SQLite

** This is an SQLite module implementing full-text search.
*/

#if defined(SQLITE_ENABLE_FTS5)

#include "fts5Int.h"

/*
** This variable is set to true when running corruption tests. Otherwise
** false. If it is false, extra assert() conditions in the fts5 code are
** activated - conditions that are only true if it is guaranteed that the
** fts5 database is not corrupt.
*/
int sqlite3_fts5_may_be_corrupt = 0;


typedef struct Fts5Table Fts5Table;
typedef struct Fts5Cursor Fts5Cursor;
typedef struct Fts5Global Fts5Global;
typedef struct Fts5Auxiliary Fts5Auxiliary;
typedef struct Fts5Auxdata Fts5Auxdata;

#endif

/*
** The assert_nc() macro is similar to the assert() macro, except that it
** is used for assert() conditions that are true only if it can be 
** guranteed that the database is not corrupt.
*/
#ifdef SQLITE_TEST
#ifdef SQLITE_DEBUG
extern int sqlite3_fts5_may_be_corrupt;
# define assert_nc(x) assert(sqlite3_fts5_may_be_corrupt || (x))
#else
# define assert_nc(x) assert(x)
#endif

typedef struct Fts5Global Fts5Global;

  char *zRankArgs;                /* Arguments to rank function */

  /* If non-NULL, points to sqlite3_vtab.base.zErrmsg. Often NULL. */
  char **pzErrmsg;
};

/* Current expected value of %_config table 'version' field */
#define FTS5_CURRENT_VERSION 1
#define FTS5_CURRENT_VERSION 2

#define FTS5_CONTENT_NORMAL   0
#define FTS5_CONTENT_NONE     1
#define FTS5_CONTENT_EXTERNAL 2

**
**     * either an 0x00 or 0x01 byte. If the value 0x01 is used, then there 
**       is an associated index-by-rowid record.
**     * the number of zero-term leaves as a varint.
**
** 5. Segment doclist indexes:
**
**   Doclist indexes are themselves b-trees, however they usually consist of
**   a single leaf record only. The format of each doclist index leaf page 
**   is:
**
**     * Flags byte. Bits are:
**         0x01: Clear if leaf is also the root page, otherwise set.
**
**     * Page number of fts index leaf page. As a varint.
**
**     * First docid on page indicated by previous field. As a varint.
**
**   A list of varints. If the first termless page contains at least one
**     * A list of varints, one for each subsequent termless page. A 
**   docid, the list begins with that docid as a varint followed by the
**   value 1 (0x01). Or, if the first termless page contains no docids,
**       positive delta if the termless page contains at least one docid, 
**   a varint containing the last docid stored on the term page followed
**   by a 0 (0x00) value.
**       or an 0x00 byte otherwise.
**
**   For each subsequent page in the doclist, either a 0x00 byte if the
**   page contains no terms, or a delta-encoded docid (always +ve) 
**   representing the first docid on the page otherwise.
**   Internal doclist index nodes are:
**
**     * Flags byte. Bits are:
**         0x01: Clear for root page, otherwise set.
**
**     * Page number of first child page. As a varint.
**
**     * Copy of first docid on page indicated by previous field. As a varint.
**
**     * A list of delta-encoded varints - the first docid on each subsequent
**       child page. 
**
*/

/*
** Rowids for the averages and structure records in the %_data table.
*/
#define FTS5_AVERAGES_ROWID     1    /* Rowid used for the averages record */
#define FTS5_STRUCTURE_ROWID   10    /* The structure record */

**
** The rowid for a node is then found using the FTS5_SEGMENT_ROWID() macro
** below. The FTS5_SEGMENT_*_BITS macros define the number of bits used
** to encode the three FTS5_SEGMENT_ROWID() arguments. This module returns
** SQLITE_FULL and fails the current operation if they ever prove too small.
*/
#define FTS5_DATA_ID_B     16     /* Max seg id number 65535 */
#define FTS5_DATA_DLI_B     1     /* Doclist-index flag (1 bit) */
#define FTS5_DATA_HEIGHT_B  5     /* Max b-tree height of 32 */
#define FTS5_DATA_PAGE_B   31     /* Max page number of 2147483648 */

#define FTS5_SEGMENT_ROWID(segid, height, pgno) (                         \
 ((i64)(segid)  << (FTS5_DATA_PAGE_B + FTS5_DATA_HEIGHT_B)) +                  \
#define fts5_dri(segid, dlidx, height, pgno) (                                 \
 ((i64)(segid)  << (FTS5_DATA_PAGE_B+FTS5_DATA_HEIGHT_B+FTS5_DATA_DLI_B)) +    \
 ((i64)(dlidx)  << (FTS5_DATA_PAGE_B + FTS5_DATA_HEIGHT_B)) +                  \
 ((i64)(height) << (FTS5_DATA_PAGE_B)) +                                       \
 ((i64)(pgno))                                                                 \
)

#define FTS5_SEGMENT_ROWID(segid, height, pgno) fts5_dri(segid, 0, height, pgno)
#define FTS5_DLIDX_ROWID(segid, height, pgno)   fts5_dri(segid, 1, height, pgno)

#if 0
/*
** The height of segment b-trees is actually limited to one less than 
** (1<<HEIGHT_BITS). This is because the rowid address space for nodes
** with such a height is used by doclist indexes.
*/
#define FTS5_SEGMENT_MAX_HEIGHT ((1 << FTS5_DATA_HEIGHT_B)-1)
#endif

/*
** Maximum segments permitted in a single index 
*/
#define FTS5_MAX_SEGMENT 2000

#if 0
/*
** The rowid for the doclist index associated with leaf page pgno of segment
** segid in index idx.
*/
#define FTS5_DOCLIST_IDX_ROWID(segid, pgno) \
#define FTS5_DOCLIST_IDX_ROWID(segid, height, pgno) \
        FTS5_SEGMENT_ROWID(segid, FTS5_SEGMENT_MAX_HEIGHT, pgno)
#endif

#ifdef SQLITE_DEBUG
int sqlite3Fts5Corrupt() { return SQLITE_CORRUPT_VTAB; }
#endif


/*
** Each time a blob is read from the %_data table, it is padded with this
** many zero bytes. This makes it easier to decode the various record formats
** without overreading if the records are corrupt.
*/
#define FTS5_DATA_ZERO_PADDING 8

typedef struct Fts5BtreeIter Fts5BtreeIter;
typedef struct Fts5BtreeIterLevel Fts5BtreeIterLevel;
typedef struct Fts5ChunkIter Fts5ChunkIter;
typedef struct Fts5Data Fts5Data;
typedef struct Fts5DlidxIter Fts5DlidxIter;
typedef struct Fts5DlidxLvl Fts5DlidxLvl;
typedef struct Fts5DlidxWriter Fts5DlidxWriter;
typedef struct Fts5MultiSegIter Fts5MultiSegIter;
typedef struct Fts5NodeIter Fts5NodeIter;
typedef struct Fts5PageWriter Fts5PageWriter;
typedef struct Fts5PosIter Fts5PosIter;
typedef struct Fts5SegIter Fts5SegIter;
typedef struct Fts5DoclistIter Fts5DoclistIter;
typedef struct Fts5SegWriter Fts5SegWriter;

** An object of type Fts5SegWriter is used to write to segments.
*/
struct Fts5PageWriter {
  int pgno;                       /* Page number for this page */
  Fts5Buffer buf;                 /* Buffer containing page data */
  Fts5Buffer term;                /* Buffer containing previous term on page */
};
struct Fts5DlidxWriter {
  int pgno;                       /* Page number for this page */
  int bPrevValid;                 /* True if iPrev is valid */
  i64 iPrev;                      /* Previous docid value written to page */
  Fts5Buffer buf;                 /* Buffer containing page data */
};
struct Fts5SegWriter {
  int iSegid;                     /* Segid to write to */
  int nWriter;                    /* Number of entries in aWriter */
  Fts5PageWriter *aWriter;        /* Array of PageWriter objects */
  i64 iPrevRowid;                 /* Previous docid written to current leaf */
  u8 bFirstRowidInDoclist;        /* True if next rowid is first in doclist */
  u8 bFirstRowidInPage;           /* True if next rowid is first in page */
  u8 bFirstTermInPage;            /* True if next term will be first in leaf */
  int nLeafWritten;               /* Number of leaf pages written */
  int nEmpty;                     /* Number of contiguous term-less nodes */
  Fts5Buffer cdlidx;               /* Doclist index */
  i64 iDlidxPrev;                 /* Previous rowid appended to dlidx */
  int bDlidxPrevValid;            /* True if iDlidxPrev is valid */

  int nDlidx;                     /* Allocated size of aDlidx[] array */
  Fts5DlidxWriter *aDlidx;        /* Array of Fts5DlidxWriter objects */
};

/*
** Object for iterating through the merged results of one or more segments,
** visiting each term/docid pair in the merged data.
**
** nSeg is always a power of two greater than or equal to the number of

**
** bEof:
**   Set to true once iterator has reached EOF.
**
** iOff:
**   Set to the current offset within record pData.
*/
struct Fts5DlidxIter {
  Fts5Data *pData;              /* Data for doclist index, if any */
  int iOff;                     /* Current offset into pDlidx */
struct Fts5DlidxLvl {
  Fts5Data *pData;              /* Data for current page of this level */
  int iOff;                     /* Current offset into pData */
  int bEof;                     /* At EOF already */
  int iFirstOff;                /* Used by reverse iterators only */
  int iFirstOff;                /* Used by reverse iterators */

  /* Output variables */
  int iLeafPgno;                /* Page number of current leaf page */
  i64 iRowid;                   /* First rowid on leaf iLeafPgno */
};
struct Fts5DlidxIter {
  int nLvl;
  int iSegid;
  Fts5DlidxLvl aLvl[1];
};



/*
** An Fts5BtreeIter object is used to iterate through all entries in the
** b-tree hierarchy belonging to a single fts5 segment. In this case the
** "b-tree hierarchy" is all b-tree nodes except leaves. Each entry in the
** b-tree hierarchy consists of the following:

** Free any memory allocated by the iterator object.
*/
static void fts5NodeIterFree(Fts5NodeIter *pIter){
  fts5BufferFree(&pIter->term);
}

/*
** The iterator passed as the first argument has the following fields set
** Advance the iterator passed as the only argument. If the end of the 
** as follows. This function sets up the rest of the iterator so that it
** points to the first rowid in the doclist-index.
**
**   pData: pointer to doclist-index record, 
** doclist-index page is reached, return non-zero.
**   iLeafPgno: page number that this doclist-index is associated with.
**
** When this function is called pIter->iLeafPgno is the page number the
** doclist is associated with (the one featuring the term).
*/
static int fts5DlidxIterFirst(Fts5DlidxIter *pIter){
  Fts5Data *pData = pIter->pData;
static int fts5DlidxLvlNext(Fts5DlidxLvl *pLvl){
  Fts5Data *pData = pLvl->pData;
  int i;
  int bPresent;

  if( pLvl->iOff==0 ){
  assert( pIter->pData );
  assert( pIter->iLeafPgno>0 );

    assert( pLvl->bEof==0 );
    pLvl->iOff = 1;
    pLvl->iOff += fts5GetVarint32(&pData->p[1], pLvl->iLeafPgno);
  /* Read the first rowid value. And the "present" flag that follows it. */
  pIter->iOff += getVarint(&pData->p[0], (u64*)&pIter->iRowid);
    pLvl->iOff += getVarint(&pData->p[pLvl->iOff], (u64*)&pLvl->iRowid);
  bPresent = pData->p[pIter->iOff++];
  if( bPresent ){
    i = 0;
    pLvl->iFirstOff = pLvl->iOff;
  }else{
    /* Count the number of leading 0x00 bytes. */
    for(i=1; pIter->iOff<pData->n; i++){ 
      if( pData->p[pIter->iOff] ) break;
    int iOff;
    for(iOff=pLvl->iOff; iOff<pData->n; iOff++){
      if( pData->p[iOff] ) break; 
      pIter->iOff++;
    }

    /* Unless we are already at the end of the doclist-index, load the first
    ** rowid value.  */
    if( pIter->iOff<pData->n ){
    if( iOff<pData->n ){
      i64 iVal;
      pLvl->iLeafPgno += (iOff - pLvl->iOff) + 1;
      pIter->iOff += getVarint(&pData->p[pIter->iOff], (u64*)&iVal);
      pIter->iRowid += iVal;
      iOff += getVarint(&pData->p[iOff], (u64*)&iVal);
      pLvl->iRowid += iVal;
      pLvl->iOff = iOff;
    }else{
      pIter->bEof = 1;
      pLvl->bEof = 1;
    }
  }
  pIter->iLeafPgno += (i+1);

  pIter->iFirstOff = pIter->iOff;
  return pIter->bEof;
  return pLvl->bEof;
}

/*
** Advance the iterator passed as the only argument.
*/
static int fts5DlidxIterNext(Fts5DlidxIter *pIter){
  Fts5Data *pData = pIter->pData;
static int fts5DlidxIterNextR(Fts5Index *p, Fts5DlidxIter *pIter, int iLvl){
  Fts5DlidxLvl *pLvl = &pIter->aLvl[iLvl];
  int iOff;

  for(iOff=pIter->iOff; iOff<pData->n; iOff++){
    if( pData->p[iOff] ) break; 
  }

  if( iOff<pData->n ){
  assert( iLvl<pIter->nLvl );
  if( fts5DlidxLvlNext(pLvl) ){
    if( (iLvl+1) < pIter->nLvl ){
      fts5DlidxIterNextR(p, pIter, iLvl+1);
      if( pLvl[1].bEof==0 ){
        fts5DataRelease(pLvl->pData);
        memset(pLvl, 0, sizeof(Fts5DlidxLvl));
        pLvl->pData = fts5DataRead(p, 
            FTS5_DLIDX_ROWID(pIter->iSegid, iLvl, pLvl[1].iLeafPgno)
        );
        if( pLvl->pData ) fts5DlidxLvlNext(pLvl);
      }
    }
  }

    i64 iVal;
    pIter->iLeafPgno += (iOff - pIter->iOff) + 1;
    iOff += getVarint(&pData->p[iOff], (u64*)&iVal);
    pIter->iRowid += iVal;
    pIter->iOff = iOff;
  return pIter->aLvl[0].bEof;
}
static int fts5DlidxIterNext(Fts5Index *p, Fts5DlidxIter *pIter){
  }else{
    pIter->bEof = 1;
  }
  return fts5DlidxIterNextR(p, pIter, 0);
}

/*
** The iterator passed as the first argument has the following fields set
** as follows. This function sets up the rest of the iterator so that it
** points to the first rowid in the doclist-index.
**
**   pData:
**     pointer to doclist-index record, 
**
** When this function is called pIter->iLeafPgno is the page number the
** doclist is associated with (the one featuring the term).
*/
static int fts5DlidxIterFirst(Fts5DlidxIter *pIter){
  int i;
  for(i=0; i<pIter->nLvl; i++){
    fts5DlidxLvlNext(&pIter->aLvl[i]);
  }
  return pIter->bEof;
  return pIter->aLvl[0].bEof;
}


static int fts5DlidxIterEof(Fts5Index *p, Fts5DlidxIter *pIter){
  return pIter->bEof;
  return p->rc!=SQLITE_OK || pIter->aLvl[0].bEof;
}

static void fts5DlidxIterLast(Fts5DlidxIter *pIter){
  if( fts5DlidxIterFirst(pIter)==0 ){
    while( 0==fts5DlidxIterNext(pIter) );
    pIter->bEof = 0;
  }
}

static int fts5DlidxIterPrev(Fts5DlidxIter *pIter){
  int iOff = pIter->iOff;
static void fts5DlidxIterLast(Fts5Index *p, Fts5DlidxIter *pIter){
  int i;

  /* Advance each level to the last entry on the last page */
  for(i=pIter->nLvl-1; p->rc==SQLITE_OK && i>=0; i--){
    Fts5DlidxLvl *pLvl = &pIter->aLvl[i];
    while( fts5DlidxLvlNext(pLvl)==0 );
    pLvl->bEof = 0;

    if( i>0 ){
      Fts5DlidxLvl *pChild = &pLvl[-1];
      fts5DataRelease(pChild->pData);
      memset(pChild, 0, sizeof(Fts5DlidxLvl));
      pChild->pData = fts5DataRead(p, 
          FTS5_DLIDX_ROWID(pIter->iSegid, i-1, pLvl->iLeafPgno)
      );
    }
  }
}

/*
** Move the iterator passed as the only argument to the previous entry.
*/
static int fts5DlidxLvlPrev(Fts5DlidxLvl *pLvl){
  int iOff = pLvl->iOff;

  assert( pIter->bEof==0 );
  if( iOff<=pIter->iFirstOff ){
    pIter->bEof = 1;
  assert( pLvl->bEof==0 );
  if( iOff<=pLvl->iFirstOff ){
    pLvl->bEof = 1;
  }else{
    u8 *a = pIter->pData->p;
    u8 *a = pLvl->pData->p;
    i64 iVal;
    int iLimit;
    int ii;
    int nZero = 0;

    /* Currently iOff points to the first byte of a varint. This block 
    ** decrements iOff until it points to the first byte of the previous 
    ** varint. Taking care not to read any memory locations that occur
    ** before the buffer in memory.  */
    iLimit = (iOff>9 ? iOff-9 : 0);
    for(iOff--; iOff>iLimit; iOff--){
      if( (a[iOff-1] & 0x80)==0 ) break;
    }

    getVarint(&a[iOff], (u64*)&iVal);
    pIter->iRowid -= iVal;
    pIter->iLeafPgno--;
    pLvl->iRowid -= iVal;
    pLvl->iLeafPgno--;

    /* Skip backwards passed any 0x00 bytes. */
    while( iOff>pIter->iFirstOff 
        && a[iOff-1]==0x00 && (a[iOff-2] & 0x80)==0 
    ){
      iOff--;
      pIter->iLeafPgno--;
    }
    pIter->iOff = iOff;
  }

  return pIter->bEof;
    /* Skip backwards past any 0x00 varints. */
    for(ii=iOff-1; ii>=pLvl->iFirstOff && a[ii]==0x00; ii--){
      nZero++;
    }
    if( ii>=pLvl->iFirstOff && (a[ii] & 0x80) ){
      /* The byte immediately before the last 0x00 byte has the 0x80 bit
      ** set. So the last 0x00 is only a varint 0 if there are 8 more 0x80
      ** bytes before a[ii]. */
      int bZero = 0;              /* True if last 0x00 counts */
      if( (ii-8)>=pLvl->iFirstOff ){
        int j;
        for(j=1; j<=8 && (a[ii-j] & 0x80); j++);
        bZero = (j>8);
      }
      if( bZero==0 ) nZero--;
    }
    pLvl->iLeafPgno -= nZero;
    pLvl->iOff = iOff - nZero;
  }

  return pLvl->bEof;
}

static int fts5DlidxIterPrevR(Fts5Index *p, Fts5DlidxIter *pIter, int iLvl){
  Fts5DlidxLvl *pLvl = &pIter->aLvl[iLvl];

  assert( iLvl<pIter->nLvl );
  if( fts5DlidxLvlPrev(pLvl) ){
    if( (iLvl+1) < pIter->nLvl ){
      fts5DlidxIterPrevR(p, pIter, iLvl+1);
      if( pLvl[1].bEof==0 ){
        fts5DataRelease(pLvl->pData);
        memset(pLvl, 0, sizeof(Fts5DlidxLvl));
        pLvl->pData = fts5DataRead(p, 
            FTS5_DLIDX_ROWID(pIter->iSegid, iLvl, pLvl[1].iLeafPgno)
        );
        if( pLvl->pData ){
          while( fts5DlidxLvlNext(pLvl)==0 );
          pLvl->bEof = 0;
        }
      }
    }
  }

  return pIter->aLvl[0].bEof;
}
static int fts5DlidxIterPrev(Fts5Index *p, Fts5DlidxIter *pIter){
  return fts5DlidxIterPrevR(p, pIter, 0);
}

/*
** Free a doclist-index iterator object allocated by fts5DlidxIterInit().
*/
static void fts5DlidxIterFree(Fts5DlidxIter *pIter){
  if( pIter ){
    int i;
    for(i=0; i<pIter->nLvl; i++){
      fts5DataRelease(pIter->aLvl[i].pData);
    }
    sqlite3_free(pIter);
  }
}

static Fts5DlidxIter *fts5DlidxIterInit(
  Fts5Index *p,                   /* Fts5 Backend to iterate within */
  int bRev,                       /* True for ORDER BY ASC */
  int iSegid,                     /* Segment id */
  int iLeafPg                     /* Leaf page number to load dlidx for */
){
  Fts5DlidxIter *pIter;
  Fts5DlidxIter *pIter = 0;
  int i;
  int bDone = 0;

  for(i=0; p->rc==SQLITE_OK && bDone==0; i++){
  pIter = (Fts5DlidxIter*)fts5IdxMalloc(p, sizeof(Fts5DlidxIter));
  if( pIter==0 ) return 0;
    int nByte = sizeof(Fts5DlidxIter) + i * sizeof(Fts5DlidxLvl);
    Fts5DlidxIter *pNew;

  pIter->pData = fts5DataRead(p, FTS5_DOCLIST_IDX_ROWID(iSegid, iLeafPg));
  if( pIter->pData==0 ){
    pNew = (Fts5DlidxIter*)sqlite3_realloc(pIter, nByte);
    if( pNew==0 ){
    sqlite3_free(pIter);
    pIter = 0;
  }else{
    pIter->iLeafPgno = iLeafPg;
      p->rc = SQLITE_NOMEM;
    }else{
      i64 iRowid = FTS5_DLIDX_ROWID(iSegid, i, iLeafPg);
      Fts5DlidxLvl *pLvl = &pNew->aLvl[i];
      pIter = pNew;
      memset(pLvl, 0, sizeof(Fts5DlidxLvl));
      pLvl->pData = fts5DataRead(p, iRowid);
      if( pLvl->pData && (pLvl->pData->p[0] & 0x0001)==0 ){
        bDone = 1;
      }
      pIter->nLvl = i+1;
    }
  }

  if( p->rc==SQLITE_OK ){
    pIter->iSegid = iSegid;
    if( bRev==0 ){
      fts5DlidxIterFirst(pIter);
    }else{
      fts5DlidxIterLast(pIter);
      fts5DlidxIterLast(p, pIter);
    }
  }

  if( p->rc!=SQLITE_OK ){
    fts5DlidxIterFree(pIter);
    pIter = 0;
  }

  return pIter;
}

/*
** Free a doclist-index iterator object allocated by fts5DlidxIterInit().
*/
static void fts5DlidxIterFree(Fts5DlidxIter *pIter){
  if( pIter ){
static i64 fts5DlidxIterRowid(Fts5DlidxIter *pIter){
  return pIter->aLvl[0].iRowid;
    fts5DataRelease(pIter->pData);
    sqlite3_free(pIter);
  }
}
static int fts5DlidxIterPgno(Fts5DlidxIter *pIter){
  return pIter->aLvl[0].iLeafPgno;
}

static void fts5LeafHeader(Fts5Data *pLeaf, int *piRowid, int *piTerm){
  *piRowid = (int)fts5GetU16(&pLeaf->p[0]);
  *piTerm = (int)fts5GetU16(&pLeaf->p[2]);
}

  int pgnoLast = 0;

  if( pDlidx ){
    /* If the doclist-iterator is already at EOF, then the current doclist
    ** contains no entries except those on the current page. */
    if( fts5DlidxIterEof(p, pDlidx)==0 ){
      int iSegid = pIter->pSeg->iSegid;
      pgnoLast = pDlidx->iLeafPgno;
      pgnoLast = fts5DlidxIterPgno(pDlidx);
      pLast = fts5DataRead(p, FTS5_SEGMENT_ROWID(iSegid, 0, pgnoLast));
    }else{
      pIter->iLeafOffset -= sqlite3Fts5GetVarintLen(pIter->nPos*2+pIter->bDel);
    }
  }else{
    int iOff;                               /* Byte offset within pLeaf */
    Fts5Data *pLeaf = pIter->pLeaf;         /* Current leaf data */

  pRes->iFirst = iRes;
  return 0;
}

/*
** Move the seg-iter so that it points to the first rowid on page iLeafPgno.
** It is an error if leaf iLeafPgno contains no rowid.
** It is an error if leaf iLeafPgno does not exist or contains no rowids.
*/
static void fts5SegIterGotoPage(
  Fts5Index *p,                   /* FTS5 backend object */
  Fts5SegIter *pIter,             /* Iterator to advance */
  int iLeafPgno
){
  assert( iLeafPgno>pIter->iLeafPgno );
  if( iLeafPgno>pIter->pSeg->pgnoLast ){
    p->rc = FTS5_CORRUPT;
  }else{
  pIter->iLeafPgno = iLeafPgno-1;
  fts5SegIterNextPage(p, pIter);
  assert( p->rc!=SQLITE_OK || pIter->iLeafPgno==iLeafPgno );
    pIter->iLeafPgno = iLeafPgno-1;
    fts5SegIterNextPage(p, pIter);
    assert( p->rc!=SQLITE_OK || pIter->iLeafPgno==iLeafPgno );

  if( p->rc==SQLITE_OK ){
    int iOff;
    u8 *a = pIter->pLeaf->p;
    int n = pIter->pLeaf->n;
    if( p->rc==SQLITE_OK ){
      int iOff;
      u8 *a = pIter->pLeaf->p;
      int n = pIter->pLeaf->n;

    iOff = fts5GetU16(&a[0]);
    if( iOff<4 || iOff>=n ){
      p->rc = FTS5_CORRUPT;
    }else{
      iOff += getVarint(&a[iOff], (u64*)&pIter->iRowid);
      pIter->iLeafOffset = iOff;
      fts5SegIterLoadNPos(p, pIter);
      iOff = fts5GetU16(&a[0]);
      if( iOff<4 || iOff>=n ){
        p->rc = FTS5_CORRUPT;
      }else{
        iOff += getVarint(&a[iOff], (u64*)&pIter->iRowid);
        pIter->iLeafOffset = iOff;
        fts5SegIterLoadNPos(p, pIter);
      }
    }
  }
}

/*
** Advance the iterator passed as the second argument until it is at or 
** past rowid iFrom. Regardless of the value of iFrom, the iterator is

  int bMove = 1;

  assert( pIter->flags & FTS5_SEGITER_ONETERM );
  assert( pIter->pDlidx );
  assert( pIter->pLeaf );

  if( bRev==0 ){
    while( fts5DlidxIterEof(p, pDlidx)==0 && iMatch>pDlidx->iRowid ){
      iLeafPgno = pDlidx->iLeafPgno;
      fts5DlidxIterNext(pDlidx);
    while( !fts5DlidxIterEof(p, pDlidx) && iMatch>fts5DlidxIterRowid(pDlidx) ){
      iLeafPgno = fts5DlidxIterPgno(pDlidx);
      fts5DlidxIterNext(p, pDlidx);
    }
    assert( iLeafPgno>=pIter->iLeafPgno || p->rc );
    assert_nc( iLeafPgno>=pIter->iLeafPgno || p->rc );
    if( iLeafPgno>pIter->iLeafPgno ){
      fts5SegIterGotoPage(p, pIter, iLeafPgno);
      bMove = 0;
    }
  }else{
    assert( iMatch<pIter->iRowid );
    while( fts5DlidxIterEof(p, pDlidx)==0 && iMatch<pDlidx->iRowid ){
      fts5DlidxIterPrev(pDlidx);
    while( !fts5DlidxIterEof(p, pDlidx) && iMatch<fts5DlidxIterRowid(pDlidx) ){
      fts5DlidxIterPrev(p, pDlidx);
    }
    iLeafPgno = pDlidx->iLeafPgno;
    iLeafPgno = fts5DlidxIterPgno(pDlidx);

    assert( fts5DlidxIterEof(p, pDlidx) || iLeafPgno<=pIter->iLeafPgno );

    if( iLeafPgno<pIter->iLeafPgno ){
      pIter->iLeafPgno = iLeafPgno+1;
      fts5SegIterReverseNewPage(p, pIter);
      bMove = 0;

  int i;
  assert( fts5BlobCompare(pOld, nOld, pNew, nNew)<0 );
  for(i=0; i<nOld; i++){
    if( pOld[i]!=pNew[i] ) break;
  }
  return i;
}

static void fts5WriteDlidxClear(
  Fts5Index *p, 
  Fts5SegWriter *pWriter,
  int bFlush                      /* If true, write dlidx to disk */
){
  int i;
  assert( bFlush==0 || (pWriter->nDlidx>0 && pWriter->aDlidx[0].buf.n>0) );
  for(i=0; i<pWriter->nDlidx; i++){
    Fts5DlidxWriter *pDlidx = &pWriter->aDlidx[i];
    if( pDlidx->buf.n==0 ) break;
    if( bFlush ){
      assert( pDlidx->pgno!=0 );
      fts5DataWrite(p, 
          FTS5_DLIDX_ROWID(pWriter->iSegid, i, pDlidx->pgno),
          pDlidx->buf.p, pDlidx->buf.n
      );
    }
    sqlite3Fts5BufferZero(&pDlidx->buf);
    pDlidx->bPrevValid = 0;
  }
}

/*
** Grow the pWriter->aDlidx[] array to at least nLvl elements in size.
** Any new array elements are zeroed before returning.
*/
static int fts5WriteDlidxGrow(
  Fts5Index *p,
  Fts5SegWriter *pWriter,
  int nLvl
){
  if( p->rc==SQLITE_OK && nLvl>=pWriter->nDlidx ){
    Fts5DlidxWriter *aDlidx = (Fts5DlidxWriter*)sqlite3_realloc(
        pWriter->aDlidx, sizeof(Fts5DlidxWriter) * nLvl
    );
    if( aDlidx==0 ){
      p->rc = SQLITE_NOMEM;
    }else{
      int nByte = sizeof(Fts5DlidxWriter) * (nLvl - pWriter->nDlidx);
      memset(&aDlidx[pWriter->nDlidx], 0, nByte);
      pWriter->aDlidx = aDlidx;
      pWriter->nDlidx = nLvl;
    }
  }
  return p->rc;
}

/*
** If an "nEmpty" record must be written to the b-tree before the next
** term, write it now. 
*/
static void fts5WriteBtreeNEmpty(Fts5Index *p, Fts5SegWriter *pWriter){
  if( pWriter->nEmpty ){
    int bFlag = 0;
    Fts5PageWriter *pPg;
    pPg = &pWriter->aWriter[1];
    if( pWriter->nEmpty>=FTS5_MIN_DLIDX_SIZE ){
      i64 iKey = FTS5_DOCLIST_IDX_ROWID(
          pWriter->iSegid, pWriter->aWriter[0].pgno - 1 - pWriter->nEmpty

    /* If there were FTS5_MIN_DLIDX_SIZE or more empty leaf pages written
    ** to the database, also write the doclist-index to disk.  */
      );
      assert( pWriter->cdlidx.n>0 );
    if( pWriter->aDlidx[0].buf.n>0 && pWriter->nEmpty>=FTS5_MIN_DLIDX_SIZE ){
      fts5DataWrite(p, iKey, pWriter->cdlidx.p, pWriter->cdlidx.n);
      bFlag = 1;
    }
    fts5WriteDlidxClear(p, pWriter, bFlag);
    fts5BufferAppendVarint(&p->rc, &pPg->buf, bFlag);
    fts5BufferAppendVarint(&p->rc, &pPg->buf, pWriter->nEmpty);
    pWriter->nEmpty = 0;
  }else{
    fts5WriteDlidxClear(p, pWriter, 0);
  }

  /* Whether or not it was written to disk, zero the doclist index at this
  ** point */
  sqlite3Fts5BufferZero(&pWriter->cdlidx);
  pWriter->bDlidxPrevValid = 0;
  assert( pWriter->nDlidx==0 || pWriter->aDlidx[0].buf.n==0 );
  assert( pWriter->nDlidx==0 || pWriter->aDlidx[0].bPrevValid==0 );
}

static void fts5WriteBtreeGrow(Fts5Index *p, Fts5SegWriter *pWriter){
  if( p->rc==SQLITE_OK ){
    Fts5PageWriter *aNew;
    Fts5PageWriter *pNew;
    int nNew = sizeof(Fts5PageWriter) * (pWriter->nWriter+1);

      fts5BufferAppendBlob(&p->rc, &pPage->buf, nTerm-nPre, pTerm+nPre);
      fts5BufferSet(&p->rc, &pPage->term, nTerm, pTerm);
      break;
    }
  }
}

/*
** This function is called when flushing a leaf page that contains no
** terms at all to disk.
*/
static void fts5WriteBtreeNoTerm(
  Fts5Index *p,                   /* FTS5 backend object */
  Fts5SegWriter *pWriter          /* Writer object */
){
  /* If there were no rowids on the leaf page either and the doclist-index
  ** has already been started, append an 0x00 byte to it.  */
  if( pWriter->bFirstRowidInPage ){
  if( pWriter->bFirstRowidInPage && pWriter->aDlidx[0].buf.n>0 ){
    /* No rowids on this page. Append an 0x00 byte to the current 
    ** doclist-index */
    if( pWriter->bDlidxPrevValid==0 ){
      i64 iRowid = pWriter->iPrevRowid;
      sqlite3Fts5BufferAppendVarint(&p->rc, &pWriter->cdlidx, iRowid);
      pWriter->bDlidxPrevValid = 1;
      pWriter->iDlidxPrev = iRowid;
    }
    sqlite3Fts5BufferAppendVarint(&p->rc, &pWriter->cdlidx, 0);
  }
    Fts5DlidxWriter *pDlidx = &pWriter->aDlidx[0];
    assert( pDlidx->bPrevValid );
    sqlite3Fts5BufferAppendVarint(&p->rc, &pDlidx->buf, 0);
  }

  /* Increment the "number of sequential leaves without a term" counter. */
  pWriter->nEmpty++;
}

static i64 fts5DlidxExtractFirstRowid(Fts5Buffer *pBuf){
  i64 iRowid;
  int iOff;

  pWriter->nEmpty++;
  iOff = 1 + getVarint(&pBuf->p[1], (u64*)&iRowid);
  getVarint(&pBuf->p[iOff], (u64*)&iRowid);
  return iRowid;
}

/*
** Rowid iRowid has just been appended to the current leaf page. As it is
** the first on its page, append an entry to the current doclist-index.
** Rowid iRowid has just been appended to the current leaf page. It is the
** first on the page. This function appends an appropriate entry to the current
** doclist-index.
*/
static void fts5WriteDlidxAppend(
  Fts5Index *p, 
  Fts5SegWriter *pWriter, 
  i64 iRowid
){
  int i;
  int bDone = 0;

  for(i=0; p->rc==SQLITE_OK && bDone==0; i++){
  i64 iVal;
  if( pWriter->bDlidxPrevValid ){
    iVal = iRowid - pWriter->iDlidxPrev;
  }else{
    sqlite3Fts5BufferAppendVarint(&p->rc, &pWriter->cdlidx, iRowid);
    iVal = 1;
  }
  sqlite3Fts5BufferAppendVarint(&p->rc, &pWriter->cdlidx, iVal);
  pWriter->bDlidxPrevValid = 1;
  pWriter->iDlidxPrev = iRowid;
    i64 iVal;
    Fts5DlidxWriter *pDlidx = &pWriter->aDlidx[i];

    if( pDlidx->buf.n>=p->pConfig->pgsz ){
      /* The current doclist-index page is full. Write it to disk and push
      ** a copy of iRowid (which will become the first rowid on the next
      ** doclist-index leaf page) up into the next level of the b-tree 
      ** hierarchy. If the node being flushed is currently the root node,
      ** also push its first rowid upwards. */
      pDlidx->buf.p[0] = 0x01;    /* Not the root node */
      fts5DataWrite(p, 
          FTS5_DLIDX_ROWID(pWriter->iSegid, i, pDlidx->pgno),
          pDlidx->buf.p, pDlidx->buf.n
      );
      fts5WriteDlidxGrow(p, pWriter, i+2);
      pDlidx = &pWriter->aDlidx[i];
      if( p->rc==SQLITE_OK && pDlidx[1].buf.n==0 ){
        i64 iFirst = fts5DlidxExtractFirstRowid(&pDlidx->buf);

        /* This was the root node. Push its first rowid up to the new root. */
        pDlidx[1].pgno = pDlidx->pgno;
        sqlite3Fts5BufferAppendVarint(&p->rc, &pDlidx[1].buf, 0);
        sqlite3Fts5BufferAppendVarint(&p->rc, &pDlidx[1].buf, pDlidx->pgno);
        sqlite3Fts5BufferAppendVarint(&p->rc, &pDlidx[1].buf, iFirst);
        pDlidx[1].bPrevValid = 1;
        pDlidx[1].iPrev = iFirst;
      }

      sqlite3Fts5BufferZero(&pDlidx->buf);
      pDlidx->bPrevValid = 0;
      pDlidx->pgno++;
    }else{
      bDone = 1;
    }

    if( pDlidx->bPrevValid ){
      iVal = iRowid - pDlidx->iPrev;
    }else{
      i64 iPgno = (i==0 ? pWriter->aWriter[0].pgno : pDlidx[-1].pgno);
      assert( pDlidx->buf.n==0 );
      sqlite3Fts5BufferAppendVarint(&p->rc, &pDlidx->buf, !bDone);
      sqlite3Fts5BufferAppendVarint(&p->rc, &pDlidx->buf, iPgno);
      iVal = iRowid;
    }

    sqlite3Fts5BufferAppendVarint(&p->rc, &pDlidx->buf, iVal);
    pDlidx->bPrevValid = 1;
    pDlidx->iPrev = iRowid;
  }
}

static void fts5WriteFlushLeaf(Fts5Index *p, Fts5SegWriter *pWriter){
  static const u8 zero[] = { 0x00, 0x00, 0x00, 0x00 };
  Fts5PageWriter *pPage = &pWriter->aWriter[0];
  i64 iRowid;

  /* Update the Fts5PageWriter.term field. */
  fts5BufferSet(&p->rc, &pPage->term, nTerm, pTerm);
  pWriter->bFirstTermInPage = 0;

  pWriter->bFirstRowidInPage = 0;
  pWriter->bFirstRowidInDoclist = 1;

  assert( p->rc || (pWriter->nDlidx>0 && pWriter->aDlidx[0].buf.n==0) );
  pWriter->aDlidx[0].pgno = pPage->pgno;

  /* If the current leaf page is full, flush it to disk. */
  if( pPage->buf.n>=p->pConfig->pgsz ){
    fts5WriteFlushLeaf(p, pWriter);
  }
}

  }
  for(i=0; i<pWriter->nWriter; i++){
    Fts5PageWriter *pPg = &pWriter->aWriter[i];
    fts5BufferFree(&pPg->term);
    fts5BufferFree(&pPg->buf);
  }
  sqlite3_free(pWriter->aWriter);

  for(i=0; i<pWriter->nDlidx; i++){
  sqlite3Fts5BufferFree(&pWriter->cdlidx);
    sqlite3Fts5BufferFree(&pWriter->aDlidx[i].buf);
  }
  sqlite3_free(pWriter->aDlidx);
}

static void fts5WriteInit(
  Fts5Index *p, 
  Fts5SegWriter *pWriter, 
  int iSegid
){
  memset(pWriter, 0, sizeof(Fts5SegWriter));
  pWriter->iSegid = iSegid;

  pWriter->aWriter = (Fts5PageWriter*)fts5IdxMalloc(p,sizeof(Fts5PageWriter));
  if( pWriter->aWriter==0 ) return;
  pWriter->aWriter = (Fts5PageWriter*)fts5IdxMalloc(p, sizeof(Fts5PageWriter));
  pWriter->aDlidx = (Fts5DlidxWriter*)fts5IdxMalloc(p, sizeof(Fts5DlidxWriter));
  if( pWriter->aDlidx==0 ) return;
  pWriter->nWriter = 1;
  pWriter->nDlidx = 1;
  pWriter->aWriter[0].pgno = 1;
  pWriter->bFirstTermInPage = 1;
}

static void fts5WriteInitForAppend(
  Fts5Index *p,                   /* FTS5 backend object */
  Fts5SegWriter *pWriter,         /* Writer to initialize */
  Fts5StructureSegment *pSeg      /* Segment object to append to */
){
  int nByte = pSeg->nHeight * sizeof(Fts5PageWriter);
  memset(pWriter, 0, sizeof(Fts5SegWriter));
  pWriter->iSegid = pSeg->iSegid;
  pWriter->aWriter = (Fts5PageWriter*)fts5IdxMalloc(p, nByte);
  pWriter->aDlidx = (Fts5DlidxWriter*)fts5IdxMalloc(p, sizeof(Fts5DlidxWriter));

  if( p->rc==SQLITE_OK ){
    int pgno = 1;
    int i;
    pWriter->nDlidx = 1;
    pWriter->nWriter = pSeg->nHeight;
    pWriter->aWriter[0].pgno = pSeg->pgnoLast+1;
    for(i=pSeg->nHeight-1; i>0; i--){
      i64 iRowid = FTS5_SEGMENT_ROWID(pWriter->iSegid, i, pgno);
      Fts5PageWriter *pPg = &pWriter->aWriter[i];
      pPg->pgno = pgno;
      fts5DataBuffer(p, &pPg->buf, iRowid);

    /* Pre-allocate the buffer used to assemble leaf pages to the target
    ** page size.  */
    assert( pgsz>0 );
    pBuf = &writer.aWriter[0].buf;
    fts5BufferGrow(&p->rc, pBuf, pgsz + 20);

    /* Begin scanning through hash table entries. */
    /* Begin scanning through hash table entries. This loop runs once for each
    ** term/doclist currently stored within the hash table. */
    if( p->rc==SQLITE_OK ){
      memset(pBuf->p, 0, 4);
      pBuf->n = 4;
      p->rc = sqlite3Fts5HashScanInit(pHash, 0, 0);
    }

    while( p->rc==SQLITE_OK && 0==sqlite3Fts5HashScanEof(pHash) ){
      const char *zTerm;
      int nTerm;
      const u8 *pDoclist;
      int nDoclist;
      const char *zTerm;          /* Buffer containing term */
      int nTerm;                  /* Size of zTerm in bytes */
      const u8 *pDoclist;         /* Pointer to doclist for this term */
      int nDoclist;               /* Size of doclist in bytes */
      int nSuffix;                /* Size of term suffix */

      sqlite3Fts5HashScanEntry(pHash, &zTerm, &pDoclist, &nDoclist);
      nTerm = strlen(zTerm);

      /* Decide if the term will fit on the current leaf. If it will not, 
      ** flush the leaf to disk here.  */
      if( (pBuf->n + nTerm + 2) > pgsz ){
        fts5WriteFlushLeaf(p, &writer);
        pBuf = &writer.aWriter[0].buf;
        if( (nTerm + 32) > pBuf->nSpace ){
          fts5BufferGrow(&p->rc, pBuf, nTerm + 32 - pBuf->n);
          if( p->rc ) break;
        }
      }

      /* Write the term to the leaf. And push it up into the b-tree hierarchy */
      /* Write the term to the leaf. And if it is the first on the leaf, and
      ** the leaf is not page number 1, push it up into the b-tree hierarchy 
      ** as well.  */
      if( writer.bFirstTermInPage==0 ){
        int nPre = fts5PrefixCompress(nTerm, zPrev, nTerm, (const u8*)zTerm);
        pBuf->n += sqlite3PutVarint(&pBuf->p[pBuf->n], nPre);
        nSuffix = nTerm - nPre;
      }else{
        fts5PutU16(&pBuf->p[2], pBuf->n);
        writer.bFirstTermInPage = 0;
        if( writer.aWriter[0].pgno!=1 ){
          int nPre = fts5PrefixCompress(nTerm, zPrev, nTerm, (const u8*)zTerm);
          fts5WriteBtreeTerm(p, &writer, nPre+1, (const u8*)zTerm);
          pBuf = &writer.aWriter[0].buf;
          assert( nPre<nTerm );
        }
        nSuffix = nTerm;
      }
      pBuf->n += sqlite3PutVarint(&pBuf->p[pBuf->n], nSuffix);
      fts5BufferSafeAppendBlob(pBuf, (const u8*)&zTerm[nTerm-nSuffix], nSuffix);

      /* We just wrote a term into page writer.aWriter[0].pgno. If a 
      ** doclist-index is to be generated for this doclist, it will be
      ** associated with this page. */
      assert( writer.nDlidx>0 && writer.aDlidx[0].buf.n==0 );
      writer.aDlidx[0].pgno = writer.aWriter[0].pgno;

      if( pgsz>=(pBuf->n + nDoclist + 1) ){
        /* The entire doclist will fit on the current leaf. */
        fts5BufferSafeAppendBlob(pBuf, pDoclist, nDoclist);
      }else{
        i64 iRowid = 0;
        i64 iDelta = 0;

  int bSz,
  Fts5Buffer *pBuf
){
  if( p->rc==SQLITE_OK ){
    Fts5ChunkIter iter;
    Fts5SegIter *pSeg = &pMulti->aSeg[ pMulti->aFirst[1].iFirst ];
    assert( fts5MultiIterEof(p, pMulti)==0 );
    static int nCall = 0;
    nCall++;

    fts5ChunkIterInit(p, pSeg, &iter);

    if( fts5ChunkIterEof(p, &iter)==0 ){
      if( bSz ){
        /* WRITEPOSLISTSIZE */
        fts5BufferAppendVarint(&p->rc, pBuf, iter.nRem * 2);

}

/*
** Return the current term.
*/
const char *sqlite3Fts5IterTerm(Fts5IndexIter *pIter, int *pn){
  int n;
  const char *z = fts5MultiIterTerm(pIter->pMulti, &n);
  const char *z = (const char*)fts5MultiIterTerm(pIter->pMulti, &n);
  *pn = n-1;
  return &z[1];
}


/*
** Return a pointer to a buffer containing a copy of the position list for

#ifdef SQLITE_DEBUG
static void fts5DlidxIterTestReverse(
  Fts5Index *p, 
  int iSegid,                     /* Segment id to load from */
  int iLeaf                       /* Load doclist-index for this leaf */
){
  Fts5DlidxIter *pDlidx = 0;
  i64 cksum1 = 13;
  i64 cksum2 = 13;
  u64 cksum1 = 13;
  u64 cksum2 = 13;

  for(pDlidx=fts5DlidxIterInit(p, 0, iSegid, iLeaf);
      fts5DlidxIterEof(p, pDlidx)==0;
      fts5DlidxIterNext(pDlidx)
      fts5DlidxIterNext(p, pDlidx)
  ){
    i64 iRowid = fts5DlidxIterRowid(pDlidx);
    int pgno = fts5DlidxIterPgno(pDlidx);
    assert( pDlidx->iLeafPgno>iLeaf );
    assert( pgno>iLeaf );
    cksum1 = (cksum1 ^ ( (i64)(pDlidx->iLeafPgno) << 32 ));
    cksum1 = (cksum1 ^ pDlidx->iRowid);
    cksum1 += iRowid + ((i64)pgno<<32);
  }
  fts5DlidxIterFree(pDlidx);
  pDlidx = 0;

  for(pDlidx=fts5DlidxIterInit(p, 1, iSegid, iLeaf);
      fts5DlidxIterEof(p, pDlidx)==0;
      fts5DlidxIterPrev(pDlidx)
      fts5DlidxIterPrev(p, pDlidx)
  ){
    assert( pDlidx->iLeafPgno>iLeaf );
    cksum2 = (cksum2 ^ ( (i64)(pDlidx->iLeafPgno) << 32 ));
    cksum2 = (cksum2 ^ pDlidx->iRowid);
    i64 iRowid = fts5DlidxIterRowid(pDlidx);
    int pgno = fts5DlidxIterPgno(pDlidx);

    assert( fts5DlidxIterPgno(pDlidx)>iLeaf );
    cksum2 += iRowid + ((i64)pgno<<32);
  }
  fts5DlidxIterFree(pDlidx);
  pDlidx = 0;

  if( p->rc==SQLITE_OK && cksum1!=cksum2 ) p->rc = FTS5_CORRUPT; 
  if( p->rc==SQLITE_OK && cksum1!=cksum2 ) p->rc = FTS5_CORRUPT;
}
#else
# define fts5DlidxIterTestReverse(x,y,z)
#endif

static void fts5IndexIntegrityCheckSegment(
  Fts5Index *p,                   /* FTS5 backend object */

      int iPrevLeaf = iter.iLeaf;
      int iSegid = pSeg->iSegid;
      int iPg;
      i64 iKey;

      for(pDlidx=fts5DlidxIterInit(p, 0, iSegid, iter.iLeaf);
          fts5DlidxIterEof(p, pDlidx)==0;
          fts5DlidxIterNext(pDlidx)
          fts5DlidxIterNext(p, pDlidx)
      ){

        /* Check any rowid-less pages that occur before the current leaf. */
        for(iPg=iPrevLeaf+1; iPg<pDlidx->iLeafPgno; iPg++){
        for(iPg=iPrevLeaf+1; iPg<fts5DlidxIterPgno(pDlidx); iPg++){
          iKey = FTS5_SEGMENT_ROWID(iSegid, 0, iPg);
          pLeaf = fts5DataRead(p, iKey);
          if( pLeaf ){
            if( fts5GetU16(&pLeaf->p[0])!=0 ) p->rc = FTS5_CORRUPT;
            fts5DataRelease(pLeaf);
          }
        }
        iPrevLeaf = pDlidx->iLeafPgno;
        iPrevLeaf = fts5DlidxIterPgno(pDlidx);

        /* Check that the leaf page indicated by the iterator really does
        ** contain the rowid suggested by the same. */
        iKey = FTS5_SEGMENT_ROWID(iSegid, 0, pDlidx->iLeafPgno);
        iKey = FTS5_SEGMENT_ROWID(iSegid, 0, iPrevLeaf);
        pLeaf = fts5DataRead(p, iKey);
        if( pLeaf ){
          i64 iRowid;
          int iRowidOff = fts5GetU16(&pLeaf->p[0]);
          getVarint(&pLeaf->p[iRowidOff], (u64*)&iRowid);
          if( iRowid!=pDlidx->iRowid ) p->rc = FTS5_CORRUPT;
          if( iRowid!=fts5DlidxIterRowid(pDlidx) ) p->rc = FTS5_CORRUPT;
          fts5DataRelease(pLeaf);
        }

      }

      for(iPg=iPrevLeaf+1; iPg<=(iter.iLeaf + iter.nEmpty); iPg++){
        iKey = FTS5_SEGMENT_ROWID(iSegid, 0, iPg);
        pLeaf = fts5DataRead(p, iKey);
        if( pLeaf ){
          if( fts5GetU16(&pLeaf->p[0])!=0 ) p->rc = FTS5_CORRUPT;

/*
** Decode a segment-data rowid from the %_data table. This function is
** the opposite of macro FTS5_SEGMENT_ROWID().
*/
static void fts5DecodeRowid(
  i64 iRowid,                     /* Rowid from %_data table */
  int *piSegid,                   /* OUT: Segment id */
  int *pbDlidx,                   /* OUT: Dlidx flag */
  int *piHeight,                  /* OUT: Height */
  int *piPgno                     /* OUT: Page number */
){
  *piPgno = (int)(iRowid & (((i64)1 << FTS5_DATA_PAGE_B) - 1));
  iRowid >>= FTS5_DATA_PAGE_B;

  *piHeight = (int)(iRowid & (((i64)1 << FTS5_DATA_HEIGHT_B) - 1));
  iRowid >>= FTS5_DATA_HEIGHT_B;

  *pbDlidx = (int)(iRowid & 0x0001);
  iRowid >>= FTS5_DATA_DLI_B;

  *piSegid = (int)(iRowid & (((i64)1 << FTS5_DATA_ID_B) - 1));
}

static void fts5DebugRowid(int *pRc, Fts5Buffer *pBuf, i64 iKey){
  int iSegid, iHeight, iPgno;     /* Rowid compenents */
  fts5DecodeRowid(iKey, &iSegid, &iHeight, &iPgno);
  int iSegid, iHeight, iPgno, bDlidx;       /* Rowid compenents */
  fts5DecodeRowid(iKey, &iSegid, &bDlidx, &iHeight, &iPgno);

  if( iSegid==0 ){
    if( iKey==FTS5_AVERAGES_ROWID ){
      sqlite3Fts5BufferAppendPrintf(pRc, pBuf, "(averages) ");
    }else{
      sqlite3Fts5BufferAppendPrintf(pRc, pBuf, 
      sqlite3Fts5BufferAppendPrintf(pRc, pBuf, "(structure)");
          "{structure idx=%d}", (int)(iKey-10)
      );
    }
  }
  else if( iHeight==FTS5_SEGMENT_MAX_HEIGHT ){
    sqlite3Fts5BufferAppendPrintf(pRc, pBuf, "(dlidx segid=%d pgno=%d)",
        iSegid, iPgno
    );
  }else{
    sqlite3Fts5BufferAppendPrintf(pRc, pBuf, "(segid=%d h=%d pgno=%d)",
        iSegid, iHeight, iPgno
  else{
    sqlite3Fts5BufferAppendPrintf(pRc, pBuf, "(%ssegid=%d h=%d pgno=%d)",
        bDlidx ? "dlidx " : "", iSegid, iHeight, iPgno
    );
  }
}

static void fts5DebugStructure(
  int *pRc,                       /* IN/OUT: error code */
  Fts5Buffer *pBuf,

*/
static void fts5DecodeFunction(
  sqlite3_context *pCtx,          /* Function call context */
  int nArg,                       /* Number of args (always 2) */
  sqlite3_value **apVal           /* Function arguments */
){
  i64 iRowid;                     /* Rowid for record being decoded */
  int iSegid,iHeight,iPgno;       /* Rowid components */
  int iSegid,iHeight,iPgno,bDlidx;/* Rowid components */
  const u8 *aBlob; int n;         /* Record to decode */
  u8 *a = 0;
  Fts5Buffer s;                   /* Build up text to return here */
  int rc = SQLITE_OK;             /* Return code */
  int nSpace = 0;

  assert( nArg==2 );
  memset(&s, 0, sizeof(Fts5Buffer));
  iRowid = sqlite3_value_int64(apVal[0]);
  n = sqlite3_value_bytes(apVal[1]);
  aBlob = sqlite3_value_blob(apVal[1]);

  nSpace = n + FTS5_DATA_ZERO_PADDING;
  a = (u8*)sqlite3Fts5MallocZero(&rc, nSpace);
  if( a==0 ) goto decode_out;
  memcpy(a, aBlob, n);
  fts5DecodeRowid(iRowid, &iSegid, &iHeight, &iPgno);
  fts5DecodeRowid(iRowid, &iSegid, &bDlidx, &iHeight, &iPgno);

  fts5DebugRowid(&rc, &s, iRowid);
  if( iHeight==FTS5_SEGMENT_MAX_HEIGHT ){
  if( bDlidx ){
    Fts5Data dlidx;
    Fts5DlidxIter iter;
    Fts5DlidxLvl lvl;

    dlidx.p = a;
    dlidx.n = n;
    dlidx.nRef = 2;

    memset(&iter, 0, sizeof(Fts5DlidxIter));
    iter.pData = &dlidx;
    iter.iLeafPgno = iPgno;
    memset(&lvl, 0, sizeof(Fts5DlidxLvl));
    lvl.pData = &dlidx;
    lvl.iLeafPgno = iPgno;

    for(fts5DlidxIterFirst(&iter); iter.bEof==0; fts5DlidxIterNext(&iter)){
    for(fts5DlidxLvlNext(&lvl); lvl.bEof==0; fts5DlidxLvlNext(&lvl)){
      sqlite3Fts5BufferAppendPrintf(&rc, &s, 
          " %d(%lld)", iter.iLeafPgno, iter.iRowid
          " %d(%lld)", lvl.iLeafPgno, lvl.iRowid
      );
    }
  }else if( iSegid==0 ){
    if( iRowid==FTS5_AVERAGES_ROWID ){
      /* todo */
    }else{
      fts5DecodeStructure(&rc, &s, a, n);

#ifdef SQLITE_ENABLE_FTS5

#include "fts5.h"
#include <string.h>
#include <assert.h>

/*
** This variable is set to true when running corruption tests. Otherwise
** false. If it is false, extra assert() conditions in the fts5 code are
** activated - conditions that are only true if it is guaranteed that the
** fts5 database is not corrupt.
*/
int sqlite3_fts5_may_be_corrupt = 0;
extern int sqlite3_fts5_may_be_corrupt;

/*************************************************************************
** This is a copy of the first part of the SqliteDb structure in 
** tclsqlite.c.  We need it here so that the get_sqlite_pointer routine
** can extract the sqlite3* pointer from an existing Tcl SQLite
** connection.
*/

  CREATE VIRTUAL TABLE t1 USING fts5(x,y);
}
do_execsql_test 2.1 {
  INSERT INTO t1 VALUES('a b c', 'd e f');
}
do_test 2.2 {
  execsql { SELECT fts5_decode(id, block) FROM t1_data WHERE id==10 }
} {/{{structure idx=0} {lvl=0 nMerge=0 {id=[0123456789]* h=1 leaves=1..1}}}/}
} {/{\(structure\) {lvl=0 nMerge=0 {id=[0123456789]* h=1 leaves=1..1}}}/}
do_execsql_test 2.3 {
  INSERT INTO t1(t1) VALUES('integrity-check');
}

#-------------------------------------------------------------------------
#
reset_db

      set y [doc]
      set z [doc]
      set rowid [expr int(rand() * 100)]
      execsql { REPLACE INTO t1(rowid,x,y,z) VALUES($rowid, $x, $y, $z) }
    }
    execsql { INSERT INTO t1(t1) VALUES('integrity-check'); }
  } {}
#  if {$i==1} break
}
#db eval {SELECT rowid, fts5_decode(rowid, block) aS r FROM t1_data} {puts $r}
#exit

#-------------------------------------------------------------------------
#
reset_db

      set rowid [expr int(rand() * 100)]
      execsql { REPLACE INTO t1(rowid,x,y,z) VALUES($rowid, $x, $y, $z) }
    }
    execsql { INSERT INTO t1(t1) VALUES('integrity-check'); }
  } {}
  if {[set_test_counter errors]} break
}


#-------------------------------------------------------------------------
#
reset_db
do_execsql_test 10.0 {
  CREATE VIRTUAL TABLE t1 USING fts5(x,y);
}

  finish_test
  return
}

do_execsql_test 1.1 {
  CREATE VIRTUAL TABLE ft1 USING fts5(x);
  SELECT * FROM ft1_config;
} {version 1}
} {version 2}

do_execsql_test 1.2 {
  INSERT INTO ft1(ft1, rank) VALUES('pgsz', 32);
  SELECT * FROM ft1_config;
} {pgsz 32 version 1}
} {pgsz 32 version 2}

do_execsql_test 1.3 {
  INSERT INTO ft1(ft1, rank) VALUES('pgsz', 64);
  SELECT * FROM ft1_config;
} {pgsz 64 version 1}
} {pgsz 64 version 2}

#--------------------------------------------------------------------------
# Test the logic for parsing the rank() function definition.
#
foreach {tn defn} {
  1 "fname()"
  2 "fname(1)"

# This file tests that FTS5 handles corrupt databases (i.e. internal
# inconsistencies in the backing tables) correctly. In this case 
# "correctly" means without crashing.
#

source [file join [file dirname [info script]] fts5_common.tcl]
set testprefix fts5corrupt2
sqlite3_fts5_may_be_corrupt 1

# Create a simple FTS5 table containing 100 documents. Each document 
# contains 10 terms, each of which start with the character "x".
#
expr srand(0)
db func rnddoc fts5_rnddoc
do_execsql_test 1.0 {
  CREATE VIRTUAL TABLE t1 USING fts5(x);
  INSERT INTO t1(t1, rank) VALUES('pgsz', 32);
  WITH ii(i) AS (SELECT 1 UNION SELECT i+1 FROM ii WHERE i<100)
  INSERT INTO t1 SELECT rnddoc(10) FROM ii;
}
set mask [expr 31 << 31]


# Test 1:
#
#   For each page in the t1_data table, open a transaction and DELETE
#   the t1_data entry. Then run:
#
#     * an integrity-check, and

    execsql ROLLBACK
  }

  do_test 4.$tn.x { expr $nCorrupt>0 } 1
}


sqlite3_fts5_may_be_corrupt 0
finish_test

      if {($i % $spc2)==0} { 
        lappend ydoc $rowid
        append doc " y" 
      }
    }
    execsql { INSERT INTO t1(rowid, x) VALUES($rowid, $doc) }
  }
  breakpoint
  execsql COMMIT

  do_test $tn.1 {
    execsql { INSERT INTO t1(t1) VALUES('integrity-check') }
  } {}
  
  do_fb_test $tn.3.1 { SELECT rowid FROM t1 WHERE t1 MATCH 'a AND x' } $xdoc

  do_fb_test $tn.5.2 { 
    SELECT rowid FROM t1 WHERE t1 MATCH 'b + c + x + y' } $ydoc
}


do_dlidx_test1 1.1     10 100 10000 0 1000
do_dlidx_test1 1.2     10 10  10000 0 128
do_dlidx_test1 1.3     10 10  100   0 36028797018963970
do_dlidx_test1 1.3     10 10  50    0 150000000000000000
do_dlidx_test1 1.3     10 10  66   0 36028797018963970
do_dlidx_test1 1.4     10 10  50    0 150000000000000000



finish_test

do_execsql_test 2.0 {
  CREATE VIRTUAL TABLE yy USING fts5(x, prefix=1);
  INSERT INTO yy VALUES('term');
}
do_execsql_test 2.1 {
  INSERT INTO yy(yy) VALUES('integrity-check');
}

#--------------------------------------------------------------------
#
do_execsql_test 3.0 {
  CREATE VIRTUAL TABLE zz USING fts5(z);
  INSERT INTO zz(zz, rank) VALUES('pgsz', 32);
  INSERT INTO zz VALUES('b b b b b b b b b b b b b b');
  INSERT INTO zz SELECT z FROM zz;
  INSERT INTO zz SELECT z FROM zz;
  INSERT INTO zz SELECT z FROM zz;
  INSERT INTO zz SELECT z FROM zz;
  INSERT INTO zz SELECT z FROM zz;
  INSERT INTO zz SELECT z FROM zz;
  INSERT INTO zz(zz) VALUES('optimize');
}

do_execsql_test 3.1 { INSERT INTO zz(zz) VALUES('integrity-check'); }


#db eval {SELECT rowid, fts5_decode(rowid, block) aS r FROM zz_data} {puts $r}
#exit


finish_test

do_catchsql_test 1.2 {
  SELECT fts5_rowid('segment')
} {1 {should be: fts5_rowid('segment', segid, height, pgno))}}

do_execsql_test 1.3 {
  SELECT fts5_rowid('segment', 1, 1, 1)
} {70866960385}
} {139586437121}

do_catchsql_test 1.4 {
  SELECT fts5_rowid('nosucharg');
} {1 {first arg to fts5_rowid() must be 'segment' or 'start-of-index'}} 


#-------------------------------------------------------------------------

do_execsql_test 1.1 {
  CREATE VIRTUAL TABLE t1 USING fts5(one);
  INSERT INTO t1 VALUES('a b c d');
} {}

do_execsql_test 1.2 {
  SELECT * FROM t1_config WHERE k='version'
} {version 1}
} {version 2}

do_execsql_test 1.3 {
  SELECT rowid FROM t1 WHERE t1 MATCH 'a';
} {1}

do_execsql_test 1.4 {
  UPDATE t1_config set v=2 WHERE k='version';
  UPDATE t1_config set v=3 WHERE k='version';
} 

do_test 1.5 {
  db close
  sqlite3 db test.db
  catchsql { SELECT * FROM t1 WHERE t1 MATCH 'a' }
} {1 {invalid fts5 file format (found 2, expected 1) - run 'rebuild'}}
} {1 {invalid fts5 file format (found 3, expected 2) - run 'rebuild'}}

breakpoint
do_test 1.6 {
  db close
  sqlite3 db test.db
  catchsql { INSERT INTO t1 VALUES('x y z') }
} {1 {invalid fts5 file format (found 2, expected 1) - run 'rebuild'}}
} {1 {invalid fts5 file format (found 3, expected 2) - run 'rebuild'}}

do_test 1.7 {
  execsql { DELETE FROM t1_config WHERE k='version' }
  db close
  sqlite3 db test.db
  catchsql { SELECT * FROM t1 WHERE t1 MATCH 'a' }
} {1 {invalid fts5 file format (found 0, expected 1) - run 'rebuild'}}
} {1 {invalid fts5 file format (found 0, expected 2) - run 'rebuild'}}


finish_test

db func loadfile loadfile

db transaction {
  set pref ""
  if {$O(prefix)!=""} { set pref ", prefix='$O(prefix)'" }
  catch {
    db eval "CREATE VIRTUAL TABLE t1 USING $O(vtab) (path, content$O(tok)$pref)"
    # db eval "INSERT INTO t1(t1, rank) VALUES('pgsz', 4050);"
  }
  if {$O(automerge)>=0} {
    if {$O(vtab) == "fts5"} {
      db eval { INSERT INTO t1(t1, rank) VALUES('automerge', $O(automerge)) }
    } else {
      db eval { INSERT INTO t1(t1) VALUES('automerge=' || $O(automerge)) }
    }

  fts3varint.test
  fts4growth.test fts4growth2.test
}

test_suite "fts5" -prefix "" -description {
  All FTS5 tests.
} -files [glob -nocomplain $::testdir/../ext/fts5/test/*.test]

test_suite "fts5-light" -prefix "" -description {
  All FTS5 tests.
} -files [
  test_set \
      [glob -nocomplain $::testdir/../ext/fts5/test/*.test] \
      -exclude *corrupt* *fault* *big* *fts5aj*
]

test_suite "nofaultsim" -prefix "" -description {
  "Very" quick test suite. Runs in less than 5 minutes on a workstation. 
  This test suite is the same as the "quick" tests, except that some files
  that test malloc and IO errors are omitted.
} -files [
  test_set $allquicktests -exclude *malloc* *ioerr* *fault*

   fts5_expr.c
   fts5_hash.c
   fts5_index.c
   fts5parse.c
   fts5_storage.c
   fts5_tokenize.c
   fts5_unicode2.c
   fts5_vocab.c

   rtree.c
   icu.c
   fts3_icu.c
} {
  copy_file tsrc/$file
}