SQLite

#
LIBOBJ+= alter.o analyze.o attach.o auth.o \
         backup.o bitvec.o btmutex.o btree.o build.o \
         callback.o complete.o ctime.o date.o delete.o expr.o fault.o fkey.o \
         fts3.o fts3_expr.o fts3_hash.o fts3_icu.o fts3_porter.o \
         fts3_snippet.o fts3_tokenizer.o fts3_tokenizer1.o fts3_write.o \
         func.o global.o hash.o \
         icu.o insert.o journal.o legacy.o loadext.o log.o \
         main.o malloc.o mem0.o mem1.o mem2.o mem3.o mem5.o \
         memjournal.o \
         mutex.o mutex_noop.o mutex_os2.o mutex_unix.o mutex_w32.o \
         notify.o opcodes.o os.o os_os2.o os_unix.o os_win.o \
         pager.o parse.o pcache.o pcache1.o pragma.o prepare.o printf.o \
         random.o resolve.o rowset.o rtree.o select.o status.o \
         table.o tokenize.o trigger.o \

  $(TOP)/src/hash.c \
  $(TOP)/src/hash.h \
  $(TOP)/src/hwtime.h \
  $(TOP)/src/insert.c \
  $(TOP)/src/journal.c \
  $(TOP)/src/legacy.c \
  $(TOP)/src/loadext.c \
  $(TOP)/src/log.c \
  $(TOP)/src/log.h \
  $(TOP)/src/main.c \
  $(TOP)/src/malloc.c \
  $(TOP)/src/mem0.c \
  $(TOP)/src/mem1.c \
  $(TOP)/src/mem2.c \
  $(TOP)/src/mem3.c \
  $(TOP)/src/mem5.c \

#TESTSRC += $(TOP)/ext/fts2/fts2_tokenizer.c
#TESTSRC += $(TOP)/ext/fts3/fts3_tokenizer.c

TESTSRC2 = \
  $(TOP)/src/attach.c $(TOP)/src/backup.c $(TOP)/src/btree.c                   \
  $(TOP)/src/build.c $(TOP)/src/date.c                                         \
  $(TOP)/src/expr.c $(TOP)/src/func.c $(TOP)/src/insert.c $(TOP)/src/log.c     \
  $(TOP)/src/mem5.c $(TOP)/src/os.c                                            \
  $(TOP)/src/os_os2.c $(TOP)/src/os_unix.c $(TOP)/src/os_win.c                 \
  $(TOP)/src/pager.c $(TOP)/src/pragma.c $(TOP)/src/prepare.c                  \
  $(TOP)/src/printf.c $(TOP)/src/random.c $(TOP)/src/pcache.c                  \
  $(TOP)/src/pcache1.c $(TOP)/src/select.c $(TOP)/src/tokenize.c               \
  $(TOP)/src/utf.c $(TOP)/src/util.c $(TOP)/src/vdbeapi.c $(TOP)/src/vdbeaux.c \
  $(TOP)/src/vdbe.c $(TOP)/src/vdbemem.c $(TOP)/src/where.c parse.c            \
  $(TOP)/ext/fts3/fts3.c $(TOP)/ext/fts3/fts3_expr.c                           \

/*
** This file contains the implementation of a log file used in 
** "journal_mode=wal" mode.
*/

#include "log.h"

#include <unistd.h>
#include <fcntl.h>
#include <sys/mman.h>

typedef struct LogSummaryHdr LogSummaryHdr;
typedef struct LogSummary LogSummary;
typedef struct LogCheckpoint LogCheckpoint;


/*
** The following structure may be used to store the same data that
** is stored in the log-summary header.
**
** Member variables iCheck1 and iCheck2 contain the checksum for the
** last frame written to the log, or 2 and 3 respectively if the log 
** is currently empty.
*/
struct LogSummaryHdr {
  u32 iChange;                    /* Counter incremented each transaction */
  u32 pgsz;                       /* Database page size in bytes */
  u32 iLastPg;                    /* Address of last valid frame in log */
  u32 nPage;                      /* Size of database in pages */
  u32 iCheck1;                    /* Checkpoint value 1 */
  u32 iCheck2;                    /* Checkpoint value 2 */
};

/* Size of serialized LogSummaryHdr object. */
#define LOGSUMMARY_HDR_NFIELD (sizeof(LogSummaryHdr) / sizeof(u32))

#define LOGSUMMARY_FRAME_OFFSET \
  (LOGSUMMARY_HDR_NFIELD + LOG_CKSM_BYTES/sizeof(u32))

/* Size of frame header */
#define LOG_FRAME_HDRSIZE 20

/*
** There is one instance of this structure for each log-summary object
** that this process has a connection to. They are stored in a linked
** list starting at pLogSummary (global variable).
**
** TODO: LogSummary.fd is a unix file descriptor. Unix APIs are used 
**       directly in this implementation because the VFS does not support
**       the required blocking file-locks.
*/
struct LogSummary {
  sqlite3_mutex *mutex;           /* Mutex used to protect this object */
  int nRef;                       /* Number of pointers to this structure */
  int fd;                         /* File descriptor open on log-summary */
  char *zPath;                    /* Path to associated WAL file */
  LogSummary *pNext;              /* Next in global list */
  int nData;                      /* Size of aData allocation/mapping */
  u32 *aData;                     /* File body */
};

/*
** List of all LogSummary objects created by this process. Protected by
** static mutex LOG_SUMMARY_MUTEX. TODO: Should have a dedicated mutex
** here instead of borrowing the LRU mutex.
*/
#define LOG_SUMMARY_MUTEX SQLITE_MUTEX_STATIC_LRU
static LogSummary *pLogSummary = 0;

struct Log {
  LogSummary *pSummary;           /* Log file summary data */
  sqlite3_vfs *pVfs;              /* The VFS used to create pFd */
  sqlite3_file *pFd;              /* File handle for log file */
  int sync_flags;                 /* Flags to use with OsSync() */
  int isLocked;                   /* True if a snapshot is held open */
  int isWriteLocked;              /* True if this is the writer connection */
  LogSummaryHdr hdr;              /* Log summary header for current snapshot */
};

/*
** This structure is used to implement an iterator that iterates through
** all frames in the log in database page order. Where two or more frames
** correspond to the same database page, the iterator visits only the 
** frame most recently written to the log.
**
** The internals of this structure are only accessed by:
**
**   logCheckpointInit() - Create a new iterator,
**   logCheckpointNext() - Step an iterator,
**   logCheckpointFree() - Free an iterator.
**
** This functionality is used by the checkpoint code (see logCheckpoint()).
*/
struct LogCheckpoint {
  int nSegment;                   /* Size of LogCheckpoint.aSummary[] array */
  int nFinal;                     /* Elements in segment nSegment-1 */
  struct LogSegment {
    int iNext;                    /* Next aIndex index */
    u8 *aIndex;                   /* Pointer to index array */
    u32 *aDbPage;                 /* Pointer to db page array */
  } aSegment[1];
};

/*
** Generate an 8 byte checksum based on the data in array aByte[] and the
** initial values of aCksum[0] and aCksum[1]. The checksum is written into
** aCksum[] before returning.
*/
#define LOG_CKSM_BYTES 8
static void logChecksumBytes(u8 *aByte, int nByte, u32 *aCksum){
  u32 *z32 = (u32 *)aByte;
  int n32 = nByte / sizeof(u32);
  int i;

  assert( LOG_CKSM_BYTES==2*sizeof(u32) );
  assert( (nByte&0x00000003)==0 );

  u32 cksum0 = aCksum[0];
  u32 cksum1 = aCksum[1];

  for(i=0; i<n32; i++){
    cksum0 = (cksum0 >> 8) + (cksum0 ^ z32[i]);
    cksum1 = (cksum1 >> 8) + (cksum1 ^ z32[i]);
  }

  aCksum[0] = cksum0;
  aCksum[1] = cksum1;
}

/*
** Argument zPath must be a nul-terminated string containing a path-name.
** This function modifies the string in-place by removing any "./" or "../" 
** elements in the path. For example, the following input:
**
**   "/home/user/plans/good/../evil/./world_domination.txt"
**
** is overwritten with the 'normalized' version:
**
**   "/home/user/plans/evil/world_domination.txt"
*/
static void logNormalizePath(char *zPath){
  int i, j;
  char *z = zPath;
  int n = strlen(z);

  while( n>1 && z[n-1]=='/' ){ n--; }
  for(i=j=0; i<n; i++){
    if( z[i]=='/' ){
      if( z[i+1]=='/' ) continue;
      if( z[i+1]=='.' && i+2<n && z[i+2]=='/' ){
        i += 1;
        continue;
      }
      if( z[i+1]=='.' && i+3<n && z[i+2]=='.' && z[i+3]=='/' ){
        while( j>0 && z[j-1]!='/' ){ j--; }
        if( j>0 ){ j--; }
        i += 2;
        continue;
      }
    }
    z[j++] = z[i];
  }
  z[j] = 0;
}

/*
** Lock the summary file pSummary->fd.
*/
static int logSummaryLock(LogSummary *pSummary){
  int rc;
  struct flock f;
  memset(&f, 0, sizeof(f));
  f.l_type = F_WRLCK;
  f.l_whence = SEEK_SET;
  f.l_start = 0;
  f.l_len = 1;
  rc = fcntl(pSummary->fd, F_SETLKW, &f);
  if( rc!=0 ){
    return SQLITE_IOERR;
  }
  return SQLITE_OK;
}

/*
** Unlock the summary file pSummary->fd.
*/
static int logSummaryUnlock(LogSummary *pSummary){
  int rc;
  struct flock f;
  memset(&f, 0, sizeof(f));
  f.l_type = F_UNLCK;
  f.l_whence = SEEK_SET;
  f.l_start = 0;
  f.l_len = 1;
  rc = fcntl(pSummary->fd, F_SETLK, &f);
  if( rc!=0 ){
    return SQLITE_IOERR;
  }
  return SQLITE_OK;
}

/*
** Memory map the first nByte bytes of the summary file opened with 
** pSummary->fd at pSummary->aData. If the summary file is smaller than
** nByte bytes in size when this function is called, ftruncate() is
** used to expand it before it is mapped.
**
** It is assumed that an exclusive lock is held on the summary file
** by the caller (to protect the ftruncate()).
*/
static int logSummaryMap(LogSummary *pSummary, int nByte){
  struct stat sStat;
  int rc;
  int fd = pSummary->fd;
  void *pMap;

  assert( pSummary->aData==0 );

  /* If the file is less than nByte bytes in size, cause it to grow. */
  rc = fstat(fd, &sStat);
  if( rc!=0 ) return SQLITE_IOERR;
  if( sStat.st_size<nByte ){
    rc = ftruncate(fd, nByte);
    if( rc!=0 ) return SQLITE_IOERR;
  }

  /* Map the file. */
  pMap = mmap(0, nByte, PROT_READ|PROT_WRITE, MAP_SHARED, fd, 0);
  if( pMap==MAP_FAILED ){
    return SQLITE_IOERR;
  }
  pSummary->aData = (u32 *)pMap;
  pSummary->nData = nByte;

  return SQLITE_OK;
}

/*
** Unmap the log-summary mapping and close the file-descriptor. If
** the isTruncate argument is non-zero, truncate the log-summary file
** region to zero bytes.
**
** Regardless of the value of isTruncate, close the file-descriptor
** opened on the log-summary file.
*/
static int logSummaryUnmap(LogSummary *pSummary, int isTruncate){
  int rc = SQLITE_OK;
  if( pSummary->aData ){
    assert( pSummary->fd>0 );
    munmap(pSummary->aData, pSummary->nData);
    pSummary->aData = 0;
    if( isTruncate ){
      rc = (ftruncate(pSummary->fd, 0) ? SQLITE_IOERR : SQLITE_OK);
    }
  }
  if( pSummary->fd>0 ){
    close(pSummary->fd);
    pSummary->fd = -1;
  }
  return rc;
}


static void logSummaryWriteHdr(LogSummary *pSummary, LogSummaryHdr *pHdr){
  u32 *aData = pSummary->aData;
  memcpy(aData, pHdr, sizeof(LogSummaryHdr));
  aData[LOGSUMMARY_HDR_NFIELD] = 1;
  aData[LOGSUMMARY_HDR_NFIELD+1] = 1;
  logChecksumBytes(
    (u8 *)aData, sizeof(LogSummaryHdr), &aData[LOGSUMMARY_HDR_NFIELD]
  );
}

/*
** This function encodes a single frame header and writes it to a buffer
** supplied by the caller. A log frame-header is made up of a series of 
** 4-byte big-endian integers, as follows:
**
**     0: Database page size in bytes.
**     4: Page number.
**     8: New database size (for commit frames, otherwise zero).
**    12: Frame checksum 1.
**    16: Frame checksum 2.
*/
static void logEncodeFrame(
  u32 *aCksum,                    /* IN/OUT: Checksum values */
  u32 iPage,                      /* Database page number for frame */
  u32 nTruncate,                  /* New db size (or 0 for non-commit frames) */
  int nData,                      /* Database page size (size of aData[]) */
  u8 *aData,                      /* Pointer to page data (for checksum) */
  u8 *aFrame                      /* OUT: Write encoded frame here */
){
  assert( LOG_FRAME_HDRSIZE==20 );

  sqlite3Put4byte(&aFrame[0], nData);
  sqlite3Put4byte(&aFrame[4], iPage);
  sqlite3Put4byte(&aFrame[8], nTruncate);

  logChecksumBytes(aFrame, 12, aCksum);
  logChecksumBytes(aData, nData, aCksum);

  sqlite3Put4byte(&aFrame[12], aCksum[0]);
  sqlite3Put4byte(&aFrame[16], aCksum[1]);
}

/*
** Return 1 and populate *piPage, *pnTruncate and aCksum if the 
** frame checksum looks Ok. Otherwise return 0.
*/
static int logDecodeFrame(
  u32 *aCksum,                    /* IN/OUT: Checksum values */
  u32 *piPage,                    /* OUT: Database page number for frame */
  u32 *pnTruncate,                /* OUT: New db size (or 0 if not commit) */
  int nData,                      /* Database page size (size of aData[]) */
  u8 *aData,                      /* Pointer to page data (for checksum) */
  u8 *aFrame                      /* Frame data */
){
  logChecksumBytes(aFrame, 12, aCksum);
  logChecksumBytes(aData, nData, aCksum);

  if( aCksum[0]!=sqlite3Get4byte(&aFrame[12]) 
   || aCksum[1]!=sqlite3Get4byte(&aFrame[16]) 
  ){
    /* Checksum failed. */
    return 0;
  }

  *piPage = sqlite3Get4byte(&aFrame[4]);
  *pnTruncate = sqlite3Get4byte(&aFrame[8]);
  return 1;
}

static void logMergesort8(
  Pgno *aContent,                 /* Pages in log */
  u8 *aBuffer,                    /* Buffer of at least *pnList items to use */
  u8 *aList,                      /* IN/OUT: List to sort */
  int *pnList                     /* IN/OUT: Number of elements in aList[] */
){
  int nList = *pnList;
  if( nList>1 ){
    int nLeft = nList / 2;        /* Elements in left list */
    int nRight = nList - nLeft;   /* Elements in right list */
    u8 *aLeft = aList;            /* Left list */
    u8 *aRight = &aList[nLeft];   /* Right list */
    int iLeft = 0;                /* Current index in aLeft */
    int iRight = 0;               /* Current index in aright */
    int iOut = 0;                 /* Current index in output buffer */

    /* TODO: Change to non-recursive version. */
    logMergesort8(aContent, aBuffer, aLeft, &nLeft);
    logMergesort8(aContent, aBuffer, aRight, &nRight);

    while( iRight<nRight || iLeft<nLeft ){
      u8 logpage;
      Pgno dbpage;

      if( (iLeft<nLeft) 
       && (iRight>=nRight || aContent[aLeft[iLeft]]<aContent[aRight[iRight]])
      ){
        logpage = aLeft[iLeft++];
      }else{
        logpage = aRight[iRight++];
      }
      dbpage = aContent[logpage];

      aBuffer[iOut++] = logpage;
      if( iLeft<nLeft && aContent[aLeft[iLeft]]==dbpage ) iLeft++;

      assert( iLeft>=nLeft || aContent[aLeft[iLeft]]>dbpage );
      assert( iRight>=nRight || aContent[aRight[iRight]]>dbpage );
    }
    memcpy(aList, aBuffer, sizeof(aList[0])*iOut);
    *pnList = iOut;
  }

#ifdef SQLITE_DEBUG
  {
    int i;
    for(i=1; i<*pnList; i++){
      assert( aContent[aList[i]] > aContent[aList[i-1]] );
    }
  }
#endif
}


/*
** Return the index in the LogSummary.aData array that corresponds to 
** frame iFrame. The log-summary file consists of a header, followed by
** alternating "map" and "index" blocks.
*/
static int logSummaryEntry(u32 iFrame){
  return ((((iFrame-1)>>8)<<6) + iFrame-1 + 2 + LOGSUMMARY_HDR_NFIELD);
}


/*
** Set an entry in the log-summary map to map log frame iFrame to db 
** page iPage. Values are always appended to the log-summary (i.e. the
** value of iFrame is always exactly one more than the value passed to
** the previous call), but that restriction is not enforced or asserted
** here.
*/
static void logSummaryAppend(LogSummary *pSummary, u32 iFrame, u32 iPage){
  u32 iSlot = logSummaryEntry(iFrame);

  /* Set the log-summary entry itself */
  pSummary->aData[iSlot] = iPage;

  /* If the frame number is a multiple of 256 (frames are numbered starting
  ** at 1), build an index of the most recently added 256 frames.
  */
  if( (iFrame&0x000000FF)==0 ){
    int i;                        /* Iterator used while initializing aIndex */
    u32 *aFrame;                  /* Pointer to array of 256 frames */
    int nIndex;                   /* Number of entries in index */
    u8 *aIndex;                   /* 256 bytes to build index in */
    u8 *aTmp;                     /* Scratch space to use while sorting */

    aFrame = &pSummary->aData[iSlot-255];
    aIndex = (u8 *)&pSummary->aData[iSlot+1];
    aTmp = &aIndex[256];

    nIndex = 256;
    for(i=0; i<256; i++) aIndex[i] = (u8)i;
    logMergesort8(aFrame, aTmp, aIndex, &nIndex);
    memset(&aIndex[nIndex], aIndex[nIndex-1], 256-nIndex);
  }
}


/*
** Recover the log-summary by reading the log file. The caller must hold 
** an exclusive lock on the log-summary file.
*/
static int logSummaryRecover(LogSummary *pSummary, sqlite3_file *pFd){
  int rc;                         /* Return Code */
  i64 nSize;                      /* Size of log file */
  LogSummaryHdr hdr;              /* Recovered log-summary header */

  memset(&hdr, 0, sizeof(hdr));

  rc = sqlite3OsFileSize(pFd, &nSize);
  if( rc!=SQLITE_OK ){
    return rc;
  }

  if( nSize>LOG_FRAME_HDRSIZE ){
    u8 aBuf[LOG_FRAME_HDRSIZE];   /* Buffer to load first frame header into */
    u8 *aFrame = 0;               /* Malloc'd buffer to load entire frame */
    int nFrame;                   /* Number of bytes at aFrame */
    u8 *aData;                    /* Pointer to data part of aFrame buffer */
    int iFrame;                   /* Index of last frame read */
    i64 iOffset;                  /* Next offset to read from log file */
    int nPgsz;                    /* Page size according to the log */
    u32 aCksum[2] = {2, 3};       /* Running checksum */

    /* Read in the first frame header in the file (to determine the 
    ** database page size).
    */
    rc = sqlite3OsRead(pFd, aBuf, LOG_FRAME_HDRSIZE, 0);
    if( rc!=SQLITE_OK ){
      return rc;
    }

    /* If the database page size is not a power of two, or is greater than
    ** SQLITE_MAX_PAGE_SIZE, conclude that the log file contains no valid data.
    */
    nPgsz = sqlite3Get4byte(&aBuf[0]);
    if( nPgsz&(nPgsz-1) || nPgsz>SQLITE_MAX_PAGE_SIZE ){
      goto finished;
    }

    /* Malloc a buffer to read frames into. */
    nFrame = nPgsz + LOG_FRAME_HDRSIZE;
    aFrame = (u8 *)sqlite3_malloc(nFrame);
    if( !aFrame ){
      return SQLITE_NOMEM;
    }
    aData = &aFrame[LOG_FRAME_HDRSIZE];

    /* Read all frames from the log file. */
    iFrame = 0;
    iOffset = 0;
    for(iOffset=0; (iOffset+nFrame)<nSize; iOffset+=nFrame){
      u32 pgno;                   /* Database page number for frame */
      u32 nTruncate;              /* dbsize field from frame header */
      int isValid;                /* True if this frame is valid */

      /* Read and decode the next log frame. */
      rc = sqlite3OsRead(pFd, aFrame, nFrame, iOffset);
      if( rc!=SQLITE_OK ) break;
      isValid = logDecodeFrame(aCksum, &pgno, &nTruncate, nPgsz, aData, aFrame);
      if( !isValid ) break;
      logSummaryAppend(pSummary, ++iFrame, pgno);

      /* If nTruncate is non-zero, this is a commit record. */
      if( nTruncate ){
        hdr.iCheck1 = aCksum[0];
        hdr.iCheck2 = aCksum[1];
        hdr.iLastPg = iFrame;
        hdr.nPage = nTruncate;
        hdr.pgsz = nPgsz;
      }
    }

    sqlite3_free(aFrame);
  }else{
    hdr.iCheck1 = 2;
    hdr.iCheck2 = 3;
  }

finished:
  logSummaryWriteHdr(pSummary, &hdr);
  return rc;
}


/*
** This function intializes the connection to the log-summary identified
** by struct pSummary.
*/
static int logSummaryInit(LogSummary *pSummary, sqlite3_file *pFd){
  int rc;                         /* Return Code */
  char *zFile;                    /* File name for summary file */

  assert( pSummary->fd<0 );
  assert( pSummary->aData==0 );
  assert( pSummary->nRef>0 );
  assert( pSummary->zPath );

  /* Open a file descriptor on the summary file. */
  zFile = sqlite3_mprintf("%s-summary", pSummary->zPath);
  if( !zFile ){
    return SQLITE_NOMEM;
  }
  pSummary->fd = open(zFile, O_RDWR|O_CREAT, S_IWUSR|S_IRUSR);
  sqlite3_free(zFile);
  if( pSummary->fd<0 ){
    return SQLITE_IOERR;
  }

  /* Grab an exclusive lock the summary file. Then mmap() it. TODO: This 
  ** code needs to be enhanced to support a growable mapping. For now, just 
  ** make the mapping very large to start with.
  */
  rc = logSummaryLock(pSummary);
  if( rc!=SQLITE_OK ) return rc;
  rc = logSummaryMap(pSummary, 512*1024);
  if( rc!=SQLITE_OK ) goto out;

  /* Grab a SHARED lock on the log file. Then try to upgrade to an EXCLUSIVE
  ** lock. If successful, then this is the first (and only) connection to
  ** the database. In this case assume the contents of the log-summary 
  ** cannot be trusted. Zero the log-summary header to make sure.
  **
  ** The SHARED lock on the log file is not released until the connection
  ** to the database is closed.
  */
  rc = sqlite3OsLock(pFd, SQLITE_LOCK_SHARED);
  if( rc!=SQLITE_OK ) goto out;
  rc = sqlite3OsLock(pFd, SQLITE_LOCK_EXCLUSIVE);
  if( rc==SQLITE_OK ){
    /* This is the first and only connection. */
    memset(pSummary->aData, 0, (LOGSUMMARY_HDR_NFIELD+2)*sizeof(u32) );
    rc = sqlite3OsUnlock(pFd, SQLITE_LOCK_SHARED);
  }else if( rc==SQLITE_BUSY ){
    rc = SQLITE_OK;
  }

 out:
  logSummaryUnlock(pSummary);
  return rc;
}

/* 
** Open a connection to the log file associated with database zDb. The
** database file does not actually have to exist. zDb is used only to
** figure out the name of the log file to open. If the log file does not 
** exist it is created by this call.
*/
int sqlite3LogOpen(
  sqlite3_vfs *pVfs,              /* vfs module to open log file with */
  const char *zDb,                /* Name of database file */
  Log **ppLog                     /* OUT: Allocated Log handle */
){
  int rc;                         /* Return Code */
  Log *pRet;                      /* Object to allocate and return */
  LogSummary *pSummary = 0;       /* Summary object */
  sqlite3_mutex *mutex = 0;       /* LOG_SUMMARY_MUTEX mutex */
  int flags;                      /* Flags passed to OsOpen() */
  char *zWal = 0;                 /* Path to WAL file */
  int nWal;                       /* Length of zWal in bytes */

  /* Zero output variables */
  assert( zDb );
  *ppLog = 0;

  /* Allocate an instance of struct Log to return. */
  pRet = (Log *)sqlite3MallocZero(sizeof(Log) + pVfs->szOsFile);
  if( !pRet ) goto out;
  pRet->pVfs = pVfs;
  pRet->pFd = (sqlite3_file *)&pRet[1];
  pRet->sync_flags = SQLITE_SYNC_NORMAL;

  /* Normalize the path name. */
  zWal = sqlite3_mprintf("%s-wal", zDb);
  if( !zWal ) goto out;
  logNormalizePath(zWal);
  flags = (SQLITE_OPEN_READWRITE|SQLITE_OPEN_CREATE|SQLITE_OPEN_MAIN_DB);
  nWal = sqlite3Strlen30(zWal);

  /* Enter the mutex that protects the linked-list of LogSummary structures */
  if( sqlite3GlobalConfig.bCoreMutex ){
    mutex = sqlite3_mutex_alloc(LOG_SUMMARY_MUTEX);
  }
  sqlite3_mutex_enter(mutex);

  /* Search for an existing log summary object in the linked list. If one 
  ** cannot be found, allocate and initialize a new object.
  */
  for(pSummary=pLogSummary; pSummary; pSummary=pSummary->pNext){
    int nPath = sqlite3Strlen30(pSummary->zPath);
    if( nWal==nPath && 0==memcmp(pSummary->zPath, zWal, nPath) ) break;
  }
  if( !pSummary ){
    int nByte = sizeof(LogSummary) + nWal + 1;
    pSummary = (LogSummary *)sqlite3MallocZero(nByte);
    if( !pSummary ){
      rc = SQLITE_NOMEM;
      goto out;
    }
    if( sqlite3GlobalConfig.bCoreMutex ){
      pSummary->mutex = sqlite3_mutex_alloc(SQLITE_MUTEX_RECURSIVE);
    }
    pSummary->zPath = (char *)&pSummary[1];
    pSummary->fd = -1;
    memcpy(pSummary->zPath, zWal, nWal);
    pSummary->pNext = pLogSummary;
    pLogSummary = pSummary;
  }
  pSummary->nRef++;
  pRet->pSummary = pSummary;

  /* Exit the mutex protecting the linked-list of LogSummary objects. */
  sqlite3_mutex_leave(mutex);
  mutex = 0;

  /* Open file handle on the log file. */
  rc = sqlite3OsOpen(pVfs, pSummary->zPath, pRet->pFd, flags, &flags);
  if( rc!=SQLITE_OK ) goto out;

  /* Object pSummary is shared between all connections to the database made
  ** by this process. So at this point it may or may not be connected to
  ** the log-summary. If it is not, connect it. Otherwise, just take the
  ** SHARED lock on the log file.
  */
  sqlite3_mutex_enter(pSummary->mutex);
  mutex = pSummary->mutex;
  if( pSummary->fd<0 ){
    rc = logSummaryInit(pSummary, pRet->pFd);
  }else{
    rc = sqlite3OsLock(pRet->pFd, SQLITE_LOCK_SHARED);
  }

 out:
  sqlite3_mutex_leave(mutex);
  sqlite3_free(zWal);
  if( rc!=SQLITE_OK ){
    assert(0);
    if( pRet ){
      sqlite3OsClose(pRet->pFd);
      sqlite3_free(pRet);
    }
    assert( !pSummary || pSummary->nRef==0 );
    sqlite3_free(pSummary);
  }
  *ppLog = pRet;
  return rc;
}

static int logCheckpointNext(
  LogCheckpoint *p,               /* Iterator */
  u32 *piPage,                    /* OUT: Next db page to write */
  u32 *piFrame                    /* OUT: Log frame to read from */
){
  u32 iMin = *piPage;
  u32 iRet = 0xFFFFFFFF;
  int i;
  int nBlock = p->nFinal;

  for(i=p->nSegment-1; i>=0; i--){
    struct LogSegment *pSegment = &p->aSegment[i];
    while( pSegment->iNext<nBlock ){
      u32 iPg = pSegment->aDbPage[pSegment->aIndex[pSegment->iNext]];
      if( iPg>iMin ){
        if( iPg<iRet ){
          iRet = iPg;
          *piFrame = i*256 + 1 + pSegment->aIndex[pSegment->iNext];
        }
        break;
      }
      pSegment->iNext++;
    }

    nBlock = 256;
  }

  *piPage = iRet;
  return (iRet==0xFFFFFFFF);
}

static LogCheckpoint *logCheckpointInit(Log *pLog){
  u32 *aData = pLog->pSummary->aData;
  LogCheckpoint *p;               /* Return value */
  int nSegment;                   /* Number of segments to merge */
  u32 iLast;                      /* Last frame in log */
  int nByte;                      /* Number of bytes to allocate */
  int i;                          /* Iterator variable */
  int nFinal;                     /* Number of unindexed entries */
  struct LogSegment *pFinal;      /* Final (unindexed) segment */
  u8 *aTmp;                       /* Temp space used by merge-sort */

  iLast = pLog->hdr.iLastPg;
  nSegment = (iLast >> 8) + 1;
  nFinal = (iLast & 0x000000FF);

  nByte = sizeof(LogCheckpoint) + (nSegment-1)*sizeof(struct LogSegment) + 512;
  p = (LogCheckpoint *)sqlite3_malloc(nByte);
  if( p ){
    memset(p, 0, nByte);
    p->nSegment = nSegment;
    p->nFinal = nFinal;
  }

  for(i=0; i<nSegment-1; i++){
    p->aSegment[i].aDbPage = &aData[logSummaryEntry(i*256+1)];
    p->aSegment[i].aIndex = (u8 *)&aData[logSummaryEntry(i*256+1)+256];
  }
  pFinal = &p->aSegment[nSegment-1];

  pFinal->aDbPage = &aData[logSummaryEntry((nSegment-1)*256+1)];
  pFinal->aIndex = (u8 *)&pFinal[1];
  aTmp = &pFinal->aIndex[256];
  for(i=0; i<nFinal; i++){
    pFinal->aIndex[i] = i;
  }
  logMergesort8(pFinal->aDbPage, aTmp, pFinal->aIndex, &nFinal);
  p->nFinal = nFinal;

  return p;
}

/* 
** Free a log iterator allocated by logCheckpointInit().
*/
static void logCheckpointFree(LogCheckpoint *p){
  sqlite3_free(p);
}

/*
** Checkpoint the contents of the log file.
*/
static int logCheckpoint(
  Log *pLog,                      /* Log connection */
  sqlite3_file *pFd,              /* File descriptor open on db file */
  u8 *zBuf                        /* Temporary buffer to use */
){
  int rc;                         /* Return code */
  int pgsz = pLog->hdr.pgsz;      /* Database page-size */
  LogCheckpoint *pIter = 0;       /* Log iterator context */
  u32 iDbpage = 0;                /* Next database page to write */
  u32 iFrame;                     /* Log frame containing data for iDbpage */

  /* Allocate the iterator */
  pIter = logCheckpointInit(pLog);
  if( !pIter ) return SQLITE_NOMEM;

  /* Sync the log file to disk */
  rc = sqlite3OsSync(pLog->pFd, pLog->sync_flags);
  if( rc!=SQLITE_OK ) goto out;

  /* Iterate through the contents of the log, copying data to the db file. */
  while( 0==logCheckpointNext(pIter, &iDbpage, &iFrame) ){
    rc = sqlite3OsRead(pLog->pFd, zBuf, pgsz, 
        (iFrame-1) * (pgsz+LOG_FRAME_HDRSIZE) + LOG_FRAME_HDRSIZE
    );
    if( rc!=SQLITE_OK ) goto out;
    rc = sqlite3OsWrite(pFd, zBuf, pgsz, (iDbpage-1)*pgsz);
    if( rc!=SQLITE_OK ) goto out;
  }

  /* Truncate the database file */
  rc = sqlite3OsTruncate(pFd, ((i64)pLog->hdr.nPage*(i64)pgsz));
  if( rc!=SQLITE_OK ) goto out;

  /* Sync the database file. If successful, update the log-summary. */
  rc = sqlite3OsSync(pFd, pLog->sync_flags);
  if( rc!=SQLITE_OK ) goto out;
  pLog->hdr.iLastPg = 0;
  pLog->hdr.iCheck1 = 2;
  pLog->hdr.iCheck2 = 3;
  logSummaryWriteHdr(pLog->pSummary, &pLog->hdr);

  /* TODO: If a crash occurs and the current log is copied into the 
  ** database there is no problem. However, if a crash occurs while
  ** writing the next transaction into the start of the log, such that:
  **
  **   * The first transaction currently in the log is left intact, but
  **   * The second (or subsequent) transaction is damaged,
  **
  ** then the database could become corrupt.
  **
  ** The easiest thing to do would be to write and sync a dummy header
  ** into the log at this point. Unfortunately, that turns out to be
  ** an unwelcome performance hit. Alternatives are...
  */
#if 0 
  memset(zBuf, 0, LOG_FRAME_HDRSIZE);
  rc = sqlite3OsWrite(pLog->pFd, zBuf, LOG_FRAME_HDRSIZE, 0);
  if( rc!=SQLITE_OK ) goto out;
  rc = sqlite3OsSync(pLog->pFd, pLog->sync_flags);
#endif

 out:
  logCheckpointFree(pIter);
  return rc;
}

/*
** Close a connection to a log file.
*/
int sqlite3LogClose(
  Log *pLog,                      /* Log to close */
  sqlite3_file *pFd,              /* Database file */
  u8 *zBuf                        /* Buffer of at least page-size bytes */
){
  int rc = SQLITE_OK;
  if( pLog ){
    LogSummary *pSummary = pLog->pSummary;
    sqlite3_mutex *mutex = 0;

    if( sqlite3GlobalConfig.bCoreMutex ){
      mutex = sqlite3_mutex_alloc(LOG_SUMMARY_MUTEX);
    }
    sqlite3_mutex_enter(mutex);

    /* Decrement the reference count on the log summary. If this is the last
    ** reference to the log summary object in this process, the object will
    ** be freed. If this is also the last connection to the database, then
    ** checkpoint the database and truncate the log and log-summary files
    ** to zero bytes in size.
    **/
    pSummary->nRef--;
    if( pSummary->nRef==0 ){
      LogSummary **pp;

      rc = logSummaryLock(pSummary);
      if( rc==SQLITE_OK ){
        int isTruncate = 0;
        int rc2 = sqlite3OsLock(pLog->pFd, SQLITE_LOCK_EXCLUSIVE);
        if( rc2==SQLITE_OK ){
          /* This is the last connection to the database (including other
          ** processes). Do three things:
          **
          **   1. Checkpoint the db.
          **   2. Truncate the log file to zero bytes.
          **   3. Truncate the log-summary file to zero bytes.
          */
          rc2 = logCheckpoint(pLog, pFd, zBuf);
          if( rc2==SQLITE_OK ){
            rc2 = sqlite3OsTruncate(pLog->pFd, 0);
          }
          isTruncate = 1;
        }else if( rc2==SQLITE_BUSY ){
          rc2 = SQLITE_OK;
        }
        logSummaryUnmap(pSummary, isTruncate);
        sqlite3OsUnlock(pLog->pFd, SQLITE_LOCK_NONE);
        rc = logSummaryUnlock(pSummary);
        if( rc2!=SQLITE_OK ) rc = rc2;
      }

      /* Remove the LogSummary object from the global list. Then free the 
      ** mutex and the object itself.
      */
      for(pp=&pLogSummary; *pp!=pSummary; pp=&(*pp)->pNext);
      *pp = (*pp)->pNext;
      sqlite3_mutex_free(pSummary->mutex);
      sqlite3_free(pSummary);
    }

    sqlite3_mutex_leave(mutex);

    /* Close the connection to the log file and free the Log handle. */
    sqlite3OsClose(pLog->pFd);
    sqlite3_free(pLog);
  }
  return rc;
}

/*
** Set the flags to pass to the sqlite3OsSync() function when syncing
** the log file.
*/
#if 0
void sqlite3LogSetSyncflags(Log *pLog, int sync_flags){
  assert( sync_flags==SQLITE_SYNC_NORMAL || sync_flags==SQLITE_SYNC_FULL );
  pLog->sync_flags = sync_flags;
}
#endif

/*
** Enter and leave the log-summary mutex. In this context, entering the
** log-summary mutex means:
**
**   1. Obtaining mutex pLog->pSummary->mutex, and
**   2. Taking an exclusive lock on the log-summary file.
**
** i.e. this mutex locks out other processes as well as other threads
** hosted in this address space.
*/
static int logEnterMutex(Log *pLog){
  LogSummary *pSummary = pLog->pSummary;
  int rc;

  sqlite3_mutex_enter(pSummary->mutex);
  rc = logSummaryLock(pSummary);
  if( rc!=SQLITE_OK ){
    sqlite3_mutex_leave(pSummary->mutex);
  }
  return rc;
}
static void logLeaveMutex(Log *pLog){
  LogSummary *pSummary = pLog->pSummary;
  logSummaryUnlock(pSummary);
  sqlite3_mutex_leave(pSummary->mutex);
}

/*
** The caller must hold a SHARED lock on the database file.
**
** If this call obtains a new read-lock and the database contents have been
** modified since the most recent call to LogCloseSnapshot() on this Log
** connection, then *pChanged is set to 1 before returning. Otherwise, it 
** is left unmodified. This is used by the pager layer to determine whether 
** or not any cached pages may be safely reused.
*/
int sqlite3LogOpenSnapshot(Log *pLog, int *pChanged){
  int rc = SQLITE_OK;
  if( pLog->isLocked==0 ){
    if( SQLITE_OK==(rc = logEnterMutex(pLog)) ){
      u32 aCksum[2] = {1, 1};
      u32 aHdr[LOGSUMMARY_HDR_NFIELD+2];
      memcpy(aHdr, pLog->pSummary->aData, sizeof(aHdr));

      /* Verify the checksum on the log-summary header. If it fails,
      ** recover the log-summary from the log file.
      */
      logChecksumBytes((u8*)aHdr, sizeof(u32)*LOGSUMMARY_HDR_NFIELD, aCksum);
      if( aCksum[0]!=aHdr[LOGSUMMARY_HDR_NFIELD]
       || aCksum[1]!=aHdr[LOGSUMMARY_HDR_NFIELD+1]
      ){
        rc = logSummaryRecover(pLog->pSummary, pLog->pFd);
        memcpy(aHdr, pLog->pSummary->aData, sizeof(aHdr));
        *pChanged = 1;
      }
      if( rc==SQLITE_OK ){
        pLog->isLocked = 1;
        if( memcmp(&pLog->hdr, aHdr, sizeof(LogSummaryHdr)) ){
          *pChanged = 1;
          memcpy(&pLog->hdr, aHdr, LOGSUMMARY_HDR_NFIELD*sizeof(u32));
        }
      }
      logLeaveMutex(pLog);
    }
  }
  return rc;
}

/*
** Unlock the current snapshot.
*/
void sqlite3LogCloseSnapshot(Log *pLog){
  pLog->isLocked = 0;
}



/* 
** Read a page from the log, if it is present. 
*/
int sqlite3LogRead(Log *pLog, Pgno pgno, int *pInLog, u8 *pOut){
  u32 iRead = 0;
  u32 *aData = pLog->pSummary->aData;
  int iFrame = (pLog->hdr.iLastPg & 0xFFFFFF00);

  /* Do a linear search of the unindexed block of page-numbers (if any) 
  ** at the end of the log-summary. An alternative to this would be to
  ** build an index in private memory each time a read transaction is
  ** opened on a new snapshot.
  */
  if( pLog->hdr.iLastPg ){
    u32 *pi = &aData[logSummaryEntry(pLog->hdr.iLastPg)];
    u32 *piStop = pi - (pLog->hdr.iLastPg & 0xFF);
    while( *pi!=pgno && pi!=piStop ) pi--;
    if( pi!=piStop ){
      iRead = (pi-piStop) + iFrame;
    }
  }
  assert( iRead==0 || aData[logSummaryEntry(iRead)]==pgno );

  while( iRead==0 && iFrame>0 ){
    int iLow = 0;
    int iHigh = 255;
    u32 *aFrame;
    u8 *aIndex;

    iFrame -= 256;
    aFrame = &aData[logSummaryEntry(iFrame+1)];
    aIndex = (u8 *)&aFrame[256];

    while( iLow<=iHigh ){
      int iTest = (iLow+iHigh)>>1;
      u32 iPg = aFrame[aIndex[iTest]];

      if( iPg==pgno ){
        iRead = iFrame + 1 + aIndex[iTest];
        break;
      }
      else if( iPg<pgno ){
        iLow = iTest+1;
      }else{
        iHigh = iTest-1;
      }
    }
  }
  assert( iRead==0 || aData[logSummaryEntry(iRead)]==pgno );

  /* If iRead is non-zero, then it is the log frame number that contains the
  ** required page. Read and return data from the log file.
  */
  if( iRead ){
    i64 iOffset = (iRead-1) * (pLog->hdr.pgsz+LOG_FRAME_HDRSIZE);
    iOffset += LOG_FRAME_HDRSIZE;
    *pInLog = 1;
    return sqlite3OsRead(pLog->pFd, pOut, pLog->hdr.pgsz, iOffset);
  }

  *pInLog = 0;
  return SQLITE_OK;
}


/* 
** Set *pPgno to the size of the database file (or zero, if unknown).
*/
void sqlite3LogMaxpgno(Log *pLog, Pgno *pPgno){
  assert( pLog->isLocked );
  *pPgno = pLog->hdr.nPage;
}

/* 
** The caller must hold at least a RESERVED lock on the database file
** when invoking this function.
**
** This function returns SQLITE_OK if the caller may write to the database.
** Otherwise, if the caller is operating on a snapshot that has already
** been overwritten by another writer, SQLITE_OBE is returned.
*/
int sqlite3LogWriteLock(Log *pLog, int op){
  assert( pLog->isLocked );
  if( op ){
    if( memcmp(&pLog->hdr, pLog->pSummary->aData, sizeof(pLog->hdr)) ){
      return SQLITE_BUSY;
    }
    pLog->isWriteLocked = 1;
  }else if( pLog->isWriteLocked ){
    memcpy(&pLog->hdr, pLog->pSummary->aData, sizeof(pLog->hdr));
    pLog->isWriteLocked = 0;
  }
  return SQLITE_OK;
}

/* 
** Write a set of frames to the log. The caller must hold at least a
** RESERVED lock on the database file.
*/
int sqlite3LogFrames(
  Log *pLog,                      /* Log handle to write to */
  int nPgsz,                      /* Database page-size in bytes */
  PgHdr *pList,                   /* List of dirty pages to write */
  Pgno nTruncate,                 /* Database size after this commit */
  int isCommit,                   /* True if this is a commit */
  int isSync                      /* True to sync the log file */
){
  /* Each frame has a 20 byte header, as follows:
  **
  **   + Pseudo-random salt (4 bytes)
  **   + Page number (4 bytes)
  **   + New database size, or 0 if not a commit frame (4 bytes)
  **   + Checksum (CHECKSUM_BYTES bytes);
  **
  ** The checksum is computed based on the following:
  **
  **   + The previous checksum, or {2, 3} for the first frame in the log.
  **   + The non-checksum fields of the frame header, and
  **   + The frame contents (page data).
  **
  ** This format must also be understood by the code in logSummaryRecover().
  ** The size of the frame header is used by LogRead() and LogCheckpoint().
  */
  int rc;                         /* Used to catch return codes */
  u32 iFrame;                     /* Next frame address */
  u8 aFrame[LOG_FRAME_HDRSIZE];
  PgHdr *p;                       /* Iterator to run through pList with. */
  u32 aCksum[2];

  PgHdr *pLast;                   /* Last frame in list */
  int nLast = 0;                  /* Number of extra copies of last page */

  assert( LOG_FRAME_HDRSIZE==(4 * 3 + LOG_CKSM_BYTES) );
  assert( pList );

  aCksum[0] = pLog->hdr.iCheck1;
  aCksum[1] = pLog->hdr.iCheck2;

  /* Write the log file. */
  iFrame = pLog->hdr.iLastPg;
  for(p=pList; p; p=p->pDirty){
    u32 nDbsize;                  /* Db-size field for frame header */
    i64 iOffset;                  /* Write offset in log file */

    iFrame++;
    iOffset = (iFrame-1) * (nPgsz+sizeof(aFrame));
    
    /* Populate and write the frame header */
    nDbsize = (isCommit && p->pDirty==0) ? nTruncate : 0;
    logEncodeFrame(aCksum, p->pgno, nDbsize, nPgsz, p->pData, aFrame);
    rc = sqlite3OsWrite(pLog->pFd, aFrame, sizeof(aFrame), iOffset);
    if( rc!=SQLITE_OK ){
      return rc;
    }

    /* Write the page data */
    rc = sqlite3OsWrite(pLog->pFd, p->pData, nPgsz, iOffset + sizeof(aFrame));
    if( rc!=SQLITE_OK ){
      return rc;
    }
    pLast = p;
  }

  /* Sync the log file if the 'isSync' flag was specified. */
  if( isSync ){
#if 0
    i64 iSegment = sqlite3OsSectorSize(pLog->pFd);
    i64 iOffset = iFrame * (nPgsz+sizeof(aFrame));

    if( iSegment<SQLITE_DEFAULT_SECTOR_SIZE ){
      iSegment = SQLITE_DEFAULT_SECTOR_SIZE;
    }
    iSegment = (((iOffset+iSegment-1)/iSegment) * iSegment);
    while( iOffset<iSegment ){
      logEncodeFrame(aCksum,pLast->pgno,nTruncate,nPgsz,pLast->pData,aFrame);
      rc = sqlite3OsWrite(pLog->pFd, aFrame, sizeof(aFrame), iOffset);
      if( rc!=SQLITE_OK ){
        return rc;
      }

      iOffset += LOG_FRAME_HDRSIZE;
      rc = sqlite3OsWrite(pLog->pFd, pLast->pData, nPgsz, iOffset); 
      if( rc!=SQLITE_OK ){
        return rc;
      }
      nLast++;
      iOffset += nPgsz;
    }
#endif

    rc = sqlite3OsSync(pLog->pFd, pLog->sync_flags);
    if( rc!=SQLITE_OK ){
      return rc;
    }
  }

  /* Append data to the log summary. It is not necessary to lock the 
  ** log-summary to do this as the RESERVED lock held on the db file
  ** guarantees that there are no other writers, and no data that may
  ** be in use by existing readers is being overwritten.
  */
  iFrame = pLog->hdr.iLastPg;
  for(p=pList; p; p=p->pDirty){
    iFrame++;
    logSummaryAppend(pLog->pSummary, iFrame, p->pgno);
  }
  while( nLast>0 ){
    iFrame++;
    nLast--;
    logSummaryAppend(pLog->pSummary, iFrame, pLast->pgno);
  }

  /* Update the private copy of the header. */
  pLog->hdr.pgsz = nPgsz;
  pLog->hdr.iLastPg = iFrame;
  if( isCommit ){
    pLog->hdr.iChange++;
    pLog->hdr.nPage = nTruncate;
  }
  pLog->hdr.iCheck1 = aCksum[0];
  pLog->hdr.iCheck2 = aCksum[1];

  /* If this is a commit, update the log-summary header too. */
  if( isCommit && SQLITE_OK==(rc = logEnterMutex(pLog)) ){
    logSummaryWriteHdr(pLog->pSummary, &pLog->hdr);
    logLeaveMutex(pLog);
  }

  return SQLITE_OK;
}

/* 
** Checkpoint the database. When this function is called the caller
** must hold an exclusive lock on the database file.
*/
int sqlite3LogCheckpoint(
  Log *pLog,                      /* Log connection */
  sqlite3_file *pFd,              /* File descriptor open on db file */
  u8 *zBuf                        /* Temporary buffer to use */
){

  /* Assert() that the caller is holding an EXCLUSIVE lock on the 
  ** database file. 
  */
#ifdef SQLITE_DEBUG
  int lock;
  sqlite3OsFileControl(pFd, SQLITE_FCNTL_LOCKSTATE, &lock);
  assert( lock>=4 );
#endif
  
  return logCheckpoint(pLog, pFd, zBuf);
}

    ** reading the database file again, make sure that the
    ** transaction counter was updated if any part of the database
    ** file changed.  If the transaction counter is not updated,
    ** other connections to the same file might not realize that
    ** the file has changed and hence might not know to flush their
    ** cache.  The use of a stale cache can lead to database corruption.
    */
#if 0
    assert( pFile->inNormalWrite==0
         || pFile->dbUpdate==0
         || pFile->transCntrChng==1 );
#endif
    pFile->inNormalWrite = 0;
#endif

    /* downgrading to a shared lock on NFS involves clearing the write lock
    ** before establishing the readlock - to avoid a race condition we downgrade
    ** the lock in 2 blocks, so that part of the range will be covered by a 
    ** write lock until the rest is covered by a read lock:

){
  unixFile *pFile = (unixFile *)id;
  int got;
  assert( id );

  /* If this is a database file (not a journal, master-journal or temp
  ** file), the bytes in the locking range should never be read or written. */
#if 0
  assert( pFile->pUnused==0
       || offset>=PENDING_BYTE+512
       || offset+amt<=PENDING_BYTE 
  );
#endif

  got = seekAndRead(pFile, offset, pBuf, amt);
  if( got==amt ){
    return SQLITE_OK;
  }else if( got<0 ){
    /* lastErrno set by seekAndRead */
    return SQLITE_IOERR_READ;

  unixFile *pFile = (unixFile*)id;
  int wrote = 0;
  assert( id );
  assert( amt>0 );

  /* If this is a database file (not a journal, master-journal or temp
  ** file), the bytes in the locking range should never be read or written. */
#if 0
  assert( pFile->pUnused==0
       || offset>=PENDING_BYTE+512
       || offset+amt<=PENDING_BYTE 
  );
#endif

#ifndef NDEBUG
  /* If we are doing a normal write to a database file (as opposed to
  ** doing a hot-journal rollback or a write to some file other than a
  ** normal database file) then record the fact that the database
  ** has changed.  If the transaction counter is modified, record that
  ** fact too.

** 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.
*/
#ifndef SQLITE_OMIT_DISKIO
#include "sqliteInt.h"
#include "log.h"

/*
******************** NOTES ON THE DESIGN OF THE PAGER ************************
**
** Within this comment block, a page is deemed to have been synced
** automatically as soon as it is written when PRAGMA synchronous=OFF.
** Otherwise, the page is not synced until the xSync method of the VFS

  void (*xCodecSizeChng)(void*,int,int); /* Notify of page size changes */
  void (*xCodecFree)(void*);             /* Destructor for the codec */
  void *pCodec;               /* First argument to xCodec... methods */
#endif
  char *pTmpSpace;            /* Pager.pageSize bytes of space for tmp use */
  PCache *pPCache;            /* Pointer to page cache object */
  sqlite3_backup *pBackup;    /* Pointer to list of ongoing backup processes */
  Log *pLog;                  /* Log used by "journal_mode=wal" */
};

/*
** 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.
*/

      rc |= sqlite3BitvecSet(p->pInSavepoint, pgno);
      testcase( rc==SQLITE_NOMEM );
      assert( rc==SQLITE_OK || rc==SQLITE_NOMEM );
    }
  }
  return rc;
}

/*
** Open a connection to the write-ahead log file for pager pPager.
*/
static int pagerOpenLog(Pager *pPager){
  if( !pPager->pLog ){
    int rc;                       /* Return code from LogOpen() */

    rc = sqlite3LogOpen(pPager->pVfs, pPager->zFilename, &pPager->pLog);
    if( rc!=SQLITE_OK ) return rc;
  }
  return SQLITE_OK;
}

/*
** Return true if this pager uses a write-ahead log instead of the usual
** rollback journal. Otherwise false.
*/
static int pagerUseLog(Pager *pPager){
  return (pPager->pLog!=0);
}

/*
** Unlock the database file. This function is a no-op if the pager
** is in exclusive mode.
**
** If the pager is currently in error state, discard the contents of 
** the cache and reset the Pager structure internal state. If there is
** an open journal-file, then the next time a shared-lock is obtained
** on the pager file (by this or any other process), it will be
** treated as a hot-journal and rolled back.
*/
static void pager_unlock(Pager *pPager){
  if( !pPager->exclusiveMode ){
    int rc = SQLITE_OK;          /* Return code */

    /* Always close the journal file when dropping the database lock.
    ** Otherwise, another connection with journal_mode=delete might
    ** delete the file out from under us.
    */
    sqlite3OsClose(pPager->jfd);
    sqlite3BitvecDestroy(pPager->pInJournal);
    pPager->pInJournal = 0;
    releaseAllSavepoints(pPager);

    /* If the file is unlocked, somebody else might change it. The
    ** values stored in Pager.dbSize etc. might become invalid if
    ** this happens.  One can argue that this doesn't need to be cleared
    ** until the change-counter check fails in PagerSharedLock().
    ** Clearing the page size cache here is being conservative.
    */
    pPager->dbSizeValid = 0;

    if( pagerUseLog(pPager) ){
      sqlite3LogCloseSnapshot(pPager->pLog);
    }
    rc = osUnlock(pPager->fd, NO_LOCK);
    if( rc ){
      pPager->errCode = rc;
    }
    IOTRACE(("UNLOCK %p\n", pPager))

    /* If Pager.errCode is set, the contents of the pager cache cannot be

  if( pPager->state<PAGER_RESERVED ){
    return SQLITE_OK;
  }
  releaseAllSavepoints(pPager);

  assert( isOpen(pPager->jfd) || pPager->pInJournal==0 );
  if( isOpen(pPager->jfd) ){
    assert( !pagerUseLog(pPager) );

    /* Finalize the journal file. */
    if( sqlite3IsMemJournal(pPager->jfd) ){
      assert( pPager->journalMode==PAGER_JOURNALMODE_MEMORY );
      sqlite3OsClose(pPager->jfd);
    }else if( pPager->journalMode==PAGER_JOURNALMODE_TRUNCATE ){
      if( pPager->journalOff==0 ){

#endif
  }
  sqlite3BitvecDestroy(pPager->pInJournal);
  pPager->pInJournal = 0;
  pPager->nRec = 0;
  sqlite3PcacheCleanAll(pPager->pPCache);

  if( pagerUseLog(pPager) ){
    rc2 = sqlite3LogWriteLock(pPager->pLog, 0);
    pPager->state = PAGER_SHARED;
  }else if( !pPager->exclusiveMode ){
    rc2 = osUnlock(pPager->fd, SHARED_LOCK);
    pPager->state = PAGER_SHARED;
    pPager->changeCountDone = 0;
  }else if( pPager->state==PAGER_SYNCED ){
    pPager->state = PAGER_EXCLUSIVE;
  }
  pPager->setMaster = 0;

    zMaster = pPager->pTmpSpace;
    rc = readMasterJournal(pPager->jfd, zMaster, pPager->pVfs->mxPathname+1);
    testcase( rc!=SQLITE_OK );
  }
  if( rc==SQLITE_OK && pPager->noSync==0 && pPager->state>=PAGER_EXCLUSIVE ){
    rc = sqlite3OsSync(pPager->fd, pPager->sync_flags);
  }
  if( rc==SQLITE_OK && pPager->noSync==0 && pPager->state>=PAGER_EXCLUSIVE ){
    rc = sqlite3OsSync(pPager->fd, pPager->sync_flags);
  }
  if( rc==SQLITE_OK ){
    rc = pager_end_transaction(pPager, zMaster[0]!='\0');
    testcase( rc!=SQLITE_OK );
  }
  if( rc==SQLITE_OK && zMaster[0] && res ){
    /* If there was a master journal and this routine will return success,
    ** see if it is possible to delete the master journal.
    */
    rc = pager_delmaster(pPager, zMaster);
    testcase( rc!=SQLITE_OK );
  }

  /* 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;
}


/*
** Read the content for page pPg out of the database file and into 
** pPg->pData. A shared lock or greater must be held on the database
** file before this function is called.
**
** If page 1 is read, then the value of Pager.dbFileVers[] is set to
** the value read from the database file.
**
** If an IO error occurs, then the IO error is returned to the caller.
** Otherwise, SQLITE_OK is returned.
*/
static int readDbPage(PgHdr *pPg){
  Pager *pPager = pPg->pPager; /* Pager object associated with page pPg */
  Pgno pgno = pPg->pgno;       /* Page number to read */
  int rc;                      /* Return code */
  i64 iOffset;                 /* Byte offset of file to read from */
  int isInLog = 0;             /* True if page is in log file */

  assert( pPager->state>=PAGER_SHARED && !MEMDB );
  assert( isOpen(pPager->fd) );

  if( NEVER(!isOpen(pPager->fd)) ){
    assert( pPager->tempFile );
    memset(pPg->pData, 0, pPager->pageSize);
    return SQLITE_OK;
  }

  if( pagerUseLog(pPager) ){
    /* Try to pull the page from the write-ahead log. */
    rc = sqlite3LogRead(pPager->pLog, pgno, &isInLog, pPg->pData);
  }
  if( rc==SQLITE_OK && !isInLog ){
    iOffset = (pgno-1)*(i64)pPager->pageSize;
    rc = sqlite3OsRead(pPager->fd, pPg->pData, pPager->pageSize, iOffset);
    if( rc==SQLITE_IOERR_SHORT_READ ){
      rc = SQLITE_OK;
    }
  }

  if( pgno==1 ){
    if( rc ){
      /* If the read is unsuccessful, set the dbFileVers[] to something
      ** that will never be a valid file version.  dbFileVers[] is a copy
      ** of bytes 24..39 of the database.  Bytes 28..31 should always be
      ** zero.  Bytes 32..35 and 35..39 should be page numbers which are
      ** never 0xffffffff.  So filling pPager->dbFileVers[] with all 0xff
      ** bytes should suffice.
      **
      ** For an encrypted database, the situation is more complex:  bytes
      ** 24..39 of the database are white noise.  But the probability of
      ** white noising equaling 16 bytes of 0xff is vanishingly small so
      ** we should still be ok.
      */
      memset(pPager->dbFileVers, 0xff, sizeof(pPager->dbFileVers));
    }else{
      u8 *dbFileVers = &((u8*)pPg->pData)[24];
      memcpy(&pPager->dbFileVers, dbFileVers, sizeof(pPager->dbFileVers));
    }
  }
  CODEC1(pPager, pPg->pData, pgno, 3, rc = SQLITE_NOMEM);

  PAGER_INCR(sqlite3_pager_readdb_count);
  PAGER_INCR(pPager->nRead);
  IOTRACE(("PGIN %p %d\n", pPager, pgno));
  PAGERTRACE(("FETCH %d page %d hash(%08x)\n",
               PAGERID(pPager), pgno, pager_pagehash(pPg)));

  return rc;
}

static int pagerRollbackLog(Pager *pPager){
  int rc = SQLITE_OK;
  PgHdr *pList = sqlite3PcacheDirtyList(pPager->pPCache);
  pPager->dbSize = pPager->dbOrigSize;
  while( pList && rc==SQLITE_OK ){
    PgHdr *pNext = pList->pDirty;
    if( sqlite3PcachePageRefcount(pList)==0 ){
      sqlite3PagerLookup(pPager, pList->pgno);
      sqlite3PcacheDrop(pList);
    }else{
      rc = readDbPage(pList);
      if( rc==SQLITE_OK ){
        pPager->xReiniter(pList);
      }
    }
    pList = pNext;
  }
  return rc;
}

/*
** Playback savepoint pSavepoint. Or, if pSavepoint==NULL, then playback
** the entire master journal file. The case pSavepoint==NULL occurs when 
** a ROLLBACK TO command is invoked on a SAVEPOINT that is a transaction 
** savepoint.
**
** When pSavepoint is not NULL (meaning a non-transaction savepoint is 

    }
  }

  /* Set the database size back to the value it was before the savepoint 
  ** being reverted was opened.
  */
  pPager->dbSize = pSavepoint ? pSavepoint->nOrig : pPager->dbOrigSize;

  if( !pSavepoint && pagerUseLog(pPager) ){
    return pagerRollbackLog(pPager);
  }

  /* Use pPager->journalOff as the effective size of the main rollback
  ** journal.  The actual file might be larger than this in
  ** PAGER_JOURNALMODE_TRUNCATE or PAGER_JOURNALMODE_PERSIST.  But anything
  ** past pPager->journalOff is off-limits to us.
  */
  szJ = pPager->journalOff;
  assert( pagerUseLog(pPager)==0 || szJ==0 );

  /* Begin by rolling back records from the main journal starting at
  ** PagerSavepoint.iOffset and continuing to the next journal header.
  ** There might be records in the main journal that have a page number
  ** greater than the current database size (pPager->dbSize) but those
  ** will be skipped automatically.  Pages are added to pDone as they
  ** are played back.
  */
  if( pSavepoint && !pagerUseLog(pPager) ){
    iHdrOff = pSavepoint->iHdrOffset ? pSavepoint->iHdrOffset : szJ;
    pPager->journalOff = pSavepoint->iOffset;
    while( rc==SQLITE_OK && pPager->journalOff<iHdrOff ){
      rc = pager_playback_one_page(pPager, &pPager->journalOff, pDone, 1, 1);
    }
    assert( rc!=SQLITE_DONE );
  }else{

** the query attempt returns an IO error, the IO error code is returned
** and *pnPage is left unchanged.
**
** Otherwise, if everything is successful, then SQLITE_OK is returned
** and *pnPage is set to the number of pages in the database.
*/
int sqlite3PagerPagecount(Pager *pPager, int *pnPage){
  Pgno nPage = 0;           /* Value to return via *pnPage */

  /* Determine the number of pages in the file. Store this in nPage. */
  if( pPager->dbSizeValid ){
    nPage = pPager->dbSize;
  }else{
    int rc;                 /* Error returned by OsFileSize() */
    i64 n = 0;              /* File size in bytes returned by OsFileSize() */

    if( pagerUseLog(pPager) ){
      sqlite3LogMaxpgno(pPager->pLog, &nPage);
    }

    if( nPage==0 ){
      assert( isOpen(pPager->fd) || pPager->tempFile );
      if( isOpen(pPager->fd) ){
        if( SQLITE_OK!=(rc = sqlite3OsFileSize(pPager->fd, &n)) ){
          pager_error(pPager, rc);
          return rc;
        }
      }
      if( n>0 && n<pPager->pageSize ){
        nPage = 1;
      }else{
        nPage = (Pgno)(n / pPager->pageSize);
      }
    }
    if( pPager->state!=PAGER_UNLOCK ){
      pPager->dbSize = nPage;
      pPager->dbFileSize = nPage;
      pPager->dbSizeValid = 1;
    }
  }

void sqlite3PagerTruncateImage(Pager *pPager, Pgno nPage){
  assert( pPager->dbSizeValid );
  assert( pPager->dbSize>=nPage );
  assert( pPager->state>=PAGER_RESERVED );
  pPager->dbSize = nPage;
  assertTruncateConstraint(pPager);
}


/*
** This function is called before attempting a hot-journal rollback. It
** syncs the journal file to disk, then sets pPager->journalHdr to the
** size of the journal file so that the pager_playback() routine knows
** that the entire journal file has been synced.
**

**
** 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.
*/
int sqlite3PagerClose(Pager *pPager){
  u8 *pTmp = (u8 *)pPager->pTmpSpace;

  disable_simulated_io_errors();
  sqlite3BeginBenignMalloc();
  pPager->errCode = 0;
  pPager->exclusiveMode = 0;
  sqlite3LogClose(pPager->pLog, pPager->fd, pTmp);
  pPager->pLog = 0;
  pager_reset(pPager);
  if( MEMDB ){
    pager_unlock(pPager);
  }else{
    /* Set Pager.journalHdr to -1 for the benefit of the pager_playback() 
    ** call which may be made from within pagerUnlockAndRollback(). If it
    ** is not -1, then the unsynced portion of an open journal file may
    ** be played back into the database. If a power failure occurs while
    ** this is happening, the database may become corrupt.
    */
    if( isOpen(pPager->jfd) ){
      pPager->errCode = pagerSyncHotJournal(pPager);
    }
    pagerUnlockAndRollback(pPager);
  }
  sqlite3EndBenignMalloc();
  enable_simulated_io_errors();
  PAGERTRACE(("CLOSE %d\n", PAGERID(pPager)));
  IOTRACE(("CLOSE %p\n", pPager))
  sqlite3OsClose(pPager->fd);
  sqlite3PageFree(pTmp);
  sqlite3PcacheClose(pPager->pPCache);

#ifdef SQLITE_HAS_CODEC
  if( pPager->xCodecFree ) pPager->xCodecFree(pPager->pCodec);
#endif

  assert( !pPager->aSavepoint && !pPager->pInJournal );

    void *pData = pPg->pData;
    i64 offset = pPager->nSubRec*(4+pPager->pageSize);
    char *pData2;

    CODEC2(pPager, pData, pPg->pgno, 7, return SQLITE_NOMEM, pData2);
    PAGERTRACE(("STMT-JOURNAL %d page %d\n", PAGERID(pPager), pPg->pgno));
  
    assert( pagerUseLog(pPager) 
         || pageInJournal(pPg) 
         || pPg->pgno>pPager->dbOrigSize 
    );
    rc = write32bits(pPager->sjfd, offset, pPg->pgno);
    if( rc==SQLITE_OK ){
      rc = sqlite3OsWrite(pPager->sjfd, pData2, pPager->pageSize, offset+4);
    }
  }
  if( rc==SQLITE_OK ){
    pPager->nSubRec++;

*/
static int pagerStress(void *p, PgHdr *pPg){
  Pager *pPager = (Pager *)p;
  int rc = SQLITE_OK;

  assert( pPg->pPager==pPager );
  assert( pPg->flags&PGHDR_DIRTY );

  if( pagerUseLog(pPager) ) return SQLITE_OK;

  /* The doNotSync flag is set by the sqlite3PagerWrite() function while it
  ** is journalling a set of two or more database pages that are stored
  ** on the same disk sector. Syncing the journal is not allowed while
  ** this is happening as it is important that all members of such a
  ** set of pages are synced to disk together. So, if the page this function
  ** is trying to make clean will require a journal sync and the doNotSync

      }
    }
  }

  return rc;
}






























































/*
** This function is called to obtain a shared lock on the database file.
** It is illegal to call sqlite3PagerAcquire() until after this function
** has been successfully called. If a shared-lock is already held when
** this function is called, it is a no-op.
**
** The following operations are also performed by this function.

    if( isOpen(pPager->jfd) || pPager->zJournal ){
      isErrorReset = 1;
    }
    pPager->errCode = SQLITE_OK;
    pager_reset(pPager);
  }






  if( pPager->noReadlock ){
    assert( pPager->readOnly );
    pPager->state = PAGER_SHARED;
  }else{
    rc = pager_wait_on_lock(pPager, SHARED_LOCK);
    if( rc!=SQLITE_OK ){
      assert( pPager->state==PAGER_UNLOCK );
      return pager_error(pPager, rc);
    }
  }
  assert( pPager->state>=SHARED_LOCK );

  if( pagerUseLog(pPager) ){
    int changed = 0;
    rc = sqlite3LogOpenSnapshot(pPager->pLog, &changed);
    if( rc==SQLITE_OK ){
      if( changed ){
        pager_reset(pPager);
        assert( pPager->errCode || pPager->dbSizeValid==0 );
      }
      pPager->state = PAGER_SHARED;
      rc = sqlite3PagerPagecount(pPager, &changed);
    }
  }else if( pPager->state==PAGER_UNLOCK || isErrorReset ){
    sqlite3_vfs * const pVfs = pPager->pVfs;
    int isHotJournal = 0;
    assert( !MEMDB );
    assert( sqlite3PcacheRefCount(pPager->pPCache)==0 );

    /* 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( !isErrorReset ){
      assert( pPager->state <= PAGER_SHARED );
      rc = hasHotJournal(pPager, &isHotJournal);

      /* Reset the journal status fields to indicates that we have no
      ** rollback journal at this time. */
      pPager->journalStarted = 0;
      pPager->journalOff = 0;
      pPager->setMaster = 0;
      pPager->journalHdr = 0;
 
      /* Make sure the journal file has been synced to disk. */
 
      /* Playback and delete the journal.  Drop the database write
      ** lock and reacquire the read lock. Purge the cache before
      ** playing back the hot-journal so that we don't end up with
      ** an inconsistent cache.  Sync the hot journal before playing
      ** it back since the process that crashed and left the hot journal

    rc = sqlite3PagerPagecount(pPager, &nMax);
    if( rc!=SQLITE_OK ){
      goto pager_acquire_err;
    }

    if( MEMDB || nMax<(int)pgno || noContent || !isOpen(pPager->fd) ){
      if( pgno>pPager->mxPgno ){
        rc = SQLITE_FULL;
        goto pager_acquire_err;
      }
      if( noContent ){
        /* Failure to set the bits in the InJournal bit-vectors is benign.
        ** It merely means that we might do some extra work to journal a 
        ** page that does not need to be journaled.  Nevertheless, be sure 
        ** to test the case where a malloc error occurs while trying to set 
        ** a bit in a bit vector.

**
** SQLITE_OK is returned if everything goes according to plan. 
** An SQLITE_IOERR_XXX error code is returned if a call to 
** sqlite3OsOpen() fails.
*/
static int openSubJournal(Pager *pPager){
  int rc = SQLITE_OK;
  if( (pagerUseLog(pPager) || isOpen(pPager->jfd)) && !isOpen(pPager->sjfd) ){
    if( pPager->journalMode==PAGER_JOURNALMODE_MEMORY || pPager->subjInMemory ){
      sqlite3MemJournalOpen(pPager->sjfd);
    }else{
      rc = pagerOpentemp(pPager, pPager->sjfd, SQLITE_OPEN_SUBJOURNAL);
    }
  }
  return rc;

    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 && pagerUseLog(pPager) ){
      /* Grab the write lock on the log file. If successful, upgrade to
      ** PAGER_EXCLUSIVE state. Otherwise, return an error code to the caller.
      ** The busy-handler is not invoked if another connection already
      ** holds the write-lock. If possible, the upper layer will call it.
      */
      rc = sqlite3LogWriteLock(pPager->pLog, 1);
      if( rc==SQLITE_OK ){
        pPager->dbOrigSize = pPager->dbSize;
      }
    }

    /* No need to open the journal file at this time.  It will be
    ** opened before it is written to.  If we defer opening the journal,
    ** we might save the work of creating a file if the transaction
    ** ends up being a no-op.
    */
  }else if( isOpen(pPager->jfd) && pPager->journalOff==0 ){
    /* This happens when the pager was in exclusive-access mode the last
    ** time a (read or write) transaction was successfully concluded
    ** by this connection. Instead of deleting the journal file it was 
    ** kept open and either was truncated to 0 bytes or its header was
    ** overwritten with zeros.
    */
    assert( pagerUseLog(pPager)==0 );
    assert( pPager->nRec==0 );
    assert( pPager->dbOrigSize==0 );
    assert( pPager->pInJournal==0 );
    rc = pager_open_journal(pPager);
  }

  PAGERTRACE(("TRANSACTION %d\n", PAGERID(pPager)));

  CHECK_PAGE(pPg);

  /* Mark the page as dirty.  If the page has already been written
  ** to the journal then we can return right away.
  */
  sqlite3PcacheMakeDirty(pPg);
  if( pageInJournal(pPg) && !subjRequiresPage(pPg) ){
    assert( !pagerUseLog(pPager) );
    pPager->dbModified = 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.
    **
    ** Higher level routines should have already started a transaction,
    ** which means they have acquired the necessary locks but the rollback
    ** journal might not yet be open.
    */
    rc = sqlite3PagerBegin(pPager, 0, pPager->subjInMemory);
    if( rc!=SQLITE_OK ){
      return rc;
    }
    if( !isOpen(pPager->jfd) 
     && pPager->journalMode!=PAGER_JOURNALMODE_OFF 
     && pPager->journalMode!=PAGER_JOURNALMODE_WAL 
    ){
      assert( pPager->useJournal );
      rc = pager_open_journal(pPager);
      if( rc!=SQLITE_OK ) return rc;
    }
    pPager->dbModified = 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( !pageInJournal(pPg) && isOpen(pPager->jfd) ){
      assert( !pagerUseLog(pPager) );
      if( pPg->pgno<=pPager->dbOrigSize ){
        u32 cksum;
        char *pData2;

        /* We should never write to the journal file the page that
        ** contains the database locks.  The following assert verifies
        ** that we do not. */

  if( MEMDB && pPager->dbModified ){
    /* If this is an in-memory db, or no pages have been written to, or this
    ** function has already been called, it is mostly a no-op.  However, any
    ** backup in progress needs to be restarted.
    */
    sqlite3BackupRestart(pPager->pBackup);
  }else if( pPager->state!=PAGER_SYNCED && pPager->dbModified ){
    if( pagerUseLog(pPager) ){
      PgHdr *pList = sqlite3PcacheDirtyList(pPager->pPCache);
      if( pList ){
        rc = sqlite3LogFrames(pPager->pLog, pPager->pageSize, pList,
            pPager->dbSize, 1, 1
        );
      }
      sqlite3PcacheCleanAll(pPager->pPCache);
    }else{
      /* The following block updates the change-counter. Exactly how it
      ** does this depends on whether or not the atomic-update optimization
      ** was enabled at compile time, and if this transaction meets the 
      ** runtime criteria to use the operation: 
      **
      **    * The file-system supports the atomic-write property for
      **      blocks of size page-size, and 
      **    * This commit is not part of a multi-file transaction, and
      **    * Exactly one page has been modified and store in the journal file.
      **
      ** If the optimization was not enabled at compile time, then the
      ** pager_incr_changecounter() function is called to update the change
      ** counter in 'indirect-mode'. If the optimization is compiled in but
      ** is not applicable to this transaction, call sqlite3JournalCreate()
      ** to make sure the journal file has actually been created, then call
      ** pager_incr_changecounter() to update the change-counter in indirect
      ** mode. 
      **
      ** Otherwise, if the optimization is both enabled and applicable,
      ** then call pager_incr_changecounter() to update the change-counter
      ** in 'direct' mode. In this case the journal file will never be
      ** created for this transaction.
      */
  #ifdef SQLITE_ENABLE_ATOMIC_WRITE
      PgHdr *pPg;
      assert( isOpen(pPager->jfd) || pPager->journalMode==PAGER_JOURNALMODE_OFF );
      if( !zMaster && isOpen(pPager->jfd) 
       && pPager->journalOff==jrnlBufferSize(pPager) 
       && pPager->dbSize>=pPager->dbFileSize
       && (0==(pPg = sqlite3PcacheDirtyList(pPager->pPCache)) || 0==pPg->pDirty)
      ){
        /* Update the db file change counter via the direct-write method. The 
        ** following call will modify the in-memory representation of page 1 
        ** to include the updated change counter and then write page 1 
        ** directly to the database file. Because of the atomic-write 
        ** property of the host file-system, this is safe.
        */
        rc = pager_incr_changecounter(pPager, 1);
      }else{
        rc = sqlite3JournalCreate(pPager->jfd);
        if( rc==SQLITE_OK ){
          rc = pager_incr_changecounter(pPager, 0);
        }
      }
  #else
      rc = pager_incr_changecounter(pPager, 0);
  #endif
      if( rc!=SQLITE_OK ) goto commit_phase_one_exit;
  
      /* If this transaction has made the database smaller, then all pages
      ** being discarded by the truncation must be written to the journal
      ** file. This can only happen in auto-vacuum mode.
      **
      ** Before reading the pages with page numbers larger than the 
      ** current value of Pager.dbSize, set dbSize back to the value
      ** that it took at the start of the transaction. Otherwise, the
      ** calls to sqlite3PagerGet() return zeroed pages instead of 
      ** reading data from the database file.
      **
      ** When journal_mode==OFF the dbOrigSize is always zero, so this
      ** block never runs if journal_mode=OFF.
      */
  #ifndef SQLITE_OMIT_AUTOVACUUM
      if( pPager->dbSize<pPager->dbOrigSize 
       && ALWAYS(pPager->journalMode!=PAGER_JOURNALMODE_OFF)
      ){
        Pgno i;                                   /* Iterator variable */
        const Pgno iSkip = PAGER_MJ_PGNO(pPager); /* Pending lock page */
        const Pgno dbSize = pPager->dbSize;       /* Database image size */ 
        pPager->dbSize = pPager->dbOrigSize;
        for( i=dbSize+1; i<=pPager->dbOrigSize; i++ ){
          if( !sqlite3BitvecTest(pPager->pInJournal, i) && i!=iSkip ){
            PgHdr *pPage;             /* Page to journal */
            rc = sqlite3PagerGet(pPager, i, &pPage);
            if( rc!=SQLITE_OK ) goto commit_phase_one_exit;
            rc = sqlite3PagerWrite(pPage);
            sqlite3PagerUnref(pPage);
            if( rc!=SQLITE_OK ) goto commit_phase_one_exit;
          }
        }
        pPager->dbSize = dbSize;
      } 
  #endif
  
      /* Write the master journal name into the journal file. If a master 
      ** journal file name has already been written to the journal file, 
      ** or if zMaster is NULL (no master journal), then this call is a no-op.
      */
      rc = writeMasterJournal(pPager, zMaster);
      if( rc!=SQLITE_OK ) goto commit_phase_one_exit;
  
      /* Sync the journal file. If the atomic-update optimization is being
      ** used, this call will not create the journal file or perform any
      ** real IO.
      */
      rc = syncJournal(pPager);
      if( rc!=SQLITE_OK ) goto commit_phase_one_exit;
  
      /* Write all dirty pages to the database file. */
      rc = pager_write_pagelist(sqlite3PcacheDirtyList(pPager->pPCache));
      if( rc!=SQLITE_OK ){
        assert( rc!=SQLITE_IOERR_BLOCKED );
        goto commit_phase_one_exit;
      }
      sqlite3PcacheCleanAll(pPager->pPCache);
  
      /* If the file on disk is not the same size as the database image,
      ** then use pager_truncate to grow or shrink the file here.
      */
      if( pPager->dbSize!=pPager->dbFileSize ){
        Pgno nNew = pPager->dbSize - (pPager->dbSize==PAGER_MJ_PGNO(pPager));
        assert( pPager->state>=PAGER_EXCLUSIVE );
        rc = pager_truncate(pPager, nNew);
        if( rc!=SQLITE_OK ) goto commit_phase_one_exit;
      }
  
      /* Finally, sync the database file. */
      if( !pPager->noSync && !noSync ){
        rc = sqlite3OsSync(pPager->fd, pPager->sync_flags);
      }
      IOTRACE(("DBSYNC %p\n", pPager))
    }

    pPager->state = PAGER_SYNCED;
  }

commit_phase_one_exit:
  return rc;
}

**   the journal file. It needs to be left in the file-system so that
**   some other process can use it to restore the database state (by
**   hot-journal rollback).
*/
int sqlite3PagerRollback(Pager *pPager){
  int rc = SQLITE_OK;                  /* Return code */
  PAGERTRACE(("ROLLBACK %d\n", PAGERID(pPager)));
  if( pagerUseLog(pPager) ){
    int rc2;
    rc = sqlite3PagerSavepoint(pPager, SAVEPOINT_ROLLBACK, -1);
    rc2 = pager_end_transaction(pPager, pPager->setMaster);
    if( rc==SQLITE_OK ) rc = rc2;
  }else if( !pPager->dbModified || !isOpen(pPager->jfd) ){
    rc = pager_end_transaction(pPager, pPager->setMaster);
  }else if( pPager->errCode && pPager->errCode!=SQLITE_FULL ){
    if( pPager->state>=PAGER_EXCLUSIVE ){
      pager_playback(pPager, 0);
    }
    rc = pPager->errCode;
  }else{

      }
    }
    /* Else this is a rollback operation, playback the specified savepoint.
    ** If this is a temp-file, it is possible that the journal file has
    ** not yet been opened. In this case there have been no changes to
    ** the database file, so the playback operation can be skipped.
    */
    else if( pagerUseLog(pPager) || isOpen(pPager->jfd) ){
      PagerSavepoint *pSavepoint = (nNew==0)?0:&pPager->aSavepoint[nNew-1];
      rc = pagerPlaybackSavepoint(pPager, pSavepoint);
      assert(rc!=SQLITE_DONE);
    }
  
  }
  return rc;

**
**    PAGER_JOURNALMODE_QUERY
**    PAGER_JOURNALMODE_DELETE
**    PAGER_JOURNALMODE_TRUNCATE
**    PAGER_JOURNALMODE_PERSIST
**    PAGER_JOURNALMODE_OFF
**    PAGER_JOURNALMODE_MEMORY
**    PAGER_JOURNALMODE_WAL
**
** If the parameter is not _QUERY, then the journal_mode is set to the
** value specified if the change is allowed.  The change is disallowed
** for the following reasons:
**
**   *  An in-memory database can only have its journal_mode set to _OFF
**      or _MEMORY.
**
**   *  The journal mode may not be changed while a transaction is active.
**
** The returned indicate the current (possibly updated) journal-mode.
*/
int sqlite3PagerJournalMode(Pager *pPager, int eMode){
  assert( eMode==PAGER_JOURNALMODE_QUERY
            || eMode==PAGER_JOURNALMODE_DELETE
            || eMode==PAGER_JOURNALMODE_TRUNCATE
            || eMode==PAGER_JOURNALMODE_PERSIST
            || eMode==PAGER_JOURNALMODE_OFF 
            || eMode==PAGER_JOURNALMODE_WAL 
            || eMode==PAGER_JOURNALMODE_MEMORY );
  assert( PAGER_JOURNALMODE_QUERY<0 );
  if( eMode>=0
   && (pPager->tempFile==0 || eMode!=PAGER_JOURNALMODE_WAL)
   && (!MEMDB || eMode==PAGER_JOURNALMODE_MEMORY||eMode==PAGER_JOURNALMODE_OFF)

   && !pPager->dbModified
   && (!isOpen(pPager->jfd) || 0==pPager->journalOff)
  ){
    if( isOpen(pPager->jfd) ){
      sqlite3OsClose(pPager->jfd);
    }
    assert( (PAGER_JOURNALMODE_TRUNCATE & 1)==1 );
    assert( (PAGER_JOURNALMODE_PERSIST & 1)==1 );
    assert( (PAGER_JOURNALMODE_DELETE & 1)==0 );
    assert( (PAGER_JOURNALMODE_MEMORY & 1)==0 );
    assert( (PAGER_JOURNALMODE_OFF & 1)==0 );
    if( (pPager->journalMode & 1)==1 && (eMode & 1)==0
         && !pPager->exclusiveMode ){
      sqlite3OsDelete(pPager->pVfs, pPager->zJournal, 0);
    }

    if( eMode==PAGER_JOURNALMODE_WAL ){
      int rc = pagerOpenLog(pPager);
      if( rc!=SQLITE_OK ){
        /* TODO: The error code should not just get dropped here. Change 
        ** this to set a flag to force the log to be opened the first time
        ** it is actually required.  */
        return (int)pPager->journalMode;
      }
    }
    pPager->journalMode = (u8)eMode;
  }
  return (int)pPager->journalMode;
}

/*
** Get/set the size-limit used for persistent journal files.

** in backup.c maintains the content of this variable. This module
** uses it opaquely as an argument to sqlite3BackupRestart() and
** sqlite3BackupUpdate() only.
*/
sqlite3_backup **sqlite3PagerBackupPtr(Pager *pPager){
  return &pPager->pBackup;
}

/*
** This function is called when the user invokes "PRAGMA checkpoint".
*/
int sqlite3PagerCheckpoint(Pager *pPager, int nMin, int nMax, int doSync){
  int rc = SQLITE_OK;
  if( pPager->pLog ){
    rc = pager_wait_on_lock(pPager, EXCLUSIVE_LOCK);
    if( rc==SQLITE_OK ){
      u8 *zBuf = (u8 *)pPager->pTmpSpace;
      rc = sqlite3LogCheckpoint(pPager->pLog, pPager->fd, zBuf);
    }
  }
  return rc;
}

#endif /* SQLITE_OMIT_DISKIO */

*/
#define PAGER_JOURNALMODE_QUERY      -1
#define PAGER_JOURNALMODE_DELETE      0   /* Commit by deleting journal file */
#define PAGER_JOURNALMODE_PERSIST     1   /* Commit by zeroing journal header */
#define PAGER_JOURNALMODE_OFF         2   /* Journal omitted.  */
#define PAGER_JOURNALMODE_TRUNCATE    3   /* Commit by truncating journal */
#define PAGER_JOURNALMODE_MEMORY      4   /* In-memory journal file */
#define PAGER_JOURNALMODE_WAL         5   /* Use write-ahead logging */

/*
** The remainder of this file contains the declarations of the functions
** that make up the Pager sub-system API. See source code comments for 
** a detailed description of each routine.
*/

        sqlite3BtreeSecureDelete(db->aDb[ii].pBt, b);
      }
    }
    b = sqlite3BtreeSecureDelete(pBt, b);
    returnSingleInt(pParse, "secure_delete", b);
  }else

  /*
  **  PRAGMA [database.]secure_delete
  **  PRAGMA [database.]secure_delete=ON/OFF
  **
  ** The first form reports the current setting for the
  ** secure_delete flag.  The second form changes the secure_delete
  ** flag setting and reports thenew value.
  */
  if( sqlite3StrICmp(zLeft,"secure_delete")==0 ){
    Btree *pBt = pDb->pBt;
    int b = -1;
    assert( pBt!=0 );
    if( zRight ){
      b = getBoolean(zRight);
    }
    if( pId2->n==0 && b>=0 ){
      int ii;
      for(ii=0; ii<db->nDb; ii++){
        sqlite3BtreeSecureDelete(db->aDb[ii].pBt, b);
      }
    }
    b = sqlite3BtreeSecureDelete(pBt, b);
    returnSingleInt(pParse, "secure_delete", b);
  }else

  /*
  **  PRAGMA [database.]page_count
  **
  ** Return the number of pages in the specified database.
  */
  if( sqlite3StrICmp(zLeft,"page_count")==0 ){
    int iReg;

  /*
  **  PRAGMA [database.]journal_mode
  **  PRAGMA [database.]journal_mode = (delete|persist|off|truncate|memory)
  */
  if( sqlite3StrICmp(zLeft,"journal_mode")==0 ){
    int eMode;
    static char * const azModeName[] = {
      "delete", "persist", "off", "truncate", "memory", "wal"
    };

    if( zRight==0 ){
      eMode = PAGER_JOURNALMODE_QUERY;
    }else{
      int n = sqlite3Strlen30(zRight);
      eMode = sizeof(azModeName)/sizeof(azModeName[0]) - 1;

      pPager = sqlite3BtreePager(pDb->pBt);
      eMode = sqlite3PagerJournalMode(pPager, eMode);
    }
    assert( eMode==PAGER_JOURNALMODE_DELETE
              || eMode==PAGER_JOURNALMODE_TRUNCATE
              || eMode==PAGER_JOURNALMODE_PERSIST
              || eMode==PAGER_JOURNALMODE_OFF
              || eMode==PAGER_JOURNALMODE_WAL
              || eMode==PAGER_JOURNALMODE_MEMORY );
    sqlite3VdbeSetNumCols(v, 1);
    sqlite3VdbeSetColName(v, 0, COLNAME_NAME, "journal_mode", SQLITE_STATIC);
    sqlite3VdbeAddOp4(v, OP_String8, 0, 1, 0, 
           azModeName[eMode], P4_STATIC);
    sqlite3VdbeAddOp2(v, OP_ResultRow, 1, 1);
  }else

    sqlite3VdbeSetColName(v, 0, COLNAME_NAME, "compile_option", SQLITE_STATIC);
    while( (zOpt = sqlite3_compileoption_get(i++))!=0 ){
      sqlite3VdbeAddOp4(v, OP_String8, 0, 1, 0, zOpt, 0);
      sqlite3VdbeAddOp2(v, OP_ResultRow, 1, 1);
    }
  }else
#endif /* SQLITE_OMIT_COMPILEOPTION_DIAGS */

  if( sqlite3StrICmp(zLeft, "checkpoint")==0 ){
    int nMin = 0;
    int nMax = 0;
    int nosync = 0;

    if( zRight ){
      char *z = zRight;
      sqlite3GetInt32(z, &nMin);
      while( sqlite3Isdigit(*z) ) z++;
      while( *z && !sqlite3Isdigit(*z) ) z++;
      sqlite3GetInt32(z, &nMax);
      while( sqlite3Isdigit(*z) ) z++;
      while( *z && !sqlite3Isdigit(*z) ) z++;
      sqlite3GetInt32(z, &nosync);
    }
    sqlite3VdbeUsesBtree(v, iDb);
    sqlite3VdbeAddOp2(v, OP_Transaction, iDb, 1);
    sqlite3VdbeAddOp3(v, OP_Checkpoint, iDb, nMin, nMax);
  }else

#if defined(SQLITE_DEBUG) || defined(SQLITE_TEST)
  /*
  ** Report the current state of file logs for all databases
  */
  if( sqlite3StrICmp(zLeft, "lock_status")==0 ){
    static const char *const azLockName[] = {

  UPDATE_MAX_BLOBSIZE(pMem);
  if( sqlite3VdbeMemTooBig(pMem) ){
    goto too_big;
  }
  break;
}

/* Opcode: Checkpoint P1 P2 P3 * P5
*/
case OP_Checkpoint: {
  Btree *pBt;                     /* Btree to checkpoint */
  int nMin = pOp->p2;             /* Minimum number of pages to copy */
  int nMax = pOp->p3;             /* Maximum number of pages to copy */
  int doNotSync = pOp->p5;        /* True to sync database */

  assert( pOp->p1>=0 && pOp->p1<db->nDb );
  assert( (p->btreeMask & (1<<pOp->p1))!=0 );
  pBt = db->aDb[pOp->p1].pBt;

  rc = sqlite3PagerCheckpoint(sqlite3BtreePager(pBt), nMin, nMax, !doNotSync);
  break;
};  

#if !defined(SQLITE_OMIT_VACUUM) && !defined(SQLITE_OMIT_ATTACH)
/* Opcode: Vacuum * * * * *
**
** Vacuum the entire database.  This opcode will cause other virtual
** machines to be created and run.  It may not be called from within
** a transaction.

set testdir [file dirname $argv0]
source $testdir/tester.tcl

proc range {args} {
  set ret [list]
  foreach {start end} $args {
    for {set i $start} {$i <= $end} {incr i} {
      lappend ret $i
    }
  }
  set ret
}

proc reopen_db {} {
  db close
  file delete -force test.db test.db-wal
  sqlite3_wal db test.db
  #register_logtest
}
proc register_logtest {{db db}} {
  register_logsummary_module $db
  execsql { CREATE VIRTUAL TABLE temp.logsummary USING logsummary } $db
  execsql { CREATE VIRTUAL TABLE temp.logcontent USING logcontent } $db
  execsql { CREATE VIRTUAL TABLE temp.loglock USING loglock } $db
}

proc sqlite3_wal {args} {
  eval sqlite3 $args
  [lindex $args 0] eval { PRAGMA journal_mode = wal }
}

#
# These are 'warm-body' tests used while developing the WAL code. They
# serve to prove that a few really simple cases work:
#
# wal-1.*: Read and write the database.
# wal-2.*: Test MVCC with one reader, one writer.
# wal-3.*: Test transaction rollback.
# wal-4.*: Test savepoint/statement rollback.
# wal-5.*: Test the temp database.
# wal-6.*: Test creating databases with different page sizes.
#

do_test wal-0.1 {
  execsql { PRAGMA journal_mode = wal }
} {wal}

do_test wal-1.0 {
  execsql { 
    BEGIN;
    CREATE TABLE t1(a, b); 
  }
  list [file exists test.db-journal] [file exists test.db-wal]
} {0 1}
do_test wal-1.1 {
  execsql COMMIT
  list [file exists test.db-journal] [file exists test.db-wal]
} {0 1}
do_test wal-1.2 {
  # There are now two pages in the log.
  file size test.db-wal
} [expr (20+1024)*2]

do_test wal-1.3 {
  execsql { SELECT * FROM sqlite_master }
} {table t1 t1 2 {CREATE TABLE t1(a, b)}}

do_test wal-1.4 {
  execsql { INSERT INTO t1 VALUES(1, 2) }
  execsql { INSERT INTO t1 VALUES(3, 4) }
  execsql { INSERT INTO t1 VALUES(5, 6) }
  execsql { INSERT INTO t1 VALUES(7, 8) }
  execsql { INSERT INTO t1 VALUES(9, 10) }
} {}

do_test wal-1.5 {
  execsql { SELECT * FROM t1 }
} {1 2 3 4 5 6 7 8 9 10}

do_test wal-2.1 {
  sqlite3_wal db2 ./test.db
  execsql { BEGIN; SELECT * FROM t1 } db2
} {1 2 3 4 5 6 7 8 9 10}

do_test wal-2.2 {
  execsql { INSERT INTO t1 VALUES(11, 12) }
  execsql { SELECT * FROM t1 }
} {1 2 3 4 5 6 7 8 9 10 11 12}

do_test wal-2.3 {
  execsql { SELECT * FROM t1 } db2
} {1 2 3 4 5 6 7 8 9 10}

do_test wal-2.4 {
  execsql { INSERT INTO t1 VALUES(13, 14) }
  execsql { SELECT * FROM t1 }
} {1 2 3 4 5 6 7 8 9 10 11 12 13 14}

do_test wal-2.5 {
  execsql { SELECT * FROM t1 } db2
} {1 2 3 4 5 6 7 8 9 10}

do_test wal-2.6 {
  execsql { COMMIT; SELECT * FROM t1 } db2
} {1 2 3 4 5 6 7 8 9 10 11 12 13 14}

do_test wal-3.1 {
  execsql { BEGIN; DELETE FROM t1 }
  execsql { SELECT * FROM t1 }
} {}
do_test wal-3.2 {
  execsql { SELECT * FROM t1 } db2
} {1 2 3 4 5 6 7 8 9 10 11 12 13 14}
do_test wal-3.3 {
  execsql { ROLLBACK }
  execsql { SELECT * FROM t1 }
} {1 2 3 4 5 6 7 8 9 10 11 12 13 14}
db2 close

do_test wal-4.1 {
  execsql {
    DELETE FROM t1;
    BEGIN;
      INSERT INTO t1 VALUES('a', 'b');
      SAVEPOINT sp;
        INSERT INTO t1 VALUES('c', 'd');
        SELECT * FROM t1;
  }
} {a b c d}
do_test wal-4.2 {
  execsql {
      ROLLBACK TO sp;
      SELECT * FROM t1;
  }
} {a b}
do_test wal-4.3 {
  execsql {
    COMMIT;
    SELECT * FROM t1;
  }
} {a b}

do_test wal-5.1 {
  execsql {
    CREATE TEMP TABLE t2(a, b);
    INSERT INTO t2 VALUES(1, 2);
  }
} {}
do_test wal-5.2 {
  execsql {
    BEGIN;
      INSERT INTO t2 VALUES(3, 4);
      SELECT * FROM t2;
  }
} {1 2 3 4}
do_test wal-5.3 {
  execsql {
    ROLLBACK;
    SELECT * FROM t2;
  }
} {1 2}
do_test wal-5.4 {
  execsql {
    CREATE TEMP TABLE t3(x UNIQUE);
    BEGIN;
      INSERT INTO t2 VALUES(3, 4);
      INSERT INTO t3 VALUES('abc');
  }
  catchsql { INSERT INTO t3 VALUES('abc') }
} {1 {column x is not unique}}
do_test wal-5.5 {
  execsql {
    COMMIT;
    SELECT * FROM t2;
  }
} {1 2 3 4}
db close


foreach sector {512 4096} {
  sqlite3_simulate_device -sectorsize $sector
  foreach pgsz {512 1024 2048 4096} {
    file delete -force test.db test.db-wal
    do_test wal-6.$sector.$pgsz.1 {
      sqlite3_wal db test.db -vfs devsym
      execsql "
        PRAGMA page_size = $pgsz ;
      "
      execsql "
        CREATE TABLE t1(a, b);
        INSERT INTO t1 VALUES(1, 2);
      "
      db close
      file size test.db
    } [expr $pgsz*2]
  
    do_test wal-6.$sector.$pgsz.2 {
      file size test.db-wal
    } {0}
  }
}

do_test wal-7.1 {
  file delete -force test.db test.db-wal
  sqlite3_wal db test.db
  execsql {
    PRAGMA page_size = 1024;
    CREATE TABLE t1(a, b);
    INSERT INTO t1 VALUES(1, 2);
  }

  list [file size test.db] [file size test.db-wal]
} [list 0 [expr (1024+20)*3]]
do_test wal-7.2 {
  execsql { PRAGMA checkpoint }
  list [file size test.db] [file size test.db-wal]
} [list 2048 [expr (1024+20)*3]]

# db close
# sqlite3_wal db test.db
# register_logsummary_module db
# # Warm-body tests of the virtual tables used for testing.
# # 
# do_test wal-8.1 {
#   execsql { CREATE VIRTUAL TABLE temp.logsummary USING logsummary }
#   execsql { CREATE VIRTUAL TABLE temp.logcontent USING logcontent }
#   execsql { CREATE VIRTUAL TABLE temp.loglock USING loglock }
#   execsql { SELECT * FROM logsummary }
# } [list [file join [pwd] test.db] 1 0 0 0 0 0 0]
# 
# do_test wal-8.2 {
#   sqlite3_wal db2 test.db
#   execsql { SELECT * FROM logsummary }
# } [list [file join [pwd] test.db] 2 0 0 0 0 0 0]
# do_test wal-8.3 {
#   db2 close
#   execsql { SELECT * FROM logsummary }
# } [list [file join [pwd] test.db] 1 0 0 0 0 0 0]
# do_test wal-8.4 {
#   execsql { INSERT INTO t1 VALUES(3, 4) }
#   execsql { SELECT * FROM logsummary }
# } [list [file join [pwd] test.db] 1 1024 2 2 3 0 0]
# do_test wal-8.5 {
#   execsql { PRAGMA checkpoint }
#   execsql { SELECT * FROM logsummary }
# } [list [file join [pwd] test.db] 1 1024 0 0 0 0 0]
# 
# do_test wal-8.6 {
#   execsql { INSERT INTO t1 VALUES(5, 6) }
#   execsql { PRAGMA checkpoint('1 1 1') }
#   execsql { SELECT * FROM logsummary }
# } [list [file join [pwd] test.db] 1 1024 2 0 3 0 0]
# do_test wal-8.7 {
#   execsql { SELECT logpage, dbpage FROM logcontent }
# } {}
# do_test wal-8.8 {
#   execsql { INSERT INTO t1 VALUES(7, 8) }
#   execsql { SELECT logpage, dbpage FROM logcontent }
# } {4 T:4 5 2}
# do_test wal-8.9 {
#   execsql { SELECT * FROM logsummary }
# } [list [file join [pwd] test.db] 1 1024 2 4 5 0 0]
# do_test wal-8.10 {
#   execsql { SELECT * FROM loglock }
# } [list [file join [pwd] test.db] 0 0 0]
# do_test wal-8.11 {
#   execsql { BEGIN; SELECT * FROM t1; }
#   execsql { SELECT * FROM loglock }
# } [list [file join [pwd] test.db] 0 0 4]
# 
# # Try making the log wrap around.
# #
# reopen_db
# 
# do_test wal-9.1 {
#   execsql {
#     BEGIN;
#     CREATE TABLE t1(a PRIMARY KEY, b);
#   }
#   for {set i 0} {$i < 100} {incr i} {
#     execsql { INSERT INTO t1 VALUES($i, randomblob(100)) }
#   }
#   execsql COMMIT
#   execsql { SELECT * FROM logsummary }
# } [list [file join [pwd] test.db] 1 1024 2 2 17 0 0]
# do_test wal-9.2 {
#   execsql { SELECT logpage, dbpage FROM logcontent }
# } {2 T:2 3 1 4 2 5 3 6 4 7 5 8 6 9 7 10 8 11 9 12 10 13 11 14 12 15 13 16 14 17 15}
# do_test wal-9.3 {
#   execsql { PRAGMA checkpoint('12, 12') }
#   execsql { SELECT * FROM logsummary }
# } [list [file join [pwd] test.db] 1 1024 2 15 17 0 0]
# do_test wal-9.4 {
#   execsql { SELECT logpage, dbpage FROM logcontent }
# } {15 13 16 14 17 15}
# do_test wal-9.5 {
#   execsql { SELECT count(*) FROM t1 }
# } {100}
# do_test wal-9.6 {
#   execsql { INSERT INTO t1 VALUES(100, randomblob(100)) }
#   execsql { SELECT * FROM logsummary }
# } [list [file join [pwd] test.db] 1 1024 2 15 20 0 0]
# 
# do_test wal-9.7 {
#   execsql { SELECT count(*) FROM t1 }
# } {101}
# do_test wal-9.8 {
#   db close
#   sqlite3_wal db test.db
#   register_logtest
#   execsql { SELECT count(*) FROM t1 }
# } {101}
# do_test wal-9.9 {
#   execsql { SELECT * FROM logsummary }
# } [list [file join [pwd] test.db] 1 0 0 0 0 0 0]
# 
# reopen_db
# do_test wal-10.1 {
#   execsql {
#     PRAGMA page_size = 1024;
#     CREATE TABLE t1(x PRIMARY KEY);
#     INSERT INTO t1 VALUES(randomblob(900));
#     INSERT INTO t1 VALUES(randomblob(900));
#     INSERT INTO t1 SELECT randomblob(900) FROM t1;         -- 4
#     INSERT INTO t1 SELECT randomblob(900) FROM t1;         -- 8
#     INSERT INTO t1 SELECT randomblob(900) FROM t1;         -- 16
#   }
#   list [file size test.db] [file size test.db-wal]
# } {0 55296}
# do_test wal-10.2 {
#   execsql { PRAGMA checkpoint('20 30') }
# } {}
# do_test wal-10.3 {
#   execsql { SELECT * FROM logsummary }
# } [list [file join [pwd] test.db] 1 1024 34 38 54 0 0]
# do_test wal-10.4 {
#   execsql { SELECT dbpage FROM logcontent }
# } {21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37}
# do_test wal-10.5 {
#   execsql { INSERT INTO t1 VALUES(randomblob(900)) }
#   execsql { SELECT * FROM logsummary }
# } [list [file join [pwd] test.db] 1 1024 34 38 6 34 54]
# do_test wal-10.6 {
#   execsql { SELECT count(*) FROM t1 WHERE x NOT NULL }
# } {17}
# do_test wal-10.8 {
#   execsql { SELECT logpage FROM logcontent }
# } [range 38 54  1 6]
# do_test wal-10.9 {
#   execsql { INSERT INTO t1 SELECT randomblob(900) FROM t1 }
#   execsql { SELECT * FROM logsummary }
# } [list [file join [pwd] test.db] 1 1024 34 38 68 34 54]
# 
# do_test wal-10.10 {
#   execsql { SELECT logpage FROM logcontent }
# } [range 38 54  1 33  55 68]
# 
# do_test wal-10.11 {
#   execsql { SELECT count(*) FROM t1 WHERE x NOT NULL }
# } {34}
# 
# do_test wal-10.12 {
#   execsql { PRAGMA checkpoint('35 35') }
# } {}
# do_test wal-10.13 {
#   execsql { SELECT logpage FROM logcontent }
# } [range 22 68]
# do_test wal-10.13a {
#   execsql { SELECT dbpage FROM logcontent }
# } [list \
#   50 51 52 53 54 55 56 57 58 59 60 61    \
#   0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 \
#   62 63 64 65 66 67 68 69 70 71 72 73 74 75 \
# ]
# 
# do_test wal-10.14 {
#   execsql { SELECT count(*) FROM t1 WHERE x NOT NULL }
# } {34}
# do_test wal-10.15 {
#   execsql { PRAGMA integrity_check }
# } {ok}
# do_test wal-10.16 {
#   execsql { PRAGMA checkpoint('20 20') }
#   execsql { SELECT * FROM logsummary }
# } [list [file join [pwd] test.db] 1 1024 7 63 68 0 0]
# do_test wal-10.17 {
#   execsql { SELECT logpage FROM logcontent }
# } [range 63 68]
# do_test wal-10.17a {
#   execsql { SELECT dbpage FROM logcontent }
# } {70 71 72 73 74 75}
# 
# do_test wal-10.18 {
#   execsql { INSERT INTO t1 SELECT randomblob(900) FROM t1 }
#   execsql { SELECT logpage FROM logcontent }
# } [range 63 147]
# integrity_check wal-10.19
# 
# do_test wal-10.20 {
#   execsql { PRAGMA checkpoint('52 52') }
#   execsql { SELECT logpage FROM logcontent }
# } [range 116 147]
# do_test wal-10.20a {
#   execsql { SELECT * FROM logsummary }
# } [list [file join [pwd] test.db] 1 1024 69 116 147 0 0]
# integrity_check wal-10.20.integrity
# 
# do_test wal-10.21 {
#   execsql { INSERT INTO t1 VALUES( randomblob(900) ) }
#   execsql { SELECT logpage FROM logcontent }
# } [range 116 152]
# do_test wal-10.22 {
#   execsql { PRAGMA integrity_check }
# } {ok}
# 
# file delete -force testX.db testX.db-wal
# file copy test.db testX.db
# file copy test.db-wal testX.db-wal
# do_test wal-10.23 {
#   sqlite3_wal db2 testX.db
#   register_logtest db2
#   execsql { SELECT logpage FROM logcontent WHERE db LIKE '%testX%' } db2
# } [range 34 54  1 33  55 152]
# 
# do_test wal-10.24 {
#   execsql { PRAGMA integrity_check } db2
# } {ok}
# db2 close
# 
# do_test wal-11.1 {
#   reopen_db
#   sqlite3_wal db2 test.db
# 
#   execsql {
#     BEGIN;
#       CREATE TABLE t1(x);
#       CREATE TABLE t2(x PRIMARY KEY);
#       INSERT INTO t1 VALUES(randomblob(900));
#       INSERT INTO t1 VALUES(randomblob(900));
#       INSERT INTO t1 SELECT randomblob(900) FROM t1;       /*  4 */
#       INSERT INTO t1 SELECT randomblob(900) FROM t1;       /*  8 */
#       INSERT INTO t1 SELECT randomblob(900) FROM t1;       /* 16 */
#       INSERT INTO t1 SELECT randomblob(900) FROM t1;       /* 32 */
#       INSERT INTO t1 SELECT randomblob(900) FROM t1;       /* 64 */
# 
#       INSERT INTO t2 VALUES('x');
#       INSERT INTO t2 VALUES('y');
#       INSERT INTO t2 VALUES('z');
#     COMMIT;
#     SELECT * FROM logsummary;
#   }
# } [list [file join [pwd] test.db] 2 1024 2 2 70 0 0]
# 
# do_test wal-11.2 {
#   execsql {
#     BEGIN; SELECT x FROM t2;
#   } db2
# } {x y z}
# do_test wal-11.2 {
#   execsql {
#     INSERT INTO t1 VALUES(randomblob(900));
#     PRAGMA checkpoint('10 100');
#     INSERT INTO t1 VALUES(randomblob(900));
#     INSERT INTO t2 VALUES('0');
#     SELECT * FROM logsummary;
#   }
# } [list [file join [pwd] test.db] 2 1024 71 71 7 71 73]
# do_test wal-12.3 {
#   execsql { PRAGMA integrity_check } db2
# } {ok}
# db2 close


# Execute some transactions in auto-vacuum mode to test database file
# truncation.
#
do_test wal-12.1 {
  reopen_db
  execsql {
    PRAGMA auto_vacuum = 1;
    PRAGMA auto_vacuum;
  }
} {1}
do_test wal-12.2 {
  execsql {
    PRAGMA page_size = 1024;
    CREATE TABLE t1(x);
    INSERT INTO t1 VALUES(randomblob(900));
    INSERT INTO t1 VALUES(randomblob(900));
    INSERT INTO t1 SELECT randomblob(900) FROM t1;       /*  4 */
    INSERT INTO t1 SELECT randomblob(900) FROM t1;       /*  8 */
    INSERT INTO t1 SELECT randomblob(900) FROM t1;       /* 16 */
    INSERT INTO t1 SELECT randomblob(900) FROM t1;       /* 32 */
    INSERT INTO t1 SELECT randomblob(900) FROM t1;       /* 64 */
    PRAGMA checkpoint;
  }
  file size test.db
} [expr 67*1024]
do_test wal-12.3 {
  execsql { 
    DELETE FROM t1 WHERE rowid<54;
    PRAGMA checkpoint('1 100000');
  }
  file size test.db
} [expr 14*1024]

# Run some "warm-body" tests to ensure that log-summary files with more
# than 256 entries (log summaries that contain index blocks) work Ok.
#
do_test wal-13.1 {
  reopen_db
  execsql {
    PRAGMA page_size = 1024;
    CREATE TABLE t1(x PRIMARY KEY);
    INSERT INTO t1 VALUES(randomblob(900));
    INSERT INTO t1 VALUES(randomblob(900));
    INSERT INTO t1 SELECT randomblob(900) FROM t1;       /*  4 */
    INSERT INTO t1 SELECT randomblob(900) FROM t1;       /*  8 */
    INSERT INTO t1 SELECT randomblob(900) FROM t1;       /* 16 */
    INSERT INTO t1 SELECT randomblob(900) FROM t1;       /* 32 */
    INSERT INTO t1 SELECT randomblob(900) FROM t1;       /* 64 */
    INSERT INTO t1 SELECT randomblob(900) FROM t1;       /* 128 */
    INSERT INTO t1 SELECT randomblob(900) FROM t1;       /* 256 */
  }
  file size test.db
} 0
do_test wal-13.2 {
  sqlite3_wal db2 test.db
  execsql {PRAGMA integrity_check } db2
} {ok}

do_test wal-13.3 {
  file delete -force test2.db test2.db-wal
  file copy test.db test2.db
  file copy test.db-wal test2.db-wal
  sqlite3_wal db3 test2.db 
  execsql {PRAGMA integrity_check } db3
} {ok}
db3 close

do_test wal-13.4 {
breakpoint
  execsql { PRAGMA checkpoint }
  db2 close
  sqlite3_wal db2 test.db
  execsql {PRAGMA integrity_check } db2
} {ok}

finish_test

# 2010 February 8
#
# 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 regression tests for SQLite library.
#

#
# These are 'warm-body' tests of database recovery used while developing 
# the WAL code. They serve to prove that a few really simple cases work:
#
# walcrash-1.*: Recover a database.
# walcrash-2.*: Recover a database where the failed transaction spanned more
#               than one page.
# walcrash-3.*: Recover multiple databases where the failed transaction 
#               was a multi-file transaction.
#

set testdir [file dirname $argv0]
source $testdir/tester.tcl
db close

set seed 0
set REPEATS 100

proc sqlite3_wal {args} {
  eval sqlite3 $args
  [lindex $args 0] eval { PRAGMA journal_mode = wal }
}

# walcrash-1.*
#
for {set i 1} {$i < $REPEATS} {incr i} {
  file delete -force test.db test.db-wal
  do_test walcrash-1.$i.1 {
    crashsql -delay 4 -file test.db-wal -seed [incr seed] {
      PRAGMA journal_mode = WAL;
      CREATE TABLE t1(a, b);
      INSERT INTO t1 VALUES(1, 1);
      INSERT INTO t1 VALUES(2, 3);
      INSERT INTO t1 VALUES(3, 6);
    }
  } {1 {child process exited abnormally}}
  do_test walcrash-1.$i.2 {
    sqlite3_wal db test.db
    execsql { SELECT sum(a)==max(b) FROM t1 }
  } {1}
  integrity_check walcrash-1.$i.3
  db close
  
  do_test walcrash-1.$i.4 {
    crashsql -delay 2 -file test.db-wal -seed [incr seed] {
      PRAGMA journal_mode = WAL;
      PRAGMA journal_mode = WAL;
      INSERT INTO t1 VALUES(4, (SELECT sum(a) FROM t1) + 4);
      INSERT INTO t1 VALUES(5, (SELECT sum(a) FROM t1) + 5);
    }
  } {1 {child process exited abnormally}}
  do_test walcrash-1.$i.5 {
    sqlite3_wal db test.db
    execsql { SELECT sum(a)==max(b) FROM t1 }
  } {1}
  integrity_check walcrash-1.$i.6
  db close
}

# walcrash-2.*
#
for {set i 1} {$i < $REPEATS} {incr i} {
  file delete -force test.db test.db-wal
  do_test walcrash-2.$i.1 {
    crashsql -delay 4 -file test.db-wal -seed [incr seed] {
      PRAGMA journal_mode = WAL;
      CREATE TABLE t1(a PRIMARY KEY, b);
      INSERT INTO t1 VALUES(1, 2);
      INSERT INTO t1 VALUES(3, 4);
      INSERT INTO t1 VALUES(5, 9);
    }
  } {1 {child process exited abnormally}}
  do_test walcrash-2.$i.2 {
    sqlite3_wal db test.db
    execsql { SELECT sum(a)==max(b) FROM t1 }
  } {1}
  integrity_check walcrash-2.$i.3
  db close
  
  do_test walcrash-2.$i.4 {
    crashsql -delay 2 -file test.db-wal -seed [incr seed] {
      PRAGMA journal_mode = WAL;
      INSERT INTO t1 VALUES(6, (SELECT sum(a) FROM t1) + 6);
      INSERT INTO t1 VALUES(7, (SELECT sum(a) FROM t1) + 7);
    }
  } {1 {child process exited abnormally}}
  do_test walcrash-2.$i.5 {
    sqlite3_wal db test.db
    execsql { SELECT sum(a)==max(b) FROM t1 }
  } {1}
  integrity_check walcrash-2.$i.6
  db close
}

# walcrash-3.*
#
# for {set i 1} {$i < $REPEATS} {incr i} {
#   file delete -force test.db test.db-wal
#   file delete -force test2.db test2.db-wal
# 
#   do_test walcrash-3.$i.1 {
#     crashsql -delay 2 -file test2.db-wal -seed [incr seed] {
#       PRAGMA journal_mode = WAL;
#       ATTACH 'test2.db' AS aux;
#       CREATE TABLE t1(a PRIMARY KEY, b);
#       CREATE TABLE aux.t2(a PRIMARY KEY, b);
#       BEGIN;
#         INSERT INTO t1 VALUES(1, 2);
#         INSERT INTO t2 VALUES(1, 2);
#       COMMIT;
#     }
#   } {1 {child process exited abnormally}}
# 
#   do_test walcrash-3.$i.2 {
#     sqlite3_wal db test.db
#     execsql { 
#       ATTACH 'test2.db' AS aux;
#       SELECT * FROM t1 EXCEPT SELECT * FROM t2;
#     }
#   } {}
#   do_test walcrash-3.$i.3 { execsql { PRAGMA main.integrity_check } } {ok}
#   do_test walcrash-3.$i.4 { execsql { PRAGMA aux.integrity_check  } } {ok}
# 
#   db close
# }

# walcrash-4.*
#
for {set i 1} {$i < $REPEATS} {incr i} {
  file delete -force test.db test.db-wal
  file delete -force test2.db test2.db-wal

  do_test walcrash-4.$i.1 {
    crashsql -delay 3 -file test.db-wal -seed [incr seed] -blocksize 4096 {
      PRAGMA journal_mode = WAL;
      PRAGMA page_size = 1024;
      CREATE TABLE t1(a PRIMARY KEY, b);
      INSERT INTO t1 VALUES(1, 2);
      INSERT INTO t1 VALUES(3, 4);
    }
  } {1 {child process exited abnormally}}

  do_test walcrash-4.$i.2 {
    sqlite3_wal db test.db
    execsql { 
      SELECT * FROM t1 WHERE a = 1;
    }
  } {1 2}
  do_test walcrash-4.$i.3 { execsql { PRAGMA main.integrity_check } } {ok}

  db close
}

# walcrash-5.*
#
for {set i 1} {$i < $REPEATS} {incr i} {
  file delete -force test.db test.db-wal
  file delete -force test2.db test2.db-wal

  do_test walcrash-5.$i.1 {
    crashsql -delay 11 -file test.db-wal -seed [incr seed] -blocksize 4096 {
      PRAGMA journal_mode = WAL;
      PRAGMA page_size = 1024;
      BEGIN;
        CREATE TABLE t1(x PRIMARY KEY);
        INSERT INTO t1 VALUES(randomblob(900));
        INSERT INTO t1 VALUES(randomblob(900));
        INSERT INTO t1 SELECT randomblob(900) FROM t1;           /* 4 */
      COMMIT;
      INSERT INTO t1 SELECT randomblob(900) FROM t1 LIMIT 4;   /* 8 */
      INSERT INTO t1 SELECT randomblob(900) FROM t1 LIMIT 4;   /* 12 */
      INSERT INTO t1 SELECT randomblob(900) FROM t1 LIMIT 4;   /* 16 */
      INSERT INTO t1 SELECT randomblob(900) FROM t1 LIMIT 4;   /* 20 */
      INSERT INTO t1 SELECT randomblob(900) FROM t1 LIMIT 4;   /* 24 */
      INSERT INTO t1 SELECT randomblob(900) FROM t1 LIMIT 4;   /* 28 */
      INSERT INTO t1 SELECT randomblob(900) FROM t1 LIMIT 4;   /* 32 */

      PRAGMA checkpoint('70 70');
      INSERT INTO t1 VALUES(randomblob(900));
      INSERT INTO t1 VALUES(randomblob(900));
      INSERT INTO t1 VALUES(randomblob(900));
    }
  } {1 {child process exited abnormally}}

  do_test walcrash-5.$i.2 {
    sqlite3_wal db test.db
    execsql { SELECT count(*)==33 OR count(*)==34 FROM t1 WHERE x != 1 }
  } {1}
  do_test walcrash-5.$i.3 { execsql { PRAGMA main.integrity_check } } {ok}

  db close
}

# walcrash-6.*
#
for {set i 1} {$i < $REPEATS} {incr i} {
  file delete -force test.db test.db-wal
  file delete -force test2.db test2.db-wal

  do_test walcrash-6.$i.1 {
    crashsql -delay 12 -file test.db-wal -seed [incr seed] -blocksize 512 {
      PRAGMA journal_mode = WAL;
      PRAGMA page_size = 1024;
      BEGIN;
        CREATE TABLE t1(x PRIMARY KEY);
        INSERT INTO t1 VALUES(randomblob(900));
        INSERT INTO t1 VALUES(randomblob(900));
        INSERT INTO t1 SELECT randomblob(900) FROM t1;           /* 4 */
      COMMIT;
      INSERT INTO t1 SELECT randomblob(900) FROM t1 LIMIT 4;   /* 8 */
      INSERT INTO t1 SELECT randomblob(900) FROM t1 LIMIT 4;   /* 12 */
      INSERT INTO t1 SELECT randomblob(900) FROM t1 LIMIT 4;   /* 16 */
      INSERT INTO t1 SELECT randomblob(900) FROM t1 LIMIT 4;   /* 20 */
      INSERT INTO t1 SELECT randomblob(900) FROM t1 LIMIT 4;   /* 24 */
      INSERT INTO t1 SELECT randomblob(900) FROM t1 LIMIT 4;   /* 28 */
      INSERT INTO t1 SELECT randomblob(900) FROM t1 LIMIT 4;   /* 32 */

      PRAGMA checkpoint('70 70');
      INSERT INTO t1 VALUES(randomblob(900));
      INSERT INTO t1 VALUES(randomblob(900));
      INSERT INTO t1 VALUES(randomblob(900));
    }
  } {1 {child process exited abnormally}}

  do_test walcrash-6.$i.2 {
    sqlite3_wal db test.db
    execsql { SELECT count(*)==34 OR count(*)==35 FROM t1 WHERE x != 1 }
  } {1}
  do_test walcrash-6.$i.3 { execsql { PRAGMA main.integrity_check } } {ok}

  db close
}

finish_test

# 2010 March 17
#
# 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.
#
#***********************************************************************
#

set testdir [file dirname $argv0]
source $testdir/tester.tcl

proc reopen_db {} {
  catch { db close }
  file delete -force test.db test.db-wal
  sqlite3 db test.db
  execsql { PRAGMA journal_mode = wal }
}

db close
save_prng_state
for {set seed 1} {$seed<10} {incr seed} {
  expr srand($seed)
  restore_prng_state
  reopen_db
  do_test walslow-1.seed=$seed.0 {
    execsql { CREATE TABLE t1(a, b) }
    execsql { CREATE INDEX i1 ON t1(a) }
    execsql { CREATE INDEX i2 ON t1(b) }
  } {}

  for {set iTest 1} {$iTest < 100} {incr iTest} {

    do_test walslow-1.seed=$seed.$iTest.1 {
      set w [expr int(rand()*2000)]
      set x [expr int(rand()*2000)]
      set y [expr int(rand()*9)+1]
      set z [expr int(rand()*2)]
      execsql { INSERT INTO t1 VALUES(randomblob($w), randomblob($x)) }
      execsql { PRAGMA integrity_check }
    } {ok}

    do_test walslow-1.seed=$seed.$iTest.2 {
      execsql "PRAGMA checkpoint('$y $y $z')"
      execsql { PRAGMA integrity_check }
    } {ok}

    do_test walslow-1.seed=$seed.$iTest.3 {
      file delete -force testX.db testX.db-wal
      file copy test.db testX.db
      file copy test.db-wal testX.db-wal
  
      sqlite3 db2 testX.db
      execsql { PRAGMA journal_mode = WAL } db2
      execsql { PRAGMA integrity_check } db2
    } {ok}
  
    do_test walslow-1.seed=$seed.$iTest.4 {
      execsql { SELECT count(*) FROM t1 WHERE a!=b } db2
    } [execsql { SELECT count(*) FROM t1 WHERE a!=b }]
    db2 close
  }
}


finish_test

# 2007 September 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.
#
#***********************************************************************
#

set testdir [file dirname $argv0]

source $testdir/tester.tcl
if {[run_thread_tests]==0} { finish_test ; return }

do_test walthread-1.1 {
  execsql {
    PRAGMA journal_mode = WAL;
    CREATE TABLE t1(x PRIMARY KEY);
    INSERT INTO t1 VALUES(randomblob(100));
    INSERT INTO t1 VALUES(randomblob(100));
    INSERT INTO t1 SELECT md5sum(x) FROM t1;
  }
} {wal}
do_test walthread-1.2 {
  execsql {
    SELECT (SELECT count(*) FROM t1), (
      SELECT md5sum(x) FROM t1 WHERE oid != (SELECT max(oid) FROM t1)
    ) == (
      SELECT x FROM t1 WHERE oid = (SELECT max(oid) FROM t1)
    )
  }
} {3 1}
do_test walthread-1.3 {
  execsql { PRAGMA integrity_check } 
} {ok}
do_test walthread-1.4 {
  execsql { PRAGMA lock_status } 
} {main unlocked temp unknown}

#--------------------------------------------------------------------------
# Start N threads. Each thread performs both read and write transactions.
# Each read transaction consists of:
#
#   1) Reading the md5sum of all but the last table row,
#   2) Running integrity check.
#   3) Reading the value stored in the last table row,
#   4) Check that the values read in steps 1 and 3 are the same, and that
#      the md5sum of all but the last table row has not changed.
#
# Each write transaction consists of:
#
#   1) Modifying the contents of t1 (inserting, updating, deleting rows).
#   2) Appending a new row to the table containing the md5sum() of all
#      rows in the table.
#
# Each of the N threads runs N read transactions followed by a single write
# transaction in a loop as fast as possible.
#
# There is also a single checkpointer thread. It runs the following loop:
#
#   1) Execute "CHECKPOINT main 32 -1 1"
#   2) Sleep for 500 ms.
#

set thread_program {
  proc rest {ms} {
    set ::rest 0
    after $ms {set ::rest 1}
    vwait ::rest
  }

  proc dosql {DB sql} {
    set res ""
    set stmt [sqlite3_prepare_v2 $DB $sql -1 dummy_tail]
    set rc [sqlite3_step $stmt]
    if {$rc eq "SQLITE_ROW"} {
      set res [sqlite3_column_text $stmt 0]
    }
    set rc [sqlite3_finalize $stmt]

    if {$rc ne "SQLITE_OK"} {
      error $rc 
    }
    return $res
  }

  proc read_transaction {DB} {
    dosql $DB BEGIN

    set md5_1 [dosql $DB {
      SELECT md5sum(x) FROM t1 WHERE rowid != (SELECT max(rowid) FROM t1)
    }]
    set check [dosql $DB { PRAGMA integrity_check }]
    set md5_2 [dosql $DB { 
      SELECT x FROM t1 WHERE rowid = (SELECT max(rowid) FROM t1)
    }]
    set md5_3 [dosql $DB {
      SELECT md5sum(x) FROM t1 WHERE rowid != (SELECT max(rowid) FROM t1)
    }]

    dosql $DB COMMIT

    if {$check ne "ok" 
     || $md5_1 ne $md5_2
     || $md5_2 ne $md5_3
    } {
      error "Failed read transaction $check $md5_1 $md5_2 $md5_3"
    }
  }

  proc write_transaction {DB} {
    dosql $DB BEGIN
    dosql $DB "INSERT INTO t1 VALUES(randomblob(100))"
    dosql $DB "INSERT INTO t1 VALUES(randomblob(100))"
    dosql $DB "INSERT INTO t1 SELECT md5sum(x) FROM t1"
    dosql $DB COMMIT
  }

  proc checkpointer {DB} {
    while { !$::finished } {
      dosql $DB "PRAGMA checkpoint(32)"
      rest 1000
    }
  }

  proc worker {DB N} {
    set j 0
    while { !$::finished } {
      for {set i 0} {$i < $N} {incr i} { read_transaction $DB }
      write_transaction $DB
      rest 1
    }
  }

  set ::finished 0
  after [expr $seconds*1000] {set ::finished 1}

  set ::DB [sqlthread open test.db]
  dosql $::DB { PRAGMA journal_mode = WAL }

  set rc [catch {
    if {$role eq "worker"} { worker $DB $N }
    if {$role eq "checkpointer"} { checkpointer $DB }
  } msg]

  sqlite3_close $::DB

  if {$rc==0} { set msg OK } 
  set msg
}

set NTHREAD 6
set SECONDS 30

#set prg "set N $NTHREAD ; set seconds $SECONDS"
set prg "set N 1 ; set seconds $SECONDS"

array unset finished
for {set i 0} {$i < $NTHREAD} {incr i} {
  thread_spawn finished($i) {set role worker} $prg $thread_program
}
thread_spawn finished(C) {set role checkpointer} $prg $thread_program
#set finished(C) 1

puts "... test runs for approximately $SECONDS seconds ..."
for {set i 0} {$i < $::NTHREAD} {incr i} {
  if {![info exists finished($i)]} {
    vwait finished($i)
  }
  do_test walthread-2.$i {
    set ::finished($i)
  } OK
}
do_test walthread-2.C {
  if {![info exists finished(C)]} { vwait finished(C) }
  set ::finished(C)
} OK

set logsize 0

set rows    [execsql { SELECT count(*) FROM t1 }]
catch { set logsize [expr [file size test.db-wal] / 1024] }
set dbsize  [expr [file size test.db] / 1024]

puts "rows=$rows db=${dbsize}K log=${logsize}K"

finish_test

   fts3.h
   fts3Int.h
   fts3_hash.h
   fts3_tokenizer.h
   hash.h
   hwtime.h
   keywordhash.h
   log.h
   mutex.h
   opcodes.h
   os_common.h
   os.h
   os_os2.h
   pager.h
   parse.h

   os_unix.c
   os_win.c

   bitvec.c
   pcache.c
   pcache1.c
   rowset.c
   log.c
   pager.c

   btmutex.c
   btree.c
   backup.c

   vdbemem.c