endif

  disable_simulated_io_errors();
  pPager->errCode = 0;
  pPager->exclusiveMode = 0;
  pager_reset(pPager);
  pagerUnlockAndRollback(pPager);
  enable_simulated_io_errors();
  PAGERTRACE2("CLOSE %d\n", PAGERID(pPager));
  IOTRACE(("CLOSE %p\n", pPager))
  assert( pPager->errCode || (pPager->journalOpen==0 && pPager->stmtOpen==0) );
  if( pPager->journalOpen ){
    sqlite3OsClose(&pPager->jfd);
  }
  sqliteFree(pPager->aInJournal);
  if( pPager->stmtOpen ){
    sqlite3OsClose(&pPager->stfd);
  }
  sqlite3OsClose(&pPager->fd);
  /* Temp files are automatically deleted by the OS
  ** if( pPager->tempFile ){
  **   sqlite3OsDelete(pPager->zFilename);
  ** }
  */

#ifdef SQLITE_ENABLE_MEMORY_MANAGEMENT
  /* Remove the pager from the linked list of pagers starting at 
  ** ThreadData.pPager if memory-management is enabled.
  */
  if( pPager==pTsd->pPager ){
    pTsd->pPager = pPager->pNext;
  }else{
    Pager *pTmp;
    for(pTmp = pTsd->pPager; pTmp->pNext!=pPager; pTmp=pTmp->pNext){}
    pTmp->pNext = pPager->pNext;
  }
#endif
  sqliteFree(pPager->aHash);
  sqliteFree(pPager->pTmpSpace);
  sqliteFree(pPager);
  return SQLITE_OK;
}

#if !defined(NDEBUG) || defined(SQLITE_TEST)
/*
** Return the page number for the given page data.
*/
SQLITE_PRIVATE Pgno sqlite3PagerPagenumber(DbPage *p){
  return p->pgno;
}
#endif

/*
** The page_ref() function increments the reference count for a page.
** If the page is currently on the freelist (the reference count is zero) then
** remove it from the freelist.
**
** For non-test systems, page_ref() is a macro that calls _page_ref()
** online of the reference count is zero.  For test systems, page_ref()
** is a real function so that we can set breakpoints and trace it.
*/
static void _page_ref(PgHdr *pPg){
  if( pPg->nRef==0 ){
    /* The page is currently on the freelist.  Remove it. */
    if( pPg==pPg->pPager->pFirstSynced ){
      PgHdr *p = pPg->pNextFree;
      while( p && p->needSync ){ p = p->pNextFree; }
      pPg->pPager->pFirstSynced = p;
    }
    if( pPg->pPrevFree ){
      pPg->pPrevFree->pNextFree = pPg->pNextFree;
    }else{
      pPg->pPager->pFirst = pPg->pNextFree;
    }
    if( pPg->pNextFree ){
      pPg->pNextFree->pPrevFree = pPg->pPrevFree;
    }else{
      pPg->pPager->pLast = pPg->pPrevFree;
    }
    pPg->pPager->nRef++;
  }
  pPg->nRef++;
  REFINFO(pPg);
}
#ifdef SQLITE_DEBUG
  static void page_ref(PgHdr *pPg){
    if( pPg->nRef==0 ){
      _page_ref(pPg);
    }else{
      pPg->nRef++;
      REFINFO(pPg);
    }
  }
#else
# define page_ref(P)   ((P)->nRef==0?_page_ref(P):(void)(P)->nRef++)
#endif

/*
** Increment the reference count for a page.  The input pointer is
** a reference to the page data.
*/
SQLITE_PRIVATE int sqlite3PagerRef(DbPage *pPg){
  page_ref(pPg);
  return SQLITE_OK;
}

/*
** Sync the journal.  In other words, make sure all the pages that have
** been written to the journal have actually reached the surface of the
** disk.  It is not safe to modify the original database file until after
** the journal has been synced.  If the original database is modified before
** the journal is synced and a power failure occurs, the unsynced journal
** data would be lost and we would be unable to completely rollback the
** database changes.  Database corruption would occur.
** 
** This routine also updates the nRec field in the header of the journal.
** (See comments on the pager_playback() routine for additional information.)
** If the sync mode is FULL, two syncs will occur.  First the whole journal
** is synced, then the nRec field is updated, then a second sync occurs.
**
** For temporary databases, we do not care if we are able to rollback
** after a power failure, so sync occurs.
**
** This routine clears the needSync field of every page current held in
** memory.
*/
static int syncJournal(Pager *pPager){
  PgHdr *pPg;
  int rc = SQLITE_OK;

  /* Sync the journal before modifying the main database
  ** (assuming there is a journal and it needs to be synced.)
  */
  if( pPager->needSync ){
    if( !pPager->tempFile ){
      assert( pPager->journalOpen );
      /* assert( !pPager->noSync ); // noSync might be set if synchronous
      ** was turned off after the transaction was started.  Ticket #615 */
#ifndef NDEBUG
      {
        /* Make sure the pPager->nRec counter we are keeping agrees
        ** with the nRec computed from the size of the journal file.
        */
        i64 jSz;
        rc = sqlite3OsFileSize(pPager->jfd, &jSz);
        if( rc!=0 ) return rc;
        assert( pPager->journalOff==jSz );
      }
#endif
      {
        /* Write the nRec value into the journal file header. If in
        ** full-synchronous mode, sync the journal first. This ensures that
        ** all data has really hit the disk before nRec is updated to mark
        ** it as a candidate for rollback. 
        */
        if( pPager->fullSync ){
          PAGERTRACE2("SYNC journal of %d\n", PAGERID(pPager));
          IOTRACE(("JSYNC %p\n", pPager))
          rc = sqlite3OsSync(pPager->jfd, 0);
          if( rc!=0 ) return rc;
        }
        rc = sqlite3OsSeek(pPager->jfd,
                           pPager->journalHdr + sizeof(aJournalMagic));
        if( rc ) return rc;
        IOTRACE(("JHDR %p %lld %d\n", pPager,
                  pPager->journalHdr + sizeof(aJournalMagic), 4))
        rc = write32bits(pPager->jfd, pPager->nRec);
        if( rc ) return rc;

        rc = sqlite3OsSeek(pPager->jfd, pPager->journalOff);
        if( rc ) return rc;
      }
      PAGERTRACE2("SYNC journal of %d\n", PAGERID(pPager));
      IOTRACE(("JSYNC %p\n", pPager))
      rc = sqlite3OsSync(pPager->jfd, pPager->full_fsync);
      if( rc!=0 ) return rc;
      pPager->journalStarted = 1;
    }
    pPager->needSync = 0;

    /* Erase the needSync flag from every page.
    */
    for(pPg=pPager->pAll; pPg; pPg=pPg->pNextAll){
      pPg->needSync = 0;
    }
    pPager->pFirstSynced = pPager->pFirst;
  }

#ifndef NDEBUG
  /* If the Pager.needSync flag is clear then the PgHdr.needSync
  ** flag must also be clear for all pages.  Verify that this
  ** invariant is true.
  */
  else{
    for(pPg=pPager->pAll; pPg; pPg=pPg->pNextAll){
      assert( pPg->needSync==0 );
    }
    assert( pPager->pFirstSynced==pPager->pFirst );
  }
#endif

  return rc;
}

/*
** Merge two lists of pages connected by pDirty and in pgno order.
** Do not both fixing the pPrevDirty pointers.
*/
static PgHdr *merge_pagelist(PgHdr *pA, PgHdr *pB){
  PgHdr result, *pTail;
  pTail = &result;
  while( pA && pB ){
    if( pA->pgno<pB->pgno ){
      pTail->pDirty = pA;
      pTail = pA;
      pA = pA->pDirty;
    }else{
      pTail->pDirty = pB;
      pTail = pB;
      pB = pB->pDirty;
    }
  }
  if( pA ){
    pTail->pDirty = pA;
  }else if( pB ){
    pTail->pDirty = pB;
  }else{
    pTail->pDirty = 0;
  }
  return result.pDirty;
}

/*
** Sort the list of pages in accending order by pgno.  Pages are
** connected by pDirty pointers.  The pPrevDirty pointers are
** corrupted by this sort.
*/
#define N_SORT_BUCKET_ALLOC 25
#define N_SORT_BUCKET       25
#ifdef SQLITE_TEST
  int sqlite3_pager_n_sort_bucket = 0;
  #undef N_SORT_BUCKET
  #define N_SORT_BUCKET \
   (sqlite3_pager_n_sort_bucket?sqlite3_pager_n_sort_bucket:N_SORT_BUCKET_ALLOC)
#endif
static PgHdr *sort_pagelist(PgHdr *pIn){
  PgHdr *a[N_SORT_BUCKET_ALLOC], *p;
  int i;
  memset(a, 0, sizeof(a));
  while( pIn ){
    p = pIn;
    pIn = p->pDirty;
    p->pDirty = 0;
    for(i=0; i<N_SORT_BUCKET-1; i++){
      if( a[i]==0 ){
        a[i] = p;
        break;
      }else{
        p = merge_pagelist(a[i], p);
        a[i] = 0;
      }
    }
    if( i==N_SORT_BUCKET-1 ){
      /* Coverage: To get here, there need to be 2^(N_SORT_BUCKET) 
      ** elements in the input list. This is possible, but impractical.
      ** Testing this line is the point of global variable
      ** sqlite3_pager_n_sort_bucket.
      */
      a[i] = merge_pagelist(a[i], p);
    }
  }
  p = a[0];
  for(i=1; i<N_SORT_BUCKET; i++){
    p = merge_pagelist(p, a[i]);
  }
  return p;
}

/*
** Given a list of pages (connected by the PgHdr.pDirty pointer) write
** every one of those pages out to the database file and mark them all
** as clean.
*/
static int pager_write_pagelist(PgHdr *pList){
  Pager *pPager;
  int rc;

  if( pList==0 ) return SQLITE_OK;
  pPager = pList->pPager;

  /* At this point there may be either a RESERVED or EXCLUSIVE lock on the
  ** database file. If there is already an EXCLUSIVE lock, the following
  ** calls to sqlite3OsLock() are no-ops.
  **
  ** Moving the lock from RESERVED to EXCLUSIVE actually involves going
  ** through an intermediate state PENDING.   A PENDING lock prevents new
  ** readers from attaching to the database but is unsufficient for us to
  ** write.  The idea of a PENDING lock is to prevent new readers from
  ** coming in while we wait for existing readers to clear.
  **
  ** While the pager is in the RESERVED state, the original database file
  ** is unchanged and we can rollback without having to playback the
  ** journal into the original database file.  Once we transition to
  ** EXCLUSIVE, it means the database file has been changed and any rollback
  ** will require a journal playback.
  */
  rc = pager_wait_on_lock(pPager, EXCLUSIVE_LOCK);
  if( rc!=SQLITE_OK ){
    return rc;
  }

  pList = sort_pagelist(pList);
  while( pList ){
    assert( pList->dirty );
    rc = sqlite3OsSeek(pPager->fd, (pList->pgno-1)*(i64)pPager->pageSize);
    if( rc ) return rc;
    /* If there are dirty pages in the page cache with page numbers greater
    ** than Pager.dbSize, this means sqlite3PagerTruncate() was called to
    ** make the file smaller (presumably by auto-vacuum code). Do not write
    ** any such pages to the file.
    */
    if( pList->pgno<=pPager->dbSize ){
      char *pData = CODEC2(pPager, PGHDR_TO_DATA(pList), pList->pgno, 6);
      PAGERTRACE4("STORE %d page %d hash(%08x)\n",
                   PAGERID(pPager), pList->pgno, pager_pagehash(pList));
      IOTRACE(("PGOUT %p %d\n", pPager, pList->pgno));
      rc = sqlite3OsWrite(pPager->fd, pData, pPager->pageSize);
      PAGER_INCR(sqlite3_pager_writedb_count);
      PAGER_INCR(pPager->nWrite);
      if( pList->pgno==1 ){
        memcpy(&pPager->dbFileVers, &pData[24], sizeof(pPager->dbFileVers));
      }
    }
#ifndef NDEBUG
    else{
      PAGERTRACE3("NOSTORE %d page %d\n", PAGERID(pPager), pList->pgno);
    }
#endif
    if( rc ) return rc;
    pList->dirty = 0;
#ifdef SQLITE_CHECK_PAGES
    pList->pageHash = pager_pagehash(pList);
#endif
    pList = pList->pDirty;
  }
  return SQLITE_OK;
}

/*
** Collect every dirty page into a dirty list and
** return a pointer to the head of that list.  All pages are
** collected even if they are still in use.
*/
static PgHdr *pager_get_all_dirty_pages(Pager *pPager){
  return pPager->pDirty;
}

/*
** Return TRUE if there is a hot journal on the given pager.
** A hot journal is one that needs to be played back.
**
** If the current size of the database file is 0 but a journal file
** exists, that is probably an old journal left over from a prior
** database with the same name.  Just delete the journal.
*/
static int hasHotJournal(Pager *pPager){
  if( !pPager->useJournal ) return 0;
  if( !sqlite3OsFileExists(pPager->zJournal) ){
    return 0;
  }
  if( sqlite3OsCheckReservedLock(pPager->fd) ){
    return 0;
  }
  if( sqlite3PagerPagecount(pPager)==0 ){
    sqlite3OsDelete(pPager->zJournal);
    return 0;
  }else{
    return 1;
  }
}

/*
** Try to find a page in the cache that can be recycled. 
**
** This routine may return SQLITE_IOERR, SQLITE_FULL or SQLITE_OK. It 
** does not set the pPager->errCode variable.
*/
static int pager_recycle(Pager *pPager, int syncOk, PgHdr **ppPg){
  PgHdr *pPg;
  *ppPg = 0;

  assert(!MEMDB);

  /* Find a page to recycle.  Try to locate a page that does not
  ** require us to do an fsync() on the journal.
  */
  pPg = pPager->pFirstSynced;

  /* If we could not find a page that does not require an fsync()
  ** on the journal file then fsync the journal file.  This is a
  ** very slow operation, so we work hard to avoid it.  But sometimes
  ** it can't be helped.
  */
  if( pPg==0 && pPager->pFirst && syncOk && !MEMDB){
    int rc = syncJournal(pPager);
    if( rc!=0 ){
      return rc;
    }
    if( pPager->fullSync ){
      /* If in full-sync mode, write a new journal header into the
      ** journal file. This is done to avoid ever modifying a journal
      ** header that is involved in the rollback of pages that have
      ** already been written to the database (in case the header is
      ** trashed when the nRec field is updated).
      */
      pPager->nRec = 0;
      assert( pPager->journalOff > 0 );
      assert( pPager->doNotSync==0 );
      rc = writeJournalHdr(pPager);
      if( rc!=0 ){
        return rc;
      }
    }
    pPg = pPager->pFirst;
  }
  if( pPg==0 ){
    return SQLITE_OK;
  }

  assert( pPg->nRef==0 );

  /* Write the page to the database file if it is dirty.
  */
  if( pPg->dirty ){
    int rc;
    assert( pPg->needSync==0 );
    makeClean(pPg);
    pPg->dirty = 1;
    pPg->pDirty = 0;
    rc = pager_write_pagelist( pPg );
    if( rc!=SQLITE_OK ){
      return rc;
    }
  }
  assert( pPg->dirty==0 );

  /* If the page we are recycling is marked as alwaysRollback, then
  ** set the global alwaysRollback flag, thus disabling the
  ** sqlite3PagerDontRollback() optimization for the rest of this transaction.
  ** It is necessary to do this because the page marked alwaysRollback
  ** might be reloaded at a later time but at that point we won't remember
  ** that is was marked alwaysRollback.  This means that all pages must
  ** be marked as alwaysRollback from here on out.
  */
  if( pPg->alwaysRollback ){
    IOTRACE(("ALWAYS_ROLLBACK %p\n", pPager))
    pPager->alwaysRollback = 1;
  }

  /* Unlink the old page from the free list and the hash table
  */
  unlinkPage(pPg);
  assert( pPg->pgno==0 );

  *ppPg = pPg;
  return SQLITE_OK;
}

/*
** This function is called to free superfluous dynamically allocated memory
** held by the pager system. Memory in use by any SQLite pager allocated
** by the current thread may be sqliteFree()ed.
**
** nReq is the number of bytes of memory required. Once this much has
** been released, the function returns. A negative value for nReq means
** free as much memory as possible. The return value is the total number 
** of bytes of memory released.
*/
#if defined(SQLITE_ENABLE_MEMORY_MANAGEMENT) && !defined(SQLITE_OMIT_DISKIO)
SQLITE_PRIVATE int sqlite3PagerReleaseMemory(int nReq){
  const ThreadData *pTsdro = sqlite3ThreadDataReadOnly();
  int nReleased = 0;
  int i;

  /* If the the global mutex is held, this subroutine becomes a
  ** o-op; zero bytes of memory are freed.  This is because
  ** some of the code invoked by this function may also
  ** try to obtain the mutex, resulting in a deadlock.
  */
  if( sqlite3OsInMutex(0) ){
    return 0;
  }

  /* Outermost loop runs for at most two iterations. First iteration we
  ** try to find memory that can be released without calling fsync(). Second
  ** iteration (which only runs if the first failed to free nReq bytes of
  ** memory) is permitted to call fsync(). This is of course much more 
  ** expensive.
  */
  for(i=0; i<=1; i++){

    /* Loop through all the SQLite pagers opened by the current thread. */
    Pager *pPager = pTsdro->pPager;
    for( ; pPager && (nReq<0 || nReleased<nReq); pPager=pPager->pNext){
      PgHdr *pPg;
      int rc;

      if( MEMDB ){
        continue;
      }

      /* For each pager, try to free as many pages as possible (without 
      ** calling fsync() if this is the first iteration of the outermost 
      ** loop).
      */
      while( SQLITE_OK==(rc = pager_recycle(pPager, i, &pPg)) && pPg) {
        /* We've found a page to free. At this point the page has been 
        ** removed from the page hash-table, free-list and synced-list 
        ** (pFirstSynced). It is still in the all pages (pAll) list. 
        ** Remove it from this list before freeing.
        **
        ** Todo: Check the Pager.pStmt list to make sure this is Ok. It 
        ** probably is though.
        */
        PgHdr *pTmp;
        assert( pPg );
        if( pPg==pPager->pAll ){
           pPager->pAll = pPg->pNextAll;
        }else{
          for( pTmp=pPager->pAll; pTmp->pNextAll!=pPg; pTmp=pTmp->pNextAll ){}
          pTmp->pNextAll = pPg->pNextAll;
        }
        nReleased += sqliteAllocSize(pPg);
        IOTRACE(("PGFREE %p %d\n", pPager, pPg->pgno));
        PAGER_INCR(sqlite3_pager_pgfree_count);
        sqliteFree(pPg);
      }

      if( rc!=SQLITE_OK ){
        /* An error occured whilst writing to the database file or 
        ** journal in pager_recycle(). The error is not returned to the 
        ** caller of this function. Instead, set the Pager.errCode variable.
        ** The error will be returned to the user (or users, in the case 
        ** of a shared pager cache) of the pager for which the error occured.
        */
        assert( (rc&0xff)==SQLITE_IOERR || rc==SQLITE_FULL );
        assert( pPager->state>=PAGER_RESERVED );
        pager_error(pPager, rc);
      }
    }
  }

  return nReleased;
}
#endif /* SQLITE_ENABLE_MEMORY_MANAGEMENT && !SQLITE_OMIT_DISKIO */

/*
** Read the content of page pPg out of the database file.
*/
static int readDbPage(Pager *pPager, PgHdr *pPg, Pgno pgno){
  int rc;
  assert( MEMDB==0 );
  rc = sqlite3OsSeek(pPager->fd, (pgno-1)*(i64)pPager->pageSize);
  if( rc==SQLITE_OK ){
    rc = sqlite3OsRead(pPager->fd, PGHDR_TO_DATA(pPg),
                          pPager->pageSize);
  }
  PAGER_INCR(sqlite3_pager_readdb_count);
  PAGER_INCR(pPager->nRead);
  IOTRACE(("PGIN %p %d\n", pPager, pgno));
  if( pgno==1 ){
    memcpy(&pPager->dbFileVers, &((u8*)PGHDR_TO_DATA(pPg))[24],
                                              sizeof(pPager->dbFileVers));
  }
  CODEC1(pPager, PGHDR_TO_DATA(pPg), pPg->pgno, 3);
  PAGERTRACE4("FETCH %d page %d hash(%08x)\n",
               PAGERID(pPager), pPg->pgno, pager_pagehash(pPg));
  return rc;
}


/*
** This function is called to obtain the shared lock required before
** data may be read from the pager cache. If the shared lock has already
** been obtained, this function is a no-op.
**
** Immediately after obtaining the shared lock (if required), this function
** checks for a hot-journal file. If one is found, an emergency rollback
** is performed immediately.
*/
static int pagerSharedLock(Pager *pPager){
  int rc = SQLITE_OK;

  if( pPager->state==PAGER_UNLOCK ){
    if( !MEMDB ){
      assert( pPager->nRef==0 );
      if( !pPager->noReadlock ){
        rc = pager_wait_on_lock(pPager, SHARED_LOCK);
        if( rc!=SQLITE_OK ){
          return pager_error(pPager, rc);
        }
        assert( pPager->state>=SHARED_LOCK );
      }
  
      /* If a journal file exists, and there is no RESERVED lock on the
      ** database file, then it either needs to be played back or deleted.
      */
      if( hasHotJournal(pPager) ){
        /* Get an EXCLUSIVE lock on the database file. At this point it is
        ** important that a RESERVED lock is not obtained on the way to the
        ** EXCLUSIVE lock. If it were, another process might open the
        ** database file, detect the RESERVED lock, and conclude that the
        ** database is safe to read while this process is still rolling it 
        ** back.
        ** 
        ** Because the intermediate RESERVED lock is not requested, the
        ** second process will get to this point in the code and fail to
        ** obtain it's own EXCLUSIVE lock on the database file.
        */
        rc = sqlite3OsLock(pPager->fd, EXCLUSIVE_LOCK);
        if( rc!=SQLITE_OK ){
          pager_unlock(pPager);
          return pager_error(pPager, rc);
        }
        pPager->state = PAGER_EXCLUSIVE;
 
        /* Open the journal for reading only.  Return SQLITE_BUSY if
        ** we are unable to open the journal file. 
        **
        ** The journal file does not need to be locked itself.  The
        ** journal file is never open unless the main database file holds
        ** a write lock, so there is never any chance of two or more
        ** processes opening the journal at the same time.
        **
	** Open the journal for read/write access. This is because in 
	** exclusive-access mode the file descriptor will be kept open and
        ** possibly used for a transaction later on. On some systems, the
        ** OsTruncate() call used in exclusive-access mode also requires
        ** a read/write file handle.
        */
        rc = SQLITE_BUSY;
        if( sqlite3OsFileExists(pPager->zJournal) ){
          int ro;
          assert( !pPager->tempFile );
          rc = sqlite3OsOpenReadWrite(pPager->zJournal, &pPager->jfd, &ro);
          assert( rc!=SQLITE_OK || pPager->jfd );
          if( ro ){
            rc = SQLITE_BUSY;
            sqlite3OsClose(&pPager->jfd);
          }
        }
        if( rc!=SQLITE_OK ){
          pager_unlock(pPager);
          return SQLITE_BUSY;
        }
        pPager->journalOpen = 1;
        pPager->journalStarted = 0;
        pPager->journalOff = 0;
        pPager->setMaster = 0;
        pPager->journalHdr = 0;
 
        /* Playback and delete the journal.  Drop the database write
        ** lock and reacquire the read lock.
        */
        rc = pager_playback(pPager, 1);
        if( rc!=SQLITE_OK ){
          return pager_error(pPager, rc);
        }
        assert(pPager->state==PAGER_SHARED || 
            (pPager->exclusiveMode && pPager->state>PAGER_SHARED)
        );
      }

      if( pPager->pAll ){
        /* The shared-lock has just been acquired on the database file
        ** and there are already pages in the cache (from a previous
        ** read or write transaction).  Check to see if the database
        ** has been modified.  If the database has changed, flush the
        ** cache.
        **
        ** Database changes is detected by looking at 15 bytes beginning
        ** at offset 24 into the file.  The first 4 of these 16 bytes are
        ** a 32-bit counter that is incremented with each change.  The
        ** other bytes change randomly with each file change when
        ** a codec is in use.
        ** 
        ** There is a vanishingly small chance that a change will not be 
        ** detected.  The chance of an undetected change is so small that
        ** it can be neglected.
        */
        char dbFileVers[sizeof(pPager->dbFileVers)];
        sqlite3PagerPagecount(pPager);

        if( pPager->errCode ){
          return pPager->errCode;
        }

        if( pPager->dbSize>0 ){
          IOTRACE(("CKVERS %p %d\n", pPager, sizeof(dbFileVers)));
          rc = sqlite3OsSeek(pPager->fd, 24);
          if( rc!=SQLITE_OK ){
            return rc;
          }
          rc = sqlite3OsRead(pPager->fd, &dbFileVers, sizeof(dbFileVers));
          if( rc!=SQLITE_OK ){
            return rc;
          }
        }else{
          memset(dbFileVers, 0, sizeof(dbFileVers));
        }

        if( memcmp(pPager->dbFileVers, dbFileVers, sizeof(dbFileVers))!=0 ){
          pager_reset(pPager);
        }
      }
    }
    assert( pPager->exclusiveMode || pPager->state<=PAGER_SHARED );
    if( pPager->state==PAGER_UNLOCK ){
      pPager->state = PAGER_SHARED;
    }
  }

  return rc;
}

/*
** Allocate a PgHdr object.   Either create a new one or reuse
** an existing one that is not otherwise in use.
**
** A new PgHdr structure is created if any of the following are
** true:
**
**     (1)  We have not exceeded our maximum allocated cache size
**          as set by the "PRAGMA cache_size" command.
**
**     (2)  There are no unused PgHdr objects available at this time.
**
**     (3)  This is an in-memory database.
**
**     (4)  There are no PgHdr objects that do not require a journal
**          file sync and a sync of the journal file is currently
**          prohibited.
**
** Otherwise, reuse an existing PgHdr.  In other words, reuse an
** existing PgHdr if all of the following are true:
**
**     (1)  We have reached or exceeded the maximum cache size
**          allowed by "PRAGMA cache_size".
**
**     (2)  There is a PgHdr available with PgHdr->nRef==0
**
**     (3)  We are not in an in-memory database
**
**     (4)  Either there is an available PgHdr that does not need
**          to be synced to disk or else disk syncing is currently
**          allowed.
*/
static int pagerAllocatePage(Pager *pPager, PgHdr **ppPg){
  int rc = SQLITE_OK;
  PgHdr *pPg;

  /* Create a new PgHdr if any of the four conditions defined 
  ** above is met: */
  if( pPager->nPage<pPager->mxPage
   || pPager->pFirst==0 
   || MEMDB
   || (pPager->pFirstSynced==0 && pPager->doNotSync)
  ){
    if( pPager->nPage>=pPager->nHash ){
      pager_resize_hash_table(pPager,
         pPager->nHash<256 ? 256 : pPager->nHash*2);
      if( pPager->nHash==0 ){
        rc = SQLITE_NOMEM;
        goto pager_allocate_out;
      }
    }
    pPg = sqliteMallocRaw( sizeof(*pPg) + pPager->pageSize
                            + sizeof(u32) + pPager->nExtra
                            + MEMDB*sizeof(PgHistory) );
    if( pPg==0 ){
      rc = SQLITE_NOMEM;
      goto pager_allocate_out;
    }
    memset(pPg, 0, sizeof(*pPg));
    if( MEMDB ){
      memset(PGHDR_TO_HIST(pPg, pPager), 0, sizeof(PgHistory));
    }
    pPg->pPager = pPager;
    pPg->pNextAll = pPager->pAll;
    pPager->pAll = pPg;
    pPager->nPage++;
  }else{
    /* Recycle an existing page with a zero ref-count. */
    rc = pager_recycle(pPager, 1, &pPg);
    if( rc==SQLITE_BUSY ){
      rc = SQLITE_IOERR_BLOCKED;
    }
    if( rc!=SQLITE_OK ){
      goto pager_allocate_out;
    }
    assert( pPager->state>=SHARED_LOCK );
    assert(pPg);
  }
  *ppPg = pPg;

pager_allocate_out:
  return rc;
}

/*
** Make sure we have the content for a page.  If the page was
** previously acquired with noContent==1, then the content was
** just initialized to zeros instead of being read from disk.
** But now we need the real data off of disk.  So make sure we
** have it.  Read it in if we do not have it already.
*/
static int pager_get_content(PgHdr *pPg){
  if( pPg->needRead ){
    int rc = readDbPage(pPg->pPager, pPg, pPg->pgno);
    if( rc==SQLITE_OK ){
      pPg->needRead = 0;
    }else{
      return rc;
    }
  }
  return SQLITE_OK;
}

/*
** Acquire a page.
**
** A read lock on the disk file is obtained when the first page is acquired. 
** This read lock is dropped when the last page is released.
**
** This routine works for any page number greater than 0.  If the database
** file is smaller than the requested page, then no actual disk
** read occurs and the memory image of the page is initialized to
** all zeros.  The extra data appended to a page is always initialized
** to zeros the first time a page is loaded into memory.
**
** The acquisition might fail for several reasons.  In all cases,
** an appropriate error code is returned and *ppPage is set to NULL.
**
** See also sqlite3PagerLookup().  Both this routine and Lookup() attempt
** to find a page in the in-memory cache first.  If the page is not already
** in memory, this routine goes to disk to read it in whereas Lookup()
** just returns 0.  This routine acquires a read-lock the first time it
** has to go to disk, and could also playback an old journal if necessary.
** Since Lookup() never goes to disk, it never has to deal with locks
** or journal files.
**
** If noContent is false, the page contents are actually read from disk.
** If noContent is true, it means that we do not care about the contents
** of the page at this time, so do not do a disk read.  Just fill in the
** page content with zeros.  But mark the fact that we have not read the
** content by setting the PgHdr.needRead flag.  Later on, if 
** sqlite3PagerWrite() is called on this page or if this routine is
** called again with noContent==0, that means that the content is needed
** and the disk read should occur at that point.
*/
SQLITE_PRIVATE int sqlite3PagerAcquire(
  Pager *pPager,      /* The pager open on the database file */
  Pgno pgno,          /* Page number to fetch */
  DbPage **ppPage,    /* Write a pointer to the page here */
  int noContent       /* Do not bother reading content from disk if true */
){
  PgHdr *pPg;
  int rc;

  assert( pPager->state==PAGER_UNLOCK || pPager->nRef>0 || pgno==1 );

  /* The maximum page number is 2^31. Return SQLITE_CORRUPT if a page
  ** number greater than this, or zero, is requested.
  */
  if( pgno>PAGER_MAX_PGNO || pgno==0 || pgno==PAGER_MJ_PGNO(pPager) ){
    return SQLITE_CORRUPT_BKPT;
  }

  /* Make sure we have not hit any critical errors.
  */ 
  assert( pPager!=0 );
  *ppPage = 0;
  if( pPager->errCode && pPager->errCode!=SQLITE_FULL ){
    return pPager->errCode;
  }

  /* If this is the first page accessed, then get a SHARED lock
  ** on the database file. pagerSharedLock() is a no-op if 
  ** a database lock is already held.
  */
  rc = pagerSharedLock(pPager);
  if( rc!=SQLITE_OK ){
    return rc;
  }
  assert( pPager->state!=PAGER_UNLOCK );

  pPg = pager_lookup(pPager, pgno);
  if( pPg==0 ){
    /* The requested page is not in the page cache. */
    int nMax;
    int h;
    PAGER_INCR(pPager->nMiss);
    rc = pagerAllocatePage(pPager, &pPg);
    if( rc!=SQLITE_OK ){
      return rc;
    }

    pPg->pgno = pgno;
    assert( !MEMDB || pgno>pPager->stmtSize );
    if( pPager->aInJournal && (int)pgno<=pPager->origDbSize ){
      sqlite3CheckMemory(pPager->aInJournal, pgno/8);
      assert( pPager->journalOpen );
      pPg->inJournal = (pPager->aInJournal[pgno/8] & (1<<(pgno&7)))!=0;
      pPg->needSync = 0;
    }else{
      pPg->inJournal = 0;
      pPg->needSync = 0;
    }

    makeClean(pPg);
    pPg->nRef = 1;
    REFINFO(pPg);

    pPager->nRef++;
    if( pPager->nExtra>0 ){
      memset(PGHDR_TO_EXTRA(pPg, pPager), 0, pPager->nExtra);
    }
    nMax = sqlite3PagerPagecount(pPager);
    if( pPager->errCode ){
      sqlite3PagerUnref(pPg);
      rc = pPager->errCode;
      return rc;
    }

    /* Populate the page with data, either by reading from the database
    ** file, or by setting the entire page to zero.
    */
    if( nMax<(int)pgno || MEMDB || (noContent && !pPager->alwaysRollback) ){
      if( pgno>pPager->mxPgno ){
        sqlite3PagerUnref(pPg);
        return SQLITE_FULL;
      }
      memset(PGHDR_TO_DATA(pPg), 0, pPager->pageSize);
      pPg->needRead = noContent && !pPager->alwaysRollback;
      IOTRACE(("ZERO %p %d\n", pPager, pgno));
    }else{
      rc = readDbPage(pPager, pPg, pgno);
      if( rc!=SQLITE_OK && rc!=SQLITE_IOERR_SHORT_READ ){
        pPg->pgno = 0;
        sqlite3PagerUnref(pPg);
        return rc;
      }
      pPg->needRead = 0;
    }

    /* Link the page into the page hash table */
    h = pgno & (pPager->nHash-1);
    assert( pgno!=0 );
    pPg->pNextHash = pPager->aHash[h];
    pPager->aHash[h] = pPg;
    if( pPg->pNextHash ){
      assert( pPg->pNextHash->pPrevHash==0 );
      pPg->pNextHash->pPrevHash = pPg;
    }

#ifdef SQLITE_CHECK_PAGES
    pPg->pageHash = pager_pagehash(pPg);
#endif
  }else{
    /* The requested page is in the page cache. */
    assert(pPager->nRef>0 || pgno==1);
    PAGER_INCR(pPager->nHit);
    if( !noContent ){
      rc = pager_get_content(pPg);
      if( rc ){
        return rc;
      }
    }
    page_ref(pPg);
  }
  *ppPage = pPg;
  return SQLITE_OK;
}

/*
** Acquire a page if it is already in the in-memory cache.  Do
** not read the page from disk.  Return a pointer to the page,
** or 0 if the page is not in cache.
**
** See also sqlite3PagerGet().  The difference between this routine
** and sqlite3PagerGet() is that _get() will go to the disk and read
** in the page if the page is not already in cache.  This routine
** returns NULL if the page is not in cache or if a disk I/O error 
** has ever happened.
*/
SQLITE_PRIVATE DbPage *sqlite3PagerLookup(Pager *pPager, Pgno pgno){
  PgHdr *pPg;

  assert( pPager!=0 );
  assert( pgno!=0 );

  if( pPager->state==PAGER_UNLOCK ){
    assert( !pPager->pAll || pPager->exclusiveMode );
    return 0;
  }
  if( pPager->errCode && pPager->errCode!=SQLITE_FULL ){
    return 0;
  }
  pPg = pager_lookup(pPager, pgno);
  if( pPg==0 ) return 0;
  page_ref(pPg);
  return pPg;
}

/*
** Release a page.
**
** If the number of references to the page drop to zero, then the
** page is added to the LRU list.  When all references to all pages
** are released, a rollback occurs and the lock on the database is
** removed.
*/
SQLITE_PRIVATE int sqlite3PagerUnref(DbPage *pPg){

  /* Decrement the reference count for this page
  */
  assert( pPg->nRef>0 );
  pPg->nRef--;
  REFINFO(pPg);

  CHECK_PAGE(pPg);

  /* When the number of references to a page reach 0, call the
  ** destructor and add the page to the freelist.
  */
  if( pPg->nRef==0 ){
    Pager *pPager;
    pPager = pPg->pPager;
    pPg->pNextFree = 0;
    pPg->pPrevFree = pPager->pLast;
    pPager->pLast = pPg;
    if( pPg->pPrevFree ){
      pPg->pPrevFree->pNextFree = pPg;
    }else{
      pPager->pFirst = pPg;
    }
    if( pPg->needSync==0 && pPager->pFirstSynced==0 ){
      pPager->pFirstSynced = pPg;
    }
    if( pPager->xDestructor ){
      pPager->xDestructor(pPg, pPager->pageSize);
    }
  
    /* When all pages reach the freelist, drop the read lock from
    ** the database file.
    */
    pPager->nRef--;
    assert( pPager->nRef>=0 );
    if( pPager->nRef==0 && (!pPager->exclusiveMode || pPager->journalOff>0) ){
      pagerUnlockAndRollback(pPager);
    }
  }
  return SQLITE_OK;
}

/*
** Create a journal file for pPager.  There should already be a RESERVED
** or EXCLUSIVE lock on the database file when this routine is called.
**
** Return SQLITE_OK if everything.  Return an error code and release the
** write lock if anything goes wrong.
*/
static int pager_open_journal(Pager *pPager){
  int rc;
  assert( !MEMDB );
  assert( pPager->state>=PAGER_RESERVED );
  assert( pPager->journalOpen==0 );
  assert( pPager->useJournal );
  assert( pPager->aInJournal==0 );
  sqlite3PagerPagecount(pPager);
  pPager->aInJournal = sqliteMalloc( pPager->dbSize/8 + 1 );
  if( pPager->aInJournal==0 ){
    rc = SQLITE_NOMEM;
    goto failed_to_open_journal;
  }
  rc = sqlite3OsOpenExclusive(pPager->zJournal, &pPager->jfd,
                                 pPager->tempFile);
  assert( rc!=SQLITE_OK || pPager->jfd );
  pPager->journalOff = 0;
  pPager->setMaster = 0;
  pPager->journalHdr = 0;
  if( rc!=SQLITE_OK ){
    if( rc==SQLITE_NOMEM ){
      sqlite3OsDelete(pPager->zJournal);
    }
    goto failed_to_open_journal;
  }
  sqlite3OsSetFullSync(pPager->jfd, pPager->full_fsync);
  sqlite3OsSetFullSync(pPager->fd, pPager->full_fsync);
  sqlite3OsOpenDirectory(pPager->jfd, pPager->zDirectory);
  pPager->journalOpen = 1;
  pPager->journalStarted = 0;
  pPager->needSync = 0;
  pPager->alwaysRollback = 0;
  pPager->nRec = 0;
  if( pPager->errCode ){
    rc = pPager->errCode;
    goto failed_to_open_journal;
  }
  pPager->origDbSize = pPager->dbSize;

  rc = writeJournalHdr(pPager);

  if( pPager->stmtAutoopen && rc==SQLITE_OK ){
    rc = sqlite3PagerStmtBegin(pPager);
  }
  if( rc!=SQLITE_OK && rc!=SQLITE_NOMEM ){
    rc = pager_end_transaction(pPager);
    if( rc==SQLITE_OK ){
      rc = SQLITE_FULL;
    }
  }
  return rc;

failed_to_open_journal:
  sqliteFree(pPager->aInJournal);
  pPager->aInJournal = 0;
  return rc;
}

/*
** Acquire a write-lock on the database.  The lock is removed when
** the any of the following happen:
**
**   *  sqlite3PagerCommitPhaseTwo() is called.
**   *  sqlite3PagerRollback() is called.
**   *  sqlite3PagerClose() is called.
**   *  sqlite3PagerUnref() is called to on every outstanding page.
**
** The first parameter to this routine is a pointer to any open page of the
** database file.  Nothing changes about the page - it is used merely to
** acquire a pointer to the Pager structure and as proof that there is
** already a read-lock on the database.
**
** The second parameter indicates how much space in bytes to reserve for a
** master journal file-name at the start of the journal when it is created.
**
** A journal file is opened if this is not a temporary file.  For temporary
** files, the opening of the journal file is deferred until there is an
** actual need to write to the journal.
**
** If the database is already reserved for writing, this routine is a no-op.
**
** If exFlag is true, go ahead and get an EXCLUSIVE lock on the file
** immediately instead of waiting until we try to flush the cache.  The
** exFlag is ignored if a transaction is already active.
*/
SQLITE_PRIVATE int sqlite3PagerBegin(DbPage *pPg, int exFlag){
  Pager *pPager = pPg->pPager;
  int rc = SQLITE_OK;
  assert( pPg->nRef>0 );
  assert( pPager->state!=PAGER_UNLOCK );
  if( pPager->state==PAGER_SHARED ){
    assert( pPager->aInJournal==0 );
    if( MEMDB ){
      pPager->state = PAGER_EXCLUSIVE;
      pPager->origDbSize = pPager->dbSize;
    }else{
      rc = sqlite3OsLock(pPager->fd, RESERVED_LOCK);
      if( rc==SQLITE_OK ){
        pPager->state = PAGER_RESERVED;
        if( exFlag ){
          rc = pager_wait_on_lock(pPager, EXCLUSIVE_LOCK);
        }
      }
      if( rc!=SQLITE_OK ){
        return rc;
      }
      pPager->dirtyCache = 0;
      PAGERTRACE2("TRANSACTION %d\n", PAGERID(pPager));
      if( pPager->useJournal && !pPager->tempFile ){
        rc = pager_open_journal(pPager);
      }
    }
  }else if( pPager->journalOpen && pPager->journalOff==0 ){
    /* This happens when the pager was in exclusive-access mode last
    ** time a (read or write) transaction was successfully concluded
    ** by this connection. Instead of deleting the journal file it was 
    ** kept open and truncated to 0 bytes.
    */
    assert( pPager->nRec==0 );
    assert( pPager->origDbSize==0 );
    assert( pPager->aInJournal==0 );
    sqlite3PagerPagecount(pPager);
    pPager->aInJournal = sqliteMalloc( pPager->dbSize/8 + 1 );
    if( !pPager->aInJournal ){
      rc = SQLITE_NOMEM;
    }else{
      pPager->origDbSize = pPager->dbSize;
      rc = writeJournalHdr(pPager);
    }
  }
  assert( !pPager->journalOpen || pPager->journalOff>0 || rc!=SQLITE_OK );
  return rc;
}

/*
** Make a page dirty.  Set its dirty flag and add it to the dirty
** page list.
*/
static void makeDirty(PgHdr *pPg){
  if( pPg->dirty==0 ){
    Pager *pPager = pPg->pPager;
    pPg->dirty = 1;
    pPg->pDirty = pPager->pDirty;
    if( pPager->pDirty ){
      pPager->pDirty->pPrevDirty = pPg;
    }
    pPg->pPrevDirty = 0;
    pPager->pDirty = pPg;
  }
}

/*
** Make a page clean.  Clear its dirty bit and remove it from the
** dirty page list.
*/
static void makeClean(PgHdr *pPg){
  if( pPg->dirty ){
    pPg->dirty = 0;
    if( pPg->pDirty ){
      pPg->pDirty->pPrevDirty = pPg->pPrevDirty;
    }
    if( pPg->pPrevDirty ){
      pPg->pPrevDirty->pDirty = pPg->pDirty;
    }else{
      pPg->pPager->pDirty = pPg->pDirty;
    }
  }
}


/*
** Mark a data page as writeable.  The page is written into the journal 
** if it is not there already.  This routine must be called before making
** changes to a page.
**
** The first time this routine is called, the pager creates a new
** journal and acquires a RESERVED lock on the database.  If the RESERVED
** lock could not be acquired, this routine returns SQLITE_BUSY.  The
** calling routine must check for that return value and be careful not to
** change any page data until this routine returns SQLITE_OK.
**
** If the journal file could not be written because the disk is full,
** then this routine returns SQLITE_FULL and does an immediate rollback.
** All subsequent write attempts also return SQLITE_FULL until there
** is a call to sqlite3PagerCommit() or sqlite3PagerRollback() to
** reset.
*/
static int pager_write(PgHdr *pPg){
  void *pData = PGHDR_TO_DATA(pPg);
  Pager *pPager = pPg->pPager;
  int rc = SQLITE_OK;

  /* Check for errors
  */
  if( pPager->errCode ){ 
    return pPager->errCode;
  }
  if( pPager->readOnly ){
    return SQLITE_PERM;
  }

  assert( !pPager->setMaster );

  CHECK_PAGE(pPg);

  /* If this page was previously acquired with noContent==1, that means
  ** we didn't really read in the content of the page.  This can happen
  ** (for example) when the page is being moved to the freelist.  But
  ** now we are (perhaps) moving the page off of the freelist for
  ** reuse and we need to know its original content so that content
  ** can be stored in the rollback journal.  So do the read at this
  ** time.
  */
  rc = pager_get_content(pPg);
  if( rc ){
    return rc;
  }

  /* Mark the page as dirty.  If the page has already been written
  ** to the journal then we can return right away.
  */
  makeDirty(pPg);
  if( pPg->inJournal && (pageInStatement(pPg) || pPager->stmtInUse==0) ){
    pPager->dirtyCache = 1;
  }else{

    /* If we get this far, it means that the page needs to be
    ** written to the transaction journal or the ckeckpoint journal
    ** or both.
    **
    ** First check to see that the transaction journal exists and
    ** create it if it does not.
    */
    assert( pPager->state!=PAGER_UNLOCK );
    rc = sqlite3PagerBegin(pPg, 0);
    if( rc!=SQLITE_OK ){
      return rc;
    }
    assert( pPager->state>=PAGER_RESERVED );
    if( !pPager->journalOpen && pPager->useJournal ){
      rc = pager_open_journal(pPager);
      if( rc!=SQLITE_OK ) return rc;
    }
    assert( pPager->journalOpen || !pPager->useJournal );
    pPager->dirtyCache = 1;
  
    /* The transaction journal now exists and we have a RESERVED or an
    ** EXCLUSIVE lock on the main database file.  Write the current page to
    ** the transaction journal if it is not there already.
    */
    if( !pPg->inJournal && (pPager->useJournal || MEMDB) ){
      if( (int)pPg->pgno <= pPager->origDbSize ){
        int szPg;
        if( MEMDB ){
          PgHistory *pHist = PGHDR_TO_HIST(pPg, pPager);
          PAGERTRACE3("JOURNAL %d page %d\n", PAGERID(pPager), pPg->pgno);
          assert( pHist->pOrig==0 );
          pHist->pOrig = sqliteMallocRaw( pPager->pageSize );
          if( pHist->pOrig ){
            memcpy(pHist->pOrig, PGHDR_TO_DATA(pPg), pPager->pageSize);
          }
        }else{
          u32 cksum, saved;
          char *pData2, *pEnd;
          /* We should never write to the journal file the page that
          ** contains the database locks.  The following assert verifies
          ** that we do not. */
          assert( pPg->pgno!=PAGER_MJ_PGNO(pPager) );
          pData2 = CODEC2(pPager, pData, pPg->pgno, 7);
          cksum = pager_cksum(pPager, (u8*)pData2);
          pEnd = pData2 + pPager->pageSize;
          pData2 -= 4;
          saved = *(u32*)pEnd;
          put32bits(pEnd, cksum);
          szPg = pPager->pageSize+8;
          put32bits(pData2, pPg->pgno);
          rc = sqlite3OsWrite(pPager->jfd, pData2, szPg);
          IOTRACE(("JOUT %p %d %lld %d\n", pPager, pPg->pgno,
                   pPager->journalOff, szPg));
          PAGER_INCR(sqlite3_pager_writej_count);
          pPager->journalOff += szPg;
          PAGERTRACE5("JOURNAL %d page %d needSync=%d hash(%08x)\n",
               PAGERID(pPager), pPg->pgno, pPg->needSync, pager_pagehash(pPg));
          *(u32*)pEnd = saved;

	  /* An error has occured writing to the journal file. The 
          ** transaction will be rolled back by the layer above.
          */
          if( rc!=SQLITE_OK ){
            return rc;
          }

          pPager->nRec++;
          assert( pPager->aInJournal!=0 );
          pPager->aInJournal[pPg->pgno/8] |= 1<<(pPg->pgno&7);
          pPg->needSync = !pPager->noSync;
          if( pPager->stmtInUse ){
            pPager->aInStmt[pPg->pgno/8] |= 1<<(pPg->pgno&7);
          }
        }
      }else{
        pPg->needSync = !pPager->journalStarted && !pPager->noSync;
        PAGERTRACE4("APPEND %d page %d needSync=%d\n",
                PAGERID(pPager), pPg->pgno, pPg->needSync);
      }
      if( pPg->needSync ){
        pPager->needSync = 1;
      }
      pPg->inJournal = 1;
    }
  
    /* If the statement journal is open and the page is not in it,
    ** then write the current page to the statement journal.  Note that
    ** the statement journal format differs from the standard journal format
    ** in that it omits the checksums and the header.
    */
    if( pPager->stmtInUse 
     && !pageInStatement(pPg) 
     && (int)pPg->pgno<=pPager->stmtSize 
    ){
      assert( pPg->inJournal || (int)pPg->pgno>pPager->origDbSize );
      if( MEMDB ){
        PgHistory *pHist = PGHDR_TO_HIST(pPg, pPager);
        assert( pHist->pStmt==0 );
        pHist->pStmt = sqliteMallocRaw( pPager->pageSize );
        if( pHist->pStmt ){
          memcpy(pHist->pStmt, PGHDR_TO_DATA(pPg), pPager->pageSize);
        }
        PAGERTRACE3("STMT-JOURNAL %d page %d\n", PAGERID(pPager), pPg->pgno);
        page_add_to_stmt_list(pPg);
      }else{
        char *pData2 = CODEC2(pPager, pData, pPg->pgno, 7)-4;
        put32bits(pData2, pPg->pgno);
        rc = sqlite3OsWrite(pPager->stfd, pData2, pPager->pageSize+4);
        PAGERTRACE3("STMT-JOURNAL %d page %d\n", PAGERID(pPager), pPg->pgno);
        if( rc!=SQLITE_OK ){
          return rc;
        }
        pPager->stmtNRec++;
        assert( pPager->aInStmt!=0 );
        pPager->aInStmt[pPg->pgno/8] |= 1<<(pPg->pgno&7);
      }
    }
  }

  /* Update the database size and return.
  */
  assert( pPager->state>=PAGER_SHARED );
  if( pPager->dbSize<(int)pPg->pgno ){
    pPager->dbSize = pPg->pgno;
    if( !MEMDB && pPager->dbSize==PENDING_BYTE/pPager->pageSize ){
      pPager->dbSize++;
    }
  }
  return rc;
}

/*
** This function is used to mark a data-page as writable. It uses 
** pager_write() to open a journal file (if it is not already open)
** and write the page *pData to the journal.
**
** The difference between this function and pager_write() is that this
** function also deals with the special case where 2 or more pages
** fit on a single disk sector. In this case all co-resident pages
** must have been written to the journal file before returning.
*/
SQLITE_PRIVATE int sqlite3PagerWrite(DbPage *pDbPage){
  int rc = SQLITE_OK;

  PgHdr *pPg = pDbPage;
  Pager *pPager = pPg->pPager;
  Pgno nPagePerSector = (pPager->sectorSize/pPager->pageSize);

  if( !MEMDB && nPagePerSector>1 ){
    Pgno nPageCount;          /* Total number of pages in database file */
    Pgno pg1;                 /* First page of the sector pPg is located on. */
    int nPage;                /* Number of pages starting at pg1 to journal */
    int ii;

    /* Set the doNotSync flag to 1. This is because we cannot allow a journal
    ** header to be written between the pages journaled by this function.
    */
    assert( pPager->doNotSync==0 );
    pPager->doNotSync = 1;

    /* This trick assumes that both the page-size and sector-size are
    ** an integer power of 2. It sets variable pg1 to the identifier
    ** of the first page of the sector pPg is located on.
    */
    pg1 = ((pPg->pgno-1) & ~(nPagePerSector-1)) + 1;

    nPageCount = sqlite3PagerPagecount(pPager);
    if( pPg->pgno>nPageCount ){
      nPage = (pPg->pgno - pg1)+1;
    }else if( (pg1+nPagePerSector-1)>nPageCount ){
      nPage = nPageCount+1-pg1;
    }else{
      nPage = nPagePerSector;
    }
    assert(nPage>0);
    assert(pg1<=pPg->pgno);
    assert((pg1+nPage)>pPg->pgno);

    for(ii=0; ii<nPage && rc==SQLITE_OK; ii++){
      Pgno pg = pg1+ii;
      if( !pPager->aInJournal || pg==pPg->pgno || 
          pg>pPager->origDbSize || !(pPager->aInJournal[pg/8]&(1<<(pg&7)))
      ) {
        if( pg!=PAGER_MJ_PGNO(pPager) ){
          PgHdr *pPage;
          rc = sqlite3PagerGet(pPager, pg, &pPage);
          if( rc==SQLITE_OK ){
            rc = pager_write(pPage);
            sqlite3PagerUnref(pPage);
          }
        }
      }
    }

    assert( pPager->doNotSync==1 );
    pPager->doNotSync = 0;
  }else{
    rc = pager_write(pDbPage);
  }
  return rc;
}

/*
** Return TRUE if the page given in the argument was previously passed
** to sqlite3PagerWrite().  In other words, return TRUE if it is ok
** to change the content of the page.
*/
#ifndef NDEBUG
SQLITE_PRIVATE int sqlite3PagerIswriteable(DbPage *pPg){
  return pPg->dirty;
}
#endif

#ifndef SQLITE_OMIT_VACUUM
/*
** Replace the content of a single page with the information in the third
** argument.
*/
SQLITE_PRIVATE int sqlite3PagerOverwrite(Pager *pPager, Pgno pgno, void *pData){
  PgHdr *pPg;
  int rc;

  rc = sqlite3PagerGet(pPager, pgno, &pPg);
  if( rc==SQLITE_OK ){
    rc = sqlite3PagerWrite(pPg);
    if( rc==SQLITE_OK ){
      memcpy(sqlite3PagerGetData(pPg), pData, pPager->pageSize);
    }
    sqlite3PagerUnref(pPg);
  }
  return rc;
}
#endif

/*
** A call to this routine tells the pager that it is not necessary to
** write the information on page pPg back to the disk, even though
** that page might be marked as dirty.
**
** The overlying software layer calls this routine when all of the data
** on the given page is unused.  The pager marks the page as clean so
** that it does not get written to disk.
**
** Tests show that this optimization, together with the
** sqlite3PagerDontRollback() below, more than double the speed
** of large INSERT operations and quadruple the speed of large DELETEs.
**
** When this routine is called, set the alwaysRollback flag to true.
** Subsequent calls to sqlite3PagerDontRollback() for the same page
** will thereafter be ignored.  This is necessary to avoid a problem
** where a page with data is added to the freelist during one part of
** a transaction then removed from the freelist during a later part
** of the same transaction and reused for some other purpose.  When it
** is first added to the freelist, this routine is called.  When reused,
** the sqlite3PagerDontRollback() routine is called.  But because the
** page contains critical data, we still need to be sure it gets
** rolled back in spite of the sqlite3PagerDontRollback() call.
*/
SQLITE_PRIVATE void sqlite3PagerDontWrite(DbPage *pDbPage){
  PgHdr *pPg = pDbPage;
  Pager *pPager = pPg->pPager;

  if( MEMDB ) return;
  pPg->alwaysRollback = 1;
  if( pPg->dirty && !pPager->stmtInUse ){
    assert( pPager->state>=PAGER_SHARED );
    if( pPager->dbSize==(int)pPg->pgno && pPager->origDbSize<pPager->dbSize ){
      /* If this pages is the last page in the file and the file has grown
      ** during the current transaction, then do NOT mark the page as clean.
      ** When the database file grows, we must make sure that the last page
      ** gets written at least once so that the disk file will be the correct
      ** size. If you do not write this page and the size of the file
      ** on the disk ends up being too small, that can lead to database
      ** corruption during the next transaction.
      */
    }else{
      PAGERTRACE3("DONT_WRITE page %d of %d\n", pPg->pgno, PAGERID(pPager));
      IOTRACE(("CLEAN %p %d\n", pPager, pPg->pgno))
      makeClean(pPg);
#ifdef SQLITE_CHECK_PAGES
      pPg->pageHash = pager_pagehash(pPg);
#endif
    }
  }
}

/*
** A call to this routine tells the pager that if a rollback occurs,
** it is not necessary to restore the data on the given page.  This
** means that the pager does not have to record the given page in the
** rollback journal.
**
** If we have not yet actually read the content of this page (if
** the PgHdr.needRead flag is set) then this routine acts as a promise
** that we will never need to read the page content in the future.
** so the needRead flag can be cleared at this point.
*/
SQLITE_PRIVATE void sqlite3PagerDontRollback(DbPage *pPg){
  Pager *pPager = pPg->pPager;

  assert( pPager->state>=PAGER_RESERVED );
  if( pPager->journalOpen==0 ) return;
  if( pPg->alwaysRollback || pPager->alwaysRollback || MEMDB ) return;
  if( !pPg->inJournal && (int)pPg->pgno <= pPager->origDbSize ){
    assert( pPager->aInJournal!=0 );
    pPager->aInJournal[pPg->pgno/8] |= 1<<(pPg->pgno&7);
    pPg->inJournal = 1;
    pPg->needRead = 0;
    if( pPager->stmtInUse ){
      pPager->aInStmt[pPg->pgno/8] |= 1<<(pPg->pgno&7);
    }
    PAGERTRACE3("DONT_ROLLBACK page %d of %d\n", pPg->pgno, PAGERID(pPager));
    IOTRACE(("GARBAGE %p %d\n", pPager, pPg->pgno))
  }
  if( pPager->stmtInUse 
   && !pageInStatement(pPg) 
   && (int)pPg->pgno<=pPager->stmtSize 
  ){
    assert( pPg->inJournal || (int)pPg->pgno>pPager->origDbSize );
    assert( pPager->aInStmt!=0 );
    pPager->aInStmt[pPg->pgno/8] |= 1<<(pPg->pgno&7);
  }
}


/*
** This routine is called to increment the database file change-counter,
** stored at byte 24 of the pager file.
*/
static int pager_incr_changecounter(Pager *pPager){
  PgHdr *pPgHdr;
  u32 change_counter;
  int rc;

  if( !pPager->changeCountDone ){
    /* Open page 1 of the file for writing. */
    rc = sqlite3PagerGet(pPager, 1, &pPgHdr);
    if( rc!=SQLITE_OK ) return rc;
    rc = sqlite3PagerWrite(pPgHdr);
    if( rc!=SQLITE_OK ) return rc;
  
    /* Increment the value just read and write it back to byte 24. */
    change_counter = sqlite3Get4byte((u8*)pPager->dbFileVers);
    change_counter++;
    put32bits(((char*)PGHDR_TO_DATA(pPgHdr))+24, change_counter);
    /* Release the page reference. */
    sqlite3PagerUnref(pPgHdr);
    pPager->changeCountDone = 1;
  }
  return SQLITE_OK;
}

/*
** Sync the database file for the pager pPager. zMaster points to the name
** of a master journal file that should be written into the individual
** journal file. zMaster may be NULL, which is interpreted as no master
** journal (a single database transaction).
**
** This routine ensures that the journal is synced, all dirty pages written
** to the database file and the database file synced. The only thing that
** remains to commit the transaction is to delete the journal file (or
** master journal file if specified).
**
** Note that if zMaster==NULL, this does not overwrite a previous value
** passed to an sqlite3PagerCommitPhaseOne() call.
**
** If parameter nTrunc is non-zero, then the pager file is truncated to
** nTrunc pages (this is used by auto-vacuum databases).
*/
SQLITE_PRIVATE int sqlite3PagerCommitPhaseOne(Pager *pPager, const char *zMaster, Pgno nTrunc){
  int rc = SQLITE_OK;

  PAGERTRACE4("DATABASE SYNC: File=%s zMaster=%s nTrunc=%d\n", 
      pPager->zFilename, zMaster, nTrunc);

  /* If this is an in-memory db, or no pages have been written to, or this
  ** function has already been called, it is a no-op.
  */
  if( pPager->state!=PAGER_SYNCED && !MEMDB && pPager->dirtyCache ){
    PgHdr *pPg;
    assert( pPager->journalOpen );

    /* If a master journal file name has already been written to the
    ** journal file, then no sync is required. This happens when it is
    ** written, then the process fails to upgrade from a RESERVED to an
    ** EXCLUSIVE lock. The next time the process tries to commit the
    ** transaction the m-j name will have already been written.
    */
    if( !pPager->setMaster ){
      rc = pager_incr_changecounter(pPager);
      if( rc!=SQLITE_OK ) goto sync_exit;
#ifndef SQLITE_OMIT_AUTOVACUUM
      if( nTrunc!=0 ){
        /* If this transaction has made the database smaller, then all pages
        ** being discarded by the truncation must be written to the journal
        ** file.
        */
        Pgno i;
        int iSkip = PAGER_MJ_PGNO(pPager);
        for( i=nTrunc+1; i<=pPager->origDbSize; i++ ){
          if( !(pPager->aInJournal[i/8] & (1<<(i&7))) && i!=iSkip ){
            rc = sqlite3PagerGet(pPager, i, &pPg);
            if( rc!=SQLITE_OK ) goto sync_exit;
            rc = sqlite3PagerWrite(pPg);
            sqlite3PagerUnref(pPg);
            if( rc!=SQLITE_OK ) goto sync_exit;
          }
        } 
      }
#endif
      rc = writeMasterJournal(pPager, zMaster);
      if( rc!=SQLITE_OK ) goto sync_exit;
      rc = syncJournal(pPager);
      if( rc!=SQLITE_OK ) goto sync_exit;
    }

#ifndef SQLITE_OMIT_AUTOVACUUM
    if( nTrunc!=0 ){
      rc = sqlite3PagerTruncate(pPager, nTrunc);
      if( rc!=SQLITE_OK ) goto sync_exit;
    }
#endif

    /* Write all dirty pages to the database file */
    pPg = pager_get_all_dirty_pages(pPager);
    rc = pager_write_pagelist(pPg);
    if( rc!=SQLITE_OK ) goto sync_exit;
    pPager->pDirty = 0;

    /* Sync the database file. */
    if( !pPager->noSync ){
      rc = sqlite3OsSync(pPager->fd, 0);
    }
    IOTRACE(("DBSYNC %p\n", pPager))

    pPager->state = PAGER_SYNCED;
  }else if( MEMDB && nTrunc!=0 ){
    rc = sqlite3PagerTruncate(pPager, nTrunc);
  }

sync_exit:
  if( rc==SQLITE_IOERR_BLOCKED ){
    /* pager_incr_changecounter() may attempt to obtain an exclusive
     * lock to spill the cache and return IOERR_BLOCKED. But since 
     * there is no chance the cache is inconsistent, it's
     * better to return SQLITE_BUSY.
     */
    rc = SQLITE_BUSY;
  }
  return rc;
}


/*
** Commit all changes to the database and release the write lock.
**
** If the commit fails for any reason, a rollback attempt is made
** and an error code is returned.  If the commit worked, SQLITE_OK
** is returned.
*/
SQLITE_PRIVATE int sqlite3PagerCommitPhaseTwo(Pager *pPager){
  int rc;
  PgHdr *pPg;

  if( pPager->errCode ){
    return pPager->errCode;
  }
  if( pPager->state<PAGER_RESERVED ){
    return SQLITE_ERROR;
  }
  PAGERTRACE2("COMMIT %d\n", PAGERID(pPager));
  if( MEMDB ){
    pPg = pager_get_all_dirty_pages(pPager);
    while( pPg ){
      PgHistory *pHist = PGHDR_TO_HIST(pPg, pPager);
      clearHistory(pHist);
      pPg->dirty = 0;
      pPg->inJournal = 0;
      pHist->inStmt = 0;
      pPg->needSync = 0;
      pHist->pPrevStmt = pHist->pNextStmt = 0;
      pPg = pPg->pDirty;
    }
    pPager->pDirty = 0;
#ifndef NDEBUG
    for(pPg=pPager->pAll; pPg; pPg=pPg->pNextAll){
      PgHistory *pHist = PGHDR_TO_HIST(pPg, pPager);
      assert( !pPg->alwaysRollback );
      assert( !pHist->pOrig );
      assert( !pHist->pStmt );
    }
#endif
    pPager->pStmt = 0;
    pPager->state = PAGER_SHARED;
    return SQLITE_OK;
  }
  assert( pPager->journalOpen || !pPager->dirtyCache );
  assert( pPager->state==PAGER_SYNCED || !pPager->dirtyCache );
  rc = pager_end_transaction(pPager);
  return pager_error(pPager, rc);
}

/*
** Rollback all changes.  The database falls back to PAGER_SHARED mode.
** All in-memory cache pages revert to their original data contents.
** The journal is deleted.
**
** This routine cannot fail unless some other process is not following
** the correct locking protocol or unless some other
** process is writing trash into the journal file (SQLITE_CORRUPT) or
** unless a prior malloc() failed (SQLITE_NOMEM).  Appropriate error
** codes are returned for all these occasions.  Otherwise,
** SQLITE_OK is returned.
*/
SQLITE_PRIVATE int sqlite3PagerRollback(Pager *pPager){
  int rc;
  PAGERTRACE2("ROLLBACK %d\n", PAGERID(pPager));
  if( MEMDB ){
    PgHdr *p;
    for(p=pPager->pAll; p; p=p->pNextAll){
      PgHistory *pHist;
      assert( !p->alwaysRollback );
      if( !p->dirty ){
        assert( !((PgHistory *)PGHDR_TO_HIST(p, pPager))->pOrig );
        assert( !((PgHistory *)PGHDR_TO_HIST(p, pPager))->pStmt );
        continue;
      }

      pHist = PGHDR_TO_HIST(p, pPager);
      if( pHist->pOrig ){
        memcpy(PGHDR_TO_DATA(p), pHist->pOrig, pPager->pageSize);
        PAGERTRACE3("ROLLBACK-PAGE %d of %d\n", p->pgno, PAGERID(pPager));
      }else{
        PAGERTRACE3("PAGE %d is clean on %d\n", p->pgno, PAGERID(pPager));
      }
      clearHistory(pHist);
      p->dirty = 0;
      p->inJournal = 0;
      pHist->inStmt = 0;
      pHist->pPrevStmt = pHist->pNextStmt = 0;
      if( pPager->xReiniter ){
        pPager->xReiniter(p, pPager->pageSize);
      }
    }
    pPager->pDirty = 0;
    pPager->pStmt = 0;
    pPager->dbSize = pPager->origDbSize;
    pager_truncate_cache(pPager);
    pPager->stmtInUse = 0;
    pPager->state = PAGER_SHARED;
    return SQLITE_OK;
  }

  if( !pPager->dirtyCache || !pPager->journalOpen ){
    rc = pager_end_transaction(pPager);
    return rc;
  }

  if( pPager->errCode && pPager->errCode!=SQLITE_FULL ){
    if( pPager->state>=PAGER_EXCLUSIVE ){
      pager_playback(pPager, 0);
    }
    return pPager->errCode;
  }
  if( pPager->state==PAGER_RESERVED ){
    int rc2;
    rc = pager_playback(pPager, 0);
    rc2 = pager_end_transaction(pPager);
    if( rc==SQLITE_OK ){
      rc = rc2;
    }
  }else{
    rc = pager_playback(pPager, 0);
  }
  /* pager_reset(pPager); */
  pPager->dbSize = -1;

  /* If an error occurs during a ROLLBACK, we can no longer trust the pager
  ** cache. So call pager_error() on the way out to make any error 
  ** persistent.
  */
  return pager_error(pPager, rc);
}

/*
** Return TRUE if the database file is opened read-only.  Return FALSE
** if the database is (in theory) writable.
*/
SQLITE_PRIVATE int sqlite3PagerIsreadonly(Pager *pPager){
  return pPager->readOnly;
}

/*
** Return the number of references to the pager.
*/
SQLITE_PRIVATE int sqlite3PagerRefcount(Pager *pPager){
  return pPager->nRef;
}

#ifdef SQLITE_TEST
/*
** This routine is used for testing and analysis only.
*/
SQLITE_PRIVATE int *sqlite3PagerStats(Pager *pPager){
  static int a[11];
  a[0] = pPager->nRef;
  a[1] = pPager->nPage;
  a[2] = pPager->mxPage;
  a[3] = pPager->dbSize;
  a[4] = pPager->state;
  a[5] = pPager->errCode;
  a[6] = pPager->nHit;
  a[7] = pPager->nMiss;
  a[8] = 0;  /* Used to be pPager->nOvfl */
  a[9] = pPager->nRead;
  a[10] = pPager->nWrite;
  return a;
}
#endif

/*
** Set the statement rollback point.
**
** This routine should be called with the transaction journal already
** open.  A new statement journal is created that can be used to rollback
** changes of a single SQL command within a larger transaction.
*/
SQLITE_PRIVATE int sqlite3PagerStmtBegin(Pager *pPager){
  int rc;
  assert( !pPager->stmtInUse );
  assert( pPager->state>=PAGER_SHARED );
  assert( pPager->dbSize>=0 );
  PAGERTRACE2("STMT-BEGIN %d\n", PAGERID(pPager));
  if( MEMDB ){
    pPager->stmtInUse = 1;
    pPager->stmtSize = pPager->dbSize;
    return SQLITE_OK;
  }
  if( !pPager->journalOpen ){
    pPager->stmtAutoopen = 1;
    return SQLITE_OK;
  }
  assert( pPager->journalOpen );
  pPager->aInStmt = sqliteMalloc( pPager->dbSize/8 + 1 );
  if( pPager->aInStmt==0 ){
    /* sqlite3OsLock(pPager->fd, SHARED_LOCK); */
    return SQLITE_NOMEM;
  }
#ifndef NDEBUG
  rc = sqlite3OsFileSize(pPager->jfd, &pPager->stmtJSize);
  if( rc ) goto stmt_begin_failed;
  assert( pPager->stmtJSize == pPager->journalOff );
#endif
  pPager->stmtJSize = pPager->journalOff;
  pPager->stmtSize = pPager->dbSize;
  pPager->stmtHdrOff = 0;
  pPager->stmtCksum = pPager->cksumInit;
  if( !pPager->stmtOpen ){
    rc = sqlite3PagerOpentemp(&pPager->stfd);
    if( rc ) goto stmt_begin_failed;
    pPager->stmtOpen = 1;
    pPager->stmtNRec = 0;
  }
  pPager->stmtInUse = 1;
  return SQLITE_OK;
 
stmt_begin_failed:
  if( pPager->aInStmt ){
    sqliteFree(pPager->aInStmt);
    pPager->aInStmt = 0;
  }
  return rc;
}

/*
** Commit a statement.
*/
SQLITE_PRIVATE int sqlite3PagerStmtCommit(Pager *pPager){
  if( pPager->stmtInUse ){
    PgHdr *pPg, *pNext;
    PAGERTRACE2("STMT-COMMIT %d\n", PAGERID(pPager));
    if( !MEMDB ){
      sqlite3OsSeek(pPager->stfd, 0);
      /* sqlite3OsTruncate(pPager->stfd, 0); */
      sqliteFree( pPager->aInStmt );
      pPager->aInStmt = 0;
    }else{
      for(pPg=pPager->pStmt; pPg; pPg=pNext){
        PgHistory *pHist = PGHDR_TO_HIST(pPg, pPager);
        pNext = pHist->pNextStmt;
        assert( pHist->inStmt );
        pHist->inStmt = 0;
        pHist->pPrevStmt = pHist->pNextStmt = 0;
        sqliteFree(pHist->pStmt);
        pHist->pStmt = 0;
      }
    }
    pPager->stmtNRec = 0;
    pPager->stmtInUse = 0;
    pPager->pStmt = 0;
  }
  pPager->stmtAutoopen = 0;
  return SQLITE_OK;
}

/*
** Rollback a statement.
*/
SQLITE_PRIVATE int sqlite3PagerStmtRollback(Pager *pPager){
  int rc;
  if( pPager->stmtInUse ){
    PAGERTRACE2("STMT-ROLLBACK %d\n", PAGERID(pPager));
    if( MEMDB ){
      PgHdr *pPg;
      PgHistory *pHist;
      for(pPg=pPager->pStmt; pPg; pPg=pHist->pNextStmt){
        pHist = PGHDR_TO_HIST(pPg, pPager);
        if( pHist->pStmt ){
          memcpy(PGHDR_TO_DATA(pPg), pHist->pStmt, pPager->pageSize);
          sqliteFree(pHist->pStmt);
          pHist->pStmt = 0;
        }
      }
      pPager->dbSize = pPager->stmtSize;
      pager_truncate_cache(pPager);
      rc = SQLITE_OK;
    }else{
      rc = pager_stmt_playback(pPager);
    }
    sqlite3PagerStmtCommit(pPager);
  }else{
    rc = SQLITE_OK;
  }
  pPager->stmtAutoopen = 0;
  return rc;
}

/*
** Return the full pathname of the database file.
*/
SQLITE_PRIVATE const char *sqlite3PagerFilename(Pager *pPager){
  return pPager->zFilename;
}

/*
** Return the directory of the database file.
*/
SQLITE_PRIVATE const char *sqlite3PagerDirname(Pager *pPager){
  return pPager->zDirectory;
}

/*
** Return the full pathname of the journal file.
*/
SQLITE_PRIVATE const char *sqlite3PagerJournalname(Pager *pPager){
  return pPager->zJournal;
}

/*
** Return true if fsync() calls are disabled for this pager.  Return FALSE
** if fsync()s are executed normally.
*/
SQLITE_PRIVATE int sqlite3PagerNosync(Pager *pPager){
  return pPager->noSync;
}

#ifdef SQLITE_HAS_CODEC
/*
** Set the codec for this pager
*/
SQLITE_PRIVATE void sqlite3PagerSetCodec(
  Pager *pPager,
  void *(*xCodec)(void*,void*,Pgno,int),
  void *pCodecArg
){
  pPager->xCodec = xCodec;
  pPager->pCodecArg = pCodecArg;
}
#endif

#ifndef SQLITE_OMIT_AUTOVACUUM
/*
** Move the page pPg to location pgno in the file. 
**
** There must be no references to the page previously located at
** pgno (which we call pPgOld) though that page is allowed to be
** in cache.  If the page previous located at pgno is not already
** in the rollback journal, it is not put there by by this routine.
**
** References to the page pPg remain valid. Updating any
** meta-data associated with pPg (i.e. data stored in the nExtra bytes
** allocated along with the page) is the responsibility of the caller.
**
** A transaction must be active when this routine is called. It used to be
** required that a statement transaction was not active, but this restriction
** has been removed (CREATE INDEX needs to move a page when a statement
** transaction is active).
*/
SQLITE_PRIVATE int sqlite3PagerMovepage(Pager *pPager, DbPage *pPg, Pgno pgno){
  PgHdr *pPgOld;  /* The page being overwritten. */
  int h;
  Pgno needSyncPgno = 0;

  assert( pPg->nRef>0 );

  PAGERTRACE5("MOVE %d page %d (needSync=%d) moves to %d\n", 
      PAGERID(pPager), pPg->pgno, pPg->needSync, pgno);
  IOTRACE(("MOVE %p %d %d\n", pPager, pPg->pgno, pgno))

  pager_get_content(pPg);
  if( pPg->needSync ){
    needSyncPgno = pPg->pgno;
    assert( pPg->inJournal || (int)pgno>pPager->origDbSize );
    assert( pPg->dirty );
    assert( pPager->needSync );
  }

  /* Unlink pPg from it's hash-chain */
  unlinkHashChain(pPager, pPg);

  /* If the cache contains a page with page-number pgno, remove it
  ** from it's hash chain. Also, if the PgHdr.needSync was set for 
  ** page pgno before the 'move' operation, it needs to be retained 
  ** for the page moved there.
  */
  pPg->needSync = 0;
  pPgOld = pager_lookup(pPager, pgno);
  if( pPgOld ){
    assert( pPgOld->nRef==0 );
    unlinkHashChain(pPager, pPgOld);
    makeClean(pPgOld);
    pPg->needSync = pPgOld->needSync;
  }else{
    pPg->needSync = 0;
  }
  if( pPager->aInJournal && (int)pgno<=pPager->origDbSize ){
    pPg->inJournal =  (pPager->aInJournal[pgno/8] & (1<<(pgno&7)))!=0;
  }else{
    pPg->inJournal = 0;
    assert( pPg->needSync==0 || (int)pgno>pPager->origDbSize );
  }

  /* Change the page number for pPg and insert it into the new hash-chain. */
  assert( pgno!=0 );
  pPg->pgno = pgno;
  h = pgno & (pPager->nHash-1);
  if( pPager->aHash[h] ){
    assert( pPager->aHash[h]->pPrevHash==0 );
    pPager->aHash[h]->pPrevHash = pPg;
  }
  pPg->pNextHash = pPager->aHash[h];
  pPager->aHash[h] = pPg;
  pPg->pPrevHash = 0;

  makeDirty(pPg);
  pPager->dirtyCache = 1;

  if( needSyncPgno ){
    /* If needSyncPgno is non-zero, then the journal file needs to be 
    ** sync()ed before any data is written to database file page needSyncPgno.
    ** Currently, no such page exists in the page-cache and the 
    ** Pager.aInJournal bit has been set. This needs to be remedied by loading
    ** the page into the pager-cache and setting the PgHdr.needSync flag.
    **
    ** The sqlite3PagerGet() call may cause the journal to sync. So make
    ** sure the Pager.needSync flag is set too.
    */
    int rc;
    PgHdr *pPgHdr;
    assert( pPager->needSync );
    rc = sqlite3PagerGet(pPager, needSyncPgno, &pPgHdr);
    if( rc!=SQLITE_OK ) return rc;
    pPager->needSync = 1;
    pPgHdr->needSync = 1;
    pPgHdr->inJournal = 1;
    makeDirty(pPgHdr);
    sqlite3PagerUnref(pPgHdr);
  }

  return SQLITE_OK;
}
#endif

/*
** Return a pointer to the data for the specified page.
*/
SQLITE_PRIVATE void *sqlite3PagerGetData(DbPage *pPg){
  return PGHDR_TO_DATA(pPg);
}

/*
** Return a pointer to the Pager.nExtra bytes of "extra" space 
** allocated along with the specified page.
*/
SQLITE_PRIVATE void *sqlite3PagerGetExtra(DbPage *pPg){
  Pager *pPager = pPg->pPager;
  return (pPager?PGHDR_TO_EXTRA(pPg, pPager):0);
}

/*
** Get/set the locking-mode for this pager. Parameter eMode must be one
** of PAGER_LOCKINGMODE_QUERY, PAGER_LOCKINGMODE_NORMAL or 
** PAGER_LOCKINGMODE_EXCLUSIVE. If the parameter is not _QUERY, then
** the locking-mode is set to the value specified.
**
** The returned value is either PAGER_LOCKINGMODE_NORMAL or
** PAGER_LOCKINGMODE_EXCLUSIVE, indicating the current (possibly updated)
** locking-mode.
*/
SQLITE_PRIVATE int sqlite3PagerLockingMode(Pager *pPager, int eMode){
  assert( eMode==PAGER_LOCKINGMODE_QUERY
            || eMode==PAGER_LOCKINGMODE_NORMAL
            || eMode==PAGER_LOCKINGMODE_EXCLUSIVE );
  assert( PAGER_LOCKINGMODE_QUERY<0 );
  assert( PAGER_LOCKINGMODE_NORMAL>=0 && PAGER_LOCKINGMODE_EXCLUSIVE>=0 );
  if( eMode>=0 && !pPager->tempFile ){
    pPager->exclusiveMode = eMode;
  }
  return (int)pPager->exclusiveMode;
}

#if defined(SQLITE_DEBUG) || defined(SQLITE_TEST)
/*
** Return the current state of the file lock for the given pager.
** The return value is one of NO_LOCK, SHARED_LOCK, RESERVED_LOCK,
** PENDING_LOCK, or EXCLUSIVE_LOCK.
*/
SQLITE_PRIVATE int sqlite3PagerLockstate(Pager *pPager){
  return sqlite3OsLockState(pPager->fd);
}
#endif

#ifdef SQLITE_DEBUG
/*
** Print a listing of all referenced pages and their ref count.
*/
SQLITE_PRIVATE void sqlite3PagerRefdump(Pager *pPager){
  PgHdr *pPg;
  for(pPg=pPager->pAll; pPg; pPg=pPg->pNextAll){
    if( pPg->nRef<=0 ) continue;
    sqlite3DebugPrintf("PAGE %3d addr=%p nRef=%d\n", 
       pPg->pgno, PGHDR_TO_DATA(pPg), pPg->nRef);
  }
}
#endif

#endif /* SQLITE_OMIT_DISKIO */

/************** End of pager.c ***********************************************/
/************** Begin file btree.c *******************************************/
/*
** 2004 April 6
**
** The author disclaims copyright to this source code.  In place of
** a legal notice, here is a blessing:
**
**    May you do good and not evil.
**    May you find forgiveness for yourself and forgive others.
**    May you share freely, never taking more than you give.
**
*************************************************************************
** $Id: sqlite3.c,v 1.4 2007/07/21 08:51:43 rmsimpson Exp $
**
** This file implements a external (disk-based) database using BTrees.
** See the header comment on "btreeInt.h" for additional information.
** Including a description of file format and an overview of operation.
*/
/************** Include btreeInt.h in the middle of btree.c ******************/
/************** Begin file btreeInt.h ****************************************/
/*
** 2004 April 6
**
** The author disclaims copyright to this source code.  In place of
** a legal notice, here is a blessing:
**
**    May you do good and not evil.
**    May you find forgiveness for yourself and forgive others.
**    May you share freely, never taking more than you give.
**
*************************************************************************
** $Id: sqlite3.c,v 1.4 2007/07/21 08:51:43 rmsimpson Exp $
**
** This file implements a external (disk-based) database using BTrees.
** For a detailed discussion of BTrees, refer to
**
**     Donald E. Knuth, THE ART OF COMPUTER PROGRAMMING, Volume 3:
**     "Sorting And Searching", pages 473-480. Addison-Wesley
**     Publishing Company, Reading, Massachusetts.
**
** The basic idea is that each page of the file contains N database
** entries and N+1 pointers to subpages.
**
**   ----------------------------------------------------------------
**   |  Ptr(0) | Key(0) | Ptr(1) | Key(1) | ... | Key(N-1) | Ptr(N) |
**   ----------------------------------------------------------------
**
** All of the keys on the page that Ptr(0) points to have values less
** than Key(0).  All of the keys on page Ptr(1) and its subpages have
** values greater than Key(0) and less than Key(1).  All of the keys
** on Ptr(N) and its subpages have values greater than Key(N-1).  And
** so forth.
**
** Finding a particular key requires reading O(log(M)) pages from the 
** disk where M is the number of entries in the tree.
**
** In this implementation, a single file can hold one or more separate 
** BTrees.  Each BTree is identified by the index of its root page.  The
** key and data for any entry are combined to form the "payload".  A
** fixed amount of payload can be carried directly on the database
** page.  If the payload is larger than the preset amount then surplus
** bytes are stored on overflow pages.  The payload for an entry
** and the preceding pointer are combined to form a "Cell".  Each 
** page has a small header which contains the Ptr(N) pointer and other
** information such as the size of key and data.
**
** FORMAT DETAILS
**
** The file is divided into pages.  The first page is called page 1,
** the second is page 2, and so forth.  A page number of zero indicates
** "no such page".  The page size can be anything between 512 and 65536.
** Each page can be either a btree page, a freelist page or an overflow
** page.
**
** The first page is always a btree page.  The first 100 bytes of the first
** page contain a special header (the "file header") that describes the file.
** The format of the file header is as follows:
**
**   OFFSET   SIZE    DESCRIPTION
**      0      16     Header string: "SQLite format 3\000"
**     16       2     Page size in bytes.  
**     18       1     File format write version
**     19       1     File format read version
**     20       1     Bytes of unused space at the end of each page
**     21       1     Max embedded payload fraction
**     22       1     Min embedded payload fraction
**     23       1     Min leaf payload fraction
**     24       4     File change counter
**     28       4     Reserved for future use
**     32       4     First freelist page
**     36       4     Number of freelist pages in the file
**     40      60     15 4-byte meta values passed to higher layers
**
** All of the integer values are big-endian (most significant byte first).
**
** The file change counter is incremented when the database is changed
** This counter allows other processes to know when the file has changed
** and thus when they need to flush their cache.
**
** The max embedded payload fraction is the amount of the total usable
** space in a page that can be consumed by a single cell for standard
** B-tree (non-LEAFDATA) tables.  A value of 255 means 100%.  The default
** is to limit the maximum cell size so that at least 4 cells will fit
** on one page.  Thus the default max embedded payload fraction is 64.
**
** If the payload for a cell is larger than the max payload, then extra
** payload is spilled to overflow pages.  Once an overflow page is allocated,
** as many bytes as possible are moved into the overflow pages without letting
** the cell size drop below the min embedded payload fraction.
**
** The min leaf payload fraction is like the min embedded payload fraction
** except that it applies to leaf nodes in a LEAFDATA tree.  The maximum
** payload fraction for a LEAFDATA tree is always 100% (or 255) and it
** not specified in the header.
**
** Each btree pages is divided into three sections:  The header, the
** cell pointer array, and the cell content area.  Page 1 also has a 100-byte
** file header that occurs before the page header.
**
**      |----------------|
**      | file header    |   100 bytes.  Page 1 only.
**      |----------------|
**      | page header    |   8 bytes for leaves.  12 bytes for interior nodes
**      |----------------|
**      | cell pointer   |   |  2 bytes per cell.  Sorted order.
**      | array          |   |  Grows downward
**      |                |   v
**      |----------------|
**      | unallocated    |
**      | space          |
**      |----------------|   ^  Grows upwards
**      | cell content   |   |  Arbitrary order interspersed with freeblocks.
**      | area           |   |  and free space fragments.
**      |----------------|
**
** The page headers looks like this:
**
**   OFFSET   SIZE     DESCRIPTION
**      0       1      Flags. 1: intkey, 2: zerodata, 4: leafdata, 8: leaf
**      1       2      byte offset to the first freeblock
**      3       2      number of cells on this page
**      5       2      first byte of the cell content area
**      7       1      number of fragmented free bytes
**      8       4      Right child (the Ptr(N) value).  Omitted on leaves.
**
** The flags define the format of this btree page.  The leaf flag means that
** this page has no children.  The zerodata flag means that this page carries
** only keys and no data.  The intkey flag means that the key is a integer
** which is stored in the key size entry of the cell header rather than in
** the payload area.
**
** The cell pointer array begins on the first byte after the page header.
** The cell pointer array contains zero or more 2-byte numbers which are
** offsets from the beginning of the page to the cell content in the cell
** content area.  The cell pointers occur in sorted order.  The system strives
** to keep free space after the last cell pointer so that new cells can
** be easily added without having to defragment the page.
**
** Cell content is stored at the very end of the page and grows toward the
** beginning of the page.
**
** Unused space within the cell content area is collected into a linked list of
** freeblocks.  Each freeblock is at least 4 bytes in size.  The byte offset
** to the first freeblock is given in the header.  Freeblocks occur in
** increasing order.  Because a freeblock must be at least 4 bytes in size,
** any group of 3 or fewer unused bytes in the cell content area cannot
** exist on the freeblock chain.  A group of 3 or fewer free bytes is called
** a fragment.  The total number of bytes in all fragments is recorded.
** in the page header at offset 7.
**
**    SIZE    DESCRIPTION
**      2     Byte offset of the next freeblock
**      2     Bytes in this freeblock
**
** Cells are of variable length.  Cells are stored in the cell content area at
** the end of the page.  Pointers to the cells are in the cell pointer array
** that immediately follows the page header.  Cells is not necessarily
** contiguous or in order, but cell pointers are contiguous and in order.
**
** Cell content makes use of variable length integers.  A variable
** length integer is 1 to 9 bytes where the lower 7 bits of each 
** byte are used.  The integer consists of all bytes that have bit 8 set and
** the first byte with bit 8 clear.  The most significant byte of the integer
** appears first.  A variable-length integer may not be more than 9 bytes long.
** As a special case, all 8 bytes of the 9th byte are used as data.  This
** allows a 64-bit integer to be encoded in 9 bytes.
**
**    0x00                      becomes  0x00000000
**    0x7f                      becomes  0x0000007f
**    0x81 0x00                 becomes  0x00000080
**    0x82 0x00                 becomes  0x00000100
**    0x80 0x7f                 becomes  0x0000007f
**    0x8a 0x91 0xd1 0xac 0x78  becomes  0x12345678
**    0x81 0x81 0x81 0x81 0x01  becomes  0x10204081
**
** Variable length integers are used for rowids and to hold the number of
** bytes of key and data in a btree cell.
**
** The content of a cell looks like this:
**
**    SIZE    DESCRIPTION
**      4     Page number of the left child. Omitted if leaf flag is set.
**     var    Number of bytes of data. Omitted if the zerodata flag is set.
**     var    Number of bytes of key. Or the key itself if intkey flag is set.
**      *     Payload
**      4     First page of the overflow chain.  Omitted if no overflow
**
** Overflow pages form a linked list.  Each page except the last is completely
** filled with data (pagesize - 4 bytes).  The last page can have as little
** as 1 byte of data.
**
**    SIZE    DESCRIPTION
**      4     Page number of next overflow page
**      *     Data
**
** Freelist pages come in two subtypes: trunk pages and leaf pages.  The
** file header points to the first in a linked list of trunk page.  Each trunk
** page points to multiple leaf pages.  The content of a leaf page is
** unspecified.  A trunk page looks like this:
**
**    SIZE    DESCRIPTION
**      4     Page number of next trunk page
**      4     Number of leaf pointers on this page
**      *     zero or more pages numbers of leaves
*/

/* Round up a number to the next larger multiple of 8.  This is used
** to force 8-byte alignment on 64-bit architectures.
*/
#define ROUND8(x)   ((x+7)&~7)


/* The following value is the maximum cell size assuming a maximum page
** size give above.
*/
#define MX_CELL_SIZE(pBt)  (pBt->pageSize-8)

/* The maximum number of cells on a single page of the database.  This
** assumes a minimum cell size of 3 bytes.  Such small cells will be
** exceedingly rare, but they are possible.
*/
#define MX_CELL(pBt) ((pBt->pageSize-8)/3)

/* Forward declarations */
typedef struct MemPage MemPage;
typedef struct BtLock BtLock;

/*
** This is a magic string that appears at the beginning of every
** SQLite database in order to identify the file as a real database.
**
** You can change this value at compile-time by specifying a
** -DSQLITE_FILE_HEADER="..." on the compiler command-line.  The
** header must be exactly 16 bytes including the zero-terminator so
** the string itself should be 15 characters long.  If you change
** the header, then your custom library will not be able to read 
** databases generated by the standard tools and the standard tools
** will not be able to read databases created by your custom library.
*/
#ifndef SQLITE_FILE_HEADER /* 123456789 123456 */
#  define SQLITE_FILE_HEADER "SQLite format 3"
#endif

/*
** Page type flags.  An ORed combination of these flags appear as the
** first byte of every BTree page.
*/
#define PTF_INTKEY    0x01
#define PTF_ZERODATA  0x02
#define PTF_LEAFDATA  0x04
#define PTF_LEAF      0x08

/*
** As each page of the file is loaded into memory, an instance of the following
** structure is appended and initialized to zero.  This structure stores
** information about the page that is decoded from the raw file page.
**
** The pParent field points back to the parent page.  This allows us to
** walk up the BTree from any leaf to the root.  Care must be taken to
** unref() the parent page pointer when this page is no longer referenced.
** The pageDestructor() routine handles that chore.
*/
struct MemPage {
  u8 isInit;           /* True if previously initialized. MUST BE FIRST! */
  u8 idxShift;         /* True if Cell indices have changed */
  u8 nOverflow;        /* Number of overflow cell bodies in aCell[] */
  u8 intKey;           /* True if intkey flag is set */
  u8 leaf;             /* True if leaf flag is set */
  u8 zeroData;         /* True if table stores keys only */
  u8 leafData;         /* True if tables stores data on leaves only */
  u8 hasData;          /* True if this page stores data */
  u8 hdrOffset;        /* 100 for page 1.  0 otherwise */
  u8 childPtrSize;     /* 0 if leaf==1.  4 if leaf==0 */
  u16 maxLocal;        /* Copy of Btree.maxLocal or Btree.maxLeaf */
  u16 minLocal;        /* Copy of Btree.minLocal or Btree.minLeaf */
  u16 cellOffset;      /* Index in aData of first cell pointer */
  u16 idxParent;       /* Index in parent of this node */
  u16 nFree;           /* Number of free bytes on the page */
  u16 nCell;           /* Number of cells on this page, local and ovfl */
  struct _OvflCell {   /* Cells that will not fit on aData[] */
    u8 *pCell;          /* Pointers to the body of the overflow cell */
    u16 idx;            /* Insert this cell before idx-th non-overflow cell */
  } aOvfl[5];
  BtShared *pBt;       /* Pointer back to BTree structure */
  u8 *aData;           /* Pointer back to the start of the page */
  DbPage *pDbPage;     /* Pager page handle */
  Pgno pgno;           /* Page number for this page */
  MemPage *pParent;    /* The parent of this page.  NULL for root */
};

/*
** The in-memory image of a disk page has the auxiliary information appended
** to the end.  EXTRA_SIZE is the number of bytes of space needed to hold
** that extra information.
*/
#define EXTRA_SIZE sizeof(MemPage)

/* Btree handle */
struct Btree {
  sqlite3 *pSqlite;
  BtShared *pBt;
  u8 inTrans;            /* TRANS_NONE, TRANS_READ or TRANS_WRITE */
};

/*
** Btree.inTrans may take one of the following values.
**
** If the shared-data extension is enabled, there may be multiple users
** of the Btree structure. At most one of these may open a write transaction,
** but any number may have active read transactions. Variable Btree.pDb 
** points to the handle that owns any current write-transaction.
*/
#define TRANS_NONE  0
#define TRANS_READ  1
#define TRANS_WRITE 2

/*
** Everything we need to know about an open database
*/
struct BtShared {
  Pager *pPager;        /* The page cache */
  BtCursor *pCursor;    /* A list of all open cursors */
  MemPage *pPage1;      /* First page of the database */
  u8 inStmt;            /* True if we are in a statement subtransaction */
  u8 readOnly;          /* True if the underlying file is readonly */
  u8 maxEmbedFrac;      /* Maximum payload as % of total page size */
  u8 minEmbedFrac;      /* Minimum payload as % of total page size */
  u8 minLeafFrac;       /* Minimum leaf payload as % of total page size */
  u8 pageSizeFixed;     /* True if the page size can no longer be changed */
#ifndef SQLITE_OMIT_AUTOVACUUM
  u8 autoVacuum;        /* True if auto-vacuum is enabled */
  u8 incrVacuum;        /* True if incr-vacuum is enabled */
  Pgno nTrunc;          /* Non-zero if the db will be truncated (incr vacuum) */
#endif
  u16 pageSize;         /* Total number of bytes on a page */
  u16 usableSize;       /* Number of usable bytes on each page */
  int maxLocal;         /* Maximum local payload in non-LEAFDATA tables */
  int minLocal;         /* Minimum local payload in non-LEAFDATA tables */
  int maxLeaf;          /* Maximum local payload in a LEAFDATA table */
  int minLeaf;          /* Minimum local payload in a LEAFDATA table */
  BusyHandler *pBusyHandler;   /* Callback for when there is lock contention */
  u8 inTransaction;     /* Transaction state */
  int nRef;             /* Number of references to this structure */
  int nTransaction;     /* Number of open transactions (read + write) */
  void *pSchema;        /* Pointer to space allocated by sqlite3BtreeSchema() */
  void (*xFreeSchema)(void*);  /* Destructor for BtShared.pSchema */
#ifndef SQLITE_OMIT_SHARED_CACHE
  BtLock *pLock;        /* List of locks held on this shared-btree struct */
  BtShared *pNext;      /* Next in ThreadData.pBtree linked list */
#endif
};

/*
** An instance of the following structure is used to hold information
** about a cell.  The parseCellPtr() function fills in this structure
** based on information extract from the raw disk page.
*/
typedef struct CellInfo CellInfo;
struct CellInfo {
  u8 *pCell;     /* Pointer to the start of cell content */
  i64 nKey;      /* The key for INTKEY tables, or number of bytes in key */
  u32 nData;     /* Number of bytes of data */
  u32 nPayload;  /* Total amount of payload */
  u16 nHeader;   /* Size of the cell content header in bytes */
  u16 nLocal;    /* Amount of payload held locally */
  u16 iOverflow; /* Offset to overflow page number.  Zero if no overflow */
  u16 nSize;     /* Size of the cell content on the main b-tree page */
};

/*
** A cursor is a pointer to a particular entry in the BTree.
** The entry is identified by its MemPage and the index in
** MemPage.aCell[] of the entry.
*/
struct BtCursor {
  Btree *pBtree;            /* The Btree to which this cursor belongs */
  BtCursor *pNext, *pPrev;  /* Forms a linked list of all cursors */
  int (*xCompare)(void*,int,const void*,int,const void*); /* Key comp func */
  void *pArg;               /* First arg to xCompare() */
  Pgno pgnoRoot;            /* The root page of this tree */
  MemPage *pPage;           /* Page that contains the entry */
  int idx;                  /* Index of the entry in pPage->aCell[] */
  CellInfo info;            /* A parse of the cell we are pointing at */
  u8 wrFlag;                /* True if writable */
  u8 eState;                /* One of the CURSOR_XXX constants (see below) */
  void *pKey;      /* Saved key that was cursor's last known position */
  i64 nKey;        /* Size of pKey, or last integer key */
  int skip;        /* (skip<0) -> Prev() is a no-op. (skip>0) -> Next() is */
#ifndef SQLITE_OMIT_INCRBLOB
  u8 isIncrblobHandle;      /* True if this cursor is an incr. io handle */
  Pgno *aOverflow;          /* Cache of overflow page locations */
#endif
};

/*
** Potential values for BtCursor.eState.
**
** CURSOR_VALID:
**   Cursor points to a valid entry. getPayload() etc. may be called.
**
** CURSOR_INVALID:
**   Cursor does not point to a valid entry. This can happen (for example) 
**   because the table is empty or because BtreeCursorFirst() has not been
**   called.
**
** CURSOR_REQUIRESEEK:
**   The table that this cursor was opened on still exists, but has been 
**   modified since the cursor was last used. The cursor position is saved
**   in variables BtCursor.pKey and BtCursor.nKey. When a cursor is in 
**   this state, restoreOrClearCursorPosition() can be called to attempt to
**   seek the cursor to the saved position.
*/
#define CURSOR_INVALID           0
#define CURSOR_VALID             1
#define CURSOR_REQUIRESEEK       2

/*
** The TRACE macro will print high-level status information about the
** btree operation when the global variable sqlite3_btree_trace is
** enabled.
*/
#if SQLITE_TEST
# define TRACE(X)   if( sqlite3_btree_trace ){ printf X; fflush(stdout); }
#else
# define TRACE(X)
#endif

/*
** Routines to read and write variable-length integers.  These used to
** be defined locally, but now we use the varint routines in the util.c
** file.
*/
#define getVarint    sqlite3GetVarint
#define getVarint32(A,B)  ((*B=*(A))<=0x7f?1:sqlite3GetVarint32(A,B))
#define putVarint    sqlite3PutVarint

/* The database page the PENDING_BYTE occupies. This page is never used.
** TODO: This macro is very similary to PAGER_MJ_PGNO() in pager.c. They
** should possibly be consolidated (presumably in pager.h).
**
** If disk I/O is omitted (meaning that the database is stored purely
** in memory) then there is no pending byte.
*/
#ifdef SQLITE_OMIT_DISKIO
# define PENDING_BYTE_PAGE(pBt)  0x7fffffff
#else
# define PENDING_BYTE_PAGE(pBt) ((PENDING_BYTE/(pBt)->pageSize)+1)
#endif

/*
** A linked list of the following structures is stored at BtShared.pLock.
** Locks are added (or upgraded from READ_LOCK to WRITE_LOCK) when a cursor 
** is opened on the table with root page BtShared.iTable. Locks are removed
** from this list when a transaction is committed or rolled back, or when
** a btree handle is closed.
*/
struct BtLock {
  Btree *pBtree;        /* Btree handle holding this lock */
  Pgno iTable;          /* Root page of table */
  u8 eLock;             /* READ_LOCK or WRITE_LOCK */
  BtLock *pNext;        /* Next in BtShared.pLock list */
};

/* Candidate values for BtLock.eLock */
#define READ_LOCK     1
#define WRITE_LOCK    2

/*
** These macros define the location of the pointer-map entry for a 
** database page. The first argument to each is the number of usable
** bytes on each page of the database (often 1024). The second is the
** page number to look up in the pointer map.
**
** PTRMAP_PAGENO returns the database page number of the pointer-map
** page that stores the required pointer. PTRMAP_PTROFFSET returns
** the offset of the requested map entry.
**
** If the pgno argument passed to PTRMAP_PAGENO is a pointer-map page,
** then pgno is returned. So (pgno==PTRMAP_PAGENO(pgsz, pgno)) can be
** used to test if pgno is a pointer-map page. PTRMAP_ISPAGE implements
** this test.
*/
#define PTRMAP_PAGENO(pBt, pgno) ptrmapPageno(pBt, pgno)
#define PTRMAP_PTROFFSET(pBt, pgno) (5*(pgno-ptrmapPageno(pBt, pgno)-1))
#define PTRMAP_ISPAGE(pBt, pgno) (PTRMAP_PAGENO((pBt),(pgno))==(pgno))

/*
** The pointer map is a lookup table that identifies the parent page for
** each child page in the database file.  The parent page is the page that
** contains a pointer to the child.  Every page in the database contains
** 0 or 1 parent pages.  (In this context 'database page' refers
** to any page that is not part of the pointer map itself.)  Each pointer map
** entry consists of a single byte 'type' and a 4 byte parent page number.
** The PTRMAP_XXX identifiers below are the valid types.
**
** The purpose of the pointer map is to facility moving pages from one
** position in the file to another as part of autovacuum.  When a page
** is moved, the pointer in its parent must be updated to point to the
** new location.  The pointer map is used to locate the parent page quickly.
**
** PTRMAP_ROOTPAGE: The database page is a root-page. The page-number is not
**                  used in this case.
**
** PTRMAP_FREEPAGE: The database page is an unused (free) page. The page-number 
**                  is not used in this case.
**
** PTRMAP_OVERFLOW1: The database page is the first page in a list of 
**                   overflow pages. The page number identifies the page that
**                   contains the cell with a pointer to this overflow page.
**
** PTRMAP_OVERFLOW2: The database page is the second or later page in a list of
**                   overflow pages. The page-number identifies the previous
**                   page in the overflow page list.
**
** PTRMAP_BTREE: The database page is a non-root btree page. The page number
**               identifies the parent page in the btree.
*/
#define PTRMAP_ROOTPAGE 1
#define PTRMAP_FREEPAGE 2
#define PTRMAP_OVERFLOW1 3
#define PTRMAP_OVERFLOW2 4
#define PTRMAP_BTREE 5

/* A bunch of assert() statements to check the transaction state variables
** of handle p (type Btree*) are internally consistent.
*/
#define btreeIntegrity(p) \
  assert( p->inTrans!=TRANS_NONE || p->pBt->nTransaction<p->pBt->nRef ); \
  assert( p->pBt->nTransaction<=p->pBt->nRef ); \
  assert( p->pBt->inTransaction!=TRANS_NONE || p->pBt->nTransaction==0 ); \
  assert( p->pBt->inTransaction>=p->inTrans ); 


/*
** The ISAUTOVACUUM macro is used within balance_nonroot() to determine
** if the database supports auto-vacuum or not. Because it is used
** within an expression that is an argument to another macro 
** (sqliteMallocRaw), it is not possible to use conditional compilation.
** So, this macro is defined instead.
*/
#ifndef SQLITE_OMIT_AUTOVACUUM
#define ISAUTOVACUUM (pBt->autoVacuum)
#else
#define ISAUTOVACUUM 0
#endif


/*
** This structure is passed around through all the sanity checking routines
** in order to keep track of some global state information.
*/
typedef struct IntegrityCk IntegrityCk;
struct IntegrityCk {
  BtShared *pBt;    /* The tree being checked out */
  Pager *pPager;    /* The associated pager.  Also accessible by pBt->pPager */
  int nPage;        /* Number of pages in the database */
  int *anRef;       /* Number of times each page is referenced */
  int mxErr;        /* Stop accumulating errors when this reaches zero */
  char *zErrMsg;    /* An error message.  NULL if no errors seen. */
  int nErr;         /* Number of messages written to zErrMsg so far */
};

/*
** Read or write a two- and four-byte big-endian integer values.
*/
#define get2byte(x)   ((x)[0]<<8 | (x)[1])
#define put2byte(p,v) ((p)[0] = (v)>>8, (p)[1] = (v))
#define get4byte sqlite3Get4byte
#define put4byte sqlite3Put4byte

/*
** Internal routines that should be accessed by the btree layer only.
*/
SQLITE_PRIVATE int sqlite3BtreeGetPage(BtShared*, Pgno, MemPage**, int);
SQLITE_PRIVATE int sqlite3BtreeInitPage(MemPage *pPage, MemPage *pParent);
SQLITE_PRIVATE void sqlite3BtreeParseCellPtr(MemPage*, u8*, CellInfo*);
SQLITE_PRIVATE void sqlite3BtreeParseCell(MemPage*, int, CellInfo*);
SQLITE_PRIVATE u8 *sqlite3BtreeFindCell(MemPage *pPage, int iCell);
SQLITE_PRIVATE int sqlite3BtreeRestoreOrClearCursorPosition(BtCursor *pCur);
SQLITE_PRIVATE void sqlite3BtreeGetTempCursor(BtCursor *pCur, BtCursor *pTempCur);
SQLITE_PRIVATE void sqlite3BtreeReleaseTempCursor(BtCursor *pCur);
SQLITE_PRIVATE int sqlite3BtreeIsRootPage(MemPage *pPage);
SQLITE_PRIVATE void sqlite3BtreeMoveToParent(BtCursor *pCur);

/************** End of btreeInt.h ********************************************/
/************** Continuing where we left off in btree.c **********************/

/*
** The header string that appears at the beginning of every
** SQLite database.
*/
static const char zMagicHeader[] = SQLITE_FILE_HEADER;


/*
** Set this global variable to 1 to enable tracing using the TRACE
** macro.
*/
#if SQLITE_TEST
int sqlite3_btree_trace=0;  /* True to enable tracing */
#endif

/*
** Forward declaration
*/
static int checkReadLocks(Btree*,Pgno,BtCursor*);


#ifdef SQLITE_OMIT_SHARED_CACHE
  /*
  ** The functions queryTableLock(), lockTable() and unlockAllTables()
  ** manipulate entries in the BtShared.pLock linked list used to store
  ** shared-cache table level locks. If the library is compiled with the
  ** shared-cache feature disabled, then there is only ever one user
  ** of each BtShared structure and so this locking is not necessary. 
  ** So define the lock related functions as no-ops.
  */
  #define queryTableLock(a,b,c) SQLITE_OK
  #define lockTable(a,b,c) SQLITE_OK
  #define unlockAllTables(a)
#else

/*
** Query to see if btree handle p may obtain a lock of type eLock 
** (READ_LOCK or WRITE_LOCK) on the table with root-page iTab. Return
** SQLITE_OK if the lock may be obtained (by calling lockTable()), or
** SQLITE_LOCKED if not.
*/
static int queryTableLock(Btree *p, Pgno iTab, u8 eLock){
  BtShared *pBt = p->pBt;
  BtLock *pIter;

  /* This is a no-op if the shared-cache is not enabled */
  if( 0==sqlite3ThreadDataReadOnly()->useSharedData ){
    return SQLITE_OK;
  }

  /* This (along with lockTable()) is where the ReadUncommitted flag is
  ** dealt with. If the caller is querying for a read-lock and the flag is
  ** set, it is unconditionally granted - even if there are write-locks
  ** on the table. If a write-lock is requested, the ReadUncommitted flag
  ** is not considered.
  **
  ** In function lockTable(), if a read-lock is demanded and the 
  ** ReadUncommitted flag is set, no entry is added to the locks list 
  ** (BtShared.pLock).
  **
  ** To summarize: If the ReadUncommitted flag is set, then read cursors do
  ** not create or respect table locks. The locking procedure for a 
  ** write-cursor does not change.
  */
  if( 
    !p->pSqlite || 
    0==(p->pSqlite->flags&SQLITE_ReadUncommitted) || 
    eLock==WRITE_LOCK ||
    iTab==MASTER_ROOT
  ){
    for(pIter=pBt->pLock; pIter; pIter=pIter->pNext){
      if( pIter->pBtree!=p && pIter->iTable==iTab && 
          (pIter->eLock!=eLock || eLock!=READ_LOCK) ){
        return SQLITE_LOCKED;
      }
    }
  }
  return SQLITE_OK;
}

/*
** Add a lock on the table with root-page iTable to the shared-btree used
** by Btree handle p. Parameter eLock must be either READ_LOCK or 
** WRITE_LOCK.
**
** SQLITE_OK is returned if the lock is added successfully. SQLITE_BUSY and
** SQLITE_NOMEM may also be returned.
*/
static int lockTable(Btree *p, Pgno iTable, u8 eLock){
  BtShared *pBt = p->pBt;
  BtLock *pLock = 0;
  BtLock *pIter;

  /* This is a no-op if the shared-cache is not enabled */
  if( 0==sqlite3ThreadDataReadOnly()->useSharedData ){
    return SQLITE_OK;
  }

  assert( SQLITE_OK==queryTableLock(p, iTable, eLock) );

  /* If the read-uncommitted flag is set and a read-lock is requested,
  ** return early without adding an entry to the BtShared.pLock list. See
  ** comment in function queryTableLock() for more info on handling 
  ** the ReadUncommitted flag.
  */
  if( 
    (p->pSqlite) && 
    (p->pSqlite->flags&SQLITE_ReadUncommitted) && 
    (eLock==READ_LOCK) &&
    iTable!=MASTER_ROOT
  ){
    return SQLITE_OK;
  }

  /* First search the list for an existing lock on this table. */
  for(pIter=pBt->pLock; pIter; pIter=pIter->pNext){
    if( pIter->iTable==iTable && pIter->pBtree==p ){
      pLock = pIter;
      break;
    }
  }

  /* If the above search did not find a BtLock struct associating Btree p
  ** with table iTable, allocate one and link it into the list.
  */
  if( !pLock ){
    pLock = (BtLock *)sqliteMalloc(sizeof(BtLock));
    if( !pLock ){
      return SQLITE_NOMEM;
    }
    pLock->iTable = iTable;
    pLock->pBtree = p;
    pLock->pNext = pBt->pLock;
    pBt->pLock = pLock;
  }

  /* Set the BtLock.eLock variable to the maximum of the current lock
  ** and the requested lock. This means if a write-lock was already held
  ** and a read-lock requested, we don't incorrectly downgrade the lock.
  */
  assert( WRITE_LOCK>READ_LOCK );
  if( eLock>pLock->eLock ){
    pLock->eLock = eLock;
  }

  return SQLITE_OK;
}

/*
** Release all the table locks (locks obtained via calls to the lockTable()
** procedure) held by Btree handle p.
*/
static void unlockAllTables(Btree *p){
  BtLock **ppIter = &p->pBt->pLock;

  /* If the shared-cache extension is not enabled, there should be no
  ** locks in the BtShared.pLock list, making this procedure a no-op. Assert
  ** that this is the case.
  */
  assert( sqlite3ThreadDataReadOnly()->useSharedData || 0==*ppIter );

  while( *ppIter ){
    BtLock *pLock = *ppIter;
    if( pLock->pBtree==p ){
      *ppIter = pLock->pNext;
      sqliteFree(pLock);
    }else{
      ppIter = &pLock->pNext;
    }
  }
}
#endif /* SQLITE_OMIT_SHARED_CACHE */

static void releasePage(MemPage *pPage);  /* Forward reference */

#ifndef SQLITE_OMIT_INCRBLOB
/*
** Invalidate the overflow page-list cache for cursor pCur, if any.
*/
static void invalidateOverflowCache(BtCursor *pCur){
  sqliteFree(pCur->aOverflow);
  pCur->aOverflow = 0;
}

/*
** Invalidate the overflow page-list cache for all cursors opened
** on the shared btree structure pBt.
*/
static void invalidateAllOverflowCache(BtShared *pBt){
  BtCursor *p;
  for(p=pBt->pCursor; p; p=p->pNext){
    invalidateOverflowCache(p);
  }
}
#else
  #define invalidateOverflowCache(x)
  #define invalidateAllOverflowCache(x)
#endif

/*
** Save the current cursor position in the variables BtCursor.nKey 
** and BtCursor.pKey. The cursor's state is set to CURSOR_REQUIRESEEK.
*/
static int saveCursorPosition(BtCursor *pCur){
  int rc;

  assert( CURSOR_VALID==pCur->eState );
  assert( 0==pCur->pKey );

  rc = sqlite3BtreeKeySize(pCur, &pCur->nKey);

  /* If this is an intKey table, then the above call to BtreeKeySize()
  ** stores the integer key in pCur->nKey. In this case this value is
  ** all that is required. Otherwise, if pCur is not open on an intKey
  ** table, then malloc space for and store the pCur->nKey bytes of key 
  ** data.
  */
  if( rc==SQLITE_OK && 0==pCur->pPage->intKey){
    void *pKey = sqliteMalloc(pCur->nKey);
    if( pKey ){
      rc = sqlite3BtreeKey(pCur, 0, pCur->nKey, pKey);
      if( rc==SQLITE_OK ){
        pCur->pKey = pKey;
      }else{
        sqliteFree(pKey);
      }
    }else{
      rc = SQLITE_NOMEM;
    }
  }
  assert( !pCur->pPage->intKey || !pCur->pKey );

  if( rc==SQLITE_OK ){
    releasePage(pCur->pPage);
    pCur->pPage = 0;
    pCur->eState = CURSOR_REQUIRESEEK;
  }

  invalidateOverflowCache(pCur);
  return rc;
}

/*
** Save the positions of all cursors except pExcept open on the table 
** with root-page iRoot. Usually, this is called just before cursor
** pExcept is used to modify the table (BtreeDelete() or BtreeInsert()).
*/
static int saveAllCursors(BtShared *pBt, Pgno iRoot, BtCursor *pExcept){
  BtCursor *p;
  for(p=pBt->pCursor; p; p=p->pNext){
    if( p!=pExcept && (0==iRoot || p->pgnoRoot==iRoot) && 
        p->eState==CURSOR_VALID ){
      int rc = saveCursorPosition(p);
      if( SQLITE_OK!=rc ){
        return rc;
      }
    }
  }
  return SQLITE_OK;
}

/*
** Clear the current cursor position.
*/
static void clearCursorPosition(BtCursor *pCur){
  sqliteFree(pCur->pKey);
  pCur->pKey = 0;
  pCur->eState = CURSOR_INVALID;
}

/*
** Restore the cursor to the position it was in (or as close to as possible)
** when saveCursorPosition() was called. Note that this call deletes the 
** saved position info stored by saveCursorPosition(), so there can be
** at most one effective restoreOrClearCursorPosition() call after each 
** saveCursorPosition().
**
** If the second argument argument - doSeek - is false, then instead of 
** returning the cursor to it's saved position, any saved position is deleted
** and the cursor state set to CURSOR_INVALID.
*/
SQLITE_PRIVATE int sqlite3BtreeRestoreOrClearCursorPosition(BtCursor *pCur){
  int rc;
  assert( pCur->eState==CURSOR_REQUIRESEEK );
#ifndef SQLITE_OMIT_INCRBLOB
  if( pCur->isIncrblobHandle ){
    return SQLITE_ABORT;
  }
#endif
  pCur->eState = CURSOR_INVALID;
  rc = sqlite3BtreeMoveto(pCur, pCur->pKey, pCur->nKey, 0, &pCur->skip);
  if( rc==SQLITE_OK ){
    sqliteFree(pCur->pKey);
    pCur->pKey = 0;
    assert( pCur->eState==CURSOR_VALID || pCur->eState==CURSOR_INVALID );
  }
  return rc;
}

#define restoreOrClearCursorPosition(p) \
  (p->eState==CURSOR_REQUIRESEEK ? \
         sqlite3BtreeRestoreOrClearCursorPosition(p) : \
         SQLITE_OK)

#ifndef SQLITE_OMIT_AUTOVACUUM
/*
** Given a page number of a regular database page, return the page
** number for the pointer-map page that contains the entry for the
** input page number.
*/
static Pgno ptrmapPageno(BtShared *pBt, Pgno pgno){
  int nPagesPerMapPage = (pBt->usableSize/5)+1;
  int iPtrMap = (pgno-2)/nPagesPerMapPage;
  int ret = (iPtrMap*nPagesPerMapPage) + 2; 
  if( ret==PENDING_BYTE_PAGE(pBt) ){
    ret++;
  }
  return ret;
}

/*
** Write an entry into the pointer map.
**
** This routine updates the pointer map entry for page number 'key'
** so that it maps to type 'eType' and parent page number 'pgno'.
** An error code is returned if something goes wrong, otherwise SQLITE_OK.
*/
static int ptrmapPut(BtShared *pBt, Pgno key, u8 eType, Pgno parent){
  DbPage *pDbPage;  /* The pointer map page */
  u8 *pPtrmap;      /* The pointer map data */
  Pgno iPtrmap;     /* The pointer map page number */
  int offset;       /* Offset in pointer map page */
  int rc;

  /* The master-journal page number must never be used as a pointer map page */
  assert( 0==PTRMAP_ISPAGE(pBt, PENDING_BYTE_PAGE(pBt)) );

  assert( pBt->autoVacuum );
  if( key==0 ){
    return SQLITE_CORRUPT_BKPT;
  }
  iPtrmap = PTRMAP_PAGENO(pBt, key);
  rc = sqlite3PagerGet(pBt->pPager, iPtrmap, &pDbPage);
  if( rc!=SQLITE_OK ){
    return rc;
  }
  offset = PTRMAP_PTROFFSET(pBt, key);
  pPtrmap = (u8 *)sqlite3PagerGetData(pDbPage);

  if( eType!=pPtrmap[offset] || get4byte(&pPtrmap[offset+1])!=parent ){
    TRACE(("PTRMAP_UPDATE: %d->(%d,%d)\n", key, eType, parent));
    rc = sqlite3PagerWrite(pDbPage);
    if( rc==SQLITE_OK ){
      pPtrmap[offset] = eType;
      put4byte(&pPtrmap[offset+1], parent);
    }
  }

  sqlite3PagerUnref(pDbPage);
  return rc;
}

/*
** Read an entry from the pointer map.
**
** This routine retrieves the pointer map entry for page 'key', writing
** the type and parent page number to *pEType and *pPgno respectively.
** An error code is returned if something goes wrong, otherwise SQLITE_OK.
*/
static int ptrmapGet(BtShared *pBt, Pgno key, u8 *pEType, Pgno *pPgno){
  DbPage *pDbPage;   /* The pointer map page */
  int iPtrmap;       /* Pointer map page index */
  u8 *pPtrmap;       /* Pointer map page data */
  int offset;        /* Offset of entry in pointer map */
  int rc;

  iPtrmap = PTRMAP_PAGENO(pBt, key);
  rc = sqlite3PagerGet(pBt->pPager, iPtrmap, &pDbPage);
  if( rc!=0 ){
    return rc;
  }
  pPtrmap = (u8 *)sqlite3PagerGetData(pDbPage);

  offset = PTRMAP_PTROFFSET(pBt, key);
  assert( pEType!=0 );
  *pEType = pPtrmap[offset];
  if( pPgno ) *pPgno = get4byte(&pPtrmap[offset+1]);

  sqlite3PagerUnref(pDbPage);
  if( *pEType<1 || *pEType>5 ) return SQLITE_CORRUPT_BKPT;
  return SQLITE_OK;
}

#endif /* SQLITE_OMIT_AUTOVACUUM */

/*
** Given a btree page and a cell index (0 means the first cell on
** the page, 1 means the second cell, and so forth) return a pointer
** to the cell content.
**
** This routine works only for pages that do not contain overflow cells.
*/
#define findCell(pPage, iCell) \
  ((pPage)->aData + get2byte(&(pPage)->aData[(pPage)->cellOffset+2*(iCell)]))
SQLITE_PRIVATE u8 *sqlite3BtreeFindCell(MemPage *pPage, int iCell){
  assert( iCell>=0 );
  assert( iCell<get2byte(&pPage->aData[pPage->hdrOffset+3]) );
  return findCell(pPage, iCell);
}

/*
** This a more complex version of sqlite3BtreeFindCell() that works for
** pages that do contain overflow cells.  See insert
*/
static u8 *findOverflowCell(MemPage *pPage, int iCell){
  int i;
  for(i=pPage->nOverflow-1; i>=0; i--){
    int k;
    struct _OvflCell *pOvfl;
    pOvfl = &pPage->aOvfl[i];
    k = pOvfl->idx;
    if( k<=iCell ){
      if( k==iCell ){
        return pOvfl->pCell;
      }
      iCell--;
    }
  }
  return findCell(pPage, iCell);
}

/*
** Parse a cell content block and fill in the CellInfo structure.  There
** are two versions of this function.  sqlite3BtreeParseCell() takes a 
** cell index as the second argument and sqlite3BtreeParseCellPtr() 
** takes a pointer to the body of the cell as its second argument.
**
** Within this file, the parseCell() macro can be called instead of
** sqlite3BtreeParseCellPtr(). Using some compilers, this will be faster.
*/
SQLITE_PRIVATE void sqlite3BtreeParseCellPtr(
  MemPage *pPage,         /* Page containing the cell */
  u8 *pCell,              /* Pointer to the cell text. */
  CellInfo *pInfo         /* Fill in this structure */
){
  int n;                  /* Number bytes in cell content header */
  u32 nPayload;           /* Number of bytes of cell payload */

  pInfo->pCell = pCell;
  assert( pPage->leaf==0 || pPage->leaf==1 );
  n = pPage->childPtrSize;
  assert( n==4-4*pPage->leaf );
  if( pPage->hasData ){
    n += getVarint32(&pCell[n], &nPayload);
  }else{
    nPayload = 0;
  }
  pInfo->nData = nPayload;
  if( pPage->intKey ){
    n += getVarint(&pCell[n], (u64 *)&pInfo->nKey);
  }else{
    u32 x;
    n += getVarint32(&pCell[n], &x);
    pInfo->nKey = x;
    nPayload += x;
  }
  pInfo->nPayload = nPayload;
  pInfo->nHeader = n;
  if( nPayload<=pPage->maxLocal ){
    /* This is the (easy) common case where the entire payload fits
    ** on the local page.  No overflow is required.
    */
    int nSize;          /* Total size of cell content in bytes */
    pInfo->nLocal = nPayload;
    pInfo->iOverflow = 0;
    nSize = nPayload + n;
    if( nSize<4 ){
      nSize = 4;        /* Minimum cell size is 4 */
    }
    pInfo->nSize = nSize;
  }else{
    /* If the payload will not fit completely on the local page, we have
    ** to decide how much to store locally and how much to spill onto
    ** overflow pages.  The strategy is to minimize the amount of unused
    ** space on overflow pages while keeping the amount of local storage
    ** in between minLocal and maxLocal.
    **
    ** Warning:  changing the way overflow payload is distributed in any
    ** way will result in an incompatible file format.
    */
    int minLocal;  /* Minimum amount of payload held locally */
    int maxLocal;  /* Maximum amount of payload held locally */
    int surplus;   /* Overflow payload available for local storage */

    minLocal = pPage->minLocal;
    maxLocal = pPage->maxLocal;
    surplus = minLocal + (nPayload - minLocal)%(pPage->pBt->usableSize - 4);
    if( surplus <= maxLocal ){
      pInfo->nLocal = surplus;
    }else{
      pInfo->nLocal = minLocal;
    }
    pInfo->iOverflow = pInfo->nLocal + n;
    pInfo->nSize = pInfo->iOverflow + 4;
  }
}
#define parseCell(pPage, iCell, pInfo) \
  sqlite3BtreeParseCellPtr((pPage), findCell((pPage), (iCell)), (pInfo))
SQLITE_PRIVATE void sqlite3BtreeParseCell(
  MemPage *pPage,         /* Page containing the cell */
  int iCell,              /* The cell index.  First cell is 0 */
  CellInfo *pInfo         /* Fill in this structure */
){
  parseCell(pPage, iCell, pInfo);
}

/*
** Compute the total number of bytes that a Cell needs in the cell
** data area of the btree-page.  The return number includes the cell
** data header and the local payload, but not any overflow page or
** the space used by the cell pointer.
*/
#ifndef NDEBUG
static int cellSize(MemPage *pPage, int iCell){
  CellInfo info;
  sqlite3BtreeParseCell(pPage, iCell, &info);
  return info.nSize;
}
#endif
static int cellSizePtr(MemPage *pPage, u8 *pCell){
  CellInfo info;
  sqlite3BtreeParseCellPtr(pPage, pCell, &info);
  return info.nSize;
}

#ifndef SQLITE_OMIT_AUTOVACUUM
/*
** If the cell pCell, part of page pPage contains a pointer
** to an overflow page, insert an entry into the pointer-map
** for the overflow page.
*/
static int ptrmapPutOvflPtr(MemPage *pPage, u8 *pCell){
  if( pCell ){
    CellInfo info;
    sqlite3BtreeParseCellPtr(pPage, pCell, &info);
    assert( (info.nData+(pPage->intKey?0:info.nKey))==info.nPayload );
    if( (info.nData+(pPage->intKey?0:info.nKey))>info.nLocal ){
      Pgno ovfl = get4byte(&pCell[info.iOverflow]);
      return ptrmapPut(pPage->pBt, ovfl, PTRMAP_OVERFLOW1, pPage->pgno);
    }
  }
  return SQLITE_OK;
}
/*
** If the cell with index iCell on page pPage contains a pointer
** to an overflow page, insert an entry into the pointer-map
** for the overflow page.
*/
static int ptrmapPutOvfl(MemPage *pPage, int iCell){
  u8 *pCell;
  pCell = findOverflowCell(pPage, iCell);
  return ptrmapPutOvflPtr(pPage, pCell);
}
#endif


/*
** Defragment the page given.  All Cells are moved to the
** end of the page and all free space is collected into one
** big FreeBlk that occurs in between the header and cell
** pointer array and the cell content area.
*/
static int defragmentPage(MemPage *pPage){
  int i;                     /* Loop counter */
  int pc;                    /* Address of a i-th cell */
  int addr;                  /* Offset of first byte after cell pointer array */
  int hdr;                   /* Offset to the page header */
  int size;                  /* Size of a cell */
  int usableSize;            /* Number of usable bytes on a page */
  int cellOffset;            /* Offset to the cell pointer array */
  int brk;                   /* Offset to the cell content area */
  int nCell;                 /* Number of cells on the page */
  unsigned char *data;       /* The page data */
  unsigned char *temp;       /* Temp area for cell content */

  assert( sqlite3PagerIswriteable(pPage->pDbPage) );
  assert( pPage->pBt!=0 );
  assert( pPage->pBt->usableSize <= SQLITE_MAX_PAGE_SIZE );
  assert( pPage->nOverflow==0 );
  temp = sqliteMalloc( pPage->pBt->pageSize );
  if( temp==0 ) return SQLITE_NOMEM;
  data = pPage->aData;
  hdr = pPage->hdrOffset;
  cellOffset = pPage->cellOffset;
  nCell = pPage->nCell;
  assert( nCell==get2byte(&data[hdr+3]) );
  usableSize = pPage->pBt->usableSize;
  brk = get2byte(&data[hdr+5]);
  memcpy(&temp[brk], &data[brk], usableSize - brk);
  brk = usableSize;
  for(i=0; i<nCell; i++){
    u8 *pAddr;     /* The i-th cell pointer */
    pAddr = &data[cellOffset + i*2];
    pc = get2byte(pAddr);
    assert( pc<pPage->pBt->usableSize );
    size = cellSizePtr(pPage, &temp[pc]);
    brk -= size;
    memcpy(&data[brk], &temp[pc], size);
    put2byte(pAddr, brk);
  }
  assert( brk>=cellOffset+2*nCell );
  put2byte(&data[hdr+5], brk);
  data[hdr+1] = 0;
  data[hdr+2] = 0;
  data[hdr+7] = 0;
  addr = cellOffset+2*nCell;
  memset(&data[addr], 0, brk-addr);
  sqliteFree(temp);
  return SQLITE_OK;
}

/*
** Allocate nByte bytes of space on a page.
**
** Return the index into pPage->aData[] of the first byte of
** the new allocation. Or return 0 if there is not enough free
** space on the page to satisfy the allocation request.
**
** If the page contains nBytes of free space but does not contain
** nBytes of contiguous free space, then this routine automatically
** calls defragementPage() to consolidate all free space before 
** allocating the new chunk.
*/
static int allocateSpace(MemPage *pPage, int nByte){
  int addr, pc, hdr;
  int size;
  int nFrag;
  int top;
  int nCell;
  int cellOffset;
  unsigned char *data;
  
  data = pPage->aData;
  assert( sqlite3PagerIswriteable(pPage->pDbPage) );
  assert( pPage->pBt );
  if( nByte<4 ) nByte = 4;
  if( pPage->nFree<nByte || pPage->nOverflow>0 ) return 0;
  pPage->nFree -= nByte;
  hdr = pPage->hdrOffset;

  nFrag = data[hdr+7];
  if( nFrag<60 ){
    /* Search the freelist looking for a slot big enough to satisfy the
    ** space request. */
    addr = hdr+1;
    while( (pc = get2byte(&data[addr]))>0 ){
      size = get2byte(&data[pc+2]);
      if( size>=nByte ){
        if( size<nByte+4 ){
          memcpy(&data[addr], &data[pc], 2);
          data[hdr+7] = nFrag + size - nByte;
          return pc;
        }else{
          put2byte(&data[pc+2], size-nByte);
          return pc + size - nByte;
        }
      }
      addr = pc;
    }
  }

  /* Allocate memory from the gap in between the cell pointer array
  ** and the cell content area.
  */
  top = get2byte(&data[hdr+5]);
  nCell = get2byte(&data[hdr+3]);
  cellOffset = pPage->cellOffset;
  if( nFrag>=60 || cellOffset + 2*nCell > top - nByte ){
    if( defragmentPage(pPage) ) return 0;
    top = get2byte(&data[hdr+5]);
  }
  top -= nByte;
  assert( cellOffset + 2*nCell <= top );
  put2byte(&data[hdr+5], top);
  return top;
}

/*
** Return a section of the pPage->aData to the freelist.
** The first byte of the new free block is pPage->aDisk[start]
** and the size of the block is "size" bytes.
**
** Most of the effort here is involved in coalesing adjacent
** free blocks into a single big free block.
*/
static void freeSpace(MemPage *pPage, int start, int size){
  int addr, pbegin, hdr;
  unsigned char *data = pPage->aData;

  assert( pPage->pBt!=0 );
  assert( sqlite3PagerIswriteable(pPage->pDbPage) );
  assert( start>=pPage->hdrOffset+6+(pPage->leaf?0:4) );
  assert( (start + size)<=pPage->pBt->usableSize );
  if( size<4 ) size = 4;

#ifdef SQLITE_SECURE_DELETE
  /* Overwrite deleted information with zeros when the SECURE_DELETE 
  ** option is enabled at compile-time */
  memset(&data[start], 0, size);
#endif

  /* Add the space back into the linked list of freeblocks */
  hdr = pPage->hdrOffset;
  addr = hdr + 1;
  while( (pbegin = get2byte(&data[addr]))<start && pbegin>0 ){
    assert( pbegin<=pPage->pBt->usableSize-4 );
    assert( pbegin>addr );
    addr = pbegin;
  }
  assert( pbegin<=pPage->pBt->usableSize-4 );
  assert( pbegin>addr || pbegin==0 );
  put2byte(&data[addr], start);
  put2byte(&data[start], pbegin);
  put2byte(&data[start+2], size);
  pPage->nFree += size;

  /* Coalesce adjacent free blocks */
  addr = pPage->hdrOffset + 1;
  while( (pbegin = get2byte(&data[addr]))>0 ){
    int pnext, psize;
    assert( pbegin>addr );
    assert( pbegin<=pPage->pBt->usableSize-4 );
    pnext = get2byte(&data[pbegin]);
    psize = get2byte(&data[pbegin+2]);
    if( pbegin + psize + 3 >= pnext && pnext>0 ){
      int frag = pnext - (pbegin+psize);
      assert( frag<=data[pPage->hdrOffset+7] );
      data[pPage->hdrOffset+7] -= frag;
      put2byte(&data[pbegin], get2byte(&data[pnext]));
      put2byte(&data[pbegin+2], pnext+get2byte(&data[pnext+2])-pbegin);
    }else{
      addr = pbegin;
    }
  }

  /* If the cell content area begins with a freeblock, remove it. */
  if( data[hdr+1]==data[hdr+5] && data[hdr+2]==data[hdr+6] ){
    int top;
    pbegin = get2byte(&data[hdr+1]);
    memcpy(&data[hdr+1], &data[pbegin], 2);
    top = get2byte(&data[hdr+5]);
    put2byte(&data[hdr+5], top + get2byte(&data[pbegin+2]));
  }
}

/*
** Decode the flags byte (the first byte of the header) for a page
** and initialize fields of the MemPage structure accordingly.
*/
static void decodeFlags(MemPage *pPage, int flagByte){
  BtShared *pBt;     /* A copy of pPage->pBt */

  assert( pPage->hdrOffset==(pPage->pgno==1 ? 100 : 0) );
  pPage->intKey = (flagByte & (PTF_INTKEY|PTF_LEAFDATA))!=0;
  pPage->zeroData = (flagByte & PTF_ZERODATA)!=0;
  pPage->leaf = (flagByte & PTF_LEAF)!=0;
  pPage->childPtrSize = 4*(pPage->leaf==0);
  pBt = pPage->pBt;
  if( flagByte & PTF_LEAFDATA ){
    pPage->leafData = 1;
    pPage->maxLocal = pBt->maxLeaf;
    pPage->minLocal = pBt->minLeaf;
  }else{
    pPage->leafData = 0;
    pPage->maxLocal = pBt->maxLocal;
    pPage->minLocal = pBt->minLocal;
  }
  pPage->hasData = !(pPage->zeroData || (!pPage->leaf && pPage->leafData));
}

/*
** Initialize the auxiliary information for a disk block.
**
** The pParent parameter must be a pointer to the MemPage which
** is the parent of the page being initialized.  The root of a
** BTree has no parent and so for that page, pParent==NULL.
**
** Return SQLITE_OK on success.  If we see that the page does
** not contain a well-formed database page, then return 
** SQLITE_CORRUPT.  Note that a return of SQLITE_OK does not
** guarantee that the page is well-formed.  It only shows that
** we failed to detect any corruption.
*/
SQLITE_PRIVATE int sqlite3BtreeInitPage(
  MemPage *pPage,        /* The page to be initialized */
  MemPage *pParent       /* The parent.  Might be NULL */
){
  int pc;            /* Address of a freeblock within pPage->aData[] */
  int hdr;           /* Offset to beginning of page header */
  u8 *data;          /* Equal to pPage->aData */
  BtShared *pBt;        /* The main btree structure */
  int usableSize;    /* Amount of usable space on each page */
  int cellOffset;    /* Offset from start of page to first cell pointer */
  int nFree;         /* Number of unused bytes on the page */
  int top;           /* First byte of the cell content area */

  pBt = pPage->pBt;
  assert( pBt!=0 );
  assert( pParent==0 || pParent->pBt==pBt );
  assert( pPage->pgno==sqlite3PagerPagenumber(pPage->pDbPage) );
  assert( pPage->aData == &((unsigned char*)pPage)[-pBt->pageSize] );
  if( pPage->pParent!=pParent && (pPage->pParent!=0 || pPage->isInit) ){
    /* The parent page should never change unless the file is corrupt */
    return SQLITE_CORRUPT_BKPT;
  }
  if( pPage->isInit ) return SQLITE_OK;
  if( pPage->pParent==0 && pParent!=0 ){
    pPage->pParent = pParent;
    sqlite3PagerRef(pParent->pDbPage);
  }
  hdr = pPage->hdrOffset;
  data = pPage->aData;
  decodeFlags(pPage, data[hdr]);
  pPage->nOverflow = 0;
  pPage->idxShift = 0;
  usableSize = pBt->usableSize;
  pPage->cellOffset = cellOffset = hdr + 12 - 4*pPage->leaf;
  top = get2byte(&data[hdr+5]);
  pPage->nCell = get2byte(&data[hdr+3]);
  if( pPage->nCell>MX_CELL(pBt) ){
    /* To many cells for a single page.  The page must be corrupt */
    return SQLITE_CORRUPT_BKPT;
  }
  if( pPage->nCell==0 && pParent!=0 && pParent->pgno!=1 ){
    /* All pages must have at least one cell, except for root pages */
    return SQLITE_CORRUPT_BKPT;
  }

  /* Compute the total free space on the page */
  pc = get2byte(&data[hdr+1]);
  nFree = data[hdr+7] + top - (cellOffset + 2*pPage->nCell);
  while( pc>0 ){
    int next, size;
    if( pc>usableSize-4 ){
      /* Free block is off the page */
      return SQLITE_CORRUPT_BKPT; 
    }
    next = get2byte(&data[pc]);
    size = get2byte(&data[pc+2]);
    if( next>0 && next<=pc+size+3 ){
      /* Free blocks must be in accending order */
      return SQLITE_CORRUPT_BKPT; 
    }
    nFree += size;
    pc = next;
  }
  pPage->nFree = nFree;
  if( nFree>=usableSize ){
    /* Free space cannot exceed total page size */
    return SQLITE_CORRUPT_BKPT; 
  }

  pPage->isInit = 1;
  return SQLITE_OK;
}

/*
** Set up a raw page so that it looks like a database page holding
** no entries.
*/
static void zeroPage(MemPage *pPage, int flags){
  unsigned char *data = pPage->aData;
  BtShared *pBt = pPage->pBt;
  int hdr = pPage->hdrOffset;
  int first;

  assert( sqlite3PagerPagenumber(pPage->pDbPage)==pPage->pgno );
  assert( &data[pBt->pageSize] == (unsigned char*)pPage );
  assert( sqlite3PagerIswriteable(pPage->pDbPage) );
  memset(&data[hdr], 0, pBt->usableSize - hdr);
  data[hdr] = flags;
  first = hdr + 8 + 4*((flags&PTF_LEAF)==0);
  memset(&data[hdr+1], 0, 4);
  data[hdr+7] = 0;
  put2byte(&data[hdr+5], pBt->usableSize);
  pPage->nFree = pBt->usableSize - first;
  decodeFlags(pPage, flags);
  pPage->hdrOffset = hdr;
  pPage->cellOffset = first;
  pPage->nOverflow = 0;
  pPage->idxShift = 0;
  pPage->nCell = 0;
  pPage->isInit = 1;
}

/*
** Get a page from the pager.  Initialize the MemPage.pBt and
** MemPage.aData elements if needed.
**
** If the noContent flag is set, it means that we do not care about
** the content of the page at this time.  So do not go to the disk
** to fetch the content.  Just fill in the content with zeros for now.
** If in the future we call sqlite3PagerWrite() on this page, that
** means we have started to be concerned about content and the disk
** read should occur at that point.
*/
SQLITE_PRIVATE int sqlite3BtreeGetPage(
  BtShared *pBt,       /* The btree */
  Pgno pgno,           /* Number of the page to fetch */
  MemPage **ppPage,    /* Return the page in this parameter */
  int noContent        /* Do not load page content if true */
){
  int rc;
  MemPage *pPage;
  DbPage *pDbPage;

  rc = sqlite3PagerAcquire(pBt->pPager, pgno, (DbPage**)&pDbPage, noContent);
  if( rc ) return rc;
  pPage = (MemPage *)sqlite3PagerGetExtra(pDbPage);
  pPage->aData = sqlite3PagerGetData(pDbPage);
  pPage->pDbPage = pDbPage;
  pPage->pBt = pBt;
  pPage->pgno = pgno;
  pPage->hdrOffset = pPage->pgno==1 ? 100 : 0;
  *ppPage = pPage;
  return SQLITE_OK;
}

/*
** Get a page from the pager and initialize it.  This routine
** is just a convenience wrapper around separate calls to
** sqlite3BtreeGetPage() and sqlite3BtreeInitPage().
*/
static int getAndInitPage(
  BtShared *pBt,          /* The database file */
  Pgno pgno,           /* Number of the page to get */
  MemPage **ppPage,    /* Write the page pointer here */
  MemPage *pParent     /* Parent of the page */
){
  int rc;
  if( pgno==0 ){
    return SQLITE_CORRUPT_BKPT; 
  }
  rc = sqlite3BtreeGetPage(pBt, pgno, ppPage, 0);
  if( rc==SQLITE_OK && (*ppPage)->isInit==0 ){
    rc = sqlite3BtreeInitPage(*ppPage, pParent);
  }
  return rc;
}

/*
** Release a MemPage.  This should be called once for each prior
** call to sqlite3BtreeGetPage.
*/
static void releasePage(MemPage *pPage){
  if( pPage ){
    assert( pPage->aData );
    assert( pPage->pBt );
    assert( &pPage->aData[pPage->pBt->pageSize]==(unsigned char*)pPage );
    sqlite3PagerUnref(pPage->pDbPage);
  }
}

/*
** This routine is called when the reference count for a page
** reaches zero.  We need to unref the pParent pointer when that
** happens.
*/
static void pageDestructor(DbPage *pData, int pageSize){
  MemPage *pPage;
  assert( (pageSize & 7)==0 );
  pPage = (MemPage *)sqlite3PagerGetExtra(pData);
  if( pPage->pParent ){
    MemPage *pParent = pPage->pParent;
    pPage->pParent = 0;
    releasePage(pParent);
  }
  pPage->isInit = 0;
}

/*
** During a rollback, when the pager reloads information into the cache
** so that the cache is restored to its original state at the start of
** the transaction, for each page restored this routine is called.
**
** This routine needs to reset the extra data section at the end of the
** page to agree with the restored data.
*/
static void pageReinit(DbPage *pData, int pageSize){
  MemPage *pPage;
  assert( (pageSize & 7)==0 );
  pPage = (MemPage *)sqlite3PagerGetExtra(pData);
  if( pPage->isInit ){
    pPage->isInit = 0;
    sqlite3BtreeInitPage(pPage, pPage->pParent);
  }
}

/*
** Open a database file.
** 
** zFilename is the name of the database file.  If zFilename is NULL
** a new database with a random name is created.  This randomly named
** database file will be deleted when sqlite3BtreeClose() is called.
*/
SQLITE_PRIVATE int sqlite3BtreeOpen(
  const char *zFilename,  /* Name of the file containing the BTree database */
  sqlite3 *pSqlite,       /* Associated database handle */
  Btree **ppBtree,        /* Pointer to new Btree object written here */
  int flags               /* Options */
){
  BtShared *pBt;          /* Shared part of btree structure */
  Btree *p;               /* Handle to return */
  int rc = SQLITE_OK;
  int nReserve;
  unsigned char zDbHeader[100];
#if !defined(SQLITE_OMIT_SHARED_CACHE) && !defined(SQLITE_OMIT_DISKIO)
  const ThreadData *pTsdro;
#endif

  /* Set the variable isMemdb to true for an in-memory database, or 
  ** false for a file-based database. This symbol is only required if
  ** either of the shared-data or autovacuum features are compiled 
  ** into the library.
  */
#if !defined(SQLITE_OMIT_SHARED_CACHE) || !defined(SQLITE_OMIT_AUTOVACUUM)
  #ifdef SQLITE_OMIT_MEMORYDB
    const int isMemdb = 0;
  #else
    const int isMemdb = zFilename && !strcmp(zFilename, ":memory:");
  #endif
#endif

  p = sqliteMalloc(sizeof(Btree));
  if( !p ){
    return SQLITE_NOMEM;
  }
  p->inTrans = TRANS_NONE;
  p->pSqlite = pSqlite;

  /* Try to find an existing Btree structure opened on zFilename. */
#if !defined(SQLITE_OMIT_SHARED_CACHE) && !defined(SQLITE_OMIT_DISKIO)
  pTsdro = sqlite3ThreadDataReadOnly();
  if( pTsdro->useSharedData && zFilename && !isMemdb ){
    char *zFullPathname = sqlite3OsFullPathname(zFilename);
    if( !zFullPathname ){
      sqliteFree(p);
      return SQLITE_NOMEM;
    }
    for(pBt=pTsdro->pBtree; pBt; pBt=pBt->pNext){
      assert( pBt->nRef>0 );
      if( 0==strcmp(zFullPathname, sqlite3PagerFilename(pBt->pPager)) ){
        p->pBt = pBt;
        *ppBtree = p;
        pBt->nRef++;
        sqliteFree(zFullPathname);
        return SQLITE_OK;
      }
    }
    sqliteFree(zFullPathname);
  }
#endif

  /*
  ** The following asserts make sure that structures used by the btree are
  ** the right size.  This is to guard against size changes that result
  ** when compiling on a different architecture.
  */
  assert( sizeof(i64)==8 || sizeof(i64)==4 );
  assert( sizeof(u64)==8 || sizeof(u64)==4 );
  assert( sizeof(u32)==4 );
  assert( sizeof(u16)==2 );
  assert( sizeof(Pgno)==4 );

  pBt = sqliteMalloc( sizeof(*pBt) );
  if( pBt==0 ){
    rc = SQLITE_NOMEM;
    goto btree_open_out;
  }
  rc = sqlite3PagerOpen(&pBt->pPager, zFilename, EXTRA_SIZE, flags);
  if( rc==SQLITE_OK ){
    rc = sqlite3PagerReadFileheader(pBt->pPager,sizeof(zDbHeader),zDbHeader);
  }
  if( rc!=SQLITE_OK ){
    goto btree_open_out;
  }
  p->pBt = pBt;

  sqlite3PagerSetDestructor(pBt->pPager, pageDestructor);
  sqlite3PagerSetReiniter(pBt->pPager, pageReinit);
  pBt->pCursor = 0;
  pBt->pPage1 = 0;
  pBt->readOnly = sqlite3PagerIsreadonly(pBt->pPager);
  pBt->pageSize = get2byte(&zDbHeader[16]);
  if( pBt->pageSize<512 || pBt->pageSize>SQLITE_MAX_PAGE_SIZE
       || ((pBt->pageSize-1)&pBt->pageSize)!=0 ){
    pBt->pageSize = SQLITE_DEFAULT_PAGE_SIZE;
    pBt->maxEmbedFrac = 64;   /* 25% */
    pBt->minEmbedFrac = 32;   /* 12.5% */
    pBt->minLeafFrac = 32;    /* 12.5% */
#ifndef SQLITE_OMIT_AUTOVACUUM
    /* If the magic name ":memory:" will create an in-memory database, then
    ** leave the autoVacuum mode at 0 (do not auto-vacuum), even if
    ** SQLITE_DEFAULT_AUTOVACUUM is true. On the other hand, if
    ** SQLITE_OMIT_MEMORYDB has been defined, then ":memory:" is just a
    ** regular file-name. In this case the auto-vacuum applies as per normal.
    */
    if( zFilename && !isMemdb ){
      pBt->autoVacuum = (SQLITE_DEFAULT_AUTOVACUUM ? 1 : 0);
      pBt->incrVacuum = (SQLITE_DEFAULT_AUTOVACUUM==2 ? 1 : 0);
    }
#endif
    nReserve = 0;
  }else{
    nReserve = zDbHeader[20];
    pBt->maxEmbedFrac = zDbHeader[21];
    pBt->minEmbedFrac = zDbHeader[22];
    pBt->minLeafFrac = zDbHeader[23];
    pBt->pageSizeFixed = 1;
#ifndef SQLITE_OMIT_AUTOVACUUM
    pBt->autoVacuum = (get4byte(&zDbHeader[36 + 4*4])?1:0);
    pBt->incrVacuum = (get4byte(&zDbHeader[36 + 7*4])?1:0);
#endif
  }
  pBt->usableSize = pBt->pageSize - nReserve;
  assert( (pBt->pageSize & 7)==0 );  /* 8-byte alignment of pageSize */
  sqlite3PagerSetPagesize(pBt->pPager, pBt->pageSize);

#if !defined(SQLITE_OMIT_SHARED_CACHE) && !defined(SQLITE_OMIT_DISKIO)
  /* Add the new btree to the linked list starting at ThreadData.pBtree.
  ** There is no chance that a malloc() may fail inside of the 
  ** sqlite3ThreadData() call, as the ThreadData structure must have already
  ** been allocated for pTsdro->useSharedData to be non-zero.
  */
  if( pTsdro->useSharedData && zFilename && !isMemdb ){
    pBt->pNext = pTsdro->pBtree;
    sqlite3ThreadData()->pBtree = pBt;
  }
#endif
  pBt->nRef = 1;
  *ppBtree = p;

btree_open_out:
  if( rc!=SQLITE_OK ){
    if( pBt && pBt->pPager ){
      sqlite3PagerClose(pBt->pPager);
    }
    sqliteFree(pBt);
    sqliteFree(p);
    *ppBtree = 0;
  }
  return rc;
}

/*
** Close an open database and invalidate all cursors.
*/
SQLITE_PRIVATE int sqlite3BtreeClose(Btree *p){
  BtShared *pBt = p->pBt;
  BtCursor *pCur;

#ifndef SQLITE_OMIT_SHARED_CACHE
  ThreadData *pTsd;
#endif

  /* Close all cursors opened via this handle.  */
  pCur = pBt->pCursor;
  while( pCur ){
    BtCursor *pTmp = pCur;
    pCur = pCur->pNext;
    if( pTmp->pBtree==p ){
      sqlite3BtreeCloseCursor(pTmp);
    }
  }

  /* Rollback any active transaction and free the handle structure.
  ** The call to sqlite3BtreeRollback() drops any table-locks held by
  ** this handle.
  */
  sqlite3BtreeRollback(p);
  sqliteFree(p);

#ifndef SQLITE_OMIT_SHARED_CACHE
  /* If there are still other outstanding references to the shared-btree
  ** structure, return now. The remainder of this procedure cleans 
  ** up the shared-btree.
  */
  assert( pBt->nRef>0 );
  pBt->nRef--;
  if( pBt->nRef ){
    return SQLITE_OK;
  }

  /* Remove the shared-btree from the thread wide list. Call 
  ** ThreadDataReadOnly() and then cast away the const property of the 
  ** pointer to avoid allocating thread data if it is not really required.
  */
  pTsd = (ThreadData *)sqlite3ThreadDataReadOnly();
  if( pTsd->pBtree==pBt ){
    assert( pTsd==sqlite3ThreadData() );
    pTsd->pBtree = pBt->pNext;
  }else{
    BtShared *pPrev;
    for(pPrev=pTsd->pBtree; pPrev && pPrev->pNext!=pBt; pPrev=pPrev->pNext){}
    if( pPrev ){
      assert( pTsd==sqlite3ThreadData() );
      pPrev->pNext = pBt->pNext;
    }
  }
#endif

  /* Close the pager and free the shared-btree structure */
  assert( !pBt->pCursor );
  sqlite3PagerClose(pBt->pPager);
  if( pBt->xFreeSchema && pBt->pSchema ){
    pBt->xFreeSchema(pBt->pSchema);
  }
  sqliteFree(pBt->pSchema);
  sqliteFree(pBt);
  return SQLITE_OK;
}

/*
** Change the busy handler callback function.
*/
SQLITE_PRIVATE int sqlite3BtreeSetBusyHandler(Btree *p, BusyHandler *pHandler){
  BtShared *pBt = p->pBt;
  pBt->pBusyHandler = pHandler;
  sqlite3PagerSetBusyhandler(pBt->pPager, pHandler);
  return SQLITE_OK;
}

/*
** Change the limit on the number of pages allowed in the cache.
**
** The maximum number of cache pages is set to the absolute
** value of mxPage.  If mxPage is negative, the pager will
** operate asynchronously - it will not stop to do fsync()s
** to insure data is written to the disk surface before
** continuing.  Transactions still work if synchronous is off,
** and the database cannot be corrupted if this program
** crashes.  But if the operating system crashes or there is
** an abrupt power failure when synchronous is off, the database
** could be left in an inconsistent and unrecoverable state.
** Synchronous is on by default so database corruption is not
** normally a worry.
*/
SQLITE_PRIVATE int sqlite3BtreeSetCacheSize(Btree *p, int mxPage){
  BtShared *pBt = p->pBt;
  sqlite3PagerSetCachesize(pBt->pPager, mxPage);
  return SQLITE_OK;
}

/*
** Change the way data is synced to disk in order to increase or decrease
** how well the database resists damage due to OS crashes and power
** failures.  Level 1 is the same as asynchronous (no syncs() occur and
** there is a high probability of damage)  Level 2 is the default.  There
** is a very low but non-zero probability of damage.  Level 3 reduces the
** probability of damage to near zero but with a write performance reduction.
*/
#ifndef SQLITE_OMIT_PAGER_PRAGMAS
SQLITE_PRIVATE int sqlite3BtreeSetSafetyLevel(Btree *p, int level, int fullSync){
  BtShared *pBt = p->pBt;
  sqlite3PagerSetSafetyLevel(pBt->pPager, level, fullSync);
  return SQLITE_OK;
}
#endif

/*
** Return TRUE if the given btree is set to safety level 1.  In other
** words, return TRUE if no sync() occurs on the disk files.
*/
SQLITE_PRIVATE int sqlite3BtreeSyncDisabled(Btree *p){
  BtShared *pBt = p->pBt;
  assert( pBt && pBt->pPager );
  return sqlite3PagerNosync(pBt->pPager);
}

#if !defined(SQLITE_OMIT_PAGER_PRAGMAS) || !defined(SQLITE_OMIT_VACUUM)
/*
** Change the default pages size and the number of reserved bytes per page.
**
** The page size must be a power of 2 between 512 and 65536.  If the page
** size supplied does not meet this constraint then the page size is not
** changed.
**
** Page sizes are constrained to be a power of two so that the region
** of the database file used for locking (beginning at PENDING_BYTE,
** the first byte past the 1GB boundary, 0x40000000) needs to occur
** at the beginning of a page.
**
** If parameter nReserve is less than zero, then the number of reserved
** bytes per page is left unchanged.
*/
SQLITE_PRIVATE int sqlite3BtreeSetPageSize(Btree *p, int pageSize, int nReserve){
  BtShared *pBt = p->pBt;
  if( pBt->pageSizeFixed ){
    return SQLITE_READONLY;
  }
  if( nReserve<0 ){
    nReserve = pBt->pageSize - pBt->usableSize;
  }
  if( pageSize>=512 && pageSize<=SQLITE_MAX_PAGE_SIZE &&
        ((pageSize-1)&pageSize)==0 ){
    assert( (pageSize & 7)==0 );
    assert( !pBt->pPage1 && !pBt->pCursor );
    pBt->pageSize = sqlite3PagerSetPagesize(pBt->pPager, pageSize);
  }
  pBt->usableSize = pBt->pageSize - nReserve;
  return SQLITE_OK;
}

/*
** Return the currently defined page size
*/
SQLITE_PRIVATE int sqlite3BtreeGetPageSize(Btree *p){
  return p->pBt->pageSize;
}
SQLITE_PRIVATE int sqlite3BtreeGetReserve(Btree *p){
  return p->pBt->pageSize - p->pBt->usableSize;
}

/*
** Set the maximum page count for a database if mxPage is positive.
** No changes are made if mxPage is 0 or negative.
** Regardless of the value of mxPage, return the maximum page count.
*/
SQLITE_PRIVATE int sqlite3BtreeMaxPageCount(Btree *p, int mxPage){
  return sqlite3PagerMaxPageCount(p->pBt->pPager, mxPage);
}
#endif /* !defined(SQLITE_OMIT_PAGER_PRAGMAS) || !defined(SQLITE_OMIT_VACUUM) */

/*
** Change the 'auto-vacuum' property of the database. If the 'autoVacuum'
** parameter is non-zero, then auto-vacuum mode is enabled. If zero, it
** is disabled. The default value for the auto-vacuum property is 
** determined by the SQLITE_DEFAULT_AUTOVACUUM macro.
*/
SQLITE_PRIVATE int sqlite3BtreeSetAutoVacuum(Btree *p, int autoVacuum){
#ifdef SQLITE_OMIT_AUTOVACUUM
  return SQLITE_READONLY;
#else
  BtShared *pBt = p->pBt;
  int av = (autoVacuum?1:0);
  if( pBt->pageSizeFixed && av!=pBt->autoVacuum ){
    return SQLITE_READONLY;
  }
  pBt->autoVacuum = av;
  return SQLITE_OK;
#endif
}

/*
** Return the value of the 'auto-vacuum' property. If auto-vacuum is 
** enabled 1 is returned. Otherwise 0.
*/
SQLITE_PRIVATE int sqlite3BtreeGetAutoVacuum(Btree *p){
#ifdef SQLITE_OMIT_AUTOVACUUM
  return BTREE_AUTOVACUUM_NONE;
#else
  return (
    (!p->pBt->autoVacuum)?BTREE_AUTOVACUUM_NONE:
    (!p->pBt->incrVacuum)?BTREE_AUTOVACUUM_FULL:
    BTREE_AUTOVACUUM_INCR
  );
#endif
}


/*
** Get a reference to pPage1 of the database file.  This will
** also acquire a readlock on that file.
**
** SQLITE_OK is returned on success.  If the file is not a
** well-formed database file, then SQLITE_CORRUPT is returned.
** SQLITE_BUSY is returned if the database is locked.  SQLITE_NOMEM
** is returned if we run out of memory. 
*/
static int lockBtree(BtShared *pBt){
  int rc, pageSize;
  MemPage *pPage1;
  if( pBt->pPage1 ) return SQLITE_OK;
  rc = sqlite3BtreeGetPage(pBt, 1, &pPage1, 0);
  if( rc!=SQLITE_OK ) return rc;
  

  /* Do some checking to help insure the file we opened really is
  ** a valid database file. 
  */
  rc = SQLITE_NOTADB;
  if( sqlite3PagerPagecount(pBt->pPager)>0 ){
    u8 *page1 = pPage1->aData;
    if( memcmp(page1, zMagicHeader, 16)!=0 ){
      goto page1_init_failed;
    }
    if( page1[18]>1 ){
      pBt->readOnly = 1;
    }
    if( page1[19]>1 ){
      goto page1_init_failed;
    }
    pageSize = get2byte(&page1[16]);
    if( ((pageSize-1)&pageSize)!=0 || pageSize<512 ){
      goto page1_init_failed;
    }
    assert( (pageSize & 7)==0 );
    pBt->pageSize = pageSize;
    pBt->usableSize = pageSize - page1[20];
    if( pBt->usableSize<500 ){
      goto page1_init_failed;
    }
    pBt->maxEmbedFrac = page1[21];
    pBt->minEmbedFrac = page1[22];
    pBt->minLeafFrac = page1[23];
#ifndef SQLITE_OMIT_AUTOVACUUM
    pBt->autoVacuum = (get4byte(&page1[36 + 4*4])?1:0);
    pBt->incrVacuum = (get4byte(&page1[36 + 7*4])?1:0);
#endif
  }

  /* maxLocal is the maximum amount of payload to store locally for
  ** a cell.  Make sure it is small enough so that at least minFanout
  ** cells can will fit on one page.  We assume a 10-byte page header.
  ** Besides the payload, the cell must store:
  **     2-byte pointer to the cell
  **     4-byte child pointer
  **     9-byte nKey value
  **     4-byte nData value
  **     4-byte overflow page pointer
  ** So a cell consists of a 2-byte poiner, a header which is as much as
  ** 17 bytes long, 0 to N bytes of payload, and an optional 4 byte overflow
  ** page pointer.
  */
  pBt->maxLocal = (pBt->usableSize-12)*pBt->maxEmbedFrac/255 - 23;
  pBt->minLocal = (pBt->usableSize-12)*pBt->minEmbedFrac/255 - 23;
  pBt->maxLeaf = pBt->usableSize - 35;
  pBt->minLeaf = (pBt->usableSize-12)*pBt->minLeafFrac/255 - 23;
  if( pBt->minLocal>pBt->maxLocal || pBt->maxLocal<0 ){
    goto page1_init_failed;
  }
  assert( pBt->maxLeaf + 23 <= MX_CELL_SIZE(pBt) );
  pBt->pPage1 = pPage1;
  return SQLITE_OK;

page1_init_failed:
  releasePage(pPage1);
  pBt->pPage1 = 0;
  return rc;
}

/*
** This routine works like lockBtree() except that it also invokes the
** busy callback if there is lock contention.
*/
static int lockBtreeWithRetry(Btree *pRef){
  int rc = SQLITE_OK;
  if( pRef->inTrans==TRANS_NONE ){
    u8 inTransaction = pRef->pBt->inTransaction;
    btreeIntegrity(pRef);
    rc = sqlite3BtreeBeginTrans(pRef, 0);
    pRef->pBt->inTransaction = inTransaction;
    pRef->inTrans = TRANS_NONE;
    if( rc==SQLITE_OK ){
      pRef->pBt->nTransaction--;
    }
    btreeIntegrity(pRef);
  }
  return rc;
}
       

/*
** If there are no outstanding cursors and we are not in the middle
** of a transaction but there is a read lock on the database, then
** this routine unrefs the first page of the database file which 
** has the effect of releasing the read lock.
**
** If there are any outstanding cursors, this routine is a no-op.
**
** If there is a transaction in progress, this routine is a no-op.
*/
static void unlockBtreeIfUnused(BtShared *pBt){
  if( pBt->inTransaction==TRANS_NONE && pBt->pCursor==0 && pBt->pPage1!=0 ){
    if( sqlite3PagerRefcount(pBt->pPager)>=1 ){
      if( pBt->pPage1->aData==0 ){
        MemPage *pPage = pBt->pPage1;
        pPage->aData = &((u8*)pPage)[-pBt->pageSize];
        pPage->pBt = pBt;
        pPage->pgno = 1;
      }
      releasePage(pBt->pPage1);
    }
    pBt->pPage1 = 0;
    pBt->inStmt = 0;
  }
}

/*
** Create a new database by initializing the first page of the
** file.
*/
static int newDatabase(BtShared *pBt){
  MemPage *pP1;
  unsigned char *data;
  int rc;
  if( sqlite3PagerPagecount(pBt->pPager)>0 ) return SQLITE_OK;
  pP1 = pBt->pPage1;
  assert( pP1!=0 );
  data = pP1->aData;
  rc = sqlite3PagerWrite(pP1->pDbPage);
  if( rc ) return rc;
  memcpy(data, zMagicHeader, sizeof(zMagicHeader));
  assert( sizeof(zMagicHeader)==16 );
  put2byte(&data[16], pBt->pageSize);
  data[18] = 1;
  data[19] = 1;
  data[20] = pBt->pageSize - pBt->usableSize;
  data[21] = pBt->maxEmbedFrac;
  data[22] = pBt->minEmbedFrac;
  data[23] = pBt->minLeafFrac;
  memset(&data[24], 0, 100-24);
  zeroPage(pP1, PTF_INTKEY|PTF_LEAF|PTF_LEAFDATA );
  pBt->pageSizeFixed = 1;
#ifndef SQLITE_OMIT_AUTOVACUUM
  assert( pBt->autoVacuum==1 || pBt->autoVacuum==0 );
  assert( pBt->incrVacuum==1 || pBt->incrVacuum==0 );
  put4byte(&data[36 + 4*4], pBt->autoVacuum);
  put4byte(&data[36 + 7*4], pBt->incrVacuum);
#endif
  return SQLITE_OK;
}

/*
** Attempt to start a new transaction. A write-transaction
** is started if the second argument is nonzero, otherwise a read-
** transaction.  If the second argument is 2 or more and exclusive
** transaction is started, meaning that no other process is allowed
** to access the database.  A preexisting transaction may not be
** upgraded to exclusive by calling this routine a second time - the
** exclusivity flag only works for a new transaction.
**
** A write-transaction must be started before attempting any 
** changes to the database.  None of the following routines 
** will work unless a transaction is started first:
**
**      sqlite3BtreeCreateTable()
**      sqlite3BtreeCreateIndex()
**      sqlite3BtreeClearTable()
**      sqlite3BtreeDropTable()
**      sqlite3BtreeInsert()
**      sqlite3BtreeDelete()
**      sqlite3BtreeUpdateMeta()
**
** If an initial attempt to acquire the lock fails because of lock contention
** and the database was previously unlocked, then invoke the busy handler
** if there is one.  But if there was previously a read-lock, do not
** invoke the busy handler - just return SQLITE_BUSY.  SQLITE_BUSY is 
** returned when there is already a read-lock in order to avoid a deadlock.
**
** Suppose there are two processes A and B.  A has a read lock and B has
** a reserved lock.  B tries to promote to exclusive but is blocked because
** of A's read lock.  A tries to promote to reserved but is blocked by B.
** One or the other of the two processes must give way or there can be
** no progress.  By returning SQLITE_BUSY and not invoking the busy callback
** when A already has a read lock, we encourage A to give up and let B
** proceed.
*/
SQLITE_PRIVATE int sqlite3BtreeBeginTrans(Btree *p, int wrflag){
  BtShared *pBt = p->pBt;
  int rc = SQLITE_OK;

  btreeIntegrity(p);

  /* If the btree is already in a write-transaction, or it
  ** is already in a read-transaction and a read-transaction
  ** is requested, this is a no-op.
  */
  if( p->inTrans==TRANS_WRITE || (p->inTrans==TRANS_READ && !wrflag) ){
    return SQLITE_OK;
  }

  /* Write transactions are not possible on a read-only database */
  if( pBt->readOnly && wrflag ){
    return SQLITE_READONLY;
  }

  /* If another database handle has already opened a write transaction 
  ** on this shared-btree structure and a second write transaction is
  ** requested, return SQLITE_BUSY.
  */
  if( pBt->inTransaction==TRANS_WRITE && wrflag ){
    return SQLITE_BUSY;
  }

  do {
    if( pBt->pPage1==0 ){
      rc = lockBtree(pBt);
    }

    if( rc==SQLITE_OK && wrflag ){
      if( pBt->readOnly ){
        rc = SQLITE_READONLY;
      }else{
        rc = sqlite3PagerBegin(pBt->pPage1->pDbPage, wrflag>1);
        if( rc==SQLITE_OK ){
          rc = newDatabase(pBt);
        }
      }
    }
  
    if( rc==SQLITE_OK ){
      if( wrflag ) pBt->inStmt = 0;
    }else{
      unlockBtreeIfUnused(pBt);
    }
  }while( rc==SQLITE_BUSY && pBt->inTransaction==TRANS_NONE &&
          sqlite3InvokeBusyHandler(pBt->pBusyHandler) );

  if( rc==SQLITE_OK ){
    if( p->inTrans==TRANS_NONE ){
      pBt->nTransaction++;
    }
    p->inTrans = (wrflag?TRANS_WRITE:TRANS_READ);
    if( p->inTrans>pBt->inTransaction ){
      pBt->inTransaction = p->inTrans;
    }
  }

  btreeIntegrity(p);
  return rc;
}

#ifndef SQLITE_OMIT_AUTOVACUUM

/*
** Set the pointer-map entries for all children of page pPage. Also, if
** pPage contains cells that point to overflow pages, set the pointer
** map entries for the overflow pages as well.
*/
static int setChildPtrmaps(MemPage *pPage){
  int i;                             /* Counter variable */
  int nCell;                         /* Number of cells in page pPage */
  int rc;                            /* Return code */
  BtShared *pBt = pPage->pBt;
  int isInitOrig = pPage->isInit;
  Pgno pgno = pPage->pgno;

  rc = sqlite3BtreeInitPage(pPage, pPage->pParent);
  if( rc!=SQLITE_OK ){
    goto set_child_ptrmaps_out;
  }
  nCell = pPage->nCell;

  for(i=0; i<nCell; i++){
    u8 *pCell = findCell(pPage, i);

    rc = ptrmapPutOvflPtr(pPage, pCell);
    if( rc!=SQLITE_OK ){
      goto set_child_ptrmaps_out;
    }

    if( !pPage->leaf ){
      Pgno childPgno = get4byte(pCell);
      rc = ptrmapPut(pBt, childPgno, PTRMAP_BTREE, pgno);
      if( rc!=SQLITE_OK ) goto set_child_ptrmaps_out;
    }
  }

  if( !pPage->leaf ){
    Pgno childPgno = get4byte(&pPage->aData[pPage->hdrOffset+8]);
    rc = ptrmapPut(pBt, childPgno, PTRMAP_BTREE, pgno);
  }

set_child_ptrmaps_out:
  pPage->isInit = isInitOrig;
  return rc;
}

/*
** Somewhere on pPage, which is guarenteed to be a btree page, not an overflow
** page, is a pointer to page iFrom. Modify this pointer so that it points to
** iTo. Parameter eType describes the type of pointer to be modified, as 
** follows:
**
** PTRMAP_BTREE:     pPage is a btree-page. The pointer points at a child 
**                   page of pPage.
**
** PTRMAP_OVERFLOW1: pPage is a btree-page. The pointer points at an overflow
**                   page pointed to by one of the cells on pPage.
**
** PTRMAP_OVERFLOW2: pPage is an overflow-page. The pointer points at the next
**                   overflow page in the list.
*/
static int modifyPagePointer(MemPage *pPage, Pgno iFrom, Pgno iTo, u8 eType){
  if( eType==PTRMAP_OVERFLOW2 ){
    /* The pointer is always the first 4 bytes of the page in this case.  */
    if( get4byte(pPage->aData)!=iFrom ){
      return SQLITE_CORRUPT_BKPT;
    }
    put4byte(pPage->aData, iTo);
  }else{
    int isInitOrig = pPage->isInit;
    int i;
    int nCell;

    sqlite3BtreeInitPage(pPage, 0);
    nCell = pPage->nCell;

    for(i=0; i<nCell; i++){
      u8 *pCell = findCell(pPage, i);
      if( eType==PTRMAP_OVERFLOW1 ){
        CellInfo info;
        sqlite3BtreeParseCellPtr(pPage, pCell, &info);
        if( info.iOverflow ){
          if( iFrom==get4byte(&pCell[info.iOverflow]) ){
            put4byte(&pCell[info.iOverflow], iTo);
            break;
          }
        }
      }else{
        if( get4byte(pCell)==iFrom ){
          put4byte(pCell, iTo);
          break;
        }
      }
    }
  
    if( i==nCell ){
      if( eType!=PTRMAP_BTREE || 
          get4byte(&pPage->aData[pPage->hdrOffset+8])!=iFrom ){
        return SQLITE_CORRUPT_BKPT;
      }
      put4byte(&pPage->aData[pPage->hdrOffset+8], iTo);
    }

    pPage->isInit = isInitOrig;
  }
  return SQLITE_OK;
}


/*
** Move the open database page pDbPage to location iFreePage in the 
** database. The pDbPage reference remains valid.
*/
static int relocatePage(
  BtShared *pBt,           /* Btree */
  MemPage *pDbPage,        /* Open page to move */
  u8 eType,                /* Pointer map 'type' entry for pDbPage */
  Pgno iPtrPage,           /* Pointer map 'page-no' entry for pDbPage */
  Pgno iFreePage           /* The location to move pDbPage to */
){
  MemPage *pPtrPage;   /* The page that contains a pointer to pDbPage */
  Pgno iDbPage = pDbPage->pgno;
  Pager *pPager = pBt->pPager;
  int rc;

  assert( eType==PTRMAP_OVERFLOW2 || eType==PTRMAP_OVERFLOW1 || 
      eType==PTRMAP_BTREE || eType==PTRMAP_ROOTPAGE );

  /* Move page iDbPage from it's current location to page number iFreePage */
  TRACE(("AUTOVACUUM: Moving %d to free page %d (ptr page %d type %d)\n", 
      iDbPage, iFreePage, iPtrPage, eType));
  rc = sqlite3PagerMovepage(pPager, pDbPage->pDbPage, iFreePage);
  if( rc!=SQLITE_OK ){
    return rc;
  }
  pDbPage->pgno = iFreePage;

  /* If pDbPage was a btree-page, then it may have child pages and/or cells
  ** that point to overflow pages. The pointer map entries for all these
  ** pages need to be changed.
  **
  ** If pDbPage is an overflow page, then the first 4 bytes may store a
  ** pointer to a subsequent overflow page. If this is the case, then
  ** the pointer map needs to be updated for the subsequent overflow page.
  */
  if( eType==PTRMAP_BTREE || eType==PTRMAP_ROOTPAGE ){
    rc = setChildPtrmaps(pDbPage);
    if( rc!=SQLITE_OK ){
      return rc;
    }
  }else{
    Pgno nextOvfl = get4byte(pDbPage->aData);
    if( nextOvfl!=0 ){
      rc = ptrmapPut(pBt, nextOvfl, PTRMAP_OVERFLOW2, iFreePage);
      if( rc!=SQLITE_OK ){
        return rc;
      }
    }
  }

  /* Fix the database pointer on page iPtrPage that pointed at iDbPage so
  ** that it points at iFreePage. Also fix the pointer map entry for
  ** iPtrPage.
  */
  if( eType!=PTRMAP_ROOTPAGE ){
    rc = sqlite3BtreeGetPage(pBt, iPtrPage, &pPtrPage, 0);
    if( rc!=SQLITE_OK ){
      return rc;
    }
    rc = sqlite3PagerWrite(pPtrPage->pDbPage);
    if( rc!=SQLITE_OK ){
      releasePage(pPtrPage);
      return rc;
    }
    rc = modifyPagePointer(pPtrPage, iDbPage, iFreePage, eType);
    releasePage(pPtrPage);
    if( rc==SQLITE_OK ){
      rc = ptrmapPut(pBt, iFreePage, eType, iPtrPage);
    }
  }
  return rc;
}

/* Forward declaration required by incrVacuumStep(). */
static int allocateBtreePage(BtShared *, MemPage **, Pgno *, Pgno, u8);

/*
** Perform a single step of an incremental-vacuum. If successful,
** return SQLITE_OK. If there is no work to do (and therefore no
** point in calling this function again), return SQLITE_DONE.
**
** More specificly, this function attempts to re-organize the 
** database so that the last page of the file currently in use
** is no longer in use.
**
** If the nFin parameter is non-zero, the implementation assumes
** that the caller will keep calling incrVacuumStep() until
** it returns SQLITE_DONE or an error, and that nFin is the
** number of pages the database file will contain after this 
** process is complete.
*/
static int incrVacuumStep(BtShared *pBt, Pgno nFin){
  Pgno iLastPg;             /* Last page in the database */
  Pgno nFreeList;           /* Number of pages still on the free-list */

  iLastPg = pBt->nTrunc;
  if( iLastPg==0 ){
    iLastPg = sqlite3PagerPagecount(pBt->pPager);
  }

  if( !PTRMAP_ISPAGE(pBt, iLastPg) && iLastPg!=PENDING_BYTE_PAGE(pBt) ){
    int rc;
    u8 eType;
    Pgno iPtrPage;

    nFreeList = get4byte(&pBt->pPage1->aData[36]);
    if( nFreeList==0 || nFin==iLastPg ){
      return SQLITE_DONE;
    }

    rc = ptrmapGet(pBt, iLastPg, &eType, &iPtrPage);
    if( rc!=SQLITE_OK ){
      return rc;
    }
    if( eType==PTRMAP_ROOTPAGE ){
      return SQLITE_CORRUPT_BKPT;
    }

    if( eType==PTRMAP_FREEPAGE ){
      if( nFin==0 ){
        /* Remove the page from the files free-list. This is not required
        ** if nFin is non-zero. In that case, the free-list will be
        ** truncated to zero after this function returns, so it doesn't 
        ** matter if it still contains some garbage entries.
        */
        Pgno iFreePg;
        MemPage *pFreePg;
        rc = allocateBtreePage(pBt, &pFreePg, &iFreePg, iLastPg, 1);
        if( rc!=SQLITE_OK ){
          return rc;
        }
        assert( iFreePg==iLastPg );
        releasePage(pFreePg);
      }
    } else {
      Pgno iFreePg;             /* Index of free page to move pLastPg to */
      MemPage *pLastPg;

      rc = sqlite3BtreeGetPage(pBt, iLastPg, &pLastPg, 0);
      if( rc!=SQLITE_OK ){
        return rc;
      }

      /* If nFin is zero, this loop runs exactly once and page pLastPg
      ** is swapped with the first free page pulled off the free list.
      **
      ** On the other hand, if nFin is greater than zero, then keep
      ** looping until a free-page located within the first nFin pages
      ** of the file is found.
      */
      do {
        MemPage *pFreePg;
        rc = allocateBtreePage(pBt, &pFreePg, &iFreePg, 0, 0);
        if( rc!=SQLITE_OK ){
          releasePage(pLastPg);
          return rc;
        }
        releasePage(pFreePg);
      }while( nFin!=0 && iFreePg>nFin );
      assert( iFreePg<iLastPg );
      
      rc = sqlite3PagerWrite(pLastPg->pDbPage);
      if( rc!=SQLITE_OK ){
        return rc;
      } 
      rc = relocatePage(pBt, pLastPg, eType, iPtrPage, iFreePg);
      releasePage(pLastPg);
      if( rc!=SQLITE_OK ){
        return rc;
      } 
    }
  }

  pBt->nTrunc = iLastPg - 1;
  while( pBt->nTrunc==PENDING_BYTE_PAGE(pBt)||PTRMAP_ISPAGE(pBt, pBt->nTrunc) ){
    pBt->nTrunc--;
  }
  return SQLITE_OK;
}

/*
** A write-transaction must be opened before calling this function.
** It performs a single unit of work towards an incremental vacuum.
**
** If the incremental vacuum is finished after this function has run,
** SQLITE_DONE is returned. If it is not finished, but no error occured,
** SQLITE_OK is returned. Otherwise an SQLite error code. 
*/
SQLITE_PRIVATE int sqlite3BtreeIncrVacuum(Btree *p){
  BtShared *pBt = p->pBt;
  assert( pBt->inTransaction==TRANS_WRITE && p->inTrans==TRANS_WRITE );
  if( !pBt->autoVacuum ){
    return SQLITE_DONE;
  }
  invalidateAllOverflowCache(pBt);
  return incrVacuumStep(pBt, 0);
}

/*
** This routine is called prior to sqlite3PagerCommit when a transaction
** is commited for an auto-vacuum database.
**
** If SQLITE_OK is returned, then *pnTrunc is set to the number of pages
** the database file should be truncated to during the commit process. 
** i.e. the database has been reorganized so that only the first *pnTrunc
** pages are in use.
*/
static int autoVacuumCommit(BtShared *pBt, Pgno *pnTrunc){
  int rc = SQLITE_OK;
  Pager *pPager = pBt->pPager;
#ifndef NDEBUG
  int nRef = sqlite3PagerRefcount(pPager);
#endif

  invalidateAllOverflowCache(pBt);
  assert(pBt->autoVacuum);
  if( !pBt->incrVacuum ){
    Pgno nFin = 0;

    if( pBt->nTrunc==0 ){
      Pgno nFree;
      Pgno nPtrmap;
      const int pgsz = pBt->pageSize;
      Pgno nOrig = sqlite3PagerPagecount(pBt->pPager);

      if( PTRMAP_ISPAGE(pBt, nOrig) ){
        return SQLITE_CORRUPT_BKPT;
      }
      if( nOrig==PENDING_BYTE_PAGE(pBt) ){
        nOrig--;
      }
      nFree = get4byte(&pBt->pPage1->aData[36]);
      nPtrmap = (nFree-nOrig+PTRMAP_PAGENO(pBt, nOrig)+pgsz/5)/(pgsz/5);
      nFin = nOrig - nFree - nPtrmap;
      if( nOrig>PENDING_BYTE_PAGE(pBt) && nFin<=PENDING_BYTE_PAGE(pBt) ){
        nFin--;
      }
      while( PTRMAP_ISPAGE(pBt, nFin) || nFin==PENDING_BYTE_PAGE(pBt) ){
        nFin--;
      }
    }

    while( rc==SQLITE_OK ){
      rc = incrVacuumStep(pBt, nFin);
    }
    if( rc==SQLITE_DONE ){
      assert(nFin==0 || pBt->nTrunc==0 || nFin<=pBt->nTrunc);
      rc = SQLITE_OK;
      if( pBt->nTrunc ){
        sqlite3PagerWrite(pBt->pPage1->pDbPage);
        put4byte(&pBt->pPage1->aData[32], 0);
        put4byte(&pBt->pPage1->aData[36], 0);
        pBt->nTrunc = nFin;
      }
    }
    if( rc!=SQLITE_OK ){
      sqlite3PagerRollback(pPager);
    }
  }

  if( rc==SQLITE_OK ){
    *pnTrunc = pBt->nTrunc;
    pBt->nTrunc = 0;
  }
  assert( nRef==sqlite3PagerRefcount(pPager) );
  return rc;
}

#endif

/*
** This routine does the first phase of a two-phase commit.  This routine
** causes a rollback journal to be created (if it does not already exist)
** and populated with enough information so that if a power loss occurs
** the database can be restored to its original state by playing back
** the journal.  Then the contents of the journal are flushed out to
** the disk.  After the journal is safely on oxide, the changes to the
** database are written into the database file and flushed to oxide.
** At the end of this call, the rollback journal still exists on the
** disk and we are still holding all locks, so the transaction has not
** committed.  See sqlite3BtreeCommit() for the second phase of the
** commit process.
**
** This call is a no-op if no write-transaction is currently active on pBt.
**
** Otherwise, sync the database file for the btree pBt. zMaster points to
** the name of a master journal file that should be written into the
** individual journal file, or is NULL, indicating no master journal file 
** (single database transaction).
**
** When this is called, the master journal should already have been
** created, populated with this journal pointer and synced to disk.
**
** Once this is routine has returned, the only thing required to commit
** the write-transaction for this database file is to delete the journal.
*/
SQLITE_PRIVATE int sqlite3BtreeCommitPhaseOne(Btree *p, const char *zMaster){
  int rc = SQLITE_OK;
  if( p->inTrans==TRANS_WRITE ){
    BtShared *pBt = p->pBt;
    Pgno nTrunc = 0;
#ifndef SQLITE_OMIT_AUTOVACUUM
    if( pBt->autoVacuum ){
      rc = autoVacuumCommit(pBt, &nTrunc); 
      if( rc!=SQLITE_OK ){
        return rc;
      }
    }
#endif
    rc = sqlite3PagerCommitPhaseOne(pBt->pPager, zMaster, nTrunc);
  }
  return rc;
}

/*
** Commit the transaction currently in progress.
**
** This routine implements the second phase of a 2-phase commit.  The
** sqlite3BtreeSync() routine does the first phase and should be invoked
** prior to calling this routine.  The sqlite3BtreeSync() routine did
** all the work of writing information out to disk and flushing the
** contents so that they are written onto the disk platter.  All this
** routine has to do is delete or truncate the rollback journal
** (which causes the transaction to commit) and drop locks.
**
** This will release the write lock on the database file.  If there
** are no active cursors, it also releases the read lock.
*/
SQLITE_PRIVATE int sqlite3BtreeCommitPhaseTwo(Btree *p){
  BtShared *pBt = p->pBt;

  btreeIntegrity(p);

  /* If the handle has a write-transaction open, commit the shared-btrees 
  ** transaction and set the shared state to TRANS_READ.
  */
  if( p->inTrans==TRANS_WRITE ){
    int rc;
    assert( pBt->inTransaction==TRANS_WRITE );
    assert( pBt->nTransaction>0 );
    rc = sqlite3PagerCommitPhaseTwo(pBt->pPager);
    if( rc!=SQLITE_OK ){
      return rc;
    }
    pBt->inTransaction = TRANS_READ;
    pBt->inStmt = 0;
  }
  unlockAllTables(p);

  /* If the handle has any kind of transaction open, decrement the transaction
  ** count of the shared btree. If the transaction count reaches 0, set
  ** the shared state to TRANS_NONE. The unlockBtreeIfUnused() call below
  ** will unlock the pager.
  */
  if( p->inTrans!=TRANS_NONE ){
    pBt->nTransaction--;
    if( 0==pBt->nTransaction ){
      pBt->inTransaction = TRANS_NONE;
    }
  }

  /* Set the handles current transaction state to TRANS_NONE and unlock
  ** the pager if this call closed the only read or write transaction.
  */
  p->inTrans = TRANS_NONE;
  unlockBtreeIfUnused(pBt);

  btreeIntegrity(p);
  return SQLITE_OK;
}

/*
** Do both phases of a commit.
*/
SQLITE_PRIVATE int sqlite3BtreeCommit(Btree *p){
  int rc;
  rc = sqlite3BtreeCommitPhaseOne(p, 0);
  if( rc==SQLITE_OK ){
    rc = sqlite3BtreeCommitPhaseTwo(p);
  }
  return rc;
}

#ifndef NDEBUG
/*
** Return the number of write-cursors open on this handle. This is for use
** in assert() expressions, so it is only compiled if NDEBUG is not
** defined.
*/
static int countWriteCursors(BtShared *pBt){
  BtCursor *pCur;
  int r = 0;
  for(pCur=pBt->pCursor; pCur; pCur=pCur->pNext){
    if( pCur->wrFlag ) r++; 
  }
  return r;
}
#endif

/*
** Rollback the transaction in progress.  All cursors will be
** invalided by this operation.  Any attempt to use a cursor
** that was open at the beginning of this operation will result
** in an error.
**
** This will release the write lock on the database file.  If there
** are no active cursors, it also releases the read lock.
*/
SQLITE_PRIVATE int sqlite3BtreeRollback(Btree *p){
  int rc;
  BtShared *pBt = p->pBt;
  MemPage *pPage1;

  rc = saveAllCursors(pBt, 0, 0);
#ifndef SQLITE_OMIT_SHARED_CACHE
  if( rc!=SQLITE_OK ){
    /* This is a horrible situation. An IO or malloc() error occured whilst
    ** trying to save cursor positions. If this is an automatic rollback (as
    ** the result of a constraint, malloc() failure or IO error) then 
    ** the cache may be internally inconsistent (not contain valid trees) so
    ** we cannot simply return the error to the caller. Instead, abort 
    ** all queries that may be using any of the cursors that failed to save.
    */
    while( pBt->pCursor ){
      sqlite3 *db = pBt->pCursor->pBtree->pSqlite;
      if( db ){
        sqlite3AbortOtherActiveVdbes(db, 0);
      }
    }
  }
#endif
  btreeIntegrity(p);
  unlockAllTables(p);

  if( p->inTrans==TRANS_WRITE ){
    int rc2;

#ifndef SQLITE_OMIT_AUTOVACUUM
    pBt->nTrunc = 0;
#endif

    assert( TRANS_WRITE==pBt->inTransaction );
    rc2 = sqlite3PagerRollback(pBt->pPager);
    if( rc2!=SQLITE_OK ){
      rc = rc2;
    }

    /* The rollback may have destroyed the pPage1->aData value.  So
    ** call sqlite3BtreeGetPage() on page 1 again to make
    ** sure pPage1->aData is set correctly. */
    if( sqlite3BtreeGetPage(pBt, 1, &pPage1, 0)==SQLITE_OK ){
      releasePage(pPage1);
    }
    assert( countWriteCursors(pBt)==0 );
    pBt->inTransaction = TRANS_READ;
  }

  if( p->inTrans!=TRANS_NONE ){
    assert( pBt->nTransaction>0 );
    pBt->nTransaction--;
    if( 0==pBt->nTransaction ){
      pBt->inTransaction = TRANS_NONE;
    }
  }

  p->inTrans = TRANS_NONE;
  pBt->inStmt = 0;
  unlockBtreeIfUnused(pBt);

  btreeIntegrity(p);
  return rc;
}

/*
** Start a statement subtransaction.  The subtransaction can
** can be rolled back independently of the main transaction.
** You must start a transaction before starting a subtransaction.
** The subtransaction is ended automatically if the main transaction
** commits or rolls back.
**
** Only one subtransaction may be active at a time.  It is an error to try
** to start a new subtransaction if another subtransaction is already active.
**
** Statement subtransactions are used around individual SQL statements
** that are contained within a BEGIN...COMMIT block.  If a constraint
** error occurs within the statement, the effect of that one statement
** can be rolled back without having to rollback the entire transaction.
*/
SQLITE_PRIVATE int sqlite3BtreeBeginStmt(Btree *p){
  int rc;
  BtShared *pBt = p->pBt;
  if( (p->inTrans!=TRANS_WRITE) || pBt->inStmt ){
    return pBt->readOnly ? SQLITE_READONLY : SQLITE_ERROR;
  }
  assert( pBt->inTransaction==TRANS_WRITE );
  rc = pBt->readOnly ? SQLITE_OK : sqlite3PagerStmtBegin(pBt->pPager);
  pBt->inStmt = 1;
  return rc;
}


/*
** Commit the statment subtransaction currently in progress.  If no
** subtransaction is active, this is a no-op.
*/
SQLITE_PRIVATE int sqlite3BtreeCommitStmt(Btree *p){
  int rc;
  BtShared *pBt = p->pBt;
  if( pBt->inStmt && !pBt->readOnly ){
    rc = sqlite3PagerStmtCommit(pBt->pPager);
  }else{
    rc = SQLITE_OK;
  }
  pBt->inStmt = 0;
  return rc;
}

/*
** Rollback the active statement subtransaction.  If no subtransaction
** is active this routine is a no-op.
**
** All cursors will be invalidated by this operation.  Any attempt
** to use a cursor that was open at the beginning of this operation
** will result in an error.
*/
SQLITE_PRIVATE int sqlite3BtreeRollbackStmt(Btree *p){
  int rc = SQLITE_OK;
  BtShared *pBt = p->pBt;
  sqlite3MallocDisallow();
  if( pBt->inStmt && !pBt->readOnly ){
    rc = sqlite3PagerStmtRollback(pBt->pPager);
    assert( countWriteCursors(pBt)==0 );
    pBt->inStmt = 0;
  }
  sqlite3MallocAllow();
  return rc;
}

/*
** Default key comparison function to be used if no comparison function
** is specified on the sqlite3BtreeCursor() call.
*/
static int dfltCompare(
  void *NotUsed,             /* User data is not used */
  int n1, const void *p1,    /* First key to compare */
  int n2, const void *p2     /* Second key to compare */
){
  int c;
  c = memcmp(p1, p2, n1<n2 ? n1 : n2);
  if( c==0 ){
    c = n1 - n2;
  }
  return c;
}

/*
** Create a new cursor for the BTree whose root is on the page
** iTable.  The act of acquiring a cursor gets a read lock on 
** the database file.
**
** If wrFlag==0, then the cursor can only be used for reading.
** If wrFlag==1, then the cursor can be used for reading or for
** writing if other conditions for writing are also met.  These
** are the conditions that must be met in order for writing to
** be allowed:
**
** 1:  The cursor must have been opened with wrFlag==1
**
** 2:  Other database connections that share the same pager cache
**     but which are not in the READ_UNCOMMITTED state may not have
**     cursors open with wrFlag==0 on the same table.  Otherwise
**     the changes made by this write cursor would be visible to
**     the read cursors in the other database connection.
**
** 3:  The database must be writable (not on read-only media)
**
** 4:  There must be an active transaction.
**
** No checking is done to make sure that page iTable really is the
** root page of a b-tree.  If it is not, then the cursor acquired
** will not work correctly.
**
** The comparison function must be logically the same for every cursor
** on a particular table.  Changing the comparison function will result
** in incorrect operations.  If the comparison function is NULL, a
** default comparison function is used.  The comparison function is
** always ignored for INTKEY tables.
*/
SQLITE_PRIVATE int sqlite3BtreeCursor(
  Btree *p,                                   /* The btree */
  int iTable,                                 /* Root page of table to open */
  int wrFlag,                                 /* 1 to write. 0 read-only */
  int (*xCmp)(void*,int,const void*,int,const void*), /* Key Comparison func */
  void *pArg,                                 /* First arg to xCompare() */
  BtCursor **ppCur                            /* Write new cursor here */
){
  int rc;
  BtCursor *pCur;
  BtShared *pBt = p->pBt;

  *ppCur = 0;
  if( wrFlag ){
    if( pBt->readOnly ){
      return SQLITE_READONLY;
    }
    if( checkReadLocks(p, iTable, 0) ){
      return SQLITE_LOCKED;
    }
  }

  if( pBt->pPage1==0 ){
    rc = lockBtreeWithRetry(p);
    if( rc!=SQLITE_OK ){
      return rc;
    }
    if( pBt->readOnly && wrFlag ){
      return SQLITE_READONLY;
    }
  }
  pCur = sqliteMalloc( sizeof(*pCur) );
  if( pCur==0 ){
    rc = SQLITE_NOMEM;
    goto create_cursor_exception;
  }
  pCur->pgnoRoot = (Pgno)iTable;
  if( iTable==1 && sqlite3PagerPagecount(pBt->pPager)==0 ){
    rc = SQLITE_EMPTY;
    goto create_cursor_exception;
  }
  rc = getAndInitPage(pBt, pCur->pgnoRoot, &pCur->pPage, 0);
  if( rc!=SQLITE_OK ){
    goto create_cursor_exception;
  }

  /* Now that no other errors can occur, finish filling in the BtCursor
  ** variables, link the cursor into the BtShared list and set *ppCur (the
  ** output argument to this function).
  */
  pCur->xCompare = xCmp ? xCmp : dfltCompare;
  pCur->pArg = pArg;
  pCur->pBtree = p;
  pCur->wrFlag = wrFlag;
  pCur->pNext = pBt->pCursor;
  if( pCur->pNext ){
    pCur->pNext->pPrev = pCur;
  }
  pBt->pCursor = pCur;
  pCur->eState = CURSOR_INVALID;
  *ppCur = pCur;

  return SQLITE_OK;
create_cursor_exception:
  if( pCur ){
    releasePage(pCur->pPage);
    sqliteFree(pCur);
  }
  unlockBtreeIfUnused(pBt);
  return rc;
}

/*
** Close a cursor.  The read lock on the database file is released
** when the last cursor is closed.
*/
SQLITE_PRIVATE int sqlite3BtreeCloseCursor(BtCursor *pCur){
  BtShared *pBt = pCur->pBtree->pBt;
  clearCursorPosition(pCur);
  if( pCur->pPrev ){
    pCur->pPrev->pNext = pCur->pNext;
  }else{
    pBt->pCursor = pCur->pNext;
  }
  if( pCur->pNext ){
    pCur->pNext->pPrev = pCur->pPrev;
  }
  releasePage(pCur->pPage);
  unlockBtreeIfUnused(pBt);
  invalidateOverflowCache(pCur);
  sqliteFree(pCur);
  return SQLITE_OK;
}

/*
** Make a temporary cursor by filling in the fields of pTempCur.
** The temporary cursor is not on the cursor list for the Btree.
*/
SQLITE_PRIVATE void sqlite3BtreeGetTempCursor(BtCursor *pCur, BtCursor *pTempCur){
  memcpy(pTempCur, pCur, sizeof(*pCur));
  pTempCur->pNext = 0;
  pTempCur->pPrev = 0;
  if( pTempCur->pPage ){
    sqlite3PagerRef(pTempCur->pPage->pDbPage);
  }
}

/*
** Delete a temporary cursor such as was made by the CreateTemporaryCursor()
** function above.
*/
SQLITE_PRIVATE void sqlite3BtreeReleaseTempCursor(BtCursor *pCur){
  if( pCur->pPage ){
    sqlite3PagerUnref(pCur->pPage->pDbPage);
  }
}

/*
** Make sure the BtCursor* given in the argument has a valid
** BtCursor.info structure.  If it is not already valid, call
** sqlite3BtreeParseCell() to fill it in.
**
** BtCursor.info is a cache of the information in the current cell.
** Using this cache reduces the number of calls to sqlite3BtreeParseCell().
**
** 2007-06-25:  There is a bug in some versions of MSVC that cause the
** compiler to crash when getCellInfo() is implemented as a macro.
** But there is a measureable speed advantage to using the macro on gcc
** (when less compiler optimizations like -Os or -O0 are used and the
** compiler is not doing agressive inlining.)  So we use a real function
** for MSVC and a macro for everything else.  Ticket #2457.
*/
#ifndef NDEBUG
  static void assertCellInfo(BtCursor *pCur){
    CellInfo info;
    memset(&info, 0, sizeof(info));
    sqlite3BtreeParseCell(pCur->pPage, pCur->idx, &info);
    assert( memcmp(&info, &pCur->info, sizeof(info))==0 );
  }
#else
  #define assertCellInfo(x)
#endif
#ifdef _MSC_VER
  /* Use a real function in MSVC to work around bugs in that compiler. */
  static void getCellInfo(BtCursor *pCur){
    if( pCur->info.nSize==0 ){
      sqlite3BtreeParseCell(pCur->pPage, pCur->idx, &pCur->info);
    }else{
      assertCellInfo(pCur);
    }
  }
#else /* if not _MSC_VER */
  /* Use a macro in all other compilers so that the function is inlined */
#define getCellInfo(pCur)                                               \
  if( pCur->info.nSize==0 ){                                            \
    sqlite3BtreeParseCell(pCur->pPage, pCur->idx, &pCur->info);         \
  }else{                                                                \
    assertCellInfo(pCur);                                               \
  }
#endif /* _MSC_VER */

/*
** Set *pSize to the size of the buffer needed to hold the value of
** the key for the current entry.  If the cursor is not pointing
** to a valid entry, *pSize is set to 0. 
**
** For a table with the INTKEY flag set, this routine returns the key
** itself, not the number of bytes in the key.
*/
SQLITE_PRIVATE int sqlite3BtreeKeySize(BtCursor *pCur, i64 *pSize){
  int rc = restoreOrClearCursorPosition(pCur);
  if( rc==SQLITE_OK ){
    assert( pCur->eState==CURSOR_INVALID || pCur->eState==CURSOR_VALID );
    if( pCur->eState==CURSOR_INVALID ){
      *pSize = 0;
    }else{
      getCellInfo(pCur);
      *pSize = pCur->info.nKey;
    }
  }
  return rc;
}

/*
** Set *pSize to the number of bytes of data in the entry the
** cursor currently points to.  Always return SQLITE_OK.
** Failure is not possible.  If the cursor is not currently
** pointing to an entry (which can happen, for example, if
** the database is empty) then *pSize is set to 0.
*/
SQLITE_PRIVATE int sqlite3BtreeDataSize(BtCursor *pCur, u32 *pSize){
  int rc = restoreOrClearCursorPosition(pCur);
  if( rc==SQLITE_OK ){
    assert( pCur->eState==CURSOR_INVALID || pCur->eState==CURSOR_VALID );
    if( pCur->eState==CURSOR_INVALID ){
      /* Not pointing at a valid entry - set *pSize to 0. */
      *pSize = 0;
    }else{
      getCellInfo(pCur);
      *pSize = pCur->info.nData;
    }
  }
  return rc;
}

/*
** Given the page number of an overflow page in the database (parameter
** ovfl), this function finds the page number of the next page in the 
** linked list of overflow pages. If possible, it uses the auto-vacuum
** pointer-map data instead of reading the content of page ovfl to do so. 
**
** If an error occurs an SQLite error code is returned. Otherwise:
**
** Unless pPgnoNext is NULL, the page number of the next overflow 
** page in the linked list is written to *pPgnoNext. If page ovfl
** is the last page in it's linked list, *pPgnoNext is set to zero. 
**
** If ppPage is not NULL, *ppPage is set to the MemPage* handle
** for page ovfl. The underlying pager page may have been requested
** with the noContent flag set, so the page data accessable via
** this handle may not be trusted.
*/
static int getOverflowPage(
  BtShared *pBt, 
  Pgno ovfl,                   /* Overflow page */
  MemPage **ppPage,            /* OUT: MemPage handle */
  Pgno *pPgnoNext              /* OUT: Next overflow page number */
){
  Pgno next = 0;
  int rc;

  /* One of these must not be NULL. Otherwise, why call this function? */
  assert(ppPage || pPgnoNext);

  /* If pPgnoNext is NULL, then this function is being called to obtain
  ** a MemPage* reference only. No page-data is required in this case.
  */
  if( !pPgnoNext ){
    return sqlite3BtreeGetPage(pBt, ovfl, ppPage, 1);
  }

#ifndef SQLITE_OMIT_AUTOVACUUM
  /* Try to find the next page in the overflow list using the
  ** autovacuum pointer-map pages. Guess that the next page in 
  ** the overflow list is page number (ovfl+1). If that guess turns 
  ** out to be wrong, fall back to loading the data of page 
  ** number ovfl to determine the next page number.
  */
  if( pBt->autoVacuum ){
    Pgno pgno;
    Pgno iGuess = ovfl+1;
    u8 eType;

    while( PTRMAP_ISPAGE(pBt, iGuess) || iGuess==PENDING_BYTE_PAGE(pBt) ){
      iGuess++;
    }

    if( iGuess<=sqlite3PagerPagecount(pBt->pPager) ){
      rc = ptrmapGet(pBt, iGuess, &eType, &pgno);
      if( rc!=SQLITE_OK ){
        return rc;
      }
      if( eType==PTRMAP_OVERFLOW2 && pgno==ovfl ){
        next = iGuess;
      }
    }
  }
#endif

  if( next==0 || ppPage ){
    MemPage *pPage = 0;

    rc = sqlite3BtreeGetPage(pBt, ovfl, &pPage, next!=0);
    assert(rc==SQLITE_OK || pPage==0);
    if( next==0 && rc==SQLITE_OK ){
      next = get4byte(pPage->aData);
    }

    if( ppPage ){
      *ppPage = pPage;
    }else{
      releasePage(pPage);
    }
  }
  *pPgnoNext = next;

  return rc;
}

/*
** Copy data from a buffer to a page, or from a page to a buffer.
**
** pPayload is a pointer to data stored on database page pDbPage.
** If argument eOp is false, then nByte bytes of data are copied
** from pPayload to the buffer pointed at by pBuf. If eOp is true,
** then sqlite3PagerWrite() is called on pDbPage and nByte bytes
** of data are copied from the buffer pBuf to pPayload.
**
** SQLITE_OK is returned on success, otherwise an error code.
*/
static int copyPayload(
  void *pPayload,           /* Pointer to page data */
  void *pBuf,               /* Pointer to buffer */
  int nByte,                /* Number of bytes to copy */
  int eOp,                  /* 0 -> copy from page, 1 -> copy to page */
  DbPage *pDbPage           /* Page containing pPayload */
){
  if( eOp ){
    /* Copy data from buffer to page (a write operation) */
    int rc = sqlite3PagerWrite(pDbPage);
    if( rc!=SQLITE_OK ){
      return rc;
    }
    memcpy(pPayload, pBuf, nByte);
  }else{
    /* Copy data from page to buffer (a read operation) */
    memcpy(pBuf, pPayload, nByte);
  }
  return SQLITE_OK;
}

/*
** This function is used to read or overwrite payload information
** for the entry that the pCur cursor is pointing to. If the eOp
** parameter is 0, this is a read operation (data copied into
** buffer pBuf). If it is non-zero, a write (data copied from
** buffer pBuf).
**
** A total of "amt" bytes are read or written beginning at "offset".
** Data is read to or from the buffer pBuf.
**
** This routine does not make a distinction between key and data.
** It just reads or writes bytes from the payload area.  Data might 
** appear on the main page or be scattered out on multiple overflow 
** pages.
**
** If the BtCursor.isIncrblobHandle flag is set, and the current
** cursor entry uses one or more overflow pages, this function
** allocates space for and lazily popluates the overflow page-list 
** cache array (BtCursor.aOverflow). Subsequent calls use this
** cache to make seeking to the supplied offset more efficient.
**
** Once an overflow page-list cache has been allocated, it may be
** invalidated if some other cursor writes to the same table, or if
** the cursor is moved to a different row. Additionally, in auto-vacuum
** mode, the following events may invalidate an overflow page-list cache.
**
**   * An incremental vacuum,
**   * A commit in auto_vacuum="full" mode,
**   * Creating a table (may require moving an overflow page).
*/
static int accessPayload(
  BtCursor *pCur,      /* Cursor pointing to entry to read from */
  int offset,          /* Begin reading this far into payload */
  int amt,             /* Read this many bytes */
  unsigned char *pBuf, /* Write the bytes into this buffer */ 
  int skipKey,         /* offset begins at data if this is true */
  int eOp              /* zero to read. non-zero to write. */
){
  unsigned char *aPayload;
  int rc = SQLITE_OK;
  u32 nKey;
  int iIdx = 0;
  MemPage *pPage = pCur->pPage;        /* Btree page of current cursor entry */
  BtShared *pBt = pCur->pBtree->pBt;   /* Btree this cursor belongs to */

  assert( pPage );
  assert( pCur->eState==CURSOR_VALID );
  assert( pCur->idx>=0 && pCur->idx<pPage->nCell );
  assert( offset>=0 );

  getCellInfo(pCur);
  aPayload = pCur->info.pCell + pCur->info.nHeader;
  nKey = (pPage->intKey ? 0 : pCur->info.nKey);

  if( skipKey ){
    offset += nKey;
  }
  if( offset+amt > nKey+pCur->info.nData ){
    /* Trying to read or write past the end of the data is an error */
    return SQLITE_ERROR;
  }

  /* Check if data must be read/written to/from the btree page itself. */
  if( offset<pCur->info.nLocal ){
    int a = amt;
    if( a+offset>pCur->info.nLocal ){
      a = pCur->info.nLocal - offset;
    }
    rc = copyPayload(&aPayload[offset], pBuf, a, eOp, pPage->pDbPage);
    offset = 0;
    pBuf += a;
    amt -= a;
  }else{
    offset -= pCur->info.nLocal;
  }

  if( rc==SQLITE_OK && amt>0 ){
    const int ovflSize = pBt->usableSize - 4;  /* Bytes content per ovfl page */
    Pgno nextPage;

    nextPage = get4byte(&aPayload[pCur->info.nLocal]);

#ifndef SQLITE_OMIT_INCRBLOB
    /* If the isIncrblobHandle flag is set and the BtCursor.aOverflow[]
    ** has not been allocated, allocate it now. The array is sized at
    ** one entry for each overflow page in the overflow chain. The
    ** page number of the first overflow page is stored in aOverflow[0],
    ** etc. A value of 0 in the aOverflow[] array means "not yet known"
    ** (the cache is lazily populated).
    */
    if( pCur->isIncrblobHandle && !pCur->aOverflow ){
      int nOvfl = (pCur->info.nPayload-pCur->info.nLocal+ovflSize-1)/ovflSize;
      pCur->aOverflow = (Pgno *)sqliteMalloc(sizeof(Pgno)*nOvfl);
      if( nOvfl && !pCur->aOverflow ){
        rc = SQLITE_NOMEM;
      }
    }

    /* If the overflow page-list cache has been allocated and the
    ** entry for the first required overflow page is valid, skip
    ** directly to it.
    */
    if( pCur->aOverflow && pCur->aOverflow[offset/ovflSize] ){
      iIdx = (offset/ovflSize);
      nextPage = pCur->aOverflow[iIdx];
      offset = (offset%ovflSize);
    }
#endif

    for( ; rc==SQLITE_OK && amt>0 && nextPage; iIdx++){

#ifndef SQLITE_OMIT_INCRBLOB
      /* If required, populate the overflow page-list cache. */
      if( pCur->aOverflow ){
        assert(!pCur->aOverflow[iIdx] || pCur->aOverflow[iIdx]==nextPage);
        pCur->aOverflow[iIdx] = nextPage;
      }
#endif

      if( offset>=ovflSize ){
        /* The only reason to read this page is to obtain the page
        ** number for the next page in the overflow chain. The page
	** data is not required. So first try to lookup the overflow
	** page-list cache, if any, then fall back to the getOverflowPage()
        ** function.
        */
#ifndef SQLITE_OMIT_INCRBLOB
        if( pCur->aOverflow && pCur->aOverflow[iIdx+1] ){
          nextPage = pCur->aOverflow[iIdx+1];
        } else 
#endif
          rc = getOverflowPage(pBt, nextPage, 0, &nextPage);
        offset -= ovflSize;
      }else{
        /* Need to read this page properly. It contains some of the
        ** range of data that is being read (eOp==0) or written (eOp!=0).
        */
        DbPage *pDbPage;
        int a = amt;
        rc = sqlite3PagerGet(pBt->pPager, nextPage, &pDbPage);
        if( rc==SQLITE_OK ){
          aPayload = sqlite3PagerGetData(pDbPage);
          nextPage = get4byte(aPayload);
          if( a + offset > ovflSize ){
            a = ovflSize - offset;
          }
          rc = copyPayload(&aPayload[offset+4], pBuf, a, eOp, pDbPage);
          sqlite3PagerUnref(pDbPage);
          offset = 0;
          amt -= a;
          pBuf += a;
        }
      }
    }
  }

  if( rc==SQLITE_OK && amt>0 ){
    return SQLITE_CORRUPT_BKPT;
  }
  return rc;
}

/*
** Read part of the key associated with cursor pCur.  Exactly
** "amt" bytes will be transfered into pBuf[].  The transfer
** begins at "offset".
**
** Return SQLITE_OK on success or an error code if anything goes
** wrong.  An error is returned if "offset+amt" is larger than
** the available payload.
*/
SQLITE_PRIVATE int sqlite3BtreeKey(BtCursor *pCur, u32 offset, u32 amt, void *pBuf){
  int rc = restoreOrClearCursorPosition(pCur);
  if( rc==SQLITE_OK ){
    assert( pCur->eState==CURSOR_VALID );
    assert( pCur->pPage!=0 );
    if( pCur->pPage->intKey ){
      return SQLITE_CORRUPT_BKPT;
    }
    assert( pCur->pPage->intKey==0 );
    assert( pCur->idx>=0 && pCur->idx<pCur->pPage->nCell );
    rc = accessPayload(pCur, offset, amt, (unsigned char*)pBuf, 0, 0);
  }
  return rc;
}

/*
** Read part of the data associated with cursor pCur.  Exactly
** "amt" bytes will be transfered into pBuf[].  The transfer
** begins at "offset".
**
** Return SQLITE_OK on success or an error code if anything goes
** wrong.  An error is returned if "offset+amt" is larger than
** the available payload.
*/
SQLITE_PRIVATE int sqlite3BtreeData(BtCursor *pCur, u32 offset, u32 amt, void *pBuf){
  int rc = restoreOrClearCursorPosition(pCur);
  if( rc==SQLITE_OK ){
    assert( pCur->eState==CURSOR_VALID );
    assert( pCur->pPage!=0 );
    assert( pCur->idx>=0 && pCur->idx<pCur->pPage->nCell );
    rc = accessPayload(pCur, offset, amt, pBuf, 1, 0);
  }
  return rc;
}

/*
** Return a pointer to payload information from the entry that the 
** pCur cursor is pointing to.  The pointer is to the beginning of
** the key if skipKey==0 and it points to the beginning of data if
** skipKey==1.  The number of bytes of available key/data is written
** into *pAmt.  If *pAmt==0, then the value returned will not be
** a valid pointer.
**
** This routine is an optimization.  It is common for the entire key
** and data to fit on the local page and for there to be no overflow
** pages.  When that is so, this routine can be used to access the
** key and data without making a copy.  If the key and/or data spills
** onto overflow pages, then accessPayload() must be used to reassembly
** the key/data and copy it into a preallocated buffer.
**
** The pointer returned by this routine looks directly into the cached
** page of the database.  The data might change or move the next time
** any btree routine is called.
*/
static const unsigned char *fetchPayload(
  BtCursor *pCur,      /* Cursor pointing to entry to read from */
  int *pAmt,           /* Write the number of available bytes here */
  int skipKey          /* read beginning at data if this is true */
){
  unsigned char *aPayload;
  MemPage *pPage;
  u32 nKey;
  int nLocal;

  assert( pCur!=0 && pCur->pPage!=0 );
  assert( pCur->eState==CURSOR_VALID );
  pPage = pCur->pPage;
  assert( pCur->idx>=0 && pCur->idx<pPage->nCell );
  getCellInfo(pCur);
  aPayload = pCur->info.pCell;
  aPayload += pCur->info.nHeader;
  if( pPage->intKey ){
    nKey = 0;
  }else{
    nKey = pCur->info.nKey;
  }
  if( skipKey ){
    aPayload += nKey;
    nLocal = pCur->info.nLocal - nKey;
  }else{
    nLocal = pCur->info.nLocal;
    if( nLocal>nKey ){
      nLocal = nKey;
    }
  }
  *pAmt = nLocal;
  return aPayload;
}


/*
** For the entry that cursor pCur is point to, return as
** many bytes of the key or data as are available on the local
** b-tree page.  Write the number of available bytes into *pAmt.
**
** The pointer returned is ephemeral.  The key/data may move
** or be destroyed on the next call to any Btree routine.
**
** These routines is used to get quick access to key and data
** in the common case where no overflow pages are used.
*/
SQLITE_PRIVATE const void *sqlite3BtreeKeyFetch(BtCursor *pCur, int *pAmt){
  if( pCur->eState==CURSOR_VALID ){
    return (const void*)fetchPayload(pCur, pAmt, 0);
  }
  return 0;
}
SQLITE_PRIVATE const void *sqlite3BtreeDataFetch(BtCursor *pCur, int *pAmt){
  if( pCur->eState==CURSOR_VALID ){
    return (const void*)fetchPayload(pCur, pAmt, 1);
  }
  return 0;
}


/*
** Move the cursor down to a new child page.  The newPgno argument is the
** page number of the child page to move to.
*/
static int moveToChild(BtCursor *pCur, u32 newPgno){
  int rc;
  MemPage *pNewPage;
  MemPage *pOldPage;
  BtShared *pBt = pCur->pBtree->pBt;

  assert( pCur->eState==CURSOR_VALID );
  rc = getAndInitPage(pBt, newPgno, &pNewPage, pCur->pPage);
  if( rc ) return rc;
  pNewPage->idxParent = pCur->idx;
  pOldPage = pCur->pPage;
  pOldPage->idxShift = 0;
  releasePage(pOldPage);
  pCur->pPage = pNewPage;
  pCur->idx = 0;
  pCur->info.nSize = 0;
  if( pNewPage->nCell<1 ){
    return SQLITE_CORRUPT_BKPT;
  }
  return SQLITE_OK;
}

/*
** Return true if the page is the virtual root of its table.
**
** The virtual root page is the root page for most tables.  But
** for the table rooted on page 1, sometime the real root page
** is empty except for the right-pointer.  In such cases the
** virtual root page is the page that the right-pointer of page
** 1 is pointing to.
*/
SQLITE_PRIVATE int sqlite3BtreeIsRootPage(MemPage *pPage){
  MemPage *pParent = pPage->pParent;
  if( pParent==0 ) return 1;
  if( pParent->pgno>1 ) return 0;
  if( get2byte(&pParent->aData[pParent->hdrOffset+3])==0 ) return 1;
  return 0;
}

/*
** Move the cursor up to the parent page.
**
** pCur->idx is set to the cell index that contains the pointer
** to the page we are coming from.  If we are coming from the
** right-most child page then pCur->idx is set to one more than
** the largest cell index.
*/
SQLITE_PRIVATE void sqlite3BtreeMoveToParent(BtCursor *pCur){
  MemPage *pParent;
  MemPage *pPage;
  int idxParent;

  assert( pCur->eState==CURSOR_VALID );
  pPage = pCur->pPage;
  assert( pPage!=0 );
  assert( !sqlite3BtreeIsRootPage(pPage) );
  pParent = pPage->pParent;
  assert( pParent!=0 );
  idxParent = pPage->idxParent;
  sqlite3PagerRef(pParent->pDbPage);
  releasePage(pPage);
  pCur->pPage = pParent;
  pCur->info.nSize = 0;
  assert( pParent->idxShift==0 );
  pCur->idx = idxParent;
}

/*
** Move the cursor to the root page
*/
static int moveToRoot(BtCursor *pCur){
  MemPage *pRoot;
  int rc = SQLITE_OK;
  BtShared *pBt = pCur->pBtree->pBt;

  if( pCur->eState==CURSOR_REQUIRESEEK ){
    clearCursorPosition(pCur);
  }
  pRoot = pCur->pPage;
  if( pRoot && pRoot->pgno==pCur->pgnoRoot ){
    assert( pRoot->isInit );
  }else{
    if( 
      SQLITE_OK!=(rc = getAndInitPage(pBt, pCur->pgnoRoot, &pRoot, 0))
    ){
      pCur->eState = CURSOR_INVALID;
      return rc;
    }
    releasePage(pCur->pPage);
    pCur->pPage = pRoot;
  }
  pCur->idx = 0;
  pCur->info.nSize = 0;
  if( pRoot->nCell==0 && !pRoot->leaf ){
    Pgno subpage;
    assert( pRoot->pgno==1 );
    subpage = get4byte(&pRoot->aData[pRoot->hdrOffset+8]);
    assert( subpage>0 );
    pCur->eState = CURSOR_VALID;
    rc = moveToChild(pCur, subpage);
  }
  pCur->eState = ((pCur->pPage->nCell>0)?CURSOR_VALID:CURSOR_INVALID);
  return rc;
}

/*
** Move the cursor down to the left-most leaf entry beneath the
** entry to which it is currently pointing.
**
** The left-most leaf is the one with the smallest key - the first
** in ascending order.
*/
static int moveToLeftmost(BtCursor *pCur){
  Pgno pgno;
  int rc;
  MemPage *pPage;

  assert( pCur->eState==CURSOR_VALID );
  while( !(pPage = pCur->pPage)->leaf ){
    assert( pCur->idx>=0 && pCur->idx<pPage->nCell );
    pgno = get4byte(findCell(pPage, pCur->idx));
    rc = moveToChild(pCur, pgno);
    if( rc ) return rc;
  }
  return SQLITE_OK;
}

/*
** Move the cursor down to the right-most leaf entry beneath the
** page to which it is currently pointing.  Notice the difference
** between moveToLeftmost() and moveToRightmost().  moveToLeftmost()
** finds the left-most entry beneath the *entry* whereas moveToRightmost()
** finds the right-most entry beneath the *page*.
**
** The right-most entry is the one with the largest key - the last
** key in ascending order.
*/
static int moveToRightmost(BtCursor *pCur){
  Pgno pgno;
  int rc;
  MemPage *pPage;

  assert( pCur->eState==CURSOR_VALID );
  while( !(pPage = pCur->pPage)->leaf ){
    pgno = get4byte(&pPage->aData[pPage->hdrOffset+8]);
    pCur->idx = pPage->nCell;
    rc = moveToChild(pCur, pgno);
    if( rc ) return rc;
  }
  pCur->idx = pPage->nCell - 1;
  pCur->info.nSize = 0;
  return SQLITE_OK;
}

/* Move the cursor to the first entry in the table.  Return SQLITE_OK
** on success.  Set *pRes to 0 if the cursor actually points to something
** or set *pRes to 1 if the table is empty.
*/
SQLITE_PRIVATE int sqlite3BtreeFirst(BtCursor *pCur, int *pRes){
  int rc;
  rc = moveToRoot(pCur);
  if( rc ) return rc;
  if( pCur->eState==CURSOR_INVALID ){
    assert( pCur->pPage->nCell==0 );
    *pRes = 1;
    return SQLITE_OK;
  }
  assert( pCur->pPage->nCell>0 );
  *pRes = 0;
  rc = moveToLeftmost(pCur);
  return rc;
}

/* Move the cursor to the last entry in the table.  Return SQLITE_OK
** on success.  Set *pRes to 0 if the cursor actually points to something
** or set *pRes to 1 if the table is empty.
*/
SQLITE_PRIVATE int sqlite3BtreeLast(BtCursor *pCur, int *pRes){
  int rc;
  rc = moveToRoot(pCur);
  if( rc ) return rc;
  if( CURSOR_INVALID==pCur->eState ){
    assert( pCur->pPage->nCell==0 );
    *pRes = 1;
    return SQLITE_OK;
  }
  assert( pCur->eState==CURSOR_VALID );
  *pRes = 0;
  rc = moveToRightmost(pCur);
  return rc;
}

/* Move the cursor so that it points to an entry near pKey/nKey.
** Return a success code.
**
** For INTKEY tables, only the nKey parameter is used.  pKey is
** ignored.  For other tables, nKey is the number of bytes of data
** in pKey.  The comparison function specified when the cursor was
** created is used to compare keys.
**
** If an exact match is not found, then the cursor is always
** left pointing at a leaf page which would hold the entry if it
** were present.  The cursor might point to an entry that comes
** before or after the key.
**
** The result of comparing the key with the entry to which the
** cursor is written to *pRes if pRes!=NULL.  The meaning of
** this value is as follows:
**
**     *pRes<0      The cursor is left pointing at an entry that
**                  is smaller than pKey or if the table is empty
**                  and the cursor is therefore left point to nothing.
**
**     *pRes==0     The cursor is left pointing at an entry that
**                  exactly matches pKey.
**
**     *pRes>0      The cursor is left pointing at an entry that
**                  is larger than pKey.
*/
SQLITE_PRIVATE int sqlite3BtreeMoveto(
  BtCursor *pCur,        /* The cursor to be moved */
  const void *pKey,      /* The key content for indices.  Not used by tables */
  i64 nKey,              /* Size of pKey.  Or the key for tables */
  int biasRight,         /* If true, bias the search to the high end */
  int *pRes              /* Search result flag */
){
  int rc;
  rc = moveToRoot(pCur);
  if( rc ) return rc;
  assert( pCur->pPage );
  assert( pCur->pPage->isInit );
  if( pCur->eState==CURSOR_INVALID ){
    *pRes = -1;
    assert( pCur->pPage->nCell==0 );
    return SQLITE_OK;
  }
  for(;;){
    int lwr, upr;
    Pgno chldPg;
    MemPage *pPage = pCur->pPage;
    int c = -1;  /* pRes return if table is empty must be -1 */
    lwr = 0;
    upr = pPage->nCell-1;
    if( !pPage->intKey && pKey==0 ){
      return SQLITE_CORRUPT_BKPT;
    }
    if( biasRight ){
      pCur->idx = upr;
    }else{
      pCur->idx = (upr+lwr)/2;
    }
    if( lwr<=upr ) for(;;){
      void *pCellKey;
      i64 nCellKey;
      pCur->info.nSize = 0;
      if( pPage->intKey ){
        u8 *pCell;
        pCell = findCell(pPage, pCur->idx) + pPage->childPtrSize;
        if( pPage->hasData ){
          u32 dummy;
          pCell += getVarint32(pCell, &dummy);
        }
        getVarint(pCell, (u64 *)&nCellKey);
        if( nCellKey<nKey ){
          c = -1;
        }else if( nCellKey>nKey ){
          c = +1;
        }else{
          c = 0;
        }
      }else{
        int available;
        pCellKey = (void *)fetchPayload(pCur, &available, 0);
        nCellKey = pCur->info.nKey;
        if( available>=nCellKey ){
          c = pCur->xCompare(pCur->pArg, nCellKey, pCellKey, nKey, pKey);
        }else{
          pCellKey = sqliteMallocRaw( nCellKey );
          if( pCellKey==0 ) return SQLITE_NOMEM;
          rc = sqlite3BtreeKey(pCur, 0, nCellKey, (void *)pCellKey);
          c = pCur->xCompare(pCur->pArg, nCellKey, pCellKey, nKey, pKey);
          sqliteFree(pCellKey);
          if( rc ) return rc;
        }
      }
      if( c==0 ){
        if( pPage->leafData && !pPage->leaf ){
          lwr = pCur->idx;
          upr = lwr - 1;
          break;
        }else{
          if( pRes ) *pRes = 0;
          return SQLITE_OK;
        }
      }
      if( c<0 ){
        lwr = pCur->idx+1;
      }else{
        upr = pCur->idx-1;
      }
      if( lwr>upr ){
        break;
      }
      pCur->idx = (lwr+upr)/2;
    }
    assert( lwr==upr+1 );
    assert( pPage->isInit );
    if( pPage->leaf ){
      chldPg = 0;
    }else if( lwr>=pPage->nCell ){
      chldPg = get4byte(&pPage->aData[pPage->hdrOffset+8]);
    }else{
      chldPg = get4byte(findCell(pPage, lwr));
    }
    if( chldPg==0 ){
      assert( pCur->idx>=0 && pCur->idx<pCur->pPage->nCell );
      if( pRes ) *pRes = c;
      return SQLITE_OK;
    }
    pCur->idx = lwr;
    pCur->info.nSize = 0;
    rc = moveToChild(pCur, chldPg);
    if( rc ){
      return rc;
    }
  }
  /* NOT REACHED */
}

/*
** Return TRUE if the cursor is not pointing at an entry of the table.
**
** TRUE will be returned after a call to sqlite3BtreeNext() moves
** past the last entry in the table or sqlite3BtreePrev() moves past
** the first entry.  TRUE is also returned if the table is empty.
*/
SQLITE_PRIVATE int sqlite3BtreeEof(BtCursor *pCur){
  /* TODO: What if the cursor is in CURSOR_REQUIRESEEK but all table entries
  ** have been deleted? This API will need to change to return an error code
  ** as well as the boolean result value.
  */
  return (CURSOR_VALID!=pCur->eState);
}

/*
** Advance the cursor to the next entry in the database.  If
** successful then set *pRes=0.  If the cursor
** was already pointing to the last entry in the database before
** this routine was called, then set *pRes=1.
*/
SQLITE_PRIVATE int sqlite3BtreeNext(BtCursor *pCur, int *pRes){
  int rc;
  MemPage *pPage;

  rc = restoreOrClearCursorPosition(pCur);
  if( rc!=SQLITE_OK ){
    return rc;
  }
  assert( pRes!=0 );
  pPage = pCur->pPage;
  if( CURSOR_INVALID==pCur->eState ){
    *pRes = 1;
    return SQLITE_OK;
  }
  if( pCur->skip>0 ){
    pCur->skip = 0;
    *pRes = 0;
    return SQLITE_OK;
  }
  pCur->skip = 0;

  assert( pPage->isInit );
  assert( pCur->idx<pPage->nCell );

  pCur->idx++;
  pCur->info.nSize = 0;
  if( pCur->idx>=pPage->nCell ){
    if( !pPage->leaf ){
      rc = moveToChild(pCur, get4byte(&pPage->aData[pPage->hdrOffset+8]));
      if( rc ) return rc;
      rc = moveToLeftmost(pCur);
      *pRes = 0;
      return rc;
    }
    do{
      if( sqlite3BtreeIsRootPage(pPage) ){
        *pRes = 1;
        pCur->eState = CURSOR_INVALID;
        return SQLITE_OK;
      }
      sqlite3BtreeMoveToParent(pCur);
      pPage = pCur->pPage;
    }while( pCur->idx>=pPage->nCell );
    *pRes = 0;
    if( pPage->leafData ){
      rc = sqlite3BtreeNext(pCur, pRes);
    }else{
      rc = SQLITE_OK;
    }
    return rc;
  }
  *pRes = 0;
  if( pPage->leaf ){
    return SQLITE_OK;
  }
  rc = moveToLeftmost(pCur);
  return rc;
}

/*
** Step the cursor to the back to the previous entry in the database.  If
** successful then set *pRes=0.  If the cursor
** was already pointing to the first entry in the database before
** this routine was called, then set *pRes=1.
*/
SQLITE_PRIVATE int sqlite3BtreePrevious(BtCursor *pCur, int *pRes){
  int rc;
  Pgno pgno;
  MemPage *pPage;

  rc = restoreOrClearCursorPosition(pCur);
  if( rc!=SQLITE_OK ){
    return rc;
  }
  if( CURSOR_INVALID==pCur->eState ){
    *pRes = 1;
    return SQLITE_OK;
  }
  if( pCur->skip<0 ){
    pCur->skip = 0;
    *pRes = 0;
    return SQLITE_OK;
  }
  pCur->skip = 0;

  pPage = pCur->pPage;
  assert( pPage->isInit );
  assert( pCur->idx>=0 );
  if( !pPage->leaf ){
    pgno = get4byte( findCell(pPage, pCur->idx) );
    rc = moveToChild(pCur, pgno);
    if( rc ) return rc;
    rc = moveToRightmost(pCur);
  }else{
    while( pCur->idx==0 ){
      if( sqlite3BtreeIsRootPage(pPage) ){
        pCur->eState = CURSOR_INVALID;
        *pRes = 1;
        return SQLITE_OK;
      }
      sqlite3BtreeMoveToParent(pCur);
      pPage = pCur->pPage;
    }
    pCur->idx--;
    pCur->info.nSize = 0;
    if( pPage->leafData && !pPage->leaf ){
      rc = sqlite3BtreePrevious(pCur, pRes);
    }else{
      rc = SQLITE_OK;
    }
  }
  *pRes = 0;
  return rc;
}

/*
** Allocate a new page from the database file.
**
** The new page is marked as dirty.  (In other words, sqlite3PagerWrite()
** has already been called on the new page.)  The new page has also
** been referenced and the calling routine is responsible for calling
** sqlite3PagerUnref() on the new page when it is done.
**
** SQLITE_OK is returned on success.  Any other return value indicates
** an error.  *ppPage and *pPgno are undefined in the event of an error.
** Do not invoke sqlite3PagerUnref() on *ppPage if an error is returned.
**
** If the "nearby" parameter is not 0, then a (feeble) effort is made to 
** locate a page close to the page number "nearby".  This can be used in an
** attempt to keep related pages close to each other in the database file,
** which in turn can make database access faster.
**
** If the "exact" parameter is not 0, and the page-number nearby exists 
** anywhere on the free-list, then it is guarenteed to be returned. This
** is only used by auto-vacuum databases when allocating a new table.
*/
static int allocateBtreePage(
  BtShared *pBt, 
  MemPage **ppPage, 
  Pgno *pPgno, 
  Pgno nearby,
  u8 exact
){
  MemPage *pPage1;
  int rc;
  int n;     /* Number of pages on the freelist */
  int k;     /* Number of leaves on the trunk of the freelist */
  MemPage *pTrunk = 0;
  MemPage *pPrevTrunk = 0;

  pPage1 = pBt->pPage1;
  n = get4byte(&pPage1->aData[36]);
  if( n>0 ){
    /* There are pages on the freelist.  Reuse one of those pages. */
    Pgno iTrunk;
    u8 searchList = 0; /* If the free-list must be searched for 'nearby' */
    
    /* If the 'exact' parameter was true and a query of the pointer-map
    ** shows that the page 'nearby' is somewhere on the free-list, then
    ** the entire-list will be searched for that page.
    */
#ifndef SQLITE_OMIT_AUTOVACUUM
    if( exact && nearby<=sqlite3PagerPagecount(pBt->pPager) ){
      u8 eType;
      assert( nearby>0 );
      assert( pBt->autoVacuum );
      rc = ptrmapGet(pBt, nearby, &eType, 0);
      if( rc ) return rc;
      if( eType==PTRMAP_FREEPAGE ){
        searchList = 1;
      }
      *pPgno = nearby;
    }
#endif

    /* Decrement the free-list count by 1. Set iTrunk to the index of the
    ** first free-list trunk page. iPrevTrunk is initially 1.
    */
    rc = sqlite3PagerWrite(pPage1->pDbPage);
    if( rc ) return rc;
    put4byte(&pPage1->aData[36], n-1);

    /* The code within this loop is run only once if the 'searchList' variable
    ** is not true. Otherwise, it runs once for each trunk-page on the
    ** free-list until the page 'nearby' is located.
    */
    do {
      pPrevTrunk = pTrunk;
      if( pPrevTrunk ){
        iTrunk = get4byte(&pPrevTrunk->aData[0]);
      }else{
        iTrunk = get4byte(&pPage1->aData[32]);
      }
      rc = sqlite3BtreeGetPage(pBt, iTrunk, &pTrunk, 0);
      if( rc ){
        pTrunk = 0;
        goto end_allocate_page;
      }

      k = get4byte(&pTrunk->aData[4]);
      if( k==0 && !searchList ){
        /* The trunk has no leaves and the list is not being searched. 
        ** So extract the trunk page itself and use it as the newly 
        ** allocated page */
        assert( pPrevTrunk==0 );
        rc = sqlite3PagerWrite(pTrunk->pDbPage);
        if( rc ){
          goto end_allocate_page;
        }
        *pPgno = iTrunk;
        memcpy(&pPage1->aData[32], &pTrunk->aData[0], 4);
        *ppPage = pTrunk;
        pTrunk = 0;
        TRACE(("ALLOCATE: %d trunk - %d free pages left\n", *pPgno, n-1));
      }else if( k>pBt->usableSize/4 - 8 ){
        /* Value of k is out of range.  Database corruption */
        rc = SQLITE_CORRUPT_BKPT;
        goto end_allocate_page;
#ifndef SQLITE_OMIT_AUTOVACUUM
      }else if( searchList && nearby==iTrunk ){
        /* The list is being searched and this trunk page is the page
        ** to allocate, regardless of whether it has leaves.
        */
        assert( *pPgno==iTrunk );
        *ppPage = pTrunk;
        searchList = 0;
        rc = sqlite3PagerWrite(pTrunk->pDbPage);
        if( rc ){
          goto end_allocate_page;
        }
        if( k==0 ){
          if( !pPrevTrunk ){
            memcpy(&pPage1->aData[32], &pTrunk->aData[0], 4);
          }else{
            memcpy(&pPrevTrunk->aData[0], &pTrunk->aData[0], 4);
          }
        }else{
          /* The trunk page is required by the caller but it contains 
          ** pointers to free-list leaves. The first leaf becomes a trunk
          ** page in this case.
          */
          MemPage *pNewTrunk;
          Pgno iNewTrunk = get4byte(&pTrunk->aData[8]);
          rc = sqlite3BtreeGetPage(pBt, iNewTrunk, &pNewTrunk, 0);
          if( rc!=SQLITE_OK ){
            goto end_allocate_page;
          }
          rc = sqlite3PagerWrite(pNewTrunk->pDbPage);
          if( rc!=SQLITE_OK ){
            releasePage(pNewTrunk);
            goto end_allocate_page;
          }
          memcpy(&pNewTrunk->aData[0], &pTrunk->aData[0], 4);
          put4byte(&pNewTrunk->aData[4], k-1);
          memcpy(&pNewTrunk->aData[8], &pTrunk->aData[12], (k-1)*4);
          releasePage(pNewTrunk);
          if( !pPrevTrunk ){
            put4byte(&pPage1->aData[32], iNewTrunk);
          }else{
            rc = sqlite3PagerWrite(pPrevTrunk->pDbPage);
            if( rc ){
              goto end_allocate_page;
            }
            put4byte(&pPrevTrunk->aData[0], iNewTrunk);
          }
        }
        pTrunk = 0;
        TRACE(("ALLOCATE: %d trunk - %d free pages left\n", *pPgno, n-1));
#endif
      }else{
        /* Extract a leaf from the trunk */
        int closest;
        Pgno iPage;
        unsigned char *aData = pTrunk->aData;
        rc = sqlite3PagerWrite(pTrunk->pDbPage);
        if( rc ){
          goto end_allocate_page;
        }
        if( nearby>0 ){
          int i, dist;
          closest = 0;
          dist = get4byte(&aData[8]) - nearby;
          if( dist<0 ) dist = -dist;
          for(i=1; i<k; i++){
            int d2 = get4byte(&aData[8+i*4]) - nearby;
            if( d2<0 ) d2 = -d2;
            if( d2<dist ){
              closest = i;
              dist = d2;
            }
          }
        }else{
          closest = 0;
        }

        iPage = get4byte(&aData[8+closest*4]);
        if( !searchList || iPage==nearby ){
          *pPgno = iPage;
          if( *pPgno>sqlite3PagerPagecount(pBt->pPager) ){
            /* Free page off the end of the file */
            return SQLITE_CORRUPT_BKPT;
          }
          TRACE(("ALLOCATE: %d was leaf %d of %d on trunk %d"
                 ": %d more free pages\n",
                 *pPgno, closest+1, k, pTrunk->pgno, n-1));
          if( closest<k-1 ){
            memcpy(&aData[8+closest*4], &aData[4+k*4], 4);
          }
          put4byte(&aData[4], k-1);
          rc = sqlite3BtreeGetPage(pBt, *pPgno, ppPage, 1);
          if( rc==SQLITE_OK ){
            sqlite3PagerDontRollback((*ppPage)->pDbPage);
            rc = sqlite3PagerWrite((*ppPage)->pDbPage);
            if( rc!=SQLITE_OK ){
              releasePage(*ppPage);
            }
          }
          searchList = 0;
        }
      }
      releasePage(pPrevTrunk);
      pPrevTrunk = 0;
    }while( searchList );
  }else{
    /* There are no pages on the freelist, so create a new page at the
    ** end of the file */
    *pPgno = sqlite3PagerPagecount(pBt->pPager) + 1;

#ifndef SQLITE_OMIT_AUTOVACUUM
    if( pBt->nTrunc ){
      /* An incr-vacuum has already run within this transaction. So the
      ** page to allocate is not from the physical end of the file, but
      ** at pBt->nTrunc. 
      */
      *pPgno = pBt->nTrunc+1;
      if( *pPgno==PENDING_BYTE_PAGE(pBt) ){
        (*pPgno)++;
      }
    }
    if( pBt->autoVacuum && PTRMAP_ISPAGE(pBt, *pPgno) ){
      /* If *pPgno refers to a pointer-map page, allocate two new pages
      ** at the end of the file instead of one. The first allocated page
      ** becomes a new pointer-map page, the second is used by the caller.
      */
      TRACE(("ALLOCATE: %d from end of file (pointer-map page)\n", *pPgno));
      assert( *pPgno!=PENDING_BYTE_PAGE(pBt) );
      (*pPgno)++;
    }
    if( pBt->nTrunc ){
      pBt->nTrunc = *pPgno;
    }
#endif

    assert( *pPgno!=PENDING_BYTE_PAGE(pBt) );
    rc = sqlite3BtreeGetPage(pBt, *pPgno, ppPage, 0);
    if( rc ) return rc;
    rc = sqlite3PagerWrite((*ppPage)->pDbPage);
    if( rc!=SQLITE_OK ){
      releasePage(*ppPage);
    }
    TRACE(("ALLOCATE: %d from end of file\n", *pPgno));
  }

  assert( *pPgno!=PENDING_BYTE_PAGE(pBt) );

end_allocate_page:
  releasePage(pTrunk);
  releasePage(pPrevTrunk);
  return rc;
}

/*
** Add a page of the database file to the freelist.
**
** sqlite3PagerUnref() is NOT called for pPage.
*/
static int freePage(MemPage *pPage){
  BtShared *pBt = pPage->pBt;
  MemPage *pPage1 = pBt->pPage1;
  int rc, n, k;

  /* Prepare the page for freeing */
  assert( pPage->pgno>1 );
  pPage->isInit = 0;
  releasePage(pPage->pParent);
  pPage->pParent = 0;

  /* Increment the free page count on pPage1 */
  rc = sqlite3PagerWrite(pPage1->pDbPage);
  if( rc ) return rc;
  n = get4byte(&pPage1->aData[36]);
  put4byte(&pPage1->aData[36], n+1);

#ifdef SQLITE_SECURE_DELETE
  /* If the SQLITE_SECURE_DELETE compile-time option is enabled, then
  ** always fully overwrite deleted information with zeros.
  */
  rc = sqlite3PagerWrite(pPage->pDbPage);
  if( rc ) return rc;
  memset(pPage->aData, 0, pPage->pBt->pageSize);
#endif

#ifndef SQLITE_OMIT_AUTOVACUUM
  /* If the database supports auto-vacuum, write an entry in the pointer-map
  ** to indicate that the page is free.
  */
  if( pBt->autoVacuum ){
    rc = ptrmapPut(pBt, pPage->pgno, PTRMAP_FREEPAGE, 0);
    if( rc ) return rc;
  }
#endif

  if( n==0 ){
    /* This is the first free page */
    rc = sqlite3PagerWrite(pPage->pDbPage);
    if( rc ) return rc;
    memset(pPage->aData, 0, 8);
    put4byte(&pPage1->aData[32], pPage->pgno);
    TRACE(("FREE-PAGE: %d first\n", pPage->pgno));
  }else{
    /* Other free pages already exist.  Retrive the first trunk page
    ** of the freelist and find out how many leaves it has. */
    MemPage *pTrunk;
    rc = sqlite3BtreeGetPage(pBt, get4byte(&pPage1->aData[32]), &pTrunk, 0);
    if( rc ) return rc;
    k = get4byte(&pTrunk->aData[4]);
    if( k>=pBt->usableSize/4 - 8 ){
      /* The trunk is full.  Turn the page being freed into a new
      ** trunk page with no leaves. */
      rc = sqlite3PagerWrite(pPage->pDbPage);
      if( rc ) return rc;
      put4byte(pPage->aData, pTrunk->pgno);
      put4byte(&pPage->aData[4], 0);
      put4byte(&pPage1->aData[32], pPage->pgno);
      TRACE(("FREE-PAGE: %d new trunk page replacing %d\n",
              pPage->pgno, pTrunk->pgno));
    }else{
      /* Add the newly freed page as a leaf on the current trunk */
      rc = sqlite3PagerWrite(pTrunk->pDbPage);
      if( rc==SQLITE_OK ){
        put4byte(&pTrunk->aData[4], k+1);
        put4byte(&pTrunk->aData[8+k*4], pPage->pgno);
#ifndef SQLITE_SECURE_DELETE
        sqlite3PagerDontWrite(pPage->pDbPage);
#endif
      }
      TRACE(("FREE-PAGE: %d leaf on trunk page %d\n",pPage->pgno,pTrunk->pgno));
    }
    releasePage(pTrunk);
  }
  return rc;
}

/*
** Free any overflow pages associated with the given Cell.
*/
static int clearCell(MemPage *pPage, unsigned char *pCell){
  BtShared *pBt = pPage->pBt;
  CellInfo info;
  Pgno ovflPgno;
  int rc;
  int nOvfl;
  int ovflPageSize;

  sqlite3BtreeParseCellPtr(pPage, pCell, &info);
  if( info.iOverflow==0 ){
    return SQLITE_OK;  /* No overflow pages. Return without doing anything */
  }
  ovflPgno = get4byte(&pCell[info.iOverflow]);
  ovflPageSize = pBt->usableSize - 4;
  nOvfl = (info.nPayload - info.nLocal + ovflPageSize - 1)/ovflPageSize;
  assert( ovflPgno==0 || nOvfl>0 );
  while( nOvfl-- ){
    MemPage *pOvfl;
    if( ovflPgno==0 || ovflPgno>sqlite3PagerPagecount(pBt->pPager) ){
      return SQLITE_CORRUPT_BKPT;
    }

    rc = getOverflowPage(pBt, ovflPgno, &pOvfl, (nOvfl==0)?0:&ovflPgno);
    if( rc ) return rc;
    rc = freePage(pOvfl);
    sqlite3PagerUnref(pOvfl->pDbPage);
    if( rc ) return rc;
  }
  return SQLITE_OK;
}

/*
** Create the byte sequence used to represent a cell on page pPage
** and write that byte sequence into pCell[].  Overflow pages are
** allocated and filled in as necessary.  The calling procedure
** is responsible for making sure sufficient space has been allocated
** for pCell[].
**
** Note that pCell does not necessary need to point to the pPage->aData
** area.  pCell might point to some temporary storage.  The cell will
** be constructed in this temporary area then copied into pPage->aData
** later.
*/
static int fillInCell(
  MemPage *pPage,                /* The page that contains the cell */
  unsigned char *pCell,          /* Complete text of the cell */
  const void *pKey, i64 nKey,    /* The key */
  const void *pData,int nData,   /* The data */
  int nZero,                     /* Extra zero bytes to append to pData */
  int *pnSize                    /* Write cell size here */
){
  int nPayload;
  const u8 *pSrc;
  int nSrc, n, rc;
  int spaceLeft;
  MemPage *pOvfl = 0;
  MemPage *pToRelease = 0;
  unsigned char *pPrior;
  unsigned char *pPayload;
  BtShared *pBt = pPage->pBt;
  Pgno pgnoOvfl = 0;
  int nHeader;
  CellInfo info;

  /* Fill in the header. */
  nHeader = 0;
  if( !pPage->leaf ){
    nHeader += 4;
  }
  if( pPage->hasData ){
    nHeader += putVarint(&pCell[nHeader], nData+nZero);
  }else{
    nData = nZero = 0;
  }
  nHeader += putVarint(&pCell[nHeader], *(u64*)&nKey);
  sqlite3BtreeParseCellPtr(pPage, pCell, &info);
  assert( info.nHeader==nHeader );
  assert( info.nKey==nKey );
  assert( info.nData==nData+nZero );
  
  /* Fill in the payload */
  nPayload = nData + nZero;
  if( pPage->intKey ){
    pSrc = pData;
    nSrc = nData;
    nData = 0;
  }else{
    nPayload += nKey;
    pSrc = pKey;
    nSrc = nKey;
  }
  *pnSize = info.nSize;
  spaceLeft = info.nLocal;
  pPayload = &pCell[nHeader];
  pPrior = &pCell[info.iOverflow];

  while( nPayload>0 ){
    if( spaceLeft==0 ){
      int isExact = 0;
#ifndef SQLITE_OMIT_AUTOVACUUM
      Pgno pgnoPtrmap = pgnoOvfl; /* Overflow page pointer-map entry page */
      if( pBt->autoVacuum ){
        do{
          pgnoOvfl++;
        } while( 
          PTRMAP_ISPAGE(pBt, pgnoOvfl) || pgnoOvfl==PENDING_BYTE_PAGE(pBt) 
        );
        if( pgnoOvfl>1 ){
          /* isExact = 1; */
        }
      }
#endif
      rc = allocateBtreePage(pBt, &pOvfl, &pgnoOvfl, pgnoOvfl, isExact);
#ifndef SQLITE_OMIT_AUTOVACUUM
      /* If the database supports auto-vacuum, and the second or subsequent
      ** overflow page is being allocated, add an entry to the pointer-map
      ** for that page now. 
      **
      ** If this is the first overflow page, then write a partial entry 
      ** to the pointer-map. If we write nothing to this pointer-map slot,
      ** then the optimistic overflow chain processing in clearCell()
      ** may misinterpret the uninitialised values and delete the
      ** wrong pages from the database.
      */
      if( pBt->autoVacuum && rc==SQLITE_OK ){
        u8 eType = (pgnoPtrmap?PTRMAP_OVERFLOW2:PTRMAP_OVERFLOW1);
        rc = ptrmapPut(pBt, pgnoOvfl, eType, pgnoPtrmap);
        if( rc ){
          releasePage(pOvfl);
        }
      }
#endif
      if( rc ){
        releasePage(pToRelease);
        return rc;
      }
      put4byte(pPrior, pgnoOvfl);
      releasePage(pToRelease);
      pToRelease = pOvfl;
      pPrior = pOvfl->aData;
      put4byte(pPrior, 0);
      pPayload = &pOvfl->aData[4];
      spaceLeft = pBt->usableSize - 4;
    }
    n = nPayload;
    if( n>spaceLeft ) n = spaceLeft;
    if( nSrc>0 ){
      if( n>nSrc ) n = nSrc;
      assert( pSrc );
      memcpy(pPayload, pSrc, n);
    }else{
      memset(pPayload, 0, n);
    }
    nPayload -= n;
    pPayload += n;
    pSrc += n;
    nSrc -= n;
    spaceLeft -= n;
    if( nSrc==0 ){
      nSrc = nData;
      pSrc = pData;
    }
  }
  releasePage(pToRelease);
  return SQLITE_OK;
}

/*
** Change the MemPage.pParent pointer on the page whose number is
** given in the second argument so that MemPage.pParent holds the
** pointer in the third argument.
*/
static int reparentPage(BtShared *pBt, Pgno pgno, MemPage *pNewParent, int idx){
  MemPage *pThis;
  DbPage *pDbPage;

  assert( pNewParent!=0 );
  if( pgno==0 ) return SQLITE_OK;
  assert( pBt->pPager!=0 );
  pDbPage = sqlite3PagerLookup(pBt->pPager, pgno);
  if( pDbPage ){
    pThis = (MemPage *)sqlite3PagerGetExtra(pDbPage);
    if( pThis->isInit ){
      assert( pThis->aData==(sqlite3PagerGetData(pDbPage)) );
      if( pThis->pParent!=pNewParent ){
        if( pThis->pParent ) sqlite3PagerUnref(pThis->pParent->pDbPage);
        pThis->pParent = pNewParent;
        sqlite3PagerRef(pNewParent->pDbPage);
      }
      pThis->idxParent = idx;
    }
    sqlite3PagerUnref(pDbPage);
  }

#ifndef SQLITE_OMIT_AUTOVACUUM
  if( pBt->autoVacuum ){
    return ptrmapPut(pBt, pgno, PTRMAP_BTREE, pNewParent->pgno);
  }
#endif
  return SQLITE_OK;
}



/*
** Change the pParent pointer of all children of pPage to point back
** to pPage.
**
** In other words, for every child of pPage, invoke reparentPage()
** to make sure that each child knows that pPage is its parent.
**
** This routine gets called after you memcpy() one page into
** another.
*/
static int reparentChildPages(MemPage *pPage){
  int i;
  BtShared *pBt = pPage->pBt;
  int rc = SQLITE_OK;

  if( pPage->leaf ) return SQLITE_OK;

  for(i=0; i<pPage->nCell; i++){
    u8 *pCell = findCell(pPage, i);
    if( !pPage->leaf ){
      rc = reparentPage(pBt, get4byte(pCell), pPage, i);
      if( rc!=SQLITE_OK ) return rc;
    }
  }
  if( !pPage->leaf ){
    rc = reparentPage(pBt, get4byte(&pPage->aData[pPage->hdrOffset+8]), 
       pPage, i);
    pPage->idxShift = 0;
  }
  return rc;
}

/*
** Remove the i-th cell from pPage.  This routine effects pPage only.
** The cell content is not freed or deallocated.  It is assumed that
** the cell content has been copied someplace else.  This routine just
** removes the reference to the cell from pPage.
**
** "sz" must be the number of bytes in the cell.
*/
static void dropCell(MemPage *pPage, int idx, int sz){
  int i;          /* Loop counter */
  int pc;         /* Offset to cell content of cell being deleted */
  u8 *data;       /* pPage->aData */
  u8 *ptr;        /* Used to move bytes around within data[] */

  assert( idx>=0 && idx<pPage->nCell );
  assert( sz==cellSize(pPage, idx) );
  assert( sqlite3PagerIswriteable(pPage->pDbPage) );
  data = pPage->aData;
  ptr = &data[pPage->cellOffset + 2*idx];
  pc = get2byte(ptr);
  assert( pc>10 && pc+sz<=pPage->pBt->usableSize );
  freeSpace(pPage, pc, sz);
  for(i=idx+1; i<pPage->nCell; i++, ptr+=2){
    ptr[0] = ptr[2];
    ptr[1] = ptr[3];
  }
  pPage->nCell--;
  put2byte(&data[pPage->hdrOffset+3], pPage->nCell);
  pPage->nFree += 2;
  pPage->idxShift = 1;
}

/*
** Insert a new cell on pPage at cell index "i".  pCell points to the
** content of the cell.
**
** If the cell content will fit on the page, then put it there.  If it
** will not fit, then make a copy of the cell content into pTemp if
** pTemp is not null.  Regardless of pTemp, allocate a new entry
** in pPage->aOvfl[] and make it point to the cell content (either
** in pTemp or the original pCell) and also record its index. 
** Allocating a new entry in pPage->aCell[] implies that 
** pPage->nOverflow is incremented.
**
** If nSkip is non-zero, then do not copy the first nSkip bytes of the
** cell. The caller will overwrite them after this function returns. If
** nSkip is non-zero, then pCell may not point to an invalid memory location 
** (but pCell+nSkip is always valid).
*/
static int insertCell(
  MemPage *pPage,   /* Page into which we are copying */
  int i,            /* New cell becomes the i-th cell of the page */
  u8 *pCell,        /* Content of the new cell */
  int sz,           /* Bytes of content in pCell */
  u8 *pTemp,        /* Temp storage space for pCell, if needed */
  u8 nSkip          /* Do not write the first nSkip bytes of the cell */
){
  int idx;          /* Where to write new cell content in data[] */
  int j;            /* Loop counter */
  int top;          /* First byte of content for any cell in data[] */
  int end;          /* First byte past the last cell pointer in data[] */
  int ins;          /* Index in data[] where new cell pointer is inserted */
  int hdr;          /* Offset into data[] of the page header */
  int cellOffset;   /* Address of first cell pointer in data[] */
  u8 *data;         /* The content of the whole page */
  u8 *ptr;          /* Used for moving information around in data[] */

  assert( i>=0 && i<=pPage->nCell+pPage->nOverflow );
  assert( sz==cellSizePtr(pPage, pCell) );
  assert( sqlite3PagerIswriteable(pPage->pDbPage) );
  if( pPage->nOverflow || sz+2>pPage->nFree ){
    if( pTemp ){
      memcpy(pTemp+nSkip, pCell+nSkip, sz-nSkip);
      pCell = pTemp;
    }
    j = pPage->nOverflow++;
    assert( j<sizeof(pPage->aOvfl)/sizeof(pPage->aOvfl[0]) );
    pPage->aOvfl[j].pCell = pCell;
    pPage->aOvfl[j].idx = i;
    pPage->nFree = 0;
  }else{
    data = pPage->aData;
    hdr = pPage->hdrOffset;
    top = get2byte(&data[hdr+5]);
    cellOffset = pPage->cellOffset;
    end = cellOffset + 2*pPage->nCell + 2;
    ins = cellOffset + 2*i;
    if( end > top - sz ){
      int rc = defragmentPage(pPage);
      if( rc!=SQLITE_OK ) return rc;
      top = get2byte(&data[hdr+5]);
      assert( end + sz <= top );
    }
    idx = allocateSpace(pPage, sz);
    assert( idx>0 );
    assert( end <= get2byte(&data[hdr+5]) );
    pPage->nCell++;
    pPage->nFree -= 2;
    memcpy(&data[idx+nSkip], pCell+nSkip, sz-nSkip);
    for(j=end-2, ptr=&data[j]; j>ins; j-=2, ptr-=2){
      ptr[0] = ptr[-2];
      ptr[1] = ptr[-1];
    }
    put2byte(&data[ins], idx);
    put2byte(&data[hdr+3], pPage->nCell);
    pPage->idxShift = 1;
#ifndef SQLITE_OMIT_AUTOVACUUM
    if( pPage->pBt->autoVacuum ){
      /* The cell may contain a pointer to an overflow page. If so, write
      ** the entry for the overflow page into the pointer map.
      */
      CellInfo info;
      sqlite3BtreeParseCellPtr(pPage, pCell, &info);
      assert( (info.nData+(pPage->intKey?0:info.nKey))==info.nPayload );
      if( (info.nData+(pPage->intKey?0:info.nKey))>info.nLocal ){
        Pgno pgnoOvfl = get4byte(&pCell[info.iOverflow]);
        int rc = ptrmapPut(pPage->pBt, pgnoOvfl, PTRMAP_OVERFLOW1, pPage->pgno);
        if( rc!=SQLITE_OK ) return rc;
      }
    }
#endif
  }

  return SQLITE_OK;
}

/*
** Add a list of cells to a page.  The page should be initially empty.
** The cells are guaranteed to fit on the page.
*/
static void assemblePage(
  MemPage *pPage,   /* The page to be assemblied */
  int nCell,        /* The number of cells to add to this page */
  u8 **apCell,      /* Pointers to cell bodies */
  int *aSize        /* Sizes of the cells */
){
  int i;            /* Loop counter */
  int totalSize;    /* Total size of all cells */
  int hdr;          /* Index of page header */
  int cellptr;      /* Address of next cell pointer */
  int cellbody;     /* Address of next cell body */
  u8 *data;         /* Data for the page */

  assert( pPage->nOverflow==0 );
  totalSize = 0;
  for(i=0; i<nCell; i++){
    totalSize += aSize[i];
  }
  assert( totalSize+2*nCell<=pPage->nFree );
  assert( pPage->nCell==0 );
  cellptr = pPage->cellOffset;
  data = pPage->aData;
  hdr = pPage->hdrOffset;
  put2byte(&data[hdr+3], nCell);
  if( nCell ){
    cellbody = allocateSpace(pPage, totalSize);
    assert( cellbody>0 );
    assert( pPage->nFree >= 2*nCell );
    pPage->nFree -= 2*nCell;
    for(i=0; i<nCell; i++){
      put2byte(&data[cellptr], cellbody);
      memcpy(&data[cellbody], apCell[i], aSize[i]);
      cellptr += 2;
      cellbody += aSize[i];
    }
    assert( cellbody==pPage->pBt->usableSize );
  }
  pPage->nCell = nCell;
}

/*
** The following parameters determine how many adjacent pages get involved
** in a balancing operation.  NN is the number of neighbors on either side
** of the page that participate in the balancing operation.  NB is the
** total number of pages that participate, including the target page and
** NN neighbors on either side.
**
** The minimum value of NN is 1 (of course).  Increasing NN above 1
** (to 2 or 3) gives a modest improvement in SELECT and DELETE performance
** in exchange for a larger degradation in INSERT and UPDATE performance.
** The value of NN appears to give the best results overall.
*/
#define NN 1             /* Number of neighbors on either side of pPage */
#define NB (NN*2+1)      /* Total pages involved in the balance */

/* Forward reference */
static int balance(MemPage*, int);

#ifndef SQLITE_OMIT_QUICKBALANCE
/*
** This version of balance() handles the common special case where
** a new entry is being inserted on the extreme right-end of the
** tree, in other words, when the new entry will become the largest
** entry in the tree.
**
** Instead of trying balance the 3 right-most leaf pages, just add
** a new page to the right-hand side and put the one new entry in
** that page.  This leaves the right side of the tree somewhat
** unbalanced.  But odds are that we will be inserting new entries
** at the end soon afterwards so the nearly empty page will quickly
** fill up.  On average.
**
** pPage is the leaf page which is the right-most page in the tree.
** pParent is its parent.  pPage must have a single overflow entry
** which is also the right-most entry on the page.
*/
static int balance_quick(MemPage *pPage, MemPage *pParent){
  int rc;
  MemPage *pNew;
  Pgno pgnoNew;
  u8 *pCell;
  int szCell;
  CellInfo info;
  BtShared *pBt = pPage->pBt;
  int parentIdx = pParent->nCell;   /* pParent new divider cell index */
  int parentSize;                   /* Size of new divider cell */
  u8 parentCell[64];                /* Space for the new divider cell */

  /* Allocate a new page. Insert the overflow cell from pPage
  ** into it. Then remove the overflow cell from pPage.
  */
  rc = allocateBtreePage(pBt, &pNew, &pgnoNew, 0, 0);
  if( rc!=SQLITE_OK ){
    return rc;
  }
  pCell = pPage->aOvfl[0].pCell;
  szCell = cellSizePtr(pPage, pCell);
  zeroPage(pNew, pPage->aData[0]);
  assemblePage(pNew, 1, &pCell, &szCell);
  pPage->nOverflow = 0;

  /* Set the parent of the newly allocated page to pParent. */
  pNew->pParent = pParent;
  sqlite3PagerRef(pParent->pDbPage);

  /* pPage is currently the right-child of pParent. Change this
  ** so that the right-child is the new page allocated above and
  ** pPage is the next-to-right child. 
  */
  assert( pPage->nCell>0 );
  pCell = findCell(pPage, pPage->nCell-1);
  sqlite3BtreeParseCellPtr(pPage, pCell, &info);
  rc = fillInCell(pParent, parentCell, 0, info.nKey, 0, 0, 0, &parentSize);
  if( rc!=SQLITE_OK ){
    return rc;
  }
  assert( parentSize<64 );
  rc = insertCell(pParent, parentIdx, parentCell, parentSize, 0, 4);
  if( rc!=SQLITE_OK ){
    return rc;
  }
  put4byte(findOverflowCell(pParent,parentIdx), pPage->pgno);
  put4byte(&pParent->aData[pParent->hdrOffset+8], pgnoNew);

#ifndef SQLITE_OMIT_AUTOVACUUM
  /* If this is an auto-vacuum database, update the pointer map
  ** with entries for the new page, and any pointer from the 
  ** cell on the page to an overflow page.
  */
  if( pBt->autoVacuum ){
    rc = ptrmapPut(pBt, pgnoNew, PTRMAP_BTREE, pParent->pgno);
    if( rc==SQLITE_OK ){
      rc = ptrmapPutOvfl(pNew, 0);
    }
    if( rc!=SQLITE_OK ){
      releasePage(pNew);
      return rc;
    }
  }
#endif

  /* Release the reference to the new page and balance the parent page,
  ** in case the divider cell inserted caused it to become overfull.
  */
  releasePage(pNew);
  return balance(pParent, 0);
}
#endif /* SQLITE_OMIT_QUICKBALANCE */

/*
** This routine redistributes Cells on pPage and up to NN*2 siblings
** of pPage so that all pages have about the same amount of free space.
** Usually NN siblings on either side of pPage is used in the balancing,
** though more siblings might come from one side if pPage is the first
** or last child of its parent.  If pPage has fewer than 2*NN siblings
** (something which can only happen if pPage is the root page or a 
** child of root) then all available siblings participate in the balancing.
**
** The number of siblings of pPage might be increased or decreased by one or
** two in an effort to keep pages nearly full but not over full. The root page
** is special and is allowed to be nearly empty. If pPage is 
** the root page, then the depth of the tree might be increased
** or decreased by one, as necessary, to keep the root page from being
** overfull or completely empty.
**
** Note that when this routine is called, some of the Cells on pPage
** might not actually be stored in pPage->aData[].  This can happen
** if the page is overfull.  Part of the job of this routine is to
** make sure all Cells for pPage once again fit in pPage->aData[].
**
** In the course of balancing the siblings of pPage, the parent of pPage
** might become overfull or underfull.  If that happens, then this routine
** is called recursively on the parent.
**
** If this routine fails for any reason, it might leave the database
** in a corrupted state.  So if this routine fails, the database should
** be rolled back.
*/
static int balance_nonroot(MemPage *pPage){
  MemPage *pParent;            /* The parent of pPage */
  BtShared *pBt;               /* The whole database */
  int nCell = 0;               /* Number of cells in apCell[] */
  int nMaxCells = 0;           /* Allocated size of apCell, szCell, aFrom. */
  int nOld;                    /* Number of pages in apOld[] */
  int nNew;                    /* Number of pages in apNew[] */
  int nDiv;                    /* Number of cells in apDiv[] */
  int i, j, k;                 /* Loop counters */
  int idx;                     /* Index of pPage in pParent->aCell[] */
  int nxDiv;                   /* Next divider slot in pParent->aCell[] */
  int rc;                      /* The return code */
  int leafCorrection;          /* 4 if pPage is a leaf.  0 if not */
  int leafData;                /* True if pPage is a leaf of a LEAFDATA tree */
  int usableSpace;             /* Bytes in pPage beyond the header */
  int pageFlags;               /* Value of pPage->aData[0] */
  int subtotal;                /* Subtotal of bytes in cells on one page */
  int iSpace = 0;              /* First unused byte of aSpace[] */
  MemPage *apOld[NB];          /* pPage and up to two siblings */
  Pgno pgnoOld[NB];            /* Page numbers for each page in apOld[] */
  MemPage *apCopy[NB];         /* Private copies of apOld[] pages */
  MemPage *apNew[NB+2];        /* pPage and up to NB siblings after balancing */
  Pgno pgnoNew[NB+2];          /* Page numbers for each page in apNew[] */
  u8 *apDiv[NB];               /* Divider cells in pParent */
  int cntNew[NB+2];            /* Index in aCell[] of cell after i-th page */
  int szNew[NB+2];             /* Combined size of cells place on i-th page */
  u8 **apCell = 0;             /* All cells begin balanced */
  int *szCell;                 /* Local size of all cells in apCell[] */
  u8 *aCopy[NB];               /* Space for holding data of apCopy[] */
  u8 *aSpace;                  /* Space to hold copies of dividers cells */
#ifndef SQLITE_OMIT_AUTOVACUUM
  u8 *aFrom = 0;
#endif

  /* 
  ** Find the parent page.
  */
  assert( pPage->isInit );
  assert( sqlite3PagerIswriteable(pPage->pDbPage) );
  pBt = pPage->pBt;
  pParent = pPage->pParent;
  assert( pParent );
  if( SQLITE_OK!=(rc = sqlite3PagerWrite(pParent->pDbPage)) ){
    return rc;
  }
  TRACE(("BALANCE: begin page %d child of %d\n", pPage->pgno, pParent->pgno));

#ifndef SQLITE_OMIT_QUICKBALANCE
  /*
  ** A special case:  If a new entry has just been inserted into a
  ** table (that is, a btree with integer keys and all data at the leaves)
  ** and the new entry is the right-most entry in the tree (it has the
  ** largest key) then use the special balance_quick() routine for
  ** balancing.  balance_quick() is much faster and results in a tighter
  ** packing of data in the common case.
  */
  if( pPage->leaf &&
      pPage->intKey &&
      pPage->leafData &&
      pPage->nOverflow==1 &&
      pPage->aOvfl[0].idx==pPage->nCell &&
      pPage->pParent->pgno!=1 &&
      get4byte(&pParent->aData[pParent->hdrOffset+8])==pPage->pgno
  ){
    /*
    ** TODO: Check the siblings to the left of pPage. It may be that
    ** they are not full and no new page is required.
    */
    return balance_quick(pPage, pParent);
  }
#endif

  /*
  ** Find the cell in the parent page whose left child points back
  ** to pPage.  The "idx" variable is the index of that cell.  If pPage
  ** is the rightmost child of pParent then set idx to pParent->nCell 
  */
  if( pParent->idxShift ){
    Pgno pgno;
    pgno = pPage->pgno;
    assert( pgno==sqlite3PagerPagenumber(pPage->pDbPage) );
    for(idx=0; idx<pParent->nCell; idx++){
      if( get4byte(findCell(pParent, idx))==pgno ){
        break;
      }
    }
    assert( idx<pParent->nCell
             || get4byte(&pParent->aData[pParent->hdrOffset+8])==pgno );
  }else{
    idx = pPage->idxParent;
  }

  /*
  ** Initialize variables so that it will be safe to jump
  ** directly to balance_cleanup at any moment.
  */
  nOld = nNew = 0;
  sqlite3PagerRef(pParent->pDbPage);

  /*
  ** Find sibling pages to pPage and the cells in pParent that divide
  ** the siblings.  An attempt is made to find NN siblings on either
  ** side of pPage.  More siblings are taken from one side, however, if
  ** pPage there are fewer than NN siblings on the other side.  If pParent
  ** has NB or fewer children then all children of pParent are taken.
  */
  nxDiv = idx - NN;
  if( nxDiv + NB > pParent->nCell ){
    nxDiv = pParent->nCell - NB + 1;
  }
  if( nxDiv<0 ){
    nxDiv = 0;
  }
  nDiv = 0;
  for(i=0, k=nxDiv; i<NB; i++, k++){
    if( k<pParent->nCell ){
      apDiv[i] = findCell(pParent, k);
      nDiv++;
      assert( !pParent->leaf );
      pgnoOld[i] = get4byte(apDiv[i]);
    }else if( k==pParent->nCell ){
      pgnoOld[i] = get4byte(&pParent->aData[pParent->hdrOffset+8]);
    }else{
      break;
    }
    rc = getAndInitPage(pBt, pgnoOld[i], &apOld[i], pParent);
    if( rc ) goto balance_cleanup;
    apOld[i]->idxParent = k;
    apCopy[i] = 0;
    assert( i==nOld );
    nOld++;
    nMaxCells += 1+apOld[i]->nCell+apOld[i]->nOverflow;
  }

  /* Make nMaxCells a multiple of 2 in order to preserve 8-byte
  ** alignment */
  nMaxCells = (nMaxCells + 1)&~1;

  /*
  ** Allocate space for memory structures
  */
  apCell = sqliteMallocRaw( 
       nMaxCells*sizeof(u8*)                           /* apCell */
     + nMaxCells*sizeof(int)                           /* szCell */
     + ROUND8(sizeof(MemPage))*NB                      /* aCopy */
     + pBt->pageSize*(5+NB)                            /* aSpace */
     + (ISAUTOVACUUM ? nMaxCells : 0)                  /* aFrom */
  );
  if( apCell==0 ){
    rc = SQLITE_NOMEM;
    goto balance_cleanup;
  }
  szCell = (int*)&apCell[nMaxCells];
  aCopy[0] = (u8*)&szCell[nMaxCells];
  assert( ((aCopy[0] - (u8*)apCell) & 7)==0 ); /* 8-byte alignment required */
  for(i=1; i<NB; i++){
    aCopy[i] = &aCopy[i-1][pBt->pageSize+ROUND8(sizeof(MemPage))];
    assert( ((aCopy[i] - (u8*)apCell) & 7)==0 ); /* 8-byte alignment required */
  }
  aSpace = &aCopy[NB-1][pBt->pageSize+ROUND8(sizeof(MemPage))];
  assert( ((aSpace - (u8*)apCell) & 7)==0 ); /* 8-byte alignment required */
#ifndef SQLITE_OMIT_AUTOVACUUM
  if( pBt->autoVacuum ){
    aFrom = &aSpace[5*pBt->pageSize];
  }
#endif
  
  /*
  ** Make copies of the content of pPage and its siblings into aOld[].
  ** The rest of this function will use data from the copies rather
  ** that the original pages since the original pages will be in the
  ** process of being overwritten.
  */
  for(i=0; i<nOld; i++){
    MemPage *p = apCopy[i] = (MemPage*)&aCopy[i][pBt->pageSize];
    p->aData = &((u8*)p)[-pBt->pageSize];
    memcpy(p->aData, apOld[i]->aData, pBt->pageSize + sizeof(MemPage));
    /* The memcpy() above changes the value of p->aData so we have to
    ** set it again. */
    p->aData = &((u8*)p)[-pBt->pageSize];
  }

  /*
  ** Load pointers to all cells on sibling pages and the divider cells
  ** into the local apCell[] array.  Make copies of the divider cells
  ** into space obtained form aSpace[] and remove the the divider Cells
  ** from pParent.
  **
  ** If the siblings are on leaf pages, then the child pointers of the
  ** divider cells are stripped from the cells before they are copied
  ** into aSpace[].  In this way, all cells in apCell[] are without
  ** child pointers.  If siblings are not leaves, then all cell in
  ** apCell[] include child pointers.  Either way, all cells in apCell[]
  ** are alike.
  **
  ** leafCorrection:  4 if pPage is a leaf.  0 if pPage is not a leaf.
  **       leafData:  1 if pPage holds key+data and pParent holds only keys.
  */
  nCell = 0;
  leafCorrection = pPage->leaf*4;
  leafData = pPage->leafData && pPage->leaf;
  for(i=0; i<nOld; i++){
    MemPage *pOld = apCopy[i];
    int limit = pOld->nCell+pOld->nOverflow;
    for(j=0; j<limit; j++){
      assert( nCell<nMaxCells );
      apCell[nCell] = findOverflowCell(pOld, j);
      szCell[nCell] = cellSizePtr(pOld, apCell[nCell]);
#ifndef SQLITE_OMIT_AUTOVACUUM
      if( pBt->autoVacuum ){
        int a;
        aFrom[nCell] = i;
        for(a=0; a<pOld->nOverflow; a++){
          if( pOld->aOvfl[a].pCell==apCell[nCell] ){
            aFrom[nCell] = 0xFF;
            break;
          }
        }
      }
#endif
      nCell++;
    }
    if( i<nOld-1 ){
      int sz = cellSizePtr(pParent, apDiv[i]);
      if( leafData ){
        /* With the LEAFDATA flag, pParent cells hold only INTKEYs that
        ** are duplicates of keys on the child pages.  We need to remove
        ** the divider cells from pParent, but the dividers cells are not
        ** added to apCell[] because they are duplicates of child cells.
        */
        dropCell(pParent, nxDiv, sz);
      }else{
        u8 *pTemp;
        assert( nCell<nMaxCells );
        szCell[nCell] = sz;
        pTemp = &aSpace[iSpace];
        iSpace += sz;
        assert( iSpace<=pBt->pageSize*5 );
        memcpy(pTemp, apDiv[i], sz);
        apCell[nCell] = pTemp+leafCorrection;
#ifndef SQLITE_OMIT_AUTOVACUUM
        if( pBt->autoVacuum ){
          aFrom[nCell] = 0xFF;
        }
#endif
        dropCell(pParent, nxDiv, sz);
        szCell[nCell] -= leafCorrection;
        assert( get4byte(pTemp)==pgnoOld[i] );
        if( !pOld->leaf ){
          assert( leafCorrection==0 );
          /* The right pointer of the child page pOld becomes the left
          ** pointer of the divider cell */
          memcpy(apCell[nCell], &pOld->aData[pOld->hdrOffset+8], 4);
        }else{
          assert( leafCorrection==4 );
          if( szCell[nCell]<4 ){
            /* Do not allow any cells smaller than 4 bytes. */
            szCell[nCell] = 4;
          }
        }
        nCell++;
      }
    }
  }

  /*
  ** Figure out the number of pages needed to hold all nCell cells.
  ** Store this number in "k".  Also compute szNew[] which is the total
  ** size of all cells on the i-th page and cntNew[] which is the index
  ** in apCell[] of the cell that divides page i from page i+1.  
  ** cntNew[k] should equal nCell.
  **
  ** Values computed by this block:
  **
  **           k: The total number of sibling pages
  **    szNew[i]: Spaced used on the i-th sibling page.
  **   cntNew[i]: Index in apCell[] and szCell[] for the first cell to
  **              the right of the i-th sibling page.
  ** usableSpace: Number of bytes of space available on each sibling.
  ** 
  */
  usableSpace = pBt->usableSize - 12 + leafCorrection;
  for(subtotal=k=i=0; i<nCell; i++){
    assert( i<nMaxCells );
    subtotal += szCell[i] + 2;
    if( subtotal > usableSpace ){
      szNew[k] = subtotal - szCell[i];
      cntNew[k] = i;
      if( leafData ){ i--; }
      subtotal = 0;
      k++;
    }
  }
  szNew[k] = subtotal;
  cntNew[k] = nCell;
  k++;

  /*
  ** The packing computed by the previous block is biased toward the siblings
  ** on the left side.  The left siblings are always nearly full, while the
  ** right-most sibling might be nearly empty.  This block of code attempts
  ** to adjust the packing of siblings to get a better balance.
  **
  ** This adjustment is more than an optimization.  The packing above might
  ** be so out of balance as to be illegal.  For example, the right-most
  ** sibling might be completely empty.  This adjustment is not optional.
  */
  for(i=k-1; i>0; i--){
    int szRight = szNew[i];  /* Size of sibling on the right */
    int szLeft = szNew[i-1]; /* Size of sibling on the left */
    int r;              /* Index of right-most cell in left sibling */
    int d;              /* Index of first cell to the left of right sibling */

    r = cntNew[i-1] - 1;
    d = r + 1 - leafData;
    assert( d<nMaxCells );
    assert( r<nMaxCells );
    while( szRight==0 || szRight+szCell[d]+2<=szLeft-(szCell[r]+2) ){
      szRight += szCell[d] + 2;
      szLeft -= szCell[r] + 2;
      cntNew[i-1]--;
      r = cntNew[i-1] - 1;
      d = r + 1 - leafData;
    }
    szNew[i] = szRight;
    szNew[i-1] = szLeft;
  }

  /* Either we found one or more cells (cntnew[0])>0) or we are the
  ** a virtual root page.  A virtual root page is when the real root
  ** page is page 1 and we are the only child of that page.
  */
  assert( cntNew[0]>0 || (pParent->pgno==1 && pParent->nCell==0) );

  /*
  ** Allocate k new pages.  Reuse old pages where possible.
  */
  assert( pPage->pgno>1 );
  pageFlags = pPage->aData[0];
  for(i=0; i<k; i++){
    MemPage *pNew;
    if( i<nOld ){
      pNew = apNew[i] = apOld[i];
      pgnoNew[i] = pgnoOld[i];
      apOld[i] = 0;
      rc = sqlite3PagerWrite(pNew->pDbPage);
      nNew++;
      if( rc ) goto balance_cleanup;
    }else{
      assert( i>0 );
      rc = allocateBtreePage(pBt, &pNew, &pgnoNew[i], pgnoNew[i-1], 0);
      if( rc ) goto balance_cleanup;
      apNew[i] = pNew;
      nNew++;
    }
    zeroPage(pNew, pageFlags);
  }

  /* Free any old pages that were not reused as new pages.
  */
  while( i<nOld ){
    rc = freePage(apOld[i]);
    if( rc ) goto balance_cleanup;
    releasePage(apOld[i]);
    apOld[i] = 0;
    i++;
  }

  /*
  ** Put the new pages in accending order.  This helps to
  ** keep entries in the disk file in order so that a scan
  ** of the table is a linear scan through the file.  That
  ** in turn helps the operating system to deliver pages
  ** from the disk more rapidly.
  **
  ** An O(n^2) insertion sort algorithm is used, but since
  ** n is never more than NB (a small constant), that should
  ** not be a problem.
  **
  ** When NB==3, this one optimization makes the database
  ** about 25% faster for large insertions and deletions.
  */
  for(i=0; i<k-1; i++){
    int minV = pgnoNew[i];
    int minI = i;
    for(j=i+1; j<k; j++){
      if( pgnoNew[j]<(unsigned)minV ){
        minI = j;
        minV = pgnoNew[j];
      }
    }
    if( minI>i ){
      int t;
      MemPage *pT;
      t = pgnoNew[i];
      pT = apNew[i];
      pgnoNew[i] = pgnoNew[minI];
      apNew[i] = apNew[minI];
      pgnoNew[minI] = t;
      apNew[minI] = pT;
    }
  }
  TRACE(("BALANCE: old: %d %d %d  new: %d(%d) %d(%d) %d(%d) %d(%d) %d(%d)\n",
    pgnoOld[0], 
    nOld>=2 ? pgnoOld[1] : 0,
    nOld>=3 ? pgnoOld[2] : 0,
    pgnoNew[0], szNew[0],
    nNew>=2 ? pgnoNew[1] : 0, nNew>=2 ? szNew[1] : 0,
    nNew>=3 ? pgnoNew[2] : 0, nNew>=3 ? szNew[2] : 0,
    nNew>=4 ? pgnoNew[3] : 0, nNew>=4 ? szNew[3] : 0,
    nNew>=5 ? pgnoNew[4] : 0, nNew>=5 ? szNew[4] : 0));

  /*
  ** Evenly distribute the data in apCell[] across the new pages.
  ** Insert divider cells into pParent as necessary.
  */
  j = 0;
  for(i=0; i<nNew; i++){
    /* Assemble the new sibling page. */
    MemPage *pNew = apNew[i];
    assert( j<nMaxCells );
    assert( pNew->pgno==pgnoNew[i] );
    assemblePage(pNew, cntNew[i]-j, &apCell[j], &szCell[j]);
    assert( pNew->nCell>0 || (nNew==1 && cntNew[0]==0) );
    assert( pNew->nOverflow==0 );

#ifndef SQLITE_OMIT_AUTOVACUUM
    /* If this is an auto-vacuum database, update the pointer map entries
    ** that point to the siblings that were rearranged. These can be: left
    ** children of cells, the right-child of the page, or overflow pages
    ** pointed to by cells.
    */
    if( pBt->autoVacuum ){
      for(k=j; k<cntNew[i]; k++){
        assert( k<nMaxCells );
        if( aFrom[k]==0xFF || apCopy[aFrom[k]]->pgno!=pNew->pgno ){
          rc = ptrmapPutOvfl(pNew, k-j);
          if( rc!=SQLITE_OK ){
            goto balance_cleanup;
          }
        }
      }
    }
#endif

    j = cntNew[i];

    /* If the sibling page assembled above was not the right-most sibling,
    ** insert a divider cell into the parent page.
    */
    if( i<nNew-1 && j<nCell ){
      u8 *pCell;
      u8 *pTemp;
      int sz;

      assert( j<nMaxCells );
      pCell = apCell[j];
      sz = szCell[j] + leafCorrection;
      if( !pNew->leaf ){
        memcpy(&pNew->aData[8], pCell, 4);
        pTemp = 0;
      }else if( leafData ){
	/* If the tree is a leaf-data tree, and the siblings are leaves, 
        ** then there is no divider cell in apCell[]. Instead, the divider 
        ** cell consists of the integer key for the right-most cell of 
        ** the sibling-page assembled above only.
        */
        CellInfo info;
        j--;
        sqlite3BtreeParseCellPtr(pNew, apCell[j], &info);
        pCell = &aSpace[iSpace];
        fillInCell(pParent, pCell, 0, info.nKey, 0, 0, 0, &sz);
        iSpace += sz;
        assert( iSpace<=pBt->pageSize*5 );
        pTemp = 0;
      }else{
        pCell -= 4;
        pTemp = &aSpace[iSpace];
        iSpace += sz;
        assert( iSpace<=pBt->pageSize*5 );
        /* Obscure case for non-leaf-data trees: If the cell at pCell was
        ** previously stored on a leaf node, and it's reported size was 4
        ** bytes, then it may actually be smaller than this 
        ** (see sqlite3BtreeParseCellPtr(), 4 bytes is the minimum size of
        ** any cell). But it's important to pass the correct size to 
        ** insertCell(), so reparse the cell now.
        **
        ** Note that this can never happen in an SQLite data file, as all
        ** cells are at least 4 bytes. It only happens in b-trees used
        ** to evaluate "IN (SELECT ...)" and similar clauses.
        */
        if( szCell[j]==4 ){
          assert(leafCorrection==4);
          sz = cellSizePtr(pParent, pCell);
        }
      }
      rc = insertCell(pParent, nxDiv, pCell, sz, pTemp, 4);
      if( rc!=SQLITE_OK ) goto balance_cleanup;
      put4byte(findOverflowCell(pParent,nxDiv), pNew->pgno);
#ifndef SQLITE_OMIT_AUTOVACUUM
      /* If this is an auto-vacuum database, and not a leaf-data tree,
      ** then update the pointer map with an entry for the overflow page
      ** that the cell just inserted points to (if any).
      */
      if( pBt->autoVacuum && !leafData ){
        rc = ptrmapPutOvfl(pParent, nxDiv);
        if( rc!=SQLITE_OK ){
          goto balance_cleanup;
        }
      }
#endif
      j++;
      nxDiv++;
    }
  }
  assert( j==nCell );
  assert( nOld>0 );
  assert( nNew>0 );
  if( (pageFlags & PTF_LEAF)==0 ){
    memcpy(&apNew[nNew-1]->aData[8], &apCopy[nOld-1]->aData[8], 4);
  }
  if( nxDiv==pParent->nCell+pParent->nOverflow ){
    /* Right-most sibling is the right-most child of pParent */
    put4byte(&pParent->aData[pParent->hdrOffset+8], pgnoNew[nNew-1]);
  }else{
    /* Right-most sibling is the left child of the first entry in pParent
    ** past the right-most divider entry */
    put4byte(findOverflowCell(pParent, nxDiv), pgnoNew[nNew-1]);
  }

  /*
  ** Reparent children of all cells.
  */
  for(i=0; i<nNew; i++){
    rc = reparentChildPages(apNew[i]);
    if( rc!=SQLITE_OK ) goto balance_cleanup;
  }
  rc = reparentChildPages(pParent);
  if( rc!=SQLITE_OK ) goto balance_cleanup;

  /*
  ** Balance the parent page.  Note that the current page (pPage) might
  ** have been added to the freelist so it might no longer be initialized.
  ** But the parent page will always be initialized.
  */
  assert( pParent->isInit );
  rc = balance(pParent, 0);
  
  /*
  ** Cleanup before returning.
  */
balance_cleanup:
  sqliteFree(apCell);
  for(i=0; i<nOld; i++){
    releasePage(apOld[i]);
  }
  for(i=0; i<nNew; i++){
    releasePage(apNew[i]);
  }
  releasePage(pParent);
  TRACE(("BALANCE: finished with %d: old=%d new=%d cells=%d\n",
          pPage->pgno, nOld, nNew, nCell));
  return rc;
}

/*
** This routine is called for the root page of a btree when the root
** page contains no cells.  This is an opportunity to make the tree
** shallower by one level.
*/
static int balance_shallower(MemPage *pPage){
  MemPage *pChild;             /* The only child page of pPage */
  Pgno pgnoChild;              /* Page number for pChild */
  int rc = SQLITE_OK;          /* Return code from subprocedures */
  BtShared *pBt;                  /* The main BTree structure */
  int mxCellPerPage;           /* Maximum number of cells per page */
  u8 **apCell;                 /* All cells from pages being balanced */
  int *szCell;                 /* Local size of all cells */

  assert( pPage->pParent==0 );
  assert( pPage->nCell==0 );
  pBt = pPage->pBt;
  mxCellPerPage = MX_CELL(pBt);
  apCell = sqliteMallocRaw( mxCellPerPage*(sizeof(u8*)+sizeof(int)) );
  if( apCell==0 ) return SQLITE_NOMEM;
  szCell = (int*)&apCell[mxCellPerPage];
  if( pPage->leaf ){
    /* The table is completely empty */
    TRACE(("BALANCE: empty table %d\n", pPage->pgno));
  }else{
    /* The root page is empty but has one child.  Transfer the
    ** information from that one child into the root page if it 
    ** will fit.  This reduces the depth of the tree by one.
    **
    ** If the root page is page 1, it has less space available than
    ** its child (due to the 100 byte header that occurs at the beginning
    ** of the database fle), so it might not be able to hold all of the 
    ** information currently contained in the child.  If this is the 
    ** case, then do not do the transfer.  Leave page 1 empty except
    ** for the right-pointer to the child page.  The child page becomes
    ** the virtual root of the tree.
    */
    pgnoChild = get4byte(&pPage->aData[pPage->hdrOffset+8]);
    assert( pgnoChild>0 );
    assert( pgnoChild<=sqlite3PagerPagecount(pPage->pBt->pPager) );
    rc = sqlite3BtreeGetPage(pPage->pBt, pgnoChild, &pChild, 0);
    if( rc ) goto end_shallow_balance;
    if( pPage->pgno==1 ){
      rc = sqlite3BtreeInitPage(pChild, pPage);
      if( rc ) goto end_shallow_balance;
      assert( pChild->nOverflow==0 );
      if( pChild->nFree>=100 ){
        /* The child information will fit on the root page, so do the
        ** copy */
        int i;
        zeroPage(pPage, pChild->aData[0]);
        for(i=0; i<pChild->nCell; i++){
          apCell[i] = findCell(pChild,i);
          szCell[i] = cellSizePtr(pChild, apCell[i]);
        }
        assemblePage(pPage, pChild->nCell, apCell, szCell);
        /* Copy the right-pointer of the child to the parent. */
        put4byte(&pPage->aData[pPage->hdrOffset+8], 
            get4byte(&pChild->aData[pChild->hdrOffset+8]));
        freePage(pChild);
        TRACE(("BALANCE: child %d transfer to page 1\n", pChild->pgno));
      }else{
        /* The child has more information that will fit on the root.
        ** The tree is already balanced.  Do nothing. */
        TRACE(("BALANCE: child %d will not fit on page 1\n", pChild->pgno));
      }
    }else{
      memcpy(pPage->aData, pChild->aData, pPage->pBt->usableSize);
      pPage->isInit = 0;
      pPage->pParent = 0;
      rc = sqlite3BtreeInitPage(pPage, 0);
      assert( rc==SQLITE_OK );
      freePage(pChild);
      TRACE(("BALANCE: transfer child %d into root %d\n",
              pChild->pgno, pPage->pgno));
    }
    rc = reparentChildPages(pPage);
    assert( pPage->nOverflow==0 );
#ifndef SQLITE_OMIT_AUTOVACUUM
    if( pBt->autoVacuum ){
      int i;
      for(i=0; i<pPage->nCell; i++){ 
        rc = ptrmapPutOvfl(pPage, i);
        if( rc!=SQLITE_OK ){
          goto end_shallow_balance;
        }
      }
    }
#endif
    if( rc!=SQLITE_OK ) goto end_shallow_balance;
    releasePage(pChild);
  }
end_shallow_balance:
  sqliteFree(apCell);
  return rc;
}


/*
** The root page is overfull
**
** When this happens, Create a new child page and copy the
** contents of the root into the child.  Then make the root
** page an empty page with rightChild pointing to the new
** child.   Finally, call balance_internal() on the new child
** to cause it to split.
*/
static int balance_deeper(MemPage *pPage){
  int rc;             /* Return value from subprocedures */
  MemPage *pChild;    /* Pointer to a new child page */
  Pgno pgnoChild;     /* Page number of the new child page */
  BtShared *pBt;         /* The BTree */
  int usableSize;     /* Total usable size of a page */
  u8 *data;           /* Content of the parent page */
  u8 *cdata;          /* Content of the child page */
  int hdr;            /* Offset to page header in parent */
  int brk;            /* Offset to content of first cell in parent */

  assert( pPage->pParent==0 );
  assert( pPage->nOverflow>0 );
  pBt = pPage->pBt;
  rc = allocateBtreePage(pBt, &pChild, &pgnoChild, pPage->pgno, 0);
  if( rc ) return rc;
  assert( sqlite3PagerIswriteable(pChild->pDbPage) );
  usableSize = pBt->usableSize;
  data = pPage->aData;
  hdr = pPage->hdrOffset;
  brk = get2byte(&data[hdr+5]);
  cdata = pChild->aData;
  memcpy(cdata, &data[hdr], pPage->cellOffset+2*pPage->nCell-hdr);
  memcpy(&cdata[brk], &data[brk], usableSize-brk);
  assert( pChild->isInit==0 );
  rc = sqlite3BtreeInitPage(pChild, pPage);
  if( rc ) goto balancedeeper_out;
  memcpy(pChild->aOvfl, pPage->aOvfl, pPage->nOverflow*sizeof(pPage->aOvfl[0]));
  pChild->nOverflow = pPage->nOverflow;
  if( pChild->nOverflow ){
    pChild->nFree = 0;
  }
  assert( pChild->nCell==pPage->nCell );
  zeroPage(pPage, pChild->aData[0] & ~PTF_LEAF);
  put4byte(&pPage->aData[pPage->hdrOffset+8], pgnoChild);
  TRACE(("BALANCE: copy root %d into %d\n", pPage->pgno, pChild->pgno));
#ifndef SQLITE_OMIT_AUTOVACUUM
  if( pBt->autoVacuum ){
    int i;
    rc = ptrmapPut(pBt, pChild->pgno, PTRMAP_BTREE, pPage->pgno);
    if( rc ) goto balancedeeper_out;
    for(i=0; i<pChild->nCell; i++){
      rc = ptrmapPutOvfl(pChild, i);
      if( rc!=SQLITE_OK ){
        return rc;
      }
    }
  }
#endif
  rc = balance_nonroot(pChild);

balancedeeper_out:
  releasePage(pChild);
  return rc;
}

/*
** Decide if the page pPage needs to be balanced.  If balancing is
** required, call the appropriate balancing routine.
*/
static int balance(MemPage *pPage, int insert){
  int rc = SQLITE_OK;
  if( pPage->pParent==0 ){
    if( pPage->nOverflow>0 ){
      rc = balance_deeper(pPage);
    }
    if( rc==SQLITE_OK && pPage->nCell==0 ){
      rc = balance_shallower(pPage);
    }
  }else{
    if( pPage->nOverflow>0 || 
        (!insert && pPage->nFree>pPage->pBt->usableSize*2/3) ){
      rc = balance_nonroot(pPage);
    }
  }
  return rc;
}

/*
** This routine checks all cursors that point to table pgnoRoot.
** If any of those cursors were opened with wrFlag==0 in a different
** database connection (a database connection that shares the pager
** cache with the current connection) and that other connection 
** is not in the ReadUncommmitted state, then this routine returns 
** SQLITE_LOCKED.
**
** In addition to checking for read-locks (where a read-lock 
** means a cursor opened with wrFlag==0) this routine also moves
** all write cursors so that they are pointing to the 
** first Cell on the root page.  This is necessary because an insert 
** or delete might change the number of cells on a page or delete
** a page entirely and we do not want to leave any cursors 
** pointing to non-existant pages or cells.
*/
static int checkReadLocks(Btree *pBtree, Pgno pgnoRoot, BtCursor *pExclude){
  BtCursor *p;
  BtShared *pBt = pBtree->pBt;
  sqlite3 *db = pBtree->pSqlite;
  for(p=pBt->pCursor; p; p=p->pNext){
    if( p==pExclude ) continue;
    if( p->eState!=CURSOR_VALID ) continue;
    if( p->pgnoRoot!=pgnoRoot ) continue;
    if( p->wrFlag==0 ){
      sqlite3 *dbOther = p->pBtree->pSqlite;
      if( dbOther==0 ||
         (dbOther!=db && (dbOther->flags & SQLITE_ReadUncommitted)==0) ){
        return SQLITE_LOCKED;
      }
    }else if( p->pPage->pgno!=p->pgnoRoot ){
      moveToRoot(p);
    }
  }
  return SQLITE_OK;
}

/*
** Insert a new record into the BTree.  The key is given by (pKey,nKey)
** and the data is given by (pData,nData).  The cursor is used only to
** define what table the record should be inserted into.  The cursor
** is left pointing at a random location.
**
** For an INTKEY table, only the nKey value of the key is used.  pKey is
** ignored.  For a ZERODATA table, the pData and nData are both ignored.
*/
SQLITE_PRIVATE int sqlite3BtreeInsert(
  BtCursor *pCur,                /* Insert data into the table of this cursor */
  const void *pKey, i64 nKey,    /* The key of the new record */
  const void *pData, int nData,  /* The data of the new record */
  int nZero,                     /* Number of extra 0 bytes to append to data */
  int appendBias                 /* True if this is likely an append */
){
  int rc;
  int loc;
  int szNew;
  MemPage *pPage;
  BtShared *pBt = pCur->pBtree->pBt;
  unsigned char *oldCell;
  unsigned char *newCell = 0;

  if( pBt->inTransaction!=TRANS_WRITE ){
    /* Must start a transaction before doing an insert */
    return pBt->readOnly ? SQLITE_READONLY : SQLITE_ERROR;
  }
  assert( !pBt->readOnly );
  if( !pCur->wrFlag ){
    return SQLITE_PERM;   /* Cursor not open for writing */
  }
  if( checkReadLocks(pCur->pBtree, pCur->pgnoRoot, pCur) ){
    return SQLITE_LOCKED; /* The table pCur points to has a read lock */
  }

  /* Save the positions of any other cursors open on this table */
  clearCursorPosition(pCur);
  if( 
    SQLITE_OK!=(rc = saveAllCursors(pBt, pCur->pgnoRoot, pCur)) ||
    SQLITE_OK!=(rc = sqlite3BtreeMoveto(pCur, pKey, nKey, appendBias, &loc))
  ){
    return rc;
  }

  pPage = pCur->pPage;
  assert( pPage->intKey || nKey>=0 );
  assert( pPage->leaf || !pPage->leafData );
  TRACE(("INSERT: table=%d nkey=%lld ndata=%d page=%d %s\n",
          pCur->pgnoRoot, nKey, nData, pPage->pgno,
          loc==0 ? "overwrite" : "new entry"));
  assert( pPage->isInit );
  rc = sqlite3PagerWrite(pPage->pDbPage);
  if( rc ) return rc;
  newCell = sqliteMallocRaw( MX_CELL_SIZE(pBt) );
  if( newCell==0 ) return SQLITE_NOMEM;
  rc = fillInCell(pPage, newCell, pKey, nKey, pData, nData, nZero, &szNew);
  if( rc ) goto end_insert;
  assert( szNew==cellSizePtr(pPage, newCell) );
  assert( szNew<=MX_CELL_SIZE(pBt) );
  if( loc==0 && CURSOR_VALID==pCur->eState ){
    int szOld;
    assert( pCur->idx>=0 && pCur->idx<pPage->nCell );
    oldCell = findCell(pPage, pCur->idx);
    if( !pPage->leaf ){
      memcpy(newCell, oldCell, 4);
    }
    szOld = cellSizePtr(pPage, oldCell);
    rc = clearCell(pPage, oldCell);
    if( rc ) goto end_insert;
    dropCell(pPage, pCur->idx, szOld);
  }else if( loc<0 && pPage->nCell>0 ){
    assert( pPage->leaf );
    pCur->idx++;
    pCur->info.nSize = 0;
  }else{
    assert( pPage->leaf );
  }
  rc = insertCell(pPage, pCur->idx, newCell, szNew, 0, 0);
  if( rc!=SQLITE_OK ) goto end_insert;
  rc = balance(pPage, 1);
  /* sqlite3BtreePageDump(pCur->pBt, pCur->pgnoRoot, 1); */
  /* fflush(stdout); */
  if( rc==SQLITE_OK ){
    moveToRoot(pCur);
  }
end_insert:
  sqliteFree(newCell);
  return rc;
}

/*
** Delete the entry that the cursor is pointing to.  The cursor
** is left pointing at a random location.
*/
SQLITE_PRIVATE int sqlite3BtreeDelete(BtCursor *pCur){
  MemPage *pPage = pCur->pPage;
  unsigned char *pCell;
  int rc;
  Pgno pgnoChild = 0;
  BtShared *pBt = pCur->pBtree->pBt;

  assert( pPage->isInit );
  if( pBt->inTransaction!=TRANS_WRITE ){
    /* Must start a transaction before doing a delete */
    return pBt->readOnly ? SQLITE_READONLY : SQLITE_ERROR;
  }
  assert( !pBt->readOnly );
  if( pCur->idx >= pPage->nCell ){
    return SQLITE_ERROR;  /* The cursor is not pointing to anything */
  }
  if( !pCur->wrFlag ){
    return SQLITE_PERM;   /* Did not open this cursor for writing */
  }
  if( checkReadLocks(pCur->pBtree, pCur->pgnoRoot, pCur) ){
    return SQLITE_LOCKED; /* The table pCur points to has a read lock */
  }

  /* Restore the current cursor position (a no-op if the cursor is not in 
  ** CURSOR_REQUIRESEEK state) and save the positions of any other cursors 
  ** open on the same table. Then call sqlite3PagerWrite() on the page
  ** that the entry will be deleted from.
  */
  if( 
    (rc = restoreOrClearCursorPosition(pCur))!=0 ||
    (rc = saveAllCursors(pBt, pCur->pgnoRoot, pCur))!=0 ||
    (rc = sqlite3PagerWrite(pPage->pDbPage))!=0
  ){
    return rc;
  }

  /* Locate the cell within it's page and leave pCell pointing to the
  ** data. The clearCell() call frees any overflow pages associated with the
  ** cell. The cell itself is still intact.
  */
  pCell = findCell(pPage, pCur->idx);
  if( !pPage->leaf ){
    pgnoChild = get4byte(pCell);
  }
  rc = clearCell(pPage, pCell);
  if( rc ) return rc;

  if( !pPage->leaf ){
    /*
    ** The entry we are about to delete is not a leaf so if we do not
    ** do something we will leave a hole on an internal page.
    ** We have to fill the hole by moving in a cell from a leaf.  The
    ** next Cell after the one to be deleted is guaranteed to exist and
    ** to be a leaf so we can use it.
    */
    BtCursor leafCur;
    unsigned char *pNext;
    int szNext;  /* The compiler warning is wrong: szNext is always 
                 ** initialized before use.  Adding an extra initialization
                 ** to silence the compiler slows down the code. */
    int notUsed;
    unsigned char *tempCell = 0;
    assert( !pPage->leafData );
    sqlite3BtreeGetTempCursor(pCur, &leafCur);
    rc = sqlite3BtreeNext(&leafCur, &notUsed);
    if( rc==SQLITE_OK ){
      rc = sqlite3PagerWrite(leafCur.pPage->pDbPage);
    }
    if( rc==SQLITE_OK ){
      TRACE(("DELETE: table=%d delete internal from %d replace from leaf %d\n",
         pCur->pgnoRoot, pPage->pgno, leafCur.pPage->pgno));
      dropCell(pPage, pCur->idx, cellSizePtr(pPage, pCell));
      pNext = findCell(leafCur.pPage, leafCur.idx);
      szNext = cellSizePtr(leafCur.pPage, pNext);
      assert( MX_CELL_SIZE(pBt)>=szNext+4 );
      tempCell = sqliteMallocRaw( MX_CELL_SIZE(pBt) );
      if( tempCell==0 ){
        rc = SQLITE_NOMEM;
      }
    }
    if( rc==SQLITE_OK ){
      rc = insertCell(pPage, pCur->idx, pNext-4, szNext+4, tempCell, 0);
    }
    if( rc==SQLITE_OK ){
      put4byte(findOverflowCell(pPage, pCur->idx), pgnoChild);
      rc = balance(pPage, 0);
    }
    if( rc==SQLITE_OK ){
      dropCell(leafCur.pPage, leafCur.idx, szNext);
      rc = balance(leafCur.pPage, 0);
    }
    sqliteFree(tempCell);
    sqlite3BtreeReleaseTempCursor(&leafCur);
  }else{
    TRACE(("DELETE: table=%d delete from leaf %d\n",
       pCur->pgnoRoot, pPage->pgno));
    dropCell(pPage, pCur->idx, cellSizePtr(pPage, pCell));
    rc = balance(pPage, 0);
  }
  if( rc==SQLITE_OK ){
    moveToRoot(pCur);
  }
  return rc;
}

/*
** Create a new BTree table.  Write into *piTable the page
** number for the root page of the new table.
**
** The type of type is determined by the flags parameter.  Only the
** following values of flags are currently in use.  Other values for
** flags might not work:
**
**     BTREE_INTKEY|BTREE_LEAFDATA     Used for SQL tables with rowid keys
**     BTREE_ZERODATA                  Used for SQL indices
*/
SQLITE_PRIVATE int sqlite3BtreeCreateTable(Btree *p, int *piTable, int flags){
  BtShared *pBt = p->pBt;
  MemPage *pRoot;
  Pgno pgnoRoot;
  int rc;
  if( pBt->inTransaction!=TRANS_WRITE ){
    /* Must start a transaction first */
    return pBt->readOnly ? SQLITE_READONLY : SQLITE_ERROR;
  }
  assert( !pBt->readOnly );

#ifdef SQLITE_OMIT_AUTOVACUUM
  rc = allocateBtreePage(pBt, &pRoot, &pgnoRoot, 1, 0);
  if( rc ) return rc;
#else
  if( pBt->autoVacuum ){
    Pgno pgnoMove;      /* Move a page here to make room for the root-page */
    MemPage *pPageMove; /* The page to move to. */

    /* Creating a new table may probably require moving an existing database
    ** to make room for the new tables root page. In case this page turns
    ** out to be an overflow page, delete all overflow page-map caches
    ** held by open cursors.
    */
    invalidateAllOverflowCache(pBt);

    /* Read the value of meta[3] from the database to determine where the
    ** root page of the new table should go. meta[3] is the largest root-page
    ** created so far, so the new root-page is (meta[3]+1).
    */
    rc = sqlite3BtreeGetMeta(p, 4, &pgnoRoot);
    if( rc!=SQLITE_OK ) return rc;
    pgnoRoot++;

    /* The new root-page may not be allocated on a pointer-map page, or the
    ** PENDING_BYTE page.
    */
    if( pgnoRoot==PTRMAP_PAGENO(pBt, pgnoRoot) ||
        pgnoRoot==PENDING_BYTE_PAGE(pBt) ){
      pgnoRoot++;
    }
    assert( pgnoRoot>=3 );

    /* Allocate a page. The page that currently resides at pgnoRoot will
    ** be moved to the allocated page (unless the allocated page happens
    ** to reside at pgnoRoot).
    */
    rc = allocateBtreePage(pBt, &pPageMove, &pgnoMove, pgnoRoot, 1);
    if( rc!=SQLITE_OK ){
      return rc;
    }

    if( pgnoMove!=pgnoRoot ){
      /* pgnoRoot is the page that will be used for the root-page of
      ** the new table (assuming an error did not occur). But we were
      ** allocated pgnoMove. If required (i.e. if it was not allocated
      ** by extending the file), the current page at position pgnoMove
      ** is already journaled.
      */
      u8 eType;
      Pgno iPtrPage;

      releasePage(pPageMove);

      /* Move the page currently at pgnoRoot to pgnoMove. */
      rc = sqlite3BtreeGetPage(pBt, pgnoRoot, &pRoot, 0);
      if( rc!=SQLITE_OK ){
        return rc;
      }
      rc = ptrmapGet(pBt, pgnoRoot, &eType, &iPtrPage);
      if( rc!=SQLITE_OK || eType==PTRMAP_ROOTPAGE || eType==PTRMAP_FREEPAGE ){
        releasePage(pRoot);
        return rc;
      }
      assert( eType!=PTRMAP_ROOTPAGE );
      assert( eType!=PTRMAP_FREEPAGE );
      rc = sqlite3PagerWrite(pRoot->pDbPage);
      if( rc!=SQLITE_OK ){
        releasePage(pRoot);
        return rc;
      }
      rc = relocatePage(pBt, pRoot, eType, iPtrPage, pgnoMove);
      releasePage(pRoot);

      /* Obtain the page at pgnoRoot */
      if( rc!=SQLITE_OK ){
        return rc;
      }
      rc = sqlite3BtreeGetPage(pBt, pgnoRoot, &pRoot, 0);
      if( rc!=SQLITE_OK ){
        return rc;
      }
      rc = sqlite3PagerWrite(pRoot->pDbPage);
      if( rc!=SQLITE_OK ){
        releasePage(pRoot);
        return rc;
      }
    }else{
      pRoot = pPageMove;
    } 

    /* Update the pointer-map and meta-data with the new root-page number. */
    rc = ptrmapPut(pBt, pgnoRoot, PTRMAP_ROOTPAGE, 0);
    if( rc ){
      releasePage(pRoot);
      return rc;
    }
    rc = sqlite3BtreeUpdateMeta(p, 4, pgnoRoot);
    if( rc ){
      releasePage(pRoot);
      return rc;
    }

  }else{
    rc = allocateBtreePage(pBt, &pRoot, &pgnoRoot, 1, 0);
    if( rc ) return rc;
  }
#endif
  assert( sqlite3PagerIswriteable(pRoot->pDbPage) );
  zeroPage(pRoot, flags | PTF_LEAF);
  sqlite3PagerUnref(pRoot->pDbPage);
  *piTable = (int)pgnoRoot;
  return SQLITE_OK;
}

/*
** Erase the given database page and all its children.  Return
** the page to the freelist.
*/
static int clearDatabasePage(
  BtShared *pBt,           /* The BTree that contains the table */
  Pgno pgno,            /* Page number to clear */
  MemPage *pParent,     /* Parent page.  NULL for the root */
  int freePageFlag      /* Deallocate page if true */
){
  MemPage *pPage = 0;
  int rc;
  unsigned char *pCell;
  int i;

  if( pgno>sqlite3PagerPagecount(pBt->pPager) ){
    return SQLITE_CORRUPT_BKPT;
  }

  rc = getAndInitPage(pBt, pgno, &pPage, pParent);
  if( rc ) goto cleardatabasepage_out;
  for(i=0; i<pPage->nCell; i++){
    pCell = findCell(pPage, i);
    if( !pPage->leaf ){
      rc = clearDatabasePage(pBt, get4byte(pCell), pPage->pParent, 1);
      if( rc ) goto cleardatabasepage_out;
    }
    rc = clearCell(pPage, pCell);
    if( rc ) goto cleardatabasepage_out;
  }
  if( !pPage->leaf ){
    rc = clearDatabasePage(pBt, get4byte(&pPage->aData[8]), pPage->pParent, 1);
    if( rc ) goto cleardatabasepage_out;
  }
  if( freePageFlag ){
    rc = freePage(pPage);
  }else if( (rc = sqlite3PagerWrite(pPage->pDbPage))==0 ){
    zeroPage(pPage, pPage->aData[0] | PTF_LEAF);
  }

cleardatabasepage_out:
  releasePage(pPage);
  return rc;
}

/*
** Delete all information from a single table in the database.  iTable is
** the page number of the root of the table.  After this routine returns,
** the root page is empty, but still exists.
**
** This routine will fail with SQLITE_LOCKED if there are any open
** read cursors on the table.  Open write cursors are moved to the
** root of the table.
*/
SQLITE_PRIVATE int sqlite3BtreeClearTable(Btree *p, int iTable){
  int rc;
  BtShared *pBt = p->pBt;
  if( p->inTrans!=TRANS_WRITE ){
    return pBt->readOnly ? SQLITE_READONLY : SQLITE_ERROR;
  }
  rc = checkReadLocks(p, iTable, 0);
  if( rc ){
    return rc;
  }

  /* Save the position of all cursors open on this table */
  if( SQLITE_OK!=(rc = saveAllCursors(pBt, iTable, 0)) ){
    return rc;
  }

  return clearDatabasePage(pBt, (Pgno)iTable, 0, 0);
}

/*
** Erase all information in a table and add the root of the table to
** the freelist.  Except, the root of the principle table (the one on
** page 1) is never added to the freelist.
**
** This routine will fail with SQLITE_LOCKED if there are any open
** cursors on the table.
**
** If AUTOVACUUM is enabled and the page at iTable is not the last
** root page in the database file, then the last root page 
** in the database file is moved into the slot formerly occupied by
** iTable and that last slot formerly occupied by the last root page
** is added to the freelist instead of iTable.  In this say, all
** root pages are kept at the beginning of the database file, which
** is necessary for AUTOVACUUM to work right.  *piMoved is set to the 
** page number that used to be the last root page in the file before
** the move.  If no page gets moved, *piMoved is set to 0.
** The last root page is recorded in meta[3] and the value of
** meta[3] is updated by this procedure.
*/
SQLITE_PRIVATE int sqlite3BtreeDropTable(Btree *p, int iTable, int *piMoved){
  int rc;
  MemPage *pPage = 0;
  BtShared *pBt = p->pBt;

  if( p->inTrans!=TRANS_WRITE ){
    return pBt->readOnly ? SQLITE_READONLY : SQLITE_ERROR;
  }

  /* It is illegal to drop a table if any cursors are open on the
  ** database. This is because in auto-vacuum mode the backend may
  ** need to move another root-page to fill a gap left by the deleted
  ** root page. If an open cursor was using this page a problem would 
  ** occur.
  */
  if( pBt->pCursor ){
    return SQLITE_LOCKED;
  }

  rc = sqlite3BtreeGetPage(pBt, (Pgno)iTable, &pPage, 0);
  if( rc ) return rc;
  rc = sqlite3BtreeClearTable(p, iTable);
  if( rc ){
    releasePage(pPage);
    return rc;
  }

  *piMoved = 0;

  if( iTable>1 ){
#ifdef SQLITE_OMIT_AUTOVACUUM
    rc = freePage(pPage);
    releasePage(pPage);
#else
    if( pBt->autoVacuum ){
      Pgno maxRootPgno;
      rc = sqlite3BtreeGetMeta(p, 4, &maxRootPgno);
      if( rc!=SQLITE_OK ){
        releasePage(pPage);
        return rc;
      }

      if( iTable==maxRootPgno ){
        /* If the table being dropped is the table with the largest root-page
        ** number in the database, put the root page on the free list. 
        */
        rc = freePage(pPage);
        releasePage(pPage);
        if( rc!=SQLITE_OK ){
          return rc;
        }
      }else{
        /* The table being dropped does not have the largest root-page
        ** number in the database. So move the page that does into the 
        ** gap left by the deleted root-page.
        */
        MemPage *pMove;
        releasePage(pPage);
        rc = sqlite3BtreeGetPage(pBt, maxRootPgno, &pMove, 0);
        if( rc!=SQLITE_OK ){
          return rc;
        }
        rc = relocatePage(pBt, pMove, PTRMAP_ROOTPAGE, 0, iTable);
        releasePage(pMove);
        if( rc!=SQLITE_OK ){
          return rc;
        }
        rc = sqlite3BtreeGetPage(pBt, maxRootPgno, &pMove, 0);
        if( rc!=SQLITE_OK ){
          return rc;
        }
        rc = freePage(pMove);
        releasePage(pMove);
        if( rc!=SQLITE_OK ){
          return rc;
        }
        *piMoved = maxRootPgno;
      }

      /* Set the new 'max-root-page' value in the database header. This
      ** is the old value less one, less one more if that happens to
      ** be a root-page number, less one again if that is the
      ** PENDING_BYTE_PAGE.
      */
      maxRootPgno--;
      if( maxRootPgno==PENDING_BYTE_PAGE(pBt) ){
        maxRootPgno--;
      }
      if( maxRootPgno==PTRMAP_PAGENO(pBt, maxRootPgno) ){
        maxRootPgno--;
      }
      assert( maxRootPgno!=PENDING_BYTE_PAGE(pBt) );

      rc = sqlite3BtreeUpdateMeta(p, 4, maxRootPgno);
    }else{
      rc = freePage(pPage);
      releasePage(pPage);
    }
#endif
  }else{
    /* If sqlite3BtreeDropTable was called on page 1. */
    zeroPage(pPage, PTF_INTKEY|PTF_LEAF );
    releasePage(pPage);
  }
  return rc;  
}


/*
** Read the meta-information out of a database file.  Meta[0]
** is the number of free pages currently in the database.  Meta[1]
** through meta[15] are available for use by higher layers.  Meta[0]
** is read-only, the others are read/write.
** 
** The schema layer numbers meta values differently.  At the schema
** layer (and the SetCookie and ReadCookie opcodes) the number of
** free pages is not visible.  So Cookie[0] is the same as Meta[1].
*/
SQLITE_PRIVATE int sqlite3BtreeGetMeta(Btree *p, int idx, u32 *pMeta){
  DbPage *pDbPage;
  int rc;
  unsigned char *pP1;
  BtShared *pBt = p->pBt;

  /* Reading a meta-data value requires a read-lock on page 1 (and hence
  ** the sqlite_master table. We grab this lock regardless of whether or
  ** not the SQLITE_ReadUncommitted flag is set (the table rooted at page
  ** 1 is treated as a special case by queryTableLock() and lockTable()).
  */
  rc = queryTableLock(p, 1, READ_LOCK);
  if( rc!=SQLITE_OK ){
    return rc;
  }

  assert( idx>=0 && idx<=15 );
  rc = sqlite3PagerGet(pBt->pPager, 1, &pDbPage);
  if( rc ) return rc;
  pP1 = (unsigned char *)sqlite3PagerGetData(pDbPage);
  *pMeta = get4byte(&pP1[36 + idx*4]);
  sqlite3PagerUnref(pDbPage);

  /* If autovacuumed is disabled in this build but we are trying to 
  ** access an autovacuumed database, then make the database readonly. 
  */
#ifdef SQLITE_OMIT_AUTOVACUUM
  if( idx==4 && *pMeta>0 ) pBt->readOnly = 1;
#endif

  /* Grab the read-lock on page 1. */
  rc = lockTable(p, 1, READ_LOCK);
  return rc;
}

/*
** Write meta-information back into the database.  Meta[0] is
** read-only and may not be written.
*/
SQLITE_PRIVATE int sqlite3BtreeUpdateMeta(Btree *p, int idx, u32 iMeta){
  BtShared *pBt = p->pBt;
  unsigned char *pP1;
  int rc;
  assert( idx>=1 && idx<=15 );
  if( p->inTrans!=TRANS_WRITE ){
    return pBt->readOnly ? SQLITE_READONLY : SQLITE_ERROR;
  }
  assert( pBt->pPage1!=0 );
  pP1 = pBt->pPage1->aData;
  rc = sqlite3PagerWrite(pBt->pPage1->pDbPage);
  if( rc ) return rc;
  put4byte(&pP1[36 + idx*4], iMeta);
  if( idx==7 ){
    assert( pBt->autoVacuum || iMeta==0 );
    assert( iMeta==0 || iMeta==1 );
    pBt->incrVacuum = iMeta;
  }
  return SQLITE_OK;
}

/*
** Return the flag byte at the beginning of the page that the cursor
** is currently pointing to.
*/
SQLITE_PRIVATE int sqlite3BtreeFlags(BtCursor *pCur){
  /* TODO: What about CURSOR_REQUIRESEEK state? Probably need to call
  ** restoreOrClearCursorPosition() here.
  */
  MemPage *pPage = pCur->pPage;
  return pPage ? pPage->aData[pPage->hdrOffset] : 0;
}


/*
** Return the pager associated with a BTree.  This routine is used for
** testing and debugging only.
*/
SQLITE_PRIVATE Pager *sqlite3BtreePager(Btree *p){
  return p->pBt->pPager;
}

#ifndef SQLITE_OMIT_INTEGRITY_CHECK
/*
** Append a message to the error message string.
*/
static void checkAppendMsg(
  IntegrityCk *pCheck,
  char *zMsg1,
  const char *zFormat,
  ...
){
  va_list ap;
  char *zMsg2;
  if( !pCheck->mxErr ) return;
  pCheck->mxErr--;
  pCheck->nErr++;
  va_start(ap, zFormat);
  zMsg2 = sqlite3VMPrintf(zFormat, ap);
  va_end(ap);
  if( zMsg1==0 ) zMsg1 = "";
  if( pCheck->zErrMsg ){
    char *zOld = pCheck->zErrMsg;
    pCheck->zErrMsg = 0;
    sqlite3SetString(&pCheck->zErrMsg, zOld, "\n", zMsg1, zMsg2, (char*)0);
    sqliteFree(zOld);
  }else{
    sqlite3SetString(&pCheck->zErrMsg, zMsg1, zMsg2, (char*)0);
  }
  sqliteFree(zMsg2);
}
#endif /* SQLITE_OMIT_INTEGRITY_CHECK */

#ifndef SQLITE_OMIT_INTEGRITY_CHECK
/*
** Add 1 to the reference count for page iPage.  If this is the second
** reference to the page, add an error message to pCheck->zErrMsg.
** Return 1 if there are 2 ore more references to the page and 0 if
** if this is the first reference to the page.
**
** Also check that the page number is in bounds.
*/
static int checkRef(IntegrityCk *pCheck, int iPage, char *zContext){
  if( iPage==0 ) return 1;
  if( iPage>pCheck->nPage || iPage<0 ){
    checkAppendMsg(pCheck, zContext, "invalid page number %d", iPage);
    return 1;
  }
  if( pCheck->anRef[iPage]==1 ){
    checkAppendMsg(pCheck, zContext, "2nd reference to page %d", iPage);
    return 1;
  }
  return  (pCheck->anRef[iPage]++)>1;
}

#ifndef SQLITE_OMIT_AUTOVACUUM
/*
** Check that the entry in the pointer-map for page iChild maps to 
** page iParent, pointer type ptrType. If not, append an error message
** to pCheck.
*/
static void checkPtrmap(
  IntegrityCk *pCheck,   /* Integrity check context */
  Pgno iChild,           /* Child page number */
  u8 eType,              /* Expected pointer map type */
  Pgno iParent,          /* Expected pointer map parent page number */
  char *zContext         /* Context description (used for error msg) */
){
  int rc;
  u8 ePtrmapType;
  Pgno iPtrmapParent;

  rc = ptrmapGet(pCheck->pBt, iChild, &ePtrmapType, &iPtrmapParent);
  if( rc!=SQLITE_OK ){
    checkAppendMsg(pCheck, zContext, "Failed to read ptrmap key=%d", iChild);
    return;
  }

  if( ePtrmapType!=eType || iPtrmapParent!=iParent ){
    checkAppendMsg(pCheck, zContext, 
      "Bad ptr map entry key=%d expected=(%d,%d) got=(%d,%d)", 
      iChild, eType, iParent, ePtrmapType, iPtrmapParent);
  }
}
#endif

/*
** Check the integrity of the freelist or of an overflow page list.
** Verify that the number of pages on the list is N.
*/
static void checkList(
  IntegrityCk *pCheck,  /* Integrity checking context */
  int isFreeList,       /* True for a freelist.  False for overflow page list */
  int iPage,            /* Page number for first page in the list */
  int N,                /* Expected number of pages in the list */
  char *zContext        /* Context for error messages */
){
  int i;
  int expected = N;
  int iFirst = iPage;
  while( N-- > 0 && pCheck->mxErr ){
    DbPage *pOvflPage;
    unsigned char *pOvflData;
    if( iPage<1 ){
      checkAppendMsg(pCheck, zContext,
         "%d of %d pages missing from overflow list starting at %d",
          N+1, expected, iFirst);
      break;
    }
    if( checkRef(pCheck, iPage, zContext) ) break;
    if( sqlite3PagerGet(pCheck->pPager, (Pgno)iPage, &pOvflPage) ){
      checkAppendMsg(pCheck, zContext, "failed to get page %d", iPage);
      break;
    }
    pOvflData = (unsigned char *)sqlite3PagerGetData(pOvflPage);
    if( isFreeList ){
      int n = get4byte(&pOvflData[4]);
#ifndef SQLITE_OMIT_AUTOVACUUM
      if( pCheck->pBt->autoVacuum ){
        checkPtrmap(pCheck, iPage, PTRMAP_FREEPAGE, 0, zContext);
      }
#endif
      if( n>pCheck->pBt->usableSize/4-8 ){
        checkAppendMsg(pCheck, zContext,
           "freelist leaf count too big on page %d", iPage);
        N--;
      }else{
        for(i=0; i<n; i++){
          Pgno iFreePage = get4byte(&pOvflData[8+i*4]);
#ifndef SQLITE_OMIT_AUTOVACUUM
          if( pCheck->pBt->autoVacuum ){
            checkPtrmap(pCheck, iFreePage, PTRMAP_FREEPAGE, 0, zContext);
          }
#endif
          checkRef(pCheck, iFreePage, zContext);
        }
        N -= n;
      }
    }
#ifndef SQLITE_OMIT_AUTOVACUUM
    else{
      /* If this database supports auto-vacuum and iPage is not the last
      ** page in this overflow list, check that the pointer-map entry for
      ** the following page matches iPage.
      */
      if( pCheck->pBt->autoVacuum && N>0 ){
        i = get4byte(pOvflData);
        checkPtrmap(pCheck, i, PTRMAP_OVERFLOW2, iPage, zContext);
      }
    }
#endif
    iPage = get4byte(pOvflData);
    sqlite3PagerUnref(pOvflPage);
  }
}
#endif /* SQLITE_OMIT_INTEGRITY_CHECK */

#ifndef SQLITE_OMIT_INTEGRITY_CHECK
/*
** Do various sanity checks on a single page of a tree.  Return
** the tree depth.  Root pages return 0.  Parents of root pages
** return 1, and so forth.
** 
** These checks are done:
**
**      1.  Make sure that cells and freeblocks do not overlap
**          but combine to completely cover the page.
**  NO  2.  Make sure cell keys are in order.
**  NO  3.  Make sure no key is less than or equal to zLowerBound.
**  NO  4.  Make sure no key is greater than or equal to zUpperBound.
**      5.  Check the integrity of overflow pages.
**      6.  Recursively call checkTreePage on all children.
**      7.  Verify that the depth of all children is the same.
**      8.  Make sure this page is at least 33% full or else it is
**          the root of the tree.
*/
static int checkTreePage(
  IntegrityCk *pCheck,  /* Context for the sanity check */
  int iPage,            /* Page number of the page to check */
  MemPage *pParent,     /* Parent page */
  char *zParentContext  /* Parent context */
){
  MemPage *pPage;
  int i, rc, depth, d2, pgno, cnt;
  int hdr, cellStart;
  int nCell;
  u8 *data;
  BtShared *pBt;
  int usableSize;
  char zContext[100];
  char *hit;

  sqlite3_snprintf(sizeof(zContext), zContext, "Page %d: ", iPage);

  /* Check that the page exists
  */
  pBt = pCheck->pBt;
  usableSize = pBt->usableSize;
  if( iPage==0 ) return 0;
  if( checkRef(pCheck, iPage, zParentContext) ) return 0;
  if( (rc = sqlite3BtreeGetPage(pBt, (Pgno)iPage, &pPage, 0))!=0 ){
    checkAppendMsg(pCheck, zContext,
       "unable to get the page. error code=%d", rc);
    return 0;
  }
  if( (rc = sqlite3BtreeInitPage(pPage, pParent))!=0 ){
    checkAppendMsg(pCheck, zContext, 
                   "sqlite3BtreeInitPage() returns error code %d", rc);
    releasePage(pPage);
    return 0;
  }

  /* Check out all the cells.
  */
  depth = 0;
  for(i=0; i<pPage->nCell && pCheck->mxErr; i++){
    u8 *pCell;
    int sz;
    CellInfo info;

    /* Check payload overflow pages
    */
    sqlite3_snprintf(sizeof(zContext), zContext,
             "On tree page %d cell %d: ", iPage, i);
    pCell = findCell(pPage,i);
    sqlite3BtreeParseCellPtr(pPage, pCell, &info);
    sz = info.nData;
    if( !pPage->intKey ) sz += info.nKey;
    assert( sz==info.nPayload );
    if( sz>info.nLocal ){
      int nPage = (sz - info.nLocal + usableSize - 5)/(usableSize - 4);
      Pgno pgnoOvfl = get4byte(&pCell[info.iOverflow]);
#ifndef SQLITE_OMIT_AUTOVACUUM
      if( pBt->autoVacuum ){
        checkPtrmap(pCheck, pgnoOvfl, PTRMAP_OVERFLOW1, iPage, zContext);
      }
#endif
      checkList(pCheck, 0, pgnoOvfl, nPage, zContext);
    }

    /* Check sanity of left child page.
    */
    if( !pPage->leaf ){
      pgno = get4byte(pCell);
#ifndef SQLITE_OMIT_AUTOVACUUM
      if( pBt->autoVacuum ){
        checkPtrmap(pCheck, pgno, PTRMAP_BTREE, iPage, zContext);
      }
#endif
      d2 = checkTreePage(pCheck,pgno,pPage,zContext);
      if( i>0 && d2!=depth ){
        checkAppendMsg(pCheck, zContext, "Child page depth differs");
      }
      depth = d2;
    }
  }
  if( !pPage->leaf ){
    pgno = get4byte(&pPage->aData[pPage->hdrOffset+8]);
    sqlite3_snprintf(sizeof(zContext), zContext, 
                     "On page %d at right child: ", iPage);
#ifndef SQLITE_OMIT_AUTOVACUUM
    if( pBt->autoVacuum ){
      checkPtrmap(pCheck, pgno, PTRMAP_BTREE, iPage, 0);
    }
#endif
    checkTreePage(pCheck, pgno, pPage, zContext);
  }
 
  /* Check for complete coverage of the page
  */
  data = pPage->aData;
  hdr = pPage->hdrOffset;
  hit = sqliteMalloc( usableSize );
  if( hit ){
    memset(hit, 1, get2byte(&data[hdr+5]));
    nCell = get2byte(&data[hdr+3]);
    cellStart = hdr + 12 - 4*pPage->leaf;
    for(i=0; i<nCell; i++){
      int pc = get2byte(&data[cellStart+i*2]);
      int size = cellSizePtr(pPage, &data[pc]);
      int j;
      if( (pc+size-1)>=usableSize || pc<0 ){
        checkAppendMsg(pCheck, 0, 
            "Corruption detected in cell %d on page %d",i,iPage,0);
      }else{
        for(j=pc+size-1; j>=pc; j--) hit[j]++;
      }
    }
    for(cnt=0, i=get2byte(&data[hdr+1]); i>0 && i<usableSize && cnt<10000; 
           cnt++){
      int size = get2byte(&data[i+2]);
      int j;
      if( (i+size-1)>=usableSize || i<0 ){
        checkAppendMsg(pCheck, 0,  
            "Corruption detected in cell %d on page %d",i,iPage,0);
      }else{
        for(j=i+size-1; j>=i; j--) hit[j]++;
      }
      i = get2byte(&data[i]);
    }
    for(i=cnt=0; i<usableSize; i++){
      if( hit[i]==0 ){
        cnt++;
      }else if( hit[i]>1 ){
        checkAppendMsg(pCheck, 0,
          "Multiple uses for byte %d of page %d", i, iPage);
        break;
      }
    }
    if( cnt!=data[hdr+7] ){
      checkAppendMsg(pCheck, 0, 
          "Fragmented space is %d byte reported as %d on page %d",
          cnt, data[hdr+7], iPage);
    }
  }
  sqliteFree(hit);

  releasePage(pPage);
  return depth+1;
}
#endif /* SQLITE_OMIT_INTEGRITY_CHECK */

#ifndef SQLITE_OMIT_INTEGRITY_CHECK
/*
** This routine does a complete check of the given BTree file.  aRoot[] is
** an array of pages numbers were each page number is the root page of
** a table.  nRoot is the number of entries in aRoot.
**
** If everything checks out, this routine returns NULL.  If something is
** amiss, an error message is written into memory obtained from malloc()
** and a pointer to that error message is returned.  The calling function
** is responsible for freeing the error message when it is done.
*/
SQLITE_PRIVATE char *sqlite3BtreeIntegrityCheck(
  Btree *p,     /* The btree to be checked */
  int *aRoot,   /* An array of root pages numbers for individual trees */
  int nRoot,    /* Number of entries in aRoot[] */
  int mxErr,    /* Stop reporting errors after this many */
  int *pnErr    /* Write number of errors seen to this variable */
){
  int i;
  int nRef;
  IntegrityCk sCheck;
  BtShared *pBt = p->pBt;

  nRef = sqlite3PagerRefcount(pBt->pPager);
  if( lockBtreeWithRetry(p)!=SQLITE_OK ){
    return sqliteStrDup("Unable to acquire a read lock on the database");
  }
  sCheck.pBt = pBt;
  sCheck.pPager = pBt->pPager;
  sCheck.nPage = sqlite3PagerPagecount(sCheck.pPager);
  sCheck.mxErr = mxErr;
  sCheck.nErr = 0;
  *pnErr = 0;
#ifndef SQLITE_OMIT_AUTOVACUUM
  if( pBt->nTrunc!=0 ){
    sCheck.nPage = pBt->nTrunc;
  }
#endif
  if( sCheck.nPage==0 ){
    unlockBtreeIfUnused(pBt);
    return 0;
  }
  sCheck.anRef = sqliteMallocRaw( (sCheck.nPage+1)*sizeof(sCheck.anRef[0]) );
  if( !sCheck.anRef ){
    unlockBtreeIfUnused(pBt);
    *pnErr = 1;
    return sqlite3MPrintf("Unable to malloc %d bytes", 
        (sCheck.nPage+1)*sizeof(sCheck.anRef[0]));
  }
  for(i=0; i<=sCheck.nPage; i++){ sCheck.anRef[i] = 0; }
  i = PENDING_BYTE_PAGE(pBt);
  if( i<=sCheck.nPage ){
    sCheck.anRef[i] = 1;
  }
  sCheck.zErrMsg = 0;

  /* Check the integrity of the freelist
  */
  checkList(&sCheck, 1, get4byte(&pBt->pPage1->aData[32]),
            get4byte(&pBt->pPage1->aData[36]), "Main freelist: ");

  /* Check all the tables.
  */
  for(i=0; i<nRoot && sCheck.mxErr; i++){
    if( aRoot[i]==0 ) continue;
#ifndef SQLITE_OMIT_AUTOVACUUM
    if( pBt->autoVacuum && aRoot[i]>1 ){
      checkPtrmap(&sCheck, aRoot[i], PTRMAP_ROOTPAGE, 0, 0);
    }
#endif
    checkTreePage(&sCheck, aRoot[i], 0, "List of tree roots: ");
  }

  /* Make sure every page in the file is referenced
  */
  for(i=1; i<=sCheck.nPage && sCheck.mxErr; i++){
#ifdef SQLITE_OMIT_AUTOVACUUM
    if( sCheck.anRef[i]==0 ){
      checkAppendMsg(&sCheck, 0, "Page %d is never used", i);
    }
#else
    /* If the database supports auto-vacuum, make sure no tables contain
    ** references to pointer-map pages.
    */
    if( sCheck.anRef[i]==0 && 
       (PTRMAP_PAGENO(pBt, i)!=i || !pBt->autoVacuum) ){
      checkAppendMsg(&sCheck, 0, "Page %d is never used", i);
    }
    if( sCheck.anRef[i]!=0 && 
       (PTRMAP_PAGENO(pBt, i)==i && pBt->autoVacuum) ){
      checkAppendMsg(&sCheck, 0, "Pointer map page %d is referenced", i);
    }
#endif
  }

  /* Make sure this analysis did not leave any unref() pages
  */
  unlockBtreeIfUnused(pBt);
  if( nRef != sqlite3PagerRefcount(pBt->pPager) ){
    checkAppendMsg(&sCheck, 0, 
      "Outstanding page count goes from %d to %d during this analysis",
      nRef, sqlite3PagerRefcount(pBt->pPager)
    );
  }

  /* Clean  up and report errors.
  */
  sqliteFree(sCheck.anRef);
  *pnErr = sCheck.nErr;
  return sCheck.zErrMsg;
}
#endif /* SQLITE_OMIT_INTEGRITY_CHECK */

/*
** Return the full pathname of the underlying database file.
*/
SQLITE_PRIVATE const char *sqlite3BtreeGetFilename(Btree *p){
  assert( p->pBt->pPager!=0 );
  return sqlite3PagerFilename(p->pBt->pPager);
}

/*
** Return the pathname of the directory that contains the database file.
*/
SQLITE_PRIVATE const char *sqlite3BtreeGetDirname(Btree *p){
  assert( p->pBt->pPager!=0 );
  return sqlite3PagerDirname(p->pBt->pPager);
}

/*
** Return the pathname of the journal file for this database. The return
** value of this routine is the same regardless of whether the journal file
** has been created or not.
*/
SQLITE_PRIVATE const char *sqlite3BtreeGetJournalname(Btree *p){
  assert( p->pBt->pPager!=0 );
  return sqlite3PagerJournalname(p->pBt->pPager);
}

#ifndef SQLITE_OMIT_VACUUM
/*
** Copy the complete content of pBtFrom into pBtTo.  A transaction
** must be active for both files.
**
** The size of file pBtFrom may be reduced by this operation.
** If anything goes wrong, the transaction on pBtFrom is rolled back.
*/
SQLITE_PRIVATE int sqlite3BtreeCopyFile(Btree *pTo, Btree *pFrom){
  int rc = SQLITE_OK;
  Pgno i, nPage, nToPage, iSkip;

  BtShared *pBtTo = pTo->pBt;
  BtShared *pBtFrom = pFrom->pBt;

  if( pTo->inTrans!=TRANS_WRITE || pFrom->inTrans!=TRANS_WRITE ){
    return SQLITE_ERROR;
  }
  if( pBtTo->pCursor ) return SQLITE_BUSY;
  nToPage = sqlite3PagerPagecount(pBtTo->pPager);
  nPage = sqlite3PagerPagecount(pBtFrom->pPager);
  iSkip = PENDING_BYTE_PAGE(pBtTo);
  for(i=1; rc==SQLITE_OK && i<=nPage; i++){
    DbPage *pDbPage;
    if( i==iSkip ) continue;
    rc = sqlite3PagerGet(pBtFrom->pPager, i, &pDbPage);
    if( rc ) break;
    rc = sqlite3PagerOverwrite(pBtTo->pPager, i, sqlite3PagerGetData(pDbPage));
    sqlite3PagerUnref(pDbPage);
  }

  /* If the file is shrinking, journal the pages that are being truncated
  ** so that they can be rolled back if the commit fails.
  */
  for(i=nPage+1; rc==SQLITE_OK && i<=nToPage; i++){
    DbPage *pDbPage;
    if( i==iSkip ) continue;
    rc = sqlite3PagerGet(pBtTo->pPager, i, &pDbPage);
    if( rc ) break;
    rc = sqlite3PagerWrite(pDbPage);
    sqlite3PagerDontWrite(pDbPage);
    /* Yeah.  It seems wierd to call DontWrite() right after Write().  But
    ** that is because the names of those procedures do not exactly 
    ** represent what they do.  Write() really means "put this page in the
    ** rollback journal and mark it as dirty so that it will be written
    ** to the database file later."  DontWrite() undoes the second part of
    ** that and prevents the page from being written to the database.  The
    ** page is still on the rollback journal, though.  And that is the whole
    ** point of this loop: to put pages on the rollback journal. */
    sqlite3PagerUnref(pDbPage);
  }
  if( !rc && nPage<nToPage ){
    rc = sqlite3PagerTruncate(pBtTo->pPager, nPage);
  }

  if( rc ){
    sqlite3BtreeRollback(pTo);
  }
  return rc;  
}
#endif /* SQLITE_OMIT_VACUUM */

/*
** Return non-zero if a transaction is active.
*/
SQLITE_PRIVATE int sqlite3BtreeIsInTrans(Btree *p){
  return (p && (p->inTrans==TRANS_WRITE));
}

/*
** Return non-zero if a statement transaction is active.
*/
SQLITE_PRIVATE int sqlite3BtreeIsInStmt(Btree *p){
  return (p->pBt && p->pBt->inStmt);
}

/*
** Return non-zero if a read (or write) transaction is active.
*/
SQLITE_PRIVATE int sqlite3BtreeIsInReadTrans(Btree *p){
  return (p && (p->inTrans!=TRANS_NONE));
}

/*
** This function returns a pointer to a blob of memory associated with
** a single shared-btree. The memory is used by client code for it's own
** purposes (for example, to store a high-level schema associated with 
** the shared-btree). The btree layer manages reference counting issues.
**
** The first time this is called on a shared-btree, nBytes bytes of memory
** are allocated, zeroed, and returned to the caller. For each subsequent 
** call the nBytes parameter is ignored and a pointer to the same blob
** of memory returned. 
**
** Just before the shared-btree is closed, the function passed as the 
** xFree argument when the memory allocation was made is invoked on the 
** blob of allocated memory. This function should not call sqliteFree()
** on the memory, the btree layer does that.
*/
SQLITE_PRIVATE void *sqlite3BtreeSchema(Btree *p, int nBytes, void(*xFree)(void *)){
  BtShared *pBt = p->pBt;
  if( !pBt->pSchema ){
    pBt->pSchema = sqliteMalloc(nBytes);
    pBt->xFreeSchema = xFree;
  }
  return pBt->pSchema;
}

/*
** Return true if another user of the same shared btree as the argument
** handle holds an exclusive lock on the sqlite_master table.
*/
SQLITE_PRIVATE int sqlite3BtreeSchemaLocked(Btree *p){
  return (queryTableLock(p, MASTER_ROOT, READ_LOCK)!=SQLITE_OK);
}


#ifndef SQLITE_OMIT_SHARED_CACHE
/*
** Obtain a lock on the table whose root page is iTab.  The
** lock is a write lock if isWritelock is true or a read lock
** if it is false.
*/
SQLITE_PRIVATE int sqlite3BtreeLockTable(Btree *p, int iTab, u8 isWriteLock){
  int rc = SQLITE_OK;
  u8 lockType = (isWriteLock?WRITE_LOCK:READ_LOCK);
  rc = queryTableLock(p, iTab, lockType);
  if( rc==SQLITE_OK ){
    rc = lockTable(p, iTab, lockType);
  }
  return rc;
}
#endif

#ifndef SQLITE_OMIT_INCRBLOB
/*
** Argument pCsr must be a cursor opened for writing on an 
** INTKEY table currently pointing at a valid table entry. 
** This function modifies the data stored as part of that entry.
** Only the data content may only be modified, it is not possible
** to change the length of the data stored.
*/
SQLITE_PRIVATE int sqlite3BtreePutData(BtCursor *pCsr, u32 offset, u32 amt, void *z){

  assert(pCsr->isIncrblobHandle);
  if( pCsr->eState==CURSOR_REQUIRESEEK ){
    return SQLITE_ABORT;
  }

  /* Check some preconditions: 
  **   (a) the cursor is open for writing,
  **   (b) there is no read-lock on the table being modified and
  **   (c) the cursor points at a valid row of an intKey table.
  */
  if( !pCsr->wrFlag ){
    return SQLITE_READONLY;
  }
  assert( !pCsr->pBtree->pBt->readOnly 
          && pCsr->pBtree->pBt->inTransaction==TRANS_WRITE );
  if( checkReadLocks(pCsr->pBtree, pCsr->pgnoRoot, pCsr) ){
    return SQLITE_LOCKED; /* The table pCur points to has a read lock */
  }
  if( pCsr->eState==CURSOR_INVALID || !pCsr->pPage->intKey ){
    return SQLITE_ERROR;
  }

  return accessPayload(pCsr, offset, amt, (unsigned char *)z, 0, 1);
}

/* 
** Set a flag on this cursor to cache the locations of pages from the 
** overflow list for the current row. This is used by cursors opened
** for incremental blob IO only.
**
** This function sets a flag only. The actual page location cache
** (stored in BtCursor.aOverflow[]) is allocated and used by function
** accessPayload() (the worker function for sqlite3BtreeData() and
** sqlite3BtreePutData()).
*/
SQLITE_PRIVATE void sqlite3BtreeCacheOverflow(BtCursor *pCur){
  assert(!pCur->isIncrblobHandle);
  assert(!pCur->aOverflow);
  pCur->isIncrblobHandle = 1;
}
#endif

/************** End of btree.c ***********************************************/
/************** Begin file vdbefifo.c ****************************************/
/*
** 2005 June 16
**
** The author disclaims copyright to this source code.  In place of
** a legal notice, here is a blessing:
**
**    May you do good and not evil.
**    May you find forgiveness for yourself and forgive others.
**    May you share freely, never taking more than you give.
**
*************************************************************************
** This file implements a FIFO queue of rowids used for processing
** UPDATE and DELETE statements.
*/

/*
** Allocate a new FifoPage and return a pointer to it.  Return NULL if
** we run out of memory.  Leave space on the page for nEntry entries.
*/
static FifoPage *allocateFifoPage(int nEntry){
  FifoPage *pPage;
  if( nEntry>32767 ){
    nEntry = 32767;
  }
  pPage = sqliteMallocRaw( sizeof(FifoPage) + sizeof(i64)*(nEntry-1) );
  if( pPage ){
    pPage->nSlot = nEntry;
    pPage->iWrite = 0;
    pPage->iRead = 0;
    pPage->pNext = 0;
  }
  return pPage;
}

/*
** Initialize a Fifo structure.
*/
SQLITE_PRIVATE void sqlite3VdbeFifoInit(Fifo *pFifo){
  memset(pFifo, 0, sizeof(*pFifo));
}

/*
** Push a single 64-bit integer value into the Fifo.  Return SQLITE_OK
** normally.   SQLITE_NOMEM is returned if we are unable to allocate
** memory.
*/
SQLITE_PRIVATE int sqlite3VdbeFifoPush(Fifo *pFifo, i64 val){
  FifoPage *pPage;
  pPage = pFifo->pLast;
  if( pPage==0 ){
    pPage = pFifo->pLast = pFifo->pFirst = allocateFifoPage(20);
    if( pPage==0 ){
      return SQLITE_NOMEM;
    }
  }else if( pPage->iWrite>=pPage->nSlot ){
    pPage->pNext = allocateFifoPage(pFifo->nEntry);
    if( pPage->pNext==0 ){
      return SQLITE_NOMEM;
    }
    pPage = pFifo->pLast = pPage->pNext;
  }
  pPage->aSlot[pPage->iWrite++] = val;
  pFifo->nEntry++;
  return SQLITE_OK;
}

/*
** Extract a single 64-bit integer value from the Fifo.  The integer
** extracted is the one least recently inserted.  If the Fifo is empty
** return SQLITE_DONE.
*/
SQLITE_PRIVATE int sqlite3VdbeFifoPop(Fifo *pFifo, i64 *pVal){
  FifoPage *pPage;
  if( pFifo->nEntry==0 ){
    return SQLITE_DONE;
  }
  assert( pFifo->nEntry>0 );
  pPage = pFifo->pFirst;
  assert( pPage!=0 );
  assert( pPage->iWrite>pPage->iRead );
  assert( pPage->iWrite<=pPage->nSlot );
  assert( pPage->iRead<pPage->nSlot );
  assert( pPage->iRead>=0 );
  *pVal = pPage->aSlot[pPage->iRead++];
  pFifo->nEntry--;
  if( pPage->iRead>=pPage->iWrite ){
    pFifo->pFirst = pPage->pNext;
    sqliteFree(pPage);
    if( pFifo->nEntry==0 ){
      assert( pFifo->pLast==pPage );
      pFifo->pLast = 0;
    }else{
      assert( pFifo->pFirst!=0 );
    }
  }else{
    assert( pFifo->nEntry>0 );
  }
  return SQLITE_OK;
}

/*
** Delete all information from a Fifo object.   Free all memory held
** by the Fifo.
*/
SQLITE_PRIVATE void sqlite3VdbeFifoClear(Fifo *pFifo){
  FifoPage *pPage, *pNextPage;
  for(pPage=pFifo->pFirst; pPage; pPage=pNextPage){
    pNextPage = pPage->pNext;
    sqliteFree(pPage);
  }
  sqlite3VdbeFifoInit(pFifo);
}

/************** End of vdbefifo.c ********************************************/
/************** Begin file vdbemem.c *****************************************/
/*
** 2004 May 26
**
** The author disclaims copyright to this source code.  In place of
** a legal notice, here is a blessing:
**
**    May you do good and not evil.
**    May you find forgiveness for yourself and forgive others.
**    May you share freely, never taking more than you give.
**
*************************************************************************
**
** This file contains code use to manipulate "Mem" structure.  A "Mem"
** stores a single value in the VDBE.  Mem is an opaque structure visible
** only within the VDBE.  Interface routines refer to a Mem using the
** name sqlite_value
*/

/*
** Call sqlite3VdbeMemExpandBlob() on the supplied value (type Mem*)
** P if required.
*/
#define expandBlob(P) (((P)->flags&MEM_Zero)?sqlite3VdbeMemExpandBlob(P):0)

/*
** If pMem is an object with a valid string representation, this routine
** ensures the internal encoding for the string representation is
** 'desiredEnc', one of SQLITE_UTF8, SQLITE_UTF16LE or SQLITE_UTF16BE.
**
** If pMem is not a string object, or the encoding of the string
** representation is already stored using the requested encoding, then this
** routine is a no-op.
**
** SQLITE_OK is returned if the conversion is successful (or not required).
** SQLITE_NOMEM may be returned if a malloc() fails during conversion
** between formats.
*/
SQLITE_PRIVATE int sqlite3VdbeChangeEncoding(Mem *pMem, int desiredEnc){
  int rc;
  if( !(pMem->flags&MEM_Str) || pMem->enc==desiredEnc ){
    return SQLITE_OK;
  }
#ifdef SQLITE_OMIT_UTF16
  return SQLITE_ERROR;
#else


  /* MemTranslate() may return SQLITE_OK or SQLITE_NOMEM. If NOMEM is returned,
  ** then the encoding of the value may not have changed.
  */
  rc = sqlite3VdbeMemTranslate(pMem, desiredEnc);
  assert(rc==SQLITE_OK    || rc==SQLITE_NOMEM);
  assert(rc==SQLITE_OK    || pMem->enc!=desiredEnc);
  assert(rc==SQLITE_NOMEM || pMem->enc==desiredEnc);
  return rc;
#endif
}

/*
** Make the given Mem object MEM_Dyn.
**
** Return SQLITE_OK on success or SQLITE_NOMEM if malloc fails.
*/
SQLITE_PRIVATE int sqlite3VdbeMemDynamicify(Mem *pMem){
  int n;
  u8 *z;
  expandBlob(pMem);
  if( (pMem->flags & (MEM_Ephem|MEM_Static|MEM_Short))==0 ){
    return SQLITE_OK;
  }
  assert( (pMem->flags & MEM_Dyn)==0 );
  n = pMem->n;
  assert( pMem->flags & (MEM_Str|MEM_Blob) );
  z = sqliteMallocRaw( n+2 );
  if( z==0 ){
    return SQLITE_NOMEM;
  }
  pMem->flags |= MEM_Dyn|MEM_Term;
  pMem->xDel = 0;
  memcpy(z, pMem->z, n );
  z[n] = 0;
  z[n+1] = 0;
  pMem->z = (char*)z;
  pMem->flags &= ~(MEM_Ephem|MEM_Static|MEM_Short);
  return SQLITE_OK;
}

/*
** If the given Mem* has a zero-filled tail, turn it into an ordinary
** blob stored in dynamically allocated space.
*/
#ifndef SQLITE_OMIT_INCRBLOB
SQLITE_PRIVATE int sqlite3VdbeMemExpandBlob(Mem *pMem){
  if( pMem->flags & MEM_Zero ){
    char *pNew;
    int nByte;
    assert( (pMem->flags & MEM_Blob)!=0 );
    nByte = pMem->n + pMem->u.i;
    if( nByte<=0 ) nByte = 1;
    pNew = sqliteMalloc(nByte);
    if( pNew==0 ){ 
      return SQLITE_NOMEM;
    }
    memcpy(pNew, pMem->z, pMem->n);
    memset(&pNew[pMem->n], 0, pMem->u.i);
    sqlite3VdbeMemRelease(pMem);
    pMem->z = pNew;
    pMem->n += pMem->u.i;
    pMem->u.i = 0;
    pMem->flags &= ~(MEM_Zero|MEM_Static|MEM_Ephem|MEM_Short|MEM_Term);
    pMem->flags |= MEM_Dyn;
  }
  return SQLITE_OK;
}
#endif


/*
** Make the given Mem object either MEM_Short or MEM_Dyn so that bytes
** of the Mem.z[] array can be modified.
**
** Return SQLITE_OK on success or SQLITE_NOMEM if malloc fails.
*/
SQLITE_PRIVATE int sqlite3VdbeMemMakeWriteable(Mem *pMem){
  int n;
  u8 *z;
  expandBlob(pMem);
  if( (pMem->flags & (MEM_Ephem|MEM_Static))==0 ){
    return SQLITE_OK;
  }
  assert( (pMem->flags & MEM_Dyn)==0 );
  assert( pMem->flags & (MEM_Str|MEM_Blob) );
  if( (n = pMem->n)+2<sizeof(pMem->zShort) ){
    z = (u8*)pMem->zShort;
    pMem->flags |= MEM_Short|MEM_Term;
  }else{
    z = sqliteMallocRaw( n+2 );
    if( z==0 ){
      return SQLITE_NOMEM;
    }
    pMem->flags |= MEM_Dyn|MEM_Term;
    pMem->xDel = 0;
  }
  memcpy(z, pMem->z, n );
  z[n] = 0;
  z[n+1] = 0;
  pMem->z = (char*)z;
  pMem->flags &= ~(MEM_Ephem|MEM_Static);
  assert(0==(1&(int)pMem->z));
  return SQLITE_OK;
}

/*
** Make sure the given Mem is \u0000 terminated.
*/
SQLITE_PRIVATE int sqlite3VdbeMemNulTerminate(Mem *pMem){
  if( (pMem->flags & MEM_Term)!=0 || (pMem->flags & MEM_Str)==0 ){
    return SQLITE_OK;   /* Nothing to do */
  }
  if( pMem->flags & (MEM_Static|MEM_Ephem) ){
    return sqlite3VdbeMemMakeWriteable(pMem);
  }else{
    char *z; 
    sqlite3VdbeMemExpandBlob(pMem);
    z = sqliteMalloc(pMem->n+2);

    if( !z ) return SQLITE_NOMEM;
    memcpy(z, pMem->z, pMem->n);
    z[pMem->n] = 0;
    z[pMem->n+1] = 0;
    if( pMem