pColl = sqlite4ExprCollSeq(pParse, pFarg->a[i].pExpr);
        }
      }
      if( pDef->flags & SQLITE4_FUNC_NEEDCOLL ){
        if( !pColl ) pColl = db->pDfltColl; 
        sqlite4VdbeAddOp4(v, OP_CollSeq, 0, 0, 0, (char *)pColl, P4_COLLSEQ);
      }
      if( pDef->bMatchinfo ){
        sqlite4VdbeAddOp1(v, OP_Mifunction, pFarg->a[0].pExpr->iTable);
      }
      sqlite4VdbeAddOp4(v, OP_Function, constMask, r1, target,
                        (char*)pDef, P4_FUNCDEF);
      sqlite4VdbeChangeP5(v, (u8)nFarg);
      if( nFarg ){
        sqlite4ReleaseTempRange(pParse, r1, nFarg);
      }
      break;

**
*************************************************************************
*/

#include "sqliteInt.h"
#include "vdbeInt.h"

/* 
** Stream numbers must be lower than this.
**
** For optimization purposes, it is assumed that a given tokenizer uses
** a set of contiguous stream numbers starting with 0. And that most
** tokens belong to stream 0.
**
** The hard limit is 63 (due to the format of "row size" records).
*/
#define SQLITE4_FTS5_NSTREAM 32

/*
** Records stored within the index:
**
** Row size record:
**   There is one "row size" record in the index for each row in the
**   indexed table. The "row size" record contains the number of tokens
**   in the associated row for each combination of a stream and column
**   number (i.e. contains the data required to find the number of
**   tokens associated with stream S present in column C of the row for
**   all S and C).
**
**   The key for the row size record is a single 0x00 byte followed by
**   a copy of the PK blob for the table row. 
**
**   The value is a series of varints. Each column of the table is
**   represented by one or more varints packed into the array.
**
**   If a column contains only stream 0 tokens, then it is represented
**   by a single varint - (nToken << 1), where nToken is the number of
**   stream 0 tokens stored in the column.
**
**   Or, if the column contains tokens from multiple streams, the first
**   varint contains a bitmask indicating which of the streams are present
**   (stored as ((bitmask << 1) | 0x01)). Following the bitmask is a
**   varint containing the number of tokens for each stream present, in
**   ascending order of stream number.
**
**   TODO: The format above is not currently implemented! Instead, there
**   is a simpler place-holder format (which consumes more space).
**
** Global size record:
**   There is a single "global size" record stored in the database. The
**   database key for this record is a single byte - 0x00.
**
**   The data for this record is a series of varint values. The first 
**   varint is the total number of rows in the table. The subsequent
**   varints make up a "row size" record containing the total number of
**   tokens for each S/C combination in all rows of the table.
**
** FTS index records:
**   The FTS index records implement the following mapping:
**
**       (token, document-pk) -> (list of instances)
**
**   The key for each index record is in the same format as the keys for
**   regular text indexes. An 0x24 byte, followed by the utf-8 representation
**   of the token, followed by 0x00, followed by the PK blob for the table
**   row.
**
**   TODO: Describe value format.
*/

/*
** Default distance value for NEAR operators.
*/
#define FTS5_DEFAULT_NEAR 10

/*
** Token types used by expression parser.

**   expr := expr NOT expr
**   expr := expr AND expr
**   expr := expr OR  expr
**   expr := LP expr RP
*/

/*
** Structure types used by this module.
*/
typedef struct Fts5Expr Fts5Expr;
typedef struct Fts5ExprNode Fts5ExprNode;
typedef struct Fts5List Fts5List;
typedef struct Fts5MatchIter Fts5MatchIter;
typedef struct Fts5Parser Fts5Parser;
typedef struct Fts5ParserToken Fts5ParserToken;
typedef struct Fts5Phrase Fts5Phrase;
typedef struct Fts5Prefix Fts5Prefix;
typedef struct Fts5Size Fts5Size;
typedef struct Fts5Str Fts5Str;
typedef struct Fts5Token Fts5Token;


struct Fts5ParserToken {
  int eType;                      /* Token type */
  int n;                          /* Size of z[] in bytes */

  Fts5ExprNode *pLeft;
  Fts5ExprNode *pRight;
  const u8 *aPk;                  /* Primary key of current entry (or null) */
  int nPk;                        /* Size of aPk[] in bytes */
};

struct Fts5Expr {
  Fts5ExprNode *pRoot;            /* Root node of expression */
  int nPhrase;                    /* Number of Fts5Str objects in query */
  Fts5Str **apPhrase;             /* All Fts5Str objects */
};

/*
** FTS5 specific cursor data.
*/
struct Fts5Cursor {
  sqlite4 *db;
  Fts5Info *pInfo;
  Fts5Expr *pExpr;                /* MATCH expression for this cursor */
  char *zExpr;                    /* Full text of MATCH expression */
  KVByteArray *aKey;              /* Buffer for primary key */
  int nKeyAlloc;                  /* Bytes allocated at aKey[] */

  KVCursor *pCsr;                 /* Cursor used to retrive values */
  Mem *aMem;                      /* Array of column values */
  int bMemValid;                  /* True if contents of aMem[] are valid */

  Fts5Size *pSz;                  /* Local size data */
  Fts5Size *pGlobal;              /* Global size data */
  i64 nGlobal;                    /* Total number of rows in table */
  int *anRow;

  Fts5MatchIter *pIter;           /* Used by mi_match_detail() */
};

/*
** A deserialized 'size record' (see above).
*/
struct Fts5Size {
  int nCol;                       /* Number of columns in indexed table */
  int nStream;                    /* Number of streams */
  i64 *aSz;                       /* Token count for each C/S */
};

/*
** This type is used when reading (decoding) an instance-list.
*/
typedef struct InstanceList InstanceList;
struct InstanceList {
  u8 *aList;
  int nList;
  int iList;

  /* The current entry */
  int iCol;
  int iStream;
  int iOff;
};

/*
** An instance of this structure is used by the sqlite4_mi_match_detail()
** API to iterate through matches. 
*/
struct Fts5MatchIter {
  int bValid;                     /* True if aList[] is current row */
  int iCurrent;                   /* Current index in aList[] (or -1) */
  int iMatch;                     /* Current iMatch value */
  InstanceList *aList;            /* One iterator for each phrase in expr */
};

/*
** Return true for EOF, or false if the next entry is valid.
*/
static int fts5InstanceListNext(InstanceList *p){
  int i = p->iList;
  int bRet = 1;
  
  while( bRet && i<p->nList ){
    u32 iVal;
    i += getVarint32(&p->aList[i], iVal);
    if( (iVal & 0x03)==0x01 ){
      p->iCol = (iVal>>2);
      p->iOff = 0;
    }
    else if( (iVal & 0x03)==0x03 ){
      p->iStream = (iVal>>2);
    }
    else{
      p->iOff += (iVal>>1);
      bRet = 0;
    }
  }
  if( bRet ){

static int fts5InstanceListEof(InstanceList *p){
  return (p->aList==0);
}

static void fts5InstanceListAppend(
  InstanceList *p,                /* Instance list to append to */
  int iCol,                       /* Column of new entry */
  int iStream,                    /* Weight of new entry */
  int iOff                        /* Offset of new entry */
){
  assert( iCol>=p->iCol );
  assert( iCol>p->iCol || iOff>=p->iOff );

  if( iCol!=p->iCol ){
    p->iList += putVarint32(&p->aList[p->iList], (iCol<<2)|0x01);
    p->iCol = iCol;
    p->iOff = 0;
  }

  if( iStream!=p->iStream ){
    p->iList += putVarint32(&p->aList[p->iList], (iStream<<2)|0x03);
    p->iStream = iStream;
  }

  p->iList += putVarint32(&p->aList[p->iList], (iOff-p->iOff)<<1);
  p->iOff = iOff;

  assert( p->iList<=p->nList );
}

}

/*
** Callback for fts5CountTokens().
*/
static int fts5CountTokensCb(
  void *pCtx, 
  int iStream, 
  int iOff, 
  const char *z, int n,
  int iSrc, int nSrc
){
  (*((int *)pCtx))++;
  return 0;
}

struct AppendTokensCtx {
  Fts5Parser *pParse;
  Fts5Str *pStr;
};

static int fts5AppendTokensCb(
  void *pCtx, 
  int iStream, 
  int iOff, 
  const char *z, int n, 
  int iSrc, int nSrc
){
  struct AppendTokensCtx *p = (struct AppendTokensCtx *)pCtx;
  Fts5Parser *pParse = p->pParse;
  Fts5Token *pToken;

    sqlite4DbFree(db, pNode);
  }
}

static void fts5ExpressionFree(sqlite4 *db, Fts5Expr *pExpr){
  if( pExpr ){
    fts5FreeExprNode(db, pExpr->pRoot);
    sqlite4DbFree(db, pExpr->apPhrase);
    sqlite4DbFree(db, pExpr);
  }
}

typedef struct ExprHier ExprHier;
struct ExprHier {
  Fts5ExprNode **ppNode;

  *pp = pNode;
  (*paHier)[*pnHier].ppNode = &pNode->pRight;
  (*paHier)[*pnHier].nOpen = 0;
  (*pnHier)++;

  return SQLITE4_OK;
}

static void fts5FindStrings(Fts5ExprNode *p, Fts5Str ***papStr){
  if( p ){
    if( p->eType==TOKEN_PRIMITIVE ){
      int i;
      Fts5Str *aStr = p->pPhrase->aStr;
      for(i=0; i<p->pPhrase->nStr; i++){
        **papStr = &aStr[i];
        (*papStr)++;
      }
    }
    fts5FindStrings(p->pLeft, papStr);
    fts5FindStrings(p->pRight, papStr);
  }
}

static int fts5ParseExpression(
  sqlite4 *db,                    /* Database handle */
  Fts5Tokenizer *pTokenizer,      /* Tokenizer module */
  sqlite4_tokenizer *p,           /* Tokenizer instance */
  int iRoot,                      /* Root page number of FTS index */
  char **azCol,                   /* Array of column names (nul-term'd) */
  int nCol,                       /* Size of array azCol[] */
  const char *zExpr,              /* FTS expression text */
  Fts5Expr **ppExpr,              /* OUT: Expression object */
  char **pzErr                    /* OUT: Error message */
){
  int rc = SQLITE4_OK;
  Fts5Parser sParse;
  int nStr = 0;
  int nExpr;
  int i;
  Fts5Expr *pExpr;

  int nHier = 0;
  int nHierAlloc = 0;
  ExprHier *aHier = 0;

            rc = SQLITE4_NOMEM;
          }else{
            pNode->eType = TOKEN_PRIMITIVE;
            pNode->pPhrase = pPhrase;
            *pp = pNode;
          }
        }
        nStr += pPhrase->nStr;
        break;
      }

      case TOKEN_AND:
      case TOKEN_OR:
      case TOKEN_NOT: {
        Fts5ExprNode **pp = aHier[nHier-1].ppNode;

  if( rc==SQLITE4_OK && *aHier[nHier-1].ppNode==0 ){
    rc = SQLITE4_ERROR;
  }
  for(i=0; rc==SQLITE4_OK && i<nHier; i++){
    if( aHier[i].nOpen>0 ) rc = SQLITE4_ERROR;
  }

  if( rc==SQLITE4_OK ){
    pExpr->nPhrase = nStr;
    pExpr->apPhrase = (Fts5Str**)sqlite4DbMallocZero(db, sizeof(Fts5Str*)*nStr);
    if( pExpr->apPhrase==0 ){
      rc = SQLITE4_NOMEM;
    }else{
      Fts5Str **a = pExpr->apPhrase;
      fts5FindStrings(pExpr->pRoot, &a);
    }
  }

  if( rc!=SQLITE4_OK ){
    fts5ExpressionFree(db, pExpr);
    *pzErr = sParse.zErr;

    pExpr = 0;
  }
  *ppExpr = pExpr;
  sqlite4DbFree(db, aHier);
  return rc;
}

/*
** Search for the Fts5Tokenizer object named zName. Return a pointer to it
** if it exists, or NULL otherwise.

          ExprListItem *pItem = &pArgs->a[j];
          if( pItem->zName ) break;
          nByte += sqlite4Strlen30(pItem->pExpr->u.zToken) + 1;
        }
        nByte += sizeof(char *) * (j-i);
        pFts->azTokenizer = (char **)sqlite4DbMallocZero(pParse->db, nByte);
        if( pFts->azTokenizer==0 ) return;
        pFts->nTokenizer = (j-i);

        pSpace = (char *)&pFts->azTokenizer[j-i];
        for(j=i; j<pArgs->nExpr; j++){
          ExprListItem *pItem = &pArgs->a[j];
          if( pItem->zName && j>i ){
            break;
          }else{

** sqlite4DbRealloc().
*/
typedef struct TokenizeCtx TokenizeCtx;
typedef struct TokenizeTerm TokenizeTerm;
struct TokenizeCtx {
  int rc;
  int iCol;
  int nCol;                       /* Number of columns in table */
  sqlite4 *db;
  int nMax;
  i64 *aSz;                       /* Number of tokens in each column/stream */
  int nStream;                    /* Number of streams in document */
  Hash hash;
};
struct TokenizeTerm {
  int iStream;                    /* Weight of previous entry */
  int iCol;                       /* Column containing previous entry */
  int iOff;                       /* Token offset of previous entry */
  int nToken;                     /* Size of token in bytes */
  int nData;                      /* Bytes of data in value */
  int nAlloc;                     /* Bytes of data allocated */
};

  a = &(((unsigned char *)&pTerm[1])[pTerm->nToken+pTerm->nData]);
  pTerm->nData += putVarint32(a, iVal);
  return pTerm;
}

static int fts5TokenizeCb(
  void *pCtx, 
  int iStream, 
  int iOff,
  const char *zToken, 
  int nToken, 
  int iSrc, 
  int nSrc
){
  TokenizeCtx *p = (TokenizeCtx *)pCtx;
  sqlite4 *db = p->db;
  TokenizeTerm *pTerm = 0;
  TokenizeTerm *pOrig = 0;

  /* TODO: Error here if iStream is out of range */

  if( nToken>p->nMax ) p->nMax = nToken;

  if( iStream>=p->nStream ){
    int nOld = p->nStream;
    int nNew = 4;
    while( nNew<=iStream ) nNew = nNew*2;
    p->aSz = (i64*)sqlite4DbReallocOrFree(db, p->aSz, nNew*p->nCol*sizeof(i64));
    if( p->aSz==0 ) goto tokenize_cb_out;
    memset(&p->aSz[nOld * p->nCol], 0, (nNew-nOld)*p->nCol*sizeof(i64));
    p->nStream = nNew;
  }
  p->aSz[iStream*p->nCol + p->iCol]++;

  pTerm = (TokenizeTerm *)sqlite4HashFind(&p->hash, zToken, nToken);
  if( pTerm==0 ){
    /* Size the initial allocation so that it fits in the lookaside buffer */
    int nAlloc = sizeof(TokenizeTerm) + nToken + 32;

    pTerm = sqlite4DbMallocZero(p->db, nAlloc);
    if( pTerm ){

        pTerm = 0;
      }
      if( pTerm==0 ) goto tokenize_cb_out;
    }
  }
  pOrig = pTerm;

  if( iStream!=pTerm->iStream ){
    pTerm = fts5TokenizeAppendInt(p, pTerm, (iStream << 2) | 0x00000003);
    if( !pTerm ) goto tokenize_cb_out;
    pTerm->iStream = iStream;
  }

  if( pTerm && p->iCol!=pTerm->iCol ){
    pTerm = fts5TokenizeAppendInt(p, pTerm, (p->iCol << 2) | 0x00000001);
    if( !pTerm ) goto tokenize_cb_out;
    pTerm->iCol = p->iCol;
    pTerm->iOff = 0;

  if( !pTerm ){
    p->rc = SQLITE4_NOMEM;
    return 1;
  }

  return 0;
}

static int fts5LoadSizeRecord(
  sqlite4 *db,                    /* Database handle */
  u8 *aKey, int nKey,             /* KVStore key */
  int nMinStream,                 /* Space for at least this many streams */
  Fts5Info *pInfo,                /* Info record */
  i64 *pnRow,                     /* non-NULL when reading global record */
  Fts5Size **ppSz                 /* OUT: Loaded size record */
){
  Fts5Size *pSz = 0;              /* Size object */
  KVCursor *pCsr = 0;             /* Cursor used to read global record */
  int rc;

  rc = sqlite4KVStoreOpenCursor(db->aDb[pInfo->iDb].pKV, &pCsr);
  if( rc==SQLITE4_OK ){
    rc = sqlite4KVCursorSeek(pCsr, aKey, nKey, 0);
    if( rc==SQLITE4_NOTFOUND ){
      if( pnRow ){
        int nByte = sizeof(Fts5Size) + sizeof(i64) * pInfo->nCol * nMinStream;
        pSz = sqlite4DbMallocZero(db, nByte);
        if( pSz==0 ){
          rc = SQLITE4_NOMEM;
        }else{
          pSz->aSz = (i64 *)&pSz[1];
          pSz->nStream = nMinStream;
          pSz->nCol = pInfo->nCol;
          *pnRow = 0;
          rc = SQLITE4_OK;
        }
      }else{
        rc = SQLITE4_CORRUPT_BKPT;
      }
    }else if( rc==SQLITE4_OK ){
      const u8 *aData = 0;
      int nData = 0;
      rc = sqlite4KVCursorData(pCsr, 0, -1, &aData, &nData);
      if( rc==SQLITE4_OK ){
        int iOff = 0;
        int nStream = 0;
        int nAlloc;

        /* If pnRow is not NULL, then this is the global record. Read the
        ** number of documents in the table from the start of the record. */
        if( pnRow ){
          iOff += sqlite4GetVarint(&aData[iOff], (u64 *)pnRow);
        }
        iOff += getVarint32(&aData[iOff], nStream);
        nAlloc = (nStream < nMinStream ? nMinStream : nStream);

        pSz = sqlite4DbMallocZero(db, 
            sizeof(Fts5Size) + sizeof(i64) * pInfo->nCol * nAlloc
        );
        if( pSz==0 ){
          rc = SQLITE4_NOMEM;
        }else{
          int iCol = 0;
          pSz->aSz = (i64 *)&pSz[1];
          pSz->nCol = pInfo->nCol;
          pSz->nStream = nAlloc;
          while( iOff<nData ){
            int i;
            i64 *aSz = &pSz->aSz[iCol*nAlloc];
            for(i=0; i<nStream; i++){
              iOff += sqlite4GetVarint(&aData[iOff], (u64*)&aSz[i]);
            }
            iCol++;
          }
        }
      }
    }
    sqlite4KVCursorClose(pCsr);
  }

  *ppSz = pSz;
  return rc;
}

static int fts5StoreSizeRecord(
  KVStore *p,
  u8 *aKey, int nKey,
  Fts5Size *pSz, 
  i64 nRow, 
  u8 *a                           /* Space to serialize record in */
){
  int iOff = 0;
  int iCol;

  if( nRow>=0 ){
    iOff += sqlite4PutVarint(&a[iOff], nRow);
  }
  iOff += sqlite4PutVarint(&a[iOff], pSz->nStream);

  for(iCol=0; iCol<pSz->nCol; iCol++){
    int i;
    for(i=0; i<pSz->nStream; i++){
      iOff += sqlite4PutVarint(&a[iOff], pSz->aSz[i*pSz->nCol+iCol]);
    }
  }

  return sqlite4KVStoreReplace(p, aKey, nKey, a, iOff);
}

static int fts5CsrLoadGlobal(Fts5Cursor *pCsr){
  int rc = SQLITE4_OK;
  if( pCsr->pGlobal==0 ){
    int nKey;
    u8 aKey[10];
    nKey = putVarint32(aKey, pCsr->pInfo->iRoot);
    aKey[nKey++] = 0x00;
    rc = fts5LoadSizeRecord(
        pCsr->db, aKey, nKey, 0, pCsr->pInfo, &pCsr->nGlobal, &pCsr->pGlobal
    );
  }
  return rc;
}

static int fts5CsrLoadSz(Fts5Cursor *pCsr){
  int rc = SQLITE4_OK;
  if( pCsr->pSz==0 ){
    sqlite4 *db = pCsr->db;
    Fts5Info *pInfo = pCsr->pInfo;
    u8 *aKey;
    int nKey = 0;
    int nPk = pCsr->pExpr->pRoot->nPk;

    aKey = (u8 *)sqlite4DbMallocZero(db, 10 + nPk);
    if( !aKey ) return SQLITE4_NOMEM;

    nKey = putVarint32(aKey, pInfo->iRoot);
    aKey[nKey++] = 0x00;
    memcpy(&aKey[nKey], pCsr->pExpr->pRoot->aPk, nPk);
    nKey += nPk;

    rc = fts5LoadSizeRecord(pCsr->db, aKey, nKey, 0, pInfo, 0, &pCsr->pSz);
    sqlite4DbFree(db, aKey);
  }

  return rc;
}


/*
** Update an fts index.
*/
int sqlite4Fts5Update(
  sqlite4 *db,                    /* Database handle */
  Fts5Info *pInfo,                /* Description of fts index to update */
  Mem *pKey,                      /* Primary key blob */
  Mem *aArg,                      /* Array of arguments (see above) */
  int bDel,                       /* True for a delete, false for insert */
  char **pzErr                    /* OUT: Error message */
){
  int i;
  int rc = SQLITE4_OK;
  KVStore *pStore;
  TokenizeCtx sCtx;


  int nTnum = 0;
  u32 dummy = 0;

  u8 *aSpace = 0;
  int nSpace = 0;

  const u8 *pPK;
  int nPK;
  HashElem *pElem;

  pStore = db->aDb[pInfo->iDb].pKV;

  memset(&sCtx, 0, sizeof(sCtx));
  sCtx.db = db;
  sCtx.nCol = pInfo->nCol;
  sqlite4HashInit(db->pEnv, &sCtx.hash, 1);

  pPK = (const u8 *)sqlite4_value_blob(pKey);
  nPK = sqlite4_value_bytes(pKey);
  
  nTnum = getVarint32(pPK, dummy);
  nPK -= nTnum;
  pPK += nTnum;

  for(i=0; rc==SQLITE4_OK && i<pInfo->nCol; i++){
    sqlite4_value *pArg = (sqlite4_value *)(&aArg[i]);
    if( pArg->flags & MEM_Str ){
      const char *zText;
      int nText;

      zText = (const char *)sqlite4_value_text(pArg);
      nText = sqlite4_value_bytes(pArg); 
      sCtx.iCol = i;
      rc = pInfo->pTokenizer->xTokenize(
          &sCtx, pInfo->p, zText, nText, fts5TokenizeCb
      );
    }
  }

  /* Allocate enough space to serialize all the stuff that needs to
  ** be inserted into the database. Specifically:
  **
  **   * Space for index record keys,
  **   * space for the size record and key for this document, and
  **   * space for the updated global size record for the document set.
  **
  ** To make it easier, the below allocates enough space to simultaneously
  ** store the largest index record key and the largest possible global
  ** size record.
  */
  nSpace = (sqlite4VarintLen(pInfo->iRoot) + 2 + sCtx.nMax + nPK) + 
           (9 * (2 + pInfo->nCol * sCtx.nStream));
  aSpace = sqlite4DbMallocRaw(db, nSpace);
  if( aSpace==0 ) rc = SQLITE4_NOMEM;

  for(pElem=sqliteHashFirst(&sCtx.hash); pElem; pElem=sqliteHashNext(pElem)){
    TokenizeTerm *pTerm = (TokenizeTerm *)sqliteHashData(pElem);
    if( rc==SQLITE4_OK ){
      int nToken = sqliteHashKeysize(pElem);
      char *zToken = (char *)sqliteHashKey(pElem);
      u8 *aKey = aSpace;
      int nKey;

      nKey = putVarint32(aKey, pInfo->iRoot);
      aKey[nKey++] = 0x24;
      memcpy(&aKey[nKey], zToken, nToken);
      nKey += nToken;
      aKey[nKey++] = 0x00;
      memcpy(&aKey[nKey], pPK, nPK);

        const KVByteArray *aData = (const KVByteArray *)&pTerm[1];
        aData += pTerm->nToken;
        rc = sqlite4KVStoreReplace(pStore, aKey, nKey, aData, pTerm->nData);
      }
    }
    sqlite4DbFree(db, pTerm);
  }

  /* Write the size record into the db */
  if( rc==SQLITE4_OK ){
    u8 *aKey = aSpace;
    int nKey;

    nKey = putVarint32(aKey, pInfo->iRoot);
    aKey[nKey++] = 0x00;
    memcpy(&aKey[nKey], pPK, nPK);
    nKey += nPK;

    if( bDel==0 ){
      Fts5Size sSz;
      sSz.nCol = pInfo->nCol;
      sSz.nStream = sCtx.nStream;
      sSz.aSz = sCtx.aSz;
      rc = fts5StoreSizeRecord(pStore, aKey, nKey, &sSz, -1, &aKey[nKey]);
    }else{
      rc = sqlite4KVStoreReplace(pStore, aKey, nKey, 0, -1);
    }
  }

  /* Update the global record */
  if( rc==SQLITE4_OK ){
    Fts5Size *pSz;                /* Deserialized global size record */
    i64 nRow;                     /* Number of rows in indexed table */
    u8 *aKey = aSpace;            /* Space to format the global record key */
    int nKey;                     /* Size of global record key in bytes */

    nKey = putVarint32(aKey, pInfo->iRoot);
    aKey[nKey++] = 0x00;
    rc = fts5LoadSizeRecord(db, aKey, nKey, sCtx.nStream, pInfo, &nRow, &pSz);
    assert( rc!=SQLITE4_OK || pSz->nStream>=sCtx.nStream );

    if( rc==SQLITE4_OK ){
      int iCol;
      for(iCol=0; iCol<pSz->nCol; iCol++){
        int iStr;
        i64 *aIn = &sCtx.aSz[iCol * sCtx.nStream];
        i64 *aOut = &pSz->aSz[iCol * pSz->nStream];
        for(iStr=0; iStr<sCtx.nStream; iStr++){
          aOut[iStr] += (aIn[iStr] * (bDel?-1:1));
        }
      }
      nRow += (bDel?-1:1);
      rc = fts5StoreSizeRecord(pStore, aKey, nKey, pSz, nRow, &aKey[nKey]);
    }

    sqlite4DbFree(db, pSz);
  }
  
  sqlite4DbFree(db, aSpace);
  sqlite4DbFree(db, sCtx.aSz);
  sqlite4HashClear(&sCtx.hash);
  return rc;
}

static Fts5Info *fts5InfoCreate(Parse *pParse, Index *pIdx, int bCol){
  sqlite4 *db = pParse->db;
  Fts5Info *pInfo;                /* p4 argument for FtsUpdate opcode */

    nByte += nCol * sizeof(char *);
  }

  pInfo = sqlite4DbMallocZero(db, nByte);
  if( pInfo ){
    pInfo->iDb = sqlite4SchemaToIndex(db, pIdx->pSchema);
    pInfo->iRoot = pIdx->tnum;
    pInfo->iTbl = sqlite4FindPrimaryKey(pIdx->pTable, 0)->tnum;
    pInfo->nCol = pIdx->pTable->nCol;
    fts5TokenizerCreate(pParse, pIdx->pFts, &pInfo->pTokenizer, &pInfo->p);

    if( pInfo->p==0 ){
      assert( pParse->nErr );
      sqlite4DbFree(db, pInfo);
      pInfo = 0;

**   * the weight assigned to the instance,
**   * the column number, and
**   * the term offset.
*/
static i64 fts5TermInstanceCksum(
  const u8 *aTerm, int nTerm,
  const u8 *aPk, int nPk,
  int iStream,
  int iCol,
  int iOff
){
  int i;
  i64 cksum = 0;

  /* Add the term to the checksum */
  for(i=0; i<nTerm; i++){
    cksum += (cksum << 3) + aTerm[i];
  }

  /* Add the primary key blob to the checksum */
  for(i=0; i<nPk; i++){
    cksum += (cksum << 3) + aPk[i];
  }

  /* Add the weight, column number and offset (in that order) to the checksum */
  cksum += (cksum << 3) + iStream;
  cksum += (cksum << 3) + iCol;
  cksum += (cksum << 3) + iOff;

  return cksum;
}

  u8 const *aVal; int nVal;       /* List of token instances */
  u8 const *aToken; int nToken;   /* Token for this entry */
  u8 const *aPk; int nPk;         /* Entry primary key blob */
  InstanceList sList;             /* Used to iterate through pVal */
  int nTnum;
  u32 tnum;


  aKey = (const u8 *)sqlite4_value_blob(pKey);
  nKey = sqlite4_value_bytes(pKey);
  aVal = (const u8 *)sqlite4_value_blob(pVal);
  nVal = sqlite4_value_bytes(pVal);

  /* Find the token and primary key blobs for this entry. */
  nTnum = getVarint32(aKey, tnum);
  if( aKey[nTnum]!=0 ){
    aToken = &aKey[nTnum+1];
    nToken = sqlite4Strlen30((const char *)aToken);
    aPk = &aToken[nToken+1];
    nPk = (&aKey[nKey] - aPk);

    fts5InstanceListInit((u8 *)aVal, nVal, &sList);
    while( 0==fts5InstanceListNext(&sList) ){
      i64 v = fts5TermInstanceCksum(
          aPk, nPk, aToken, nToken, sList.iStream, sList.iCol, sList.iOff
          );
      cksum = cksum ^ v;
    }
  }

  *piCksum = cksum;
  return SQLITE4_OK;
}

typedef struct CksumCtx CksumCtx;
struct CksumCtx {
  const u8 *pPK;
  int nPK;
  int iCol;
  i64 cksum;
};

static int fts5CksumCb(
  void *pCtx, 
  int iStream, 
  int iOff,
  const char *zToken, 
  int nToken, 
  int iSrc, 
  int nSrc
){
  CksumCtx *p = (CksumCtx *)pCtx;
  i64 cksum;

  cksum = fts5TermInstanceCksum(p->pPK, p->nPK, 
      (const u8 *)zToken, nToken, iStream, p->iCol, iOff
  );

  p->cksum = (p->cksum ^ cksum);
  return 0;
}

int sqlite4Fts5RowCksum(

      fts5InstanceListNext(&in2);
    }else if( in1.iCol<in2.iCol || (in1.iCol==in2.iCol && in1.iOff<in2.iOff) ){
      pAdv = &in1;
    }else{
      pAdv = &in2;
    }

    fts5InstanceListAppend(&out, pAdv->iCol, pAdv->iStream, pAdv->iOff);
    fts5InstanceListNext(pAdv);
  }

  if( bFree ){
    sqlite4DbFree(db, p1->aData);
    sqlite4DbFree(db, p2->aData);
  }

/*
** Open a cursor for each token in the expression.
*/
static int fts5OpenCursors(sqlite4 *db, Fts5Info *pInfo, Fts5Cursor *pCsr){
  return fts5OpenExprCursors(db, pInfo, pCsr->pExpr->pRoot);
}

void sqlite4Fts5Close(Fts5Cursor *pCsr){
  if( pCsr ){
    sqlite4 *db = pCsr->db;
    if( pCsr->aMem ){
      int i;
      for(i=0; i<pCsr->pInfo->nCol; i++){
        sqlite4DbFree(db, pCsr->aMem[i].zMalloc);
      }
      sqlite4DbFree(db, pCsr->aMem);
    }

    sqlite4KVCursorClose(pCsr->pCsr);
    fts5ExpressionFree(db, pCsr->pExpr);
    sqlite4DbFree(db, pCsr->pIter);
    sqlite4DbFree(db, pCsr->aKey);
    sqlite4DbFree(db, pCsr->anRow);
    sqlite4DbFree(db, pCsr);
  }
}

static int fts5TokenAdvanceToMatch(
  InstanceList *p,
  InstanceList *pFirst,

      return 0;
    }
  }

  return (p->iCol==pFirst->iCol && p->iOff==iReq);
}

static int fts5StringFindInstances(sqlite4 *db, int iCol, Fts5Str *pStr){

  int i;
  int rc = SQLITE4_OK;
  int bEof = 0;
  int nByte = sizeof(InstanceList) * pStr->nToken;
  InstanceList *aIn;
  InstanceList out;

  while( rc==SQLITE4_OK && bEof==0 ){
    for(i=1; i<pStr->nToken; i++){
      int bMatch = fts5TokenAdvanceToMatch(&aIn[i], &aIn[0], i, &bEof);
      if( bMatch==0 || bEof ) break;
    }
    if( i==pStr->nToken && (iCol<0 || aIn[0].iCol==iCol) ){
      /* Record a match here */
      fts5InstanceListAppend(&out, aIn[0].iCol, aIn[0].iStream, aIn[0].iOff);
    }
    bEof = fts5InstanceListNext(&aIn[0]);
  }

  pStr->nList = out.iList;
  sqlite4DbFree(db, aIn);

  return rc;
}

static int fts5IsNear(InstanceList *p1, InstanceList *p2, int nNear){
  if( p1->iCol==p2->iCol && p1->iOff<p2->iOff && (p1->iOff+nNear)>=p2->iOff ){
    return 1;
  }
  return 0;
}

static int fts5StringNearTrim(

  Fts5Str *pTrim,                 /* Trim this instance list */
  Fts5Str *pNext,                 /* According to this one */
  int nNear
){
  if( pNext->nList==0 ){
    pTrim->nList = 0;
  }else{

    while( bEof==0 ){
      if( fts5IsNear(&near, &in, nTrail) 
       || fts5IsNear(&in, &near, nLead)
      ){
        /* The current position is a match. Append an entry to the output
        ** and advance the input cursor. */
        fts5InstanceListAppend(&out, in.iCol, in.iStream, in.iOff);
        bEof = fts5InstanceListNext(&in);
      }else{
        if( near.iCol<in.iCol || (near.iCol==in.iCol && near.iOff<in.iOff) ){
          bEof = fts5InstanceListNext(&near);
        }else if( near.iCol==in.iCol && near.iOff==in.iOff ){
          bEof = fts5InstanceListNext(&in);
          if( fts5IsNear(&near, &in, nTrail) ){
            fts5InstanceListAppend(&out, near.iCol, near.iStream, near.iOff);
          }
        }else{
          bEof = fts5InstanceListNext(&in);
        }
      }
    }

** is set to true before returning.
**
** If the cursors do not point to a match, then *ppAdvance is set to
** the token of the individual cursor that should be advanced before
** retrying this function.
*/
static int fts5PhraseIsMatch(
  sqlite4 *db,                    /* Database handle */
  Fts5Phrase *pPhrase,            /* Phrase to test */
  int *pbMatch,                   /* OUT: True for a match, false otherwise */
  Fts5Token **ppAdvance           /* OUT: Token to advance before retrying */
){
  const u8 *aPk1 = 0;
  int nPk1 = 0;
  int rc = SQLITE4_OK;

  /* At this point, it is established that all of the token cursors in the
  ** phrase point to an entry with the same primary key. Now figure out if
  ** the various string constraints are met. Along the way, synthesize a 
  ** position list for each Fts5Str object.  */
  for(i=0; rc==SQLITE4_OK && i<pPhrase->nStr; i++){
    Fts5Str *pStr = &pPhrase->aStr[i];
    rc = fts5StringFindInstances(db, pPhrase->iCol, pStr);
  }

  /* Trim the instance lists according to any NEAR constraints.  */
  for(i=1; rc==SQLITE4_OK && i<pPhrase->nStr; i++){
    int n = pPhrase->aiNear[i-1];
    rc = fts5StringNearTrim(&pPhrase->aStr[i], &pPhrase->aStr[i-1], n);
  }
  for(i=pPhrase->nStr-1; rc==SQLITE4_OK && i>0; i--){
    int n = pPhrase->aiNear[i-1];
    rc = fts5StringNearTrim(&pPhrase->aStr[i-1], &pPhrase->aStr[i], n);
  }

  *pbMatch = (pPhrase->aStr[0].nList>0);
  return rc;
}

static int fts5PhraseAdvanceToMatch(sqlite4 *db, Fts5Phrase *pPhrase){
  int rc;
  do {
    int bMatch;
    Fts5Token *pAdvance = 0;
    rc = fts5PhraseIsMatch(db, pPhrase, &bMatch, &pAdvance);
    if( rc!=SQLITE4_OK || bMatch ) break;
    rc = fts5TokenAdvance(db, pAdvance);
  }while( rc==SQLITE4_OK );
  return rc;
}

static int fts5ExprAdvance(sqlite4 *db, Fts5ExprNode *p, int bFirst){
  int rc = SQLITE4_OK;

  switch( p->eType ){
    case TOKEN_PRIMITIVE: {
      Fts5Phrase *pPhrase = p->pPhrase;
      if( bFirst==0 ){
        rc = fts5TokenAdvance(db, &pPhrase->aStr[0].aToken[0]);
      }
      if( rc==SQLITE4_OK ) rc = fts5PhraseAdvanceToMatch(db, pPhrase);
      if( rc==SQLITE4_OK ){
        rc = fts5TokenPk(&pPhrase->aStr[0].aToken[0], &p->aPk, &p->nPk);
      }else{
        p->aPk = 0;
        p->nPk = 0;
        if( rc==SQLITE4_NOTFOUND ) rc = SQLITE4_OK;
      }
      break;
    }

    case TOKEN_AND:
      p->aPk = 0;
      p->nPk = 0;
      rc = fts5ExprAdvance(db, p->pLeft, bFirst);
      if( rc==SQLITE4_OK ) rc = fts5ExprAdvance(db, p->pRight, bFirst);
      while( rc==SQLITE4_OK && p->pLeft->aPk && p->pRight->aPk ){
        int res = fts5KeyCompare(
            p->pLeft->aPk, p->pLeft->nPk, p->pRight->aPk, p->pRight->nPk
        );
        if( res<0 ){
          rc = fts5ExprAdvance(db, p->pLeft, 0);
        }else if( res>0 ){
          rc = fts5ExprAdvance(db, p->pRight, 0);
        }else{
          p->aPk = p->pLeft->aPk;
          p->nPk = p->pLeft->nPk;
          break;
        }
      }
      break;

    case TOKEN_OR: {
      int res = 0;
      if( bFirst==0 ){
        res = fts5KeyCompare(
            p->pLeft->aPk, p->pLeft->nPk, p->pRight->aPk, p->pRight->nPk
        );
      }
        
      if( res<=0 ) rc = fts5ExprAdvance(db, p->pLeft, bFirst);
      if( rc==SQLITE4_OK && res>=0 ){
        rc = fts5ExprAdvance(db, p->pRight, bFirst);
      }

      res = fts5KeyCompare(
          p->pLeft->aPk, p->pLeft->nPk, p->pRight->aPk, p->pRight->nPk
      );
      if( res>0 ){
        p->aPk = p->pRight->aPk;

    default: assert( p->eType==TOKEN_NOT );

      p->aPk = 0;
      p->nPk = 0;

      rc = fts5ExprAdvance(db, p->pLeft, bFirst);
      if( bFirst && rc==SQLITE4_OK ){
        rc = fts5ExprAdvance(db, p->pRight, bFirst);
      }

      while( rc==SQLITE4_OK && p->pLeft->aPk && p->pRight->aPk ){
        int res = fts5KeyCompare(
            p->pLeft->aPk, p->pLeft->nPk, p->pRight->aPk, p->pRight->nPk
        );
        if( res<0 ){
          break;
        }else if( res>0 ){
          rc = fts5ExprAdvance(db, p->pRight, 0);
        }else{
          rc = fts5ExprAdvance(db, p->pLeft, 0);
        }
      }

      p->aPk = p->pLeft->aPk;
      p->nPk = p->pLeft->nPk;
      break;
  }

  assert( rc!=SQLITE4_NOTFOUND );
  return rc;
}

int sqlite4Fts5Next(Fts5Cursor *pCsr){
  sqlite4DbFree(pCsr->db, pCsr->pSz);
  pCsr->pSz = 0;
  pCsr->bMemValid = 0;
  return fts5ExprAdvance(pCsr->db, pCsr->pExpr->pRoot, 0);
}

int sqlite4Fts5Open(
  sqlite4 *db,                    /* Database handle */
  Fts5Info *pInfo,                /* Index description */
  const char *zMatch,             /* Match expression */
  int bDesc,                      /* True to iterate in desc. order of PK */
  Fts5Cursor **ppCsr,             /* OUT: New FTS cursor object */
  char **pzErr                    /* OUT: Error message */
){
  int rc = SQLITE4_OK;
  Fts5Cursor *pCsr;
  int nMatch = sqlite4Strlen30(zMatch);

  pCsr = sqlite4DbMallocZero(db, sizeof(Fts5Cursor) + nMatch + 1);

  if( !pCsr ){
    rc = SQLITE4_NOMEM;
  }else{
    pCsr->zExpr = (char *)&pCsr[1];
    memcpy(pCsr->zExpr, zMatch, nMatch);
    pCsr->pInfo = pInfo;
    pCsr->db = db;
    rc = fts5ParseExpression(db, pInfo->pTokenizer, pInfo->p, 
        pInfo->iRoot, pInfo->azCol, pInfo->nCol, zMatch, &pCsr->pExpr, pzErr
    );
  }

  if( rc==SQLITE4_OK ){
    /* Open a KV cursor for each term in the expression. Set each cursor
    ** to point to the first entry in the range it will scan.  */
    rc = fts5OpenCursors(db, pInfo, pCsr);
  }
  if( rc!=SQLITE4_OK ){
    sqlite4Fts5Close(pCsr);
    pCsr = 0;
  }else{
    rc = fts5ExprAdvance(db, pCsr->pExpr->pRoot, 1);
  }
  *ppCsr = pCsr;
  return rc;
}

/*
** Return true if the cursor passed as the second argument currently points
** to a valid entry, or false otherwise.
*/
int sqlite4Fts5Valid(Fts5Cursor *pCsr){
  return( pCsr->pExpr->pRoot->aPk!=0 );
}

int sqlite4Fts5Pk(
  Fts5Cursor *pCsr, 
  int iTbl,
  KVByteArray **paKey,
  KVSize *pnKey
){
  int i;
  int nReq;
  const u8 *aPk;
  int nPk;

  i = putVarint32(pCsr->aKey, iTbl);
  memcpy(&pCsr->aKey[i], aPk, nPk);

  *paKey = pCsr->aKey;
  *pnKey = nReq;
  return SQLITE4_OK;
}

int sqlite4_mi_column_count(sqlite4_context *pCtx, int *pn){
  int rc = SQLITE4_OK;
  if( pCtx->pFts ){
    *pn = pCtx->pFts->pInfo->nCol;
  }else{
    rc = SQLITE4_MISUSE;
  }
  return rc;
}

int sqlite4_mi_phrase_count(sqlite4_context *pCtx, int *pn){
  int rc = SQLITE4_OK;
  if( pCtx->pFts ){
    *pn = pCtx->pFts->pExpr->nPhrase;
  }else{
    rc = SQLITE4_MISUSE;
  }
  return rc;
}

int sqlite4_mi_phrase_token_count(sqlite4_context *pCtx, int iP, int *pn){
  int rc = SQLITE4_OK;
  if( pCtx->pFts ){
    Fts5Expr *pExpr = pCtx->pFts->pExpr;
    if( iP>pExpr->nPhrase || iP<0 ){
      *pn = 0;
    }else{
      *pn = pExpr->apPhrase[iP]->nToken;
    }
  }else{
    rc = SQLITE4_MISUSE;
  }
  return rc;
}

int sqlite4_mi_stream_count(sqlite4_context *pCtx, int *pn){
  int rc = SQLITE4_OK;
  Fts5Cursor *pCsr = pCtx->pFts;
  if( pCsr ){
    rc = fts5CsrLoadGlobal(pCtx->pFts);
    if( rc==SQLITE4_OK ) *pn = pCsr->pGlobal->nStream;
  }else{
    rc = SQLITE4_MISUSE;
  }
  return rc;
}

static int fts5GetSize(Fts5Size *pSz, int iC, int iS){
  int nToken = 0;
  int i;

  if( iC<0 && iS<0 ){
    int nFin = pSz->nCol * pSz->nStream;
    for(i=0; i<nFin; i++) nToken += pSz->aSz[i];
  }else if( iC<0 ){
    for(i=0; i<pSz->nCol; i++) nToken += pSz->aSz[i*pSz->nStream + iS];
  }else if( iS<0 ){
    for(i=0; i<pSz->nStream; i++) nToken += pSz->aSz[pSz->nStream*iC + i];
  }else if( iC<pSz->nCol && iS<pSz->nStream ){
    nToken = pSz->aSz[iC * pSz->nStream + iS];
  }

  return nToken;
}

int sqlite4_mi_size(sqlite4_context *pCtx, int iC, int iS, int *pn){
  int rc = SQLITE4_OK;
  Fts5Cursor *pCsr = pCtx->pFts;

  if( pCsr==0 ){
    rc = SQLITE4_MISUSE;
  }else{
    rc = fts5CsrLoadSz(pCsr);
    if( rc==SQLITE4_OK ){
      *pn = fts5GetSize(pCsr->pSz, iC, iS);
    }
  }
  return rc;
}

int sqlite4_mi_total_size(sqlite4_context *pCtx, int iC, int iS, int *pn){
  int rc = SQLITE4_OK;
  Fts5Cursor *pCsr = pCtx->pFts;

  if( pCsr==0 ){
    rc = SQLITE4_MISUSE;
  }else{
    rc = fts5CsrLoadGlobal(pCsr);
    if( rc==SQLITE4_OK ){
      *pn = fts5GetSize(pCsr->pGlobal, iC, iS);
    }
  }
  return rc;
}

int sqlite4_mi_total_rows(sqlite4_context *pCtx, int *pn){
  int rc = SQLITE4_OK;
  Fts5Cursor *pCsr = pCtx->pFts;
  if( pCsr==0 ){
    rc = SQLITE4_MISUSE;
  }else{
    rc = fts5CsrLoadGlobal(pCsr);
    if( rc==SQLITE4_OK ) *pn = pCsr->nGlobal;
  }
  return rc;
}

int sqlite4_mi_column_value(
  sqlite4_context *pCtx, 
  int iCol, 
  sqlite4_value **ppVal
){
  int rc = SQLITE4_OK;
  Fts5Cursor *pCsr = pCtx->pFts;
  if( pCsr==0 ){
    rc = SQLITE4_MISUSE;
  }else{
    if( pCsr->bMemValid==0 ){
      sqlite4 *db = pCsr->db;

      Fts5Info *pInfo = pCsr->pInfo;
      if( pCsr->aMem==0 ){
        int nByte = sizeof(Mem) * pInfo->nCol;
        pCsr->aMem = (Mem *)sqlite4DbMallocZero(db, nByte);
        if( pCsr->aMem==0 ){
          rc = SQLITE4_NOMEM;
        }else{
          int i;
          for(i=0; i<pInfo->nCol; i++){
            pCsr->aMem[i].db = db;
          }
        }
      }

      if( pCsr->pCsr==0 && rc==SQLITE4_OK ){
        KVStore *pStore = db->aDb[pInfo->iDb].pKV;
        rc = sqlite4KVStoreOpenCursor(pStore, &pCsr->pCsr);
      }

      if( rc==SQLITE4_OK ){
        u8 *aKey = 0; int nKey;     /* Primary key for current row */
        const u8 *aData; int nData; /* Data record for current row */

        rc = sqlite4Fts5Pk(pCsr, pInfo->iTbl, &aKey, &nKey);
        if( rc==SQLITE4_OK ){
          rc = sqlite4KVCursorSeek(pCsr->pCsr, aKey, nKey, 0);
          if( rc==SQLITE4_NOTFOUND ){
            rc = SQLITE4_CORRUPT_BKPT;
          }
        }

        if( rc==SQLITE4_OK ){
          rc = sqlite4KVCursorData(pCsr->pCsr, 0, -1, &aData, &nData);
        }

        if( rc==SQLITE4_OK ){
          int i;
          ValueDecoder *pCodec;   /* The decoder object */

          rc = sqlite4VdbeCreateDecoder(db, aData, nData, pInfo->nCol, &pCodec);
          for(i=0; rc==SQLITE4_OK && i<pInfo->nCol; i++){
            rc = sqlite4VdbeDecodeValue(pCodec, i, 0, &pCsr->aMem[i]);
          }
          sqlite4VdbeDestroyDecoder(pCodec);
        }

        if( rc==SQLITE4_OK ) pCsr->bMemValid = 1;
      }
    }

    if( rc==SQLITE4_OK ){
      assert( pCsr->bMemValid );
      *ppVal = &pCsr->aMem[iCol];
    }
  }

  return rc;
}

int sqlite4_mi_tokenize(
  sqlite4_context *pCtx,
  const char *zText,
  int nText,
  void *p,
  int(*x)(void *, int, int, const char *, int, int, int)
){
  int rc = SQLITE4_OK;
  Fts5Cursor *pCsr = pCtx->pFts;

  if( pCsr==0 ){
    rc = SQLITE4_MISUSE;
  }else{
    Fts5Info *pInfo = pCsr->pInfo;
    rc = pInfo->pTokenizer->xTokenize(p, pInfo->p, zText, nText, x);
  }
  return rc;
}

static Fts5Str *fts5FindStr(Fts5ExprNode *p, int *piStr){
  Fts5Str *pRet = 0;
  if( p->eType==TOKEN_PRIMITIVE ){
    int iStr = *piStr;
    if( iStr<p->pPhrase->nStr ){
      pRet = &p->pPhrase->aStr[iStr];
    }else{
      *piStr = iStr - p->pPhrase->nStr;
    }
  }else{
    pRet = fts5FindStr(p->pLeft, piStr);
    if( pRet==0 ) pRet = fts5FindStr(p->pRight, piStr);
  }
  return pRet;
}

int sqlite4_mi_match_count(
  sqlite4_context *pCtx, 
  int iC,
  int iS,
  int iPhrase,
  int *pnMatch
){
  int rc = SQLITE4_OK;
  Fts5Cursor *pCsr = pCtx->pFts;
  if( pCsr ){
    int nMatch = 0;
    Fts5Str *pStr;
    int iCopy = iPhrase;
    InstanceList sList;

    pStr = fts5FindStr(pCsr->pExpr->pRoot, &iCopy);
    assert( pStr );

    fts5InstanceListInit(pStr->aList, pStr->nList, &sList);
    while( 0==fts5InstanceListNext(&sList) ){
      if( (iC<0 || sList.iCol==iC) && (iS<0 || sList.iStream==iS) ) nMatch++;
    }
    *pnMatch = nMatch;
  }else{
    rc = SQLITE4_MISUSE;
  }
  return rc;
}

int sqlite4_mi_match_offset(
  sqlite4_context *pCtx, 
  int iCol, 
  int iPhrase, 
  int iMatch, 
  int *piOff
){
  return SQLITE4_OK;
}

int sqlite4_mi_total_match_count(
  sqlite4_context *pCtx,
  int iCol,
  int iPhrase,
  int *pnMatch,
  int *pnDoc,
  int *pnRelevant
){
  return SQLITE4_OK;
}

static void fts5StrLoadRowcounts(Fts5Str *pStr, int nStream, int *anRow){
  u32 mask = 0;
  int iPrevCol = 0;
  InstanceList sList;

  fts5InstanceListInit(pStr->aList, pStr->nList, &sList);
  while( 0==fts5InstanceListNext(&sList) ){
    if( sList.iCol!=iPrevCol ) mask = 0;
    if( (mask & (1<<sList.iStream))==0 ){
      anRow[sList.iCol * nStream + sList.iStream]++;
      mask |= (1<<sList.iStream);
      iPrevCol = sList.iCol;
    }
  }
}

static int fts5ExprLoadRowcounts(
  sqlite4 *db, 
  Fts5Info *pInfo,
  int nStream,
  Fts5ExprNode *pNode, 
  int **panRow
){
  int rc = SQLITE4_OK;

  if( pNode ){
    if( pNode->eType==TOKEN_PRIMITIVE ){
      int *anRow = *panRow;
      Fts5Phrase *pPhrase = pNode->pPhrase;

      rc = fts5ExprAdvance(db, pNode, 1);
      while( rc==SQLITE4_OK ){
        int nIncr =  pInfo->nCol * nStream;      /* Values for each Fts5Str */
        int i;
        for(i=0; i<pPhrase->nStr; i++){
          fts5StrLoadRowcounts(&pPhrase->aStr[i], nStream, &anRow[i*nIncr]);
        }
        rc = fts5ExprAdvance(db, pNode, 0);
      }

      *panRow = &anRow[pInfo->nCol * nStream * pPhrase->nStr];
    }

    if( rc==SQLITE4_OK ){
      rc = fts5ExprLoadRowcounts(db, pInfo, nStream, pNode->pLeft, panRow);
    }
    if( rc==SQLITE4_OK ){
      rc = fts5ExprLoadRowcounts(db, pInfo, nStream, pNode->pRight, panRow);
    }
  }

  return rc;
}

static int fts5CsrLoadRowcounts(Fts5Cursor *pCsr){
  int rc = SQLITE4_OK;

  if( pCsr->anRow==0 ){
    int nStream = pCsr->pGlobal->nStream;
    sqlite4 *db = pCsr->db;
    Fts5Expr *pCopy;
    Fts5Expr *pExpr = pCsr->pExpr;
    Fts5Info *pInfo = pCsr->pInfo;
    int *anRow;

    pCsr->anRow = anRow = (int *)sqlite4DbMallocZero(db, 
        pExpr->nPhrase * pInfo->nCol * pCsr->pGlobal->nStream * sizeof(int)
    );
    if( !anRow ) return SQLITE4_NOMEM;

    rc = fts5ParseExpression(db, pInfo->pTokenizer, pInfo->p, 
        pInfo->iRoot, pInfo->azCol, pInfo->nCol, pCsr->zExpr, &pCopy, 0
    );
    if( rc==SQLITE4_OK ){
      rc = fts5OpenExprCursors(db, pInfo, pExpr->pRoot);
    }
    if( rc==SQLITE4_OK ){
      rc = fts5ExprLoadRowcounts(db, pInfo, nStream, pCopy->pRoot, &anRow);
    }

    fts5ExpressionFree(db, pCopy);
  }

  return rc;
}

int sqlite4_mi_row_count(
  sqlite4_context *pCtx,          /* Context object passed to mi function */
  int iC,                         /* Specific column (or -ve for all columns) */
  int iS,                         /* Specific stream (or -ve for all streams) */
  int iP,                         /* Specific phrase */
  int *pn                         /* Total number of rows containing C/S/P */
){
  int rc = SQLITE4_OK;
  Fts5Cursor *pCsr = pCtx->pFts;
  if( pCsr==0 ){
    rc = SQLITE4_MISUSE;
  }else{
    rc = fts5CsrLoadGlobal(pCsr);
    if( rc==SQLITE4_OK ) rc = fts5CsrLoadRowcounts(pCsr);

    if( rc==SQLITE4_OK ){
      int i;
      int nRow = 0;
      int nStream = pCsr->pGlobal->nStream;
      int nCol = pCsr->pInfo->nCol;
      int *aRow = &pCsr->anRow[iP * nStream * nCol];

      if( iC<0 && iS<0 ){
        int nFin = nCol * nStream;
        for(i=0; i<nFin; i++) nRow += aRow[i];
      }else if( iC<0 ){
        for(i=0; i<nCol; i++) nRow += aRow[i*nStream + iS];
      }else if( iS<0 ){
        for(i=0; i<nStream; i++) nRow += aRow[nStream*iC + iS];
      }else if( iC<nCol && iS<nStream ){
        nRow = aRow[iC * nStream + iS];
      }

      *pn = nRow;
    }
  }
  return rc;
}

static void fts5IterSetCurrent(Fts5MatchIter *pIter, int nList){
  InstanceList *pBest = 0;
  int i;

  for(i=0; i<nList; i++){
    InstanceList *p = &pIter->aList[i];
    if( fts5InstanceListEof(p)==0 ){
      if( (pBest==0)
       || (p->iCol<pBest->iCol)
       || (p->iCol==pBest->iCol && p->iOff<pBest->iOff)
      ){
        pBest = p;
      }
    }
  }

  if( pBest==0 ){
    pIter->iCurrent = -1;
  }else{
    pIter->iCurrent = pBest - pIter->aList;
  }
}

static void fts5InitExprIterator(
  const u8 *aPk, 
  int nPk, 
  Fts5ExprNode *p,
  Fts5MatchIter *pIter
){
  if( p ){
    if( p->eType==TOKEN_PRIMITIVE ){
      if( p->nPk==nPk && 0==memcmp(aPk, p->aPk, nPk) ){
        int i;
        for(i=0; i<p->pPhrase->nStr; i++){
          Fts5Str *pStr = &p->pPhrase->aStr[i];
          InstanceList *pList = &pIter->aList[pIter->iCurrent++];
          fts5InstanceListInit(pStr->aList, pStr->nList, pList);
          fts5InstanceListNext(pList);
        }
      }else{
        memset(&pIter->aList[pIter->iCurrent], 0, sizeof(InstanceList));
        pIter->iCurrent += p->pPhrase->nStr;
      }
    }
    fts5InitExprIterator(aPk, nPk, p->pLeft, pIter);
    fts5InitExprIterator(aPk, nPk, p->pRight, pIter);
  }
}

static void fts5InitIterator(Fts5Cursor *pCsr){
  Fts5MatchIter *pIter = pCsr->pIter;
  Fts5ExprNode *pRoot = pCsr->pExpr->pRoot;

  pIter->iCurrent = 0;
  fts5InitExprIterator(pRoot->aPk, pRoot->nPk, pRoot, pIter);
  pIter->iMatch = 0;
  pIter->bValid = 1;
  fts5IterSetCurrent(pIter, pCsr->pExpr->nPhrase);
}

int sqlite4_mi_match_detail(
  sqlite4_context *pCtx,          /* Context object passed to mi function */
  int iMatch,                     /* Index of match */
  int *piOff,                     /* OUT: Token offset of match */
  int *piC,                       /* OUT: Column number of match iMatch */
  int *piS,                       /* OUT: Stream number of match iMatch */
  int *piP                        /* OUT: Phrase number of match iMatch */
){
  int rc = SQLITE4_OK;
  Fts5Cursor *pCsr = pCtx->pFts;
  if( pCsr==0 ){
    rc = SQLITE4_MISUSE;
  }else{
    int nPhrase = pCsr->pExpr->nPhrase;
    Fts5MatchIter *pIter = pCsr->pIter;
    if( pIter==0 ){
      pCsr->pIter = pIter = (Fts5MatchIter *)sqlite4DbMallocZero(
          pCsr->db, sizeof(Fts5MatchIter) + sizeof(InstanceList)*nPhrase
      );
      if( pIter ){
        pIter->aList = (InstanceList *)&pIter[1];
      }else{
        rc = SQLITE4_NOMEM;
      }
    }

    if( rc==SQLITE4_OK && (pIter->bValid==0 || iMatch<pIter->iMatch) ){
      fts5InitIterator(pCsr);
#if 0
      int i;
      for(i=0; i<pCsr->pExpr->nPhrase; i++){
        Fts5Str *pStr = pCsr->pExpr->apPhrase[i];
        fts5InstanceListInit(pStr->aList, pStr->nList, &pIter->aList[i]);
        fts5InstanceListNext(&pIter->aList[i]);
      }
      pIter->iMatch = 0;
      fts5IterSetCurrent(pIter, pCsr->pExpr->nPhrase);
#endif
    }

    if( rc==SQLITE4_OK ){
      assert( pIter->iMatch<=iMatch );
      while( pIter->iCurrent>=0 && pIter->iMatch<iMatch ){
        fts5InstanceListNext(&pIter->aList[pIter->iCurrent]);
        fts5IterSetCurrent(pIter, pCsr->pExpr->nPhrase);
        pIter->iMatch++;
      }
      if( pIter->iCurrent<0 ){
        rc = SQLITE4_NOTFOUND;
      }else{
        InstanceList *p = &pIter->aList[pIter->iCurrent];
        *piOff = p->iOff;
        *piC = p->iCol;
        *piS = p->iStream;
        *piP = pIter->iCurrent;
      }
    }
  }
  return rc;
}

/**************************************************************************
***************************************************************************
** Below this point is test code.
*/
#ifdef SQLITE4_TEST
static int fts5PrintExprNode(sqlite4 *, const char **, Fts5ExprNode *, char **);

/*
** 2012 December 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.
**
*************************************************************************
*/

/*
** The BM25 and BM25F implementations in this file are based on information
** found in:
**
**   Stephen Robertson and Hugo Zaragoza: "The Probablistic Relevance
**   Framework: BM25 and Beyond", 2009.
*/

#include "sqliteInt.h"
#include <math.h>                 /* temporary: For log() */

static char fts5Tolower(char c){
  if( c>='A' && c<='Z' ) c = c + ('a' - 'A');
  return c;
}

static int fts5SimpleCreate(
  void *pCtx, 
  const char **azArg, 
  int nArg, 
  sqlite4_tokenizer **pp
){
  *pp = (sqlite4_tokenizer *)pCtx;
  return SQLITE4_OK;
}

static int fts5SimpleDestroy(sqlite4_tokenizer *p){
  return SQLITE4_OK;
}

typedef struct Fts5RankCtx Fts5RankCtx;
struct Fts5RankCtx {
  sqlite4 *db;
  double avgdl;                   /* Average document size in tokens */
  int nPhrase;                    /* Number of phrases in query */
  double *aIdf;                   /* IDF weights for each phrase in query */
};

static void fts5RankFreeCtx(void *pCtx){
  if( pCtx ){
    Fts5RankCtx *p = (Fts5RankCtx *)pCtx;
    sqlite4DbFree(p->db, p);
  }
}

/*
** A BM25 based ranking function for fts5.
**
** This is based on the information in the Robertson/Zaragoza paper 
** referenced above. As there is no way to provide relevance feedback 
** IDF weights (equation 3.3 in R/Z) are used instead of RSJ for each phrase.
** The rest of the implementation is as presented in equation 3.15.
**
** R and Z observe that the experimental evidence suggests that reasonable
** values for free parameters "b" and "k1" are often in the ranges 
** (0.5 < b < 0.8) and (1.2 < k1 < 2), although the optimal values depend
** on the nature of both the documents and queries. The implementation
** below sets each parameter to the midpoint of the suggested range.
*/
static void fts5Rank(sqlite4_context *pCtx, int nArg, sqlite4_value **apArg){
  const double b = 0.65;
  const double k1 = 1.6;

  int rc = SQLITE4_OK;            /* Error code */
  Fts5RankCtx *p;                 /* Structure to store reusable values */
  int i;                          /* Used to iterate through phrases */
  double rank = 0.0;              /* UDF return value */

  p = sqlite4_get_auxdata(pCtx, 0);
  if( p==0 ){
    sqlite4 *db = sqlite4_context_db_handle(pCtx);
    int nPhrase;                  /* Number of phrases in query expression */
    int nByte;                    /* Number of bytes of data to allocate */

    sqlite4_mi_phrase_count(pCtx, &nPhrase);
    nByte = sizeof(Fts5RankCtx) + nPhrase * sizeof(double);
    p = (Fts5RankCtx *)sqlite4DbMallocZero(db, nByte);
    sqlite4_set_auxdata(pCtx, 0, (void *)p, fts5RankFreeCtx);
    p = sqlite4_get_auxdata(pCtx, 0);

    if( !p ){
      rc = SQLITE4_NOMEM;
    }else{
      int N;                      /* Total number of docs in collection */
      int ni;                     /* Number of docs with phrase i */

      p->db = db;
      p->nPhrase = nPhrase;
      p->aIdf = (double *)&p[1];

      /* Determine the IDF weight for each phrase in the query. */
      rc = sqlite4_mi_total_rows(pCtx, &N);
      for(i=0; rc==SQLITE4_OK && i<nPhrase; i++){
        rc = sqlite4_mi_row_count(pCtx, -1, -1, i, &ni);
        if( rc==SQLITE4_OK ){
          p->aIdf[i] = log((0.5 + N - ni) / (0.5 + ni));
        }
      }

      /* Determine the average document length */
      if( rc==SQLITE4_OK ){
        int nTotal;
        rc = sqlite4_mi_total_size(pCtx, -1, -1, &nTotal);
        if( rc==SQLITE4_OK ){
          p->avgdl = (double)nTotal / (double)N;
        }
      }
    }
  }

  for(i=0; rc==SQLITE4_OK && i<p->nPhrase; i++){
    int tf;                     /* Occurences of phrase i in row (term freq.) */
    int dl;                     /* Tokens in this row (document length) */
    double L;                   /* Normalized document length */
    double prank;               /* Contribution to rank of this phrase */

    /* Set variable tf to the total number of occurrences of phrase iPhrase
    ** in this row (within any column). And dl to the number of tokens in
    ** the current row (again, in any column).  */
    rc = sqlite4_mi_match_count(pCtx, -1, -1, i, &tf); 
    if( rc==SQLITE4_OK ) rc = sqlite4_mi_size(pCtx, -1, -1, &dl); 

    /* Calculate the normalized document length */
    L = (double)dl / p->avgdl;

    /* Calculate the contribution to the rank made by this phrase. Then
    ** add it to variable rank.  */
    prank = (p->aIdf[i] * tf) / (k1 * ( (1.0 - b) + b * L) + tf);
    rank += prank;
  }

  if( rc==SQLITE4_OK ){
    sqlite4_result_double(pCtx, rank);
  }else{
    sqlite4_result_error_code(pCtx, rc);
  }
}

typedef struct Snippet Snippet;
typedef struct SnippetText SnippetText;

struct Snippet {
  int iCol;
  int iOff;
  u64 hlmask;
};

struct SnippetText {
  char *zOut;                     /* Pointer to snippet text */
  int nOut;                       /* Size of zOut in bytes */
  int nAlloc;                     /* Bytes of space allocated at zOut */
};

typedef struct SnippetCtx SnippetCtx;
struct SnippetCtx {
  sqlite4 *db;                    /* Database handle */
  int nToken;                     /* Number of tokens in snippet */
  int iOff;                       /* First token in snippet */
  u64 mask;                       /* Snippet mask. Highlight these terms */
  const char *zStart;
  const char *zEnd;
  const char *zEllipses;

  SnippetText *pOut;

  int iFrom;
  int iTo;
  const char *zText;              /* Document to extract snippet from */
  int rc;                         /* Set to NOMEM if OOM is encountered */
};

static void fts5SnippetAppend(SnippetCtx *p, const char *z, int n){
  if( p->rc==SQLITE4_OK ){
    SnippetText *pOut = p->pOut;
    if( n<0 ) n = strlen(z);
    if( (pOut->nOut + n) > pOut->nAlloc ){
      int nNew = (pOut->nOut+n) * 2;

      pOut->zOut = sqlite4DbReallocOrFree(p->db, pOut->zOut, nNew);
      if( pOut->zOut==0 ){
        p->rc = SQLITE4_NOMEM;
        return;
      }
      pOut->nAlloc = sqlite4DbMallocSize(p->db, pOut->zOut);
    }

    memcpy(&pOut->zOut[pOut->nOut], z, n);
    pOut->nOut += n;
  }
}

static int fts5SnippetCb(
  void *pCtx, 
  int iStream, 
  int iOff, 
  const char *z, int n,
  int iSrc, int nSrc
){
  SnippetCtx *p = (SnippetCtx *)pCtx;

  if( iOff<p->iOff ){
    return 0;
  }else if( iOff>=(p->iOff + p->nToken) ){
    fts5SnippetAppend(p, &p->zText[p->iFrom], p->iTo - p->iFrom);
    fts5SnippetAppend(p, "...", 3);
    p->iFrom = -1;
    return 1;
  }else{
    int bHighlight;               /* True to highlight term */

    bHighlight = (p->mask & (1 << (iOff-p->iOff)));

    if( p->iFrom==0 && p->iOff!=0 ){
      p->iFrom = iSrc;
      if( p->pOut->nOut==0 ) fts5SnippetAppend(p, p->zEllipses, -1);
    }

    if( bHighlight ){
      fts5SnippetAppend(p, &p->zText[p->iFrom], iSrc - p->iFrom);
      fts5SnippetAppend(p, p->zStart, -1);
      fts5SnippetAppend(p, &p->zText[iSrc], nSrc);
      fts5SnippetAppend(p, p->zEnd, -1);
      p->iTo = p->iFrom = iSrc+nSrc;
    }else{
      p->iTo = iSrc + nSrc;
    }
  }

  return 0;
}

static int fts5SnippetText(
  sqlite4_context *pCtx, 
  Snippet *pSnip,
  SnippetText *pText,
  int nToken,
  const char *zStart,
  const char *zEnd,
  const char *zEllipses
){
  int rc;
  sqlite4_value *pVal = 0;

  u64 mask = pSnip->hlmask;
  int iOff = pSnip->iOff;
  int iCol = pSnip->iCol;

  rc = sqlite4_mi_column_value(pCtx, iCol, &pVal);
  if( rc==SQLITE4_OK ){
    SnippetCtx sCtx;
    int nText;

    nText = sqlite4_value_bytes(pVal);
    memset(&sCtx, 0, sizeof(sCtx));
    sCtx.zText = (const char *)sqlite4_value_text(pVal);
    sCtx.db = sqlite4_context_db_handle(pCtx);
    sCtx.nToken = nToken;
    sCtx.iOff = iOff;
    sCtx.mask = mask;
    sCtx.zStart = zStart;
    sCtx.zEnd = zEnd;
    sCtx.zEllipses = zEllipses;
    sCtx.pOut = pText;

    sqlite4_mi_tokenize(pCtx, sCtx.zText, nText, &sCtx, fts5SnippetCb);
    if( sCtx.rc==SQLITE4_OK && sCtx.iFrom>0 ){
      fts5SnippetAppend(&sCtx, &sCtx.zText[sCtx.iFrom], nText - sCtx.iFrom);
    }
    rc = sCtx.rc;
  }

  return rc;
}

static int fts5BestSnippet(
  sqlite4_context *pCtx,          /* Context snippet() was called in */
  int iColumn,                    /* In this column (-1 means any column) */
  u64 *pMask,                     /* IN/OUT: Mask of high-priority phrases */
  int nToken,                     /* Number of tokens in requested snippet */
  Snippet *pSnip                  /* Populate this object */
){
  sqlite4 *db = sqlite4_context_db_handle(pCtx);
  int nPhrase;
  int rc = SQLITE4_OK;
  int i;
  int iPrev = 0;
  int iPrevCol = 0;
  u64 *aMask;
  u64 mask = *pMask;
  u64 allmask = 0;

  int iBestOff = nToken-1;
  int iBestCol = (iColumn >= 0 ? iColumn : 0);
  int nBest = 0;
  u64 hlmask = 0;                 /* Highlight mask associated with iBestOff */
  u64 missmask = 0;               /* Mask of missing terms in iBestOff snip. */

  sqlite4_mi_phrase_count(pCtx, &nPhrase);
  aMask = sqlite4DbMallocZero(db, sizeof(u64) * nPhrase);
  if( !aMask ) return SQLITE4_NOMEM;

  /* Iterate through all matches for all phrases */
  for(i=0; rc==SQLITE4_OK; i++){
    int iOff;
    int iCol;
    int iStream;
    int iPhrase;

    rc = sqlite4_mi_match_detail(pCtx, i, &iOff, &iCol, &iStream, &iPhrase);
    if( rc==SQLITE4_OK ){
      u64 tmask = 0;
      u64 miss = 0;
      int iMask;
      int nShift; 
      int nScore = 0;

      int nPTok;
      int iPTok;

      if( iColumn>=0 && iColumn!=iCol ) continue;

      allmask |= (1 << iPhrase);

      nShift = ((iPrevCol==iCol) ? (iOff-iPrev) : 100);

      for(iMask=0; iMask<nPhrase; iMask++){
        if( nShift<64){
          aMask[iMask] = aMask[iMask] >> nShift;
        }else{
          aMask[iMask] = 0;
        }
      }
      sqlite4_mi_phrase_token_count(pCtx, iPhrase, &nPTok);
      for(iPTok=0; iPTok<nPTok; iPTok++){
        aMask[iPhrase] = aMask[iPhrase] | (1<<(nToken-1+iPTok));
      }

      for(iMask=0; iMask<nPhrase; iMask++){
        if( aMask[iMask] ){
          nScore += (((1 << iMask) & mask) ? 100 : 1);
        }else{
          miss |= (1 << iMask);
        }
        tmask = tmask | aMask[iMask];
      }

      if( nScore>nBest ){
        hlmask = tmask;
        missmask = miss;
        nBest = nScore;
        iBestOff = iOff;
        iBestCol = iCol;
      }

      iPrev = iOff;
      iPrevCol = iCol;
    }
  }
  if( rc==SQLITE4_NOTFOUND ) rc = SQLITE4_OK;

  pSnip->iOff = iBestOff-nToken+1;
  pSnip->iCol = iBestCol;
  pSnip->hlmask = hlmask;
  *pMask = mask & missmask & allmask;

  sqlite4DbFree(db, aMask);
  return rc;
}

static void fts5SnippetImprove(
  sqlite4_context *pCtx, 
  int nToken,                     /* Size of required snippet */
  int nSz,                        /* Total size of column in tokens */
  Snippet *pSnip
){
  int i;
  int nLead = 0;
  int nShift = 0;

  u64 mask = pSnip->hlmask;
  int iOff = pSnip->iOff;

  if( mask==0 ) return;
  assert( mask & (1 << (nToken-1)) );

  for(i=0; (mask & (1<<i))==0; i++);
  nLead = i;

  nShift = (nLead/2);
  if( iOff+nShift > nSz-nToken ) nShift = (nSz-nToken) - iOff;
  if( iOff+nShift < 0 ) nShift = -1 * iOff;

  iOff += nShift;
  mask = mask >> nShift;

  pSnip->iOff = iOff;
  pSnip->hlmask = mask;
}

static void fts5Snippet(sqlite4_context *pCtx, int nArg, sqlite4_value **apArg){
  Snippet aSnip[4];
  int nSnip;
  int iCol = -1;
  int nToken = -15;
  int rc;
  int nPhrase;

  const char *zStart = "<b>";
  const char *zEnd = "</b>";
  const char *zEllipses = "...";

  if( nArg>0 ) zStart = (const char *)sqlite4_value_text(apArg[0]);
  if( nArg>1 ) zEnd = (const char *)sqlite4_value_text(apArg[1]);
  if( nArg>2 ) zEllipses = (const char *)sqlite4_value_text(apArg[2]);
  if( nArg>3 ) iCol = sqlite4_value_int(apArg[3]);
  if( nArg>4 ) nToken = sqlite4_value_int(apArg[4]);

  rc = sqlite4_mi_phrase_count(pCtx, &nPhrase);
  for(nSnip=1; rc==SQLITE4_OK && nSnip<5; nSnip = ((nSnip==2) ? 3 : (nSnip+1))){
    int nTok;
    int i;
    u64 mask = ((u64)1 << nPhrase) - 1;

    if( nToken<0 ){
      nTok = nToken * -1;
    }else{
      nTok = (nToken + (nSnip-1)) / nSnip;
    }

    memset(aSnip, 0, sizeof(aSnip));
    for(i=0; rc==SQLITE4_OK && i<nSnip; i++){
      rc = fts5BestSnippet(pCtx, iCol, &mask, nTok, &aSnip[i]);
    }
    if( mask==0 || nSnip==4 ){
      SnippetText text = {0, 0, 0};
      for(i=0; rc==SQLITE4_OK && i<nSnip; i++){
        int nSz;
        rc = sqlite4_mi_size(pCtx, aSnip[i].iCol, -1, &nSz);
        if( rc==SQLITE4_OK ){
          fts5SnippetImprove(pCtx, nTok, nSz, &aSnip[i]);
          rc = fts5SnippetText(
              pCtx, &aSnip[i], &text, nTok, zStart, zEnd, zEllipses
          );
        }
      }
      sqlite4_result_text(pCtx, text.zOut, text.nOut, SQLITE4_TRANSIENT);
      sqlite4DbFree(sqlite4_context_db_handle(pCtx), text.zOut);
      break;
    }
  }

  if( rc!=SQLITE4_OK ){
    sqlite4_result_error_code(pCtx, rc);
  }
}

static int fts5SimpleTokenize(

  void *pCtx, sqlite4_tokenizer *p,
  const char *zDoc,
  int nDoc,
  int(*x)(void*, int, int, const char*, int, int, int)
){
  sqlite4_env *pEnv = (sqlite4_env *)p;
  char *aBuf;
  int nBuf;

  sqlite4_env *pEnv = sqlite4_db_env(db);

  rc = sqlite4_create_tokenizer(db, "simple", (void *)pEnv, 
      fts5SimpleCreate, fts5SimpleTokenize, fts5SimpleDestroy
  );
  if( rc!=SQLITE4_OK ) return rc;

  rc = sqlite4_create_mi_function(db, "rank", 0, SQLITE4_UTF8, 0, fts5Rank, 0);
  if( rc!=SQLITE4_OK ) return rc;

  rc = sqlite4_create_mi_function(
      db, "snippet", -1, SQLITE4_UTF8, 0, fts5Snippet, 0
  );
  return rc;
}

  if( pArg && pArg->nRef==0 ){
    assert( rc!=SQLITE4_OK );
    xDestroy(p);
    sqlite4DbFree(db, pArg);
  }

 out:
  rc = sqlite4ApiExit(db, rc);
  sqlite4_mutex_leave(db->mutex);
  return rc;
}

int sqlite4_create_mi_function(
  sqlite4 *db,
  const char *zFunc,
  int nArg,
  int enc,
  void *p,
  void (*xFunc)(sqlite4_context*,int,sqlite4_value **),
  void (*xDestroy)(void *)
){
  int rc;
  int n;

  n = nArg + (nArg>=0);
  sqlite4_mutex_enter(db->mutex);
  rc = sqlite4_create_function_v2(db, zFunc, n, enc, p, xFunc, 0,0,xDestroy);
  if( rc==SQLITE4_OK ){
    FuncDef *p = sqlite4FindFunction(db, zFunc, -1, n, enc, 0);
    p->bMatchinfo = 1;
  }
  rc = sqlite4ApiExit(db, rc);
  sqlite4_mutex_leave(db->mutex);
  return rc;
}

#ifndef SQLITE4_OMIT_UTF16
int sqlite4_create_function16(

    testcase( iCol==BMS );
    testcase( iCol==BMS-1 );
    pItem->colUsed |= ((Bitmask)1)<<(iCol>=BMS ? BMS-1 : iCol);
    ExprSetProperty(p, EP_Resolved);
  }
  return p;
}

static void resolveMatchArg(Parse *pParse, NameContext *pNC, Expr *pExpr){
  SrcList *pSrc = pNC->pSrcList;
  SrcListItem *pItem;
  char *zLhs;
  int i;
  Index *pIdx;

  if( pExpr->op!=TK_ID || pSrc==0 || pExpr==0 ){
    sqlite4ErrorMsg(pParse, "first argument xxx must be a table name");
    return;
  }
  zLhs = pExpr->u.zToken;

  for(i=0; i<pSrc->nSrc; i++){
    pItem = &pSrc->a[i];
    if( pItem->zAlias && sqlite4StrICmp(zLhs, pItem->zAlias)==0 ) break;
    if( pItem->zAlias==0 && sqlite4StrICmp(zLhs, pItem->zName)==0 ) break;
  }
  if( i==pSrc->nSrc ){
    sqlite4ErrorMsg(pParse, "no such table: %s", zLhs);
    return;
  }

  pExpr->op = TK_NULL;
  pExpr->iTable = pItem->iCursor;
  ExprSetProperty(pExpr, EP_Resolved);
}

static void resolveMatch(Parse *pParse, NameContext *pNC, Expr *pExpr){
  Expr *pLeft = pExpr->pLeft;
  SrcList *pSrc = pNC->pSrcList;
  SrcListItem *pItem;
  char *zLhs;
  int i;

             nId, zId);
        pNC->nErr++;
      }
      if( is_agg ){
        pExpr->op = TK_AGG_FUNCTION;
        pNC->hasAgg = 1;
      }

      if( pParse->nErr==0 ){
        if( pDef->bMatchinfo ){
          resolveMatchArg(pParse, pNC, n>0 ? pList->a[0].pExpr : 0);
        }
        if( is_agg ) pNC->allowAgg = 0;
        sqlite4WalkExprList(pWalker, pList);
        if( is_agg ) pNC->allowAgg = 1;
      }

      /* FIX ME:  Compute pExpr->affinity based on the expected return
      ** type of the function 
      */
      return WRC_Prune;
    }
#ifndef SQLITE4_OMIT_SUBQUERY
    case TK_SELECT:

  int (*xTokenize)(void*, sqlite4_tokenizer*,
      const char*, int, 
      int(*x)(void *ctx, int iWeight, int iOff, 
              const char *zToken, int nToken, int iSrc, int nSrc)
  ),
  int (*xDestroy)(sqlite4_tokenizer *)
);

/*
** CAPI4REF: Register a matchinfo function.
*/
int sqlite4_create_mi_function(
  sqlite4 *db,
  const char *zFunc,
  int nArg,
  int enc,
  void *p,
  void (*xFunc)(sqlite4_context*,int,sqlite4_value **),
  void (*xDestroy)(void *)
);

/*
** CAPIREF: Matchinfo APIs.
**
** Special functions that may be called from within matchinfo UDFs. All
** return an SQLite error code - SQLITE4_OK if successful, or some other
** error code otherwise.
**
** sqlite4_mi_column_count():
**   Set *pn to the number of columns in the queried table.
**
** sqlite4_mi_phrase_count():
**   Set *pn to the number of phrases in the query.
**
** sqlite4_mi_stream_count():
**   Set *pn to the number of streams in the FTS index.
**
** sqlite4_mi_phrase_token_count():
**   Set *pn to the number of tokens in phrase iP of the query.
**
** sqlite4_mi_size():
**   Set *pn to the number of tokens belonging to stream iS in the value 
**   stored in column iC of the current row. 
**
**   Either or both of iS and iC may be negative. If iC is negative, then the
**   output value is the total number of tokens for the specified stream (or
**   streams) across all table columns. Similarly, if iS is negative, the 
**   output value is the total number of tokens in the specified column or 
**   columns, regardless of stream.
**
** sqlite4_mi_total_size():
**   Similar to sqlite4_mi_size(), except the output parameter is set to
**   the total number of tokens belonging to the specified column(s) 
**   and stream(s) in all rows of the table, not just the current row.
**
** sqlite4_mi_total_rows():
**   Set *pn to the total number of rows in the indexed table.
**
** sqlite4_mi_row_count():
**   Set the output parameter to the total number of rows in the table that
**   contain at least one instance of the phrase identified by parameter
**   iP in the column(s) and stream(s) identified by parameters iC and iS.
**
** sqlite4_mi_match_count():
**   Set the output parameter to the total number of occurences of phrase
**   iP in the current row that belong to the column(s) and stream(s) 
**   identified by parameters iC and iS.
**
**   Parameter iP may also be negative. In this case, the output value is
**   set to the total number of occurrences of all query phrases in the
**   current row, subject to the constraints imposed by iC and iS.
**
** sqlite4_mi_match_detail():
**   This function is used to access the details of the iMatch'th match
**   (of any phrase) in the current row. Matches are sorted in order of
**   occurrence. If parameter iMatch is equal to or greater than the number 
**   of matches in the current row, SQLITE_NOTFOUND is returned. Otherwise,
**   unless an error occurs, SQLITE4_OK is returned and the *piOff, *piC, *piS,
**   and *piP output parameters are set to the token offset, column number,
**   stream number and phrase number respectively.
**
**   It is anticipated that this function be used to iterate through matches 
**   in order of occurrence. It is optimized so that it is fastest when 
**   called with the iMatch parameter set to 0, P or P+1, where P is the 
**   iMatch value passed to the previous call.
**
** sqlite4_mi_column_value():
**   Set *ppVal to point to an sqlite4_value object containing the value
**   read from column iCol of the current row. This object is valid until
**   the function callback returns.
*/
int sqlite4_mi_column_count(sqlite4_context *, int *pn);
int sqlite4_mi_phrase_count(sqlite4_context *, int *pn);
int sqlite4_mi_stream_count(sqlite4_context *, int *pn);
int sqlite4_mi_phrase_token_count(sqlite4_context *, int iP, int *pn);

int sqlite4_mi_total_size(sqlite4_context *, int iC, int iS, int *pn);
int sqlite4_mi_total_rows(sqlite4_context *, int *pn);

int sqlite4_mi_row_count(sqlite4_context *, int iC, int iS, int iP, int *pn);

int sqlite4_mi_size(sqlite4_context *, int iC, int iS, int *pn);
int sqlite4_mi_match_count(sqlite4_context *, int iC, int iS, int iP, int *pn);
int sqlite4_mi_match_detail(
    sqlite4_context *, int iMatch, int *piOff, int *piC, int *piS, int *piP
);
int sqlite4_mi_column_value(sqlite4_context *, int iC, sqlite4_value **ppVal);

int sqlite4_mi_tokenize(sqlite4_context *, const char *, int, void *,
  int(*x)(void *, int, int, const char *, int, int, int)
);



/*
** Undo the hack that converts floating point types to integer for
** builds on processors without floating point support.
*/
#ifdef SQLITE4_OMIT_FLOATING_POINT
# undef double

  FuncDef *pSameName;  /* Next with a different name but the same hash */
  void (*xFunc)(sqlite4_context*,int,sqlite4_value**); /* Regular function */
  void (*xStep)(sqlite4_context*,int,sqlite4_value**); /* Aggregate step */
  void (*xFinalize)(sqlite4_context*);                /* Aggregate finalizer */
  char *zName;         /* SQL name of the function. */
  FuncDef *pNextName;  /* Next function with a different name */
  FuncDestructor *pDestructor;   /* Reference counted destructor function */
  u8 bMatchinfo;       /* True for matchinfo function */
};

/*
** A table of SQL functions.  
**
** The content is a linked list of FuncDef structures with branches.  When
** there are two or more FuncDef objects with the same name, they are 

/*
** An instance of this structure is used as the p4 argument to some fts5
** related vdbe opcodes.
*/
struct Fts5Info {
  int iDb;                        /* Database containing this index */
  int iRoot;                      /* Root page number of index */
  int iTbl;                       /* Root page number of indexed table */
  int nCol;                       /* Number of columns in indexed table */
  char **azCol;                   /* Column names for table */
  Fts5Tokenizer *pTokenizer;      /* Tokenizer module */
  sqlite4_tokenizer *p;           /* Tokenizer instance */
};

int sqlite4WalkExpr(Walker*, Expr*);

void sqlite4Fts5FreeInfo(sqlite4 *db, Fts5Info *);
void sqlite4Fts5CodeUpdate(Parse *, Index *pIdx, int iRegPk, int iRegData, int);
void sqlite4Fts5CodeCksum(Parse *, Index *, int, int, int);
void sqlite4Fts5CodeQuery(Parse *, Index *, int, int, int);

int sqlite4Fts5Pk(Fts5Cursor *, int, KVByteArray **, KVSize *);
int sqlite4Fts5Next(Fts5Cursor *pCsr);

int sqlite4Fts5EntryCksum(sqlite4 *, Fts5Info *, Mem *, Mem *, i64 *);
int sqlite4Fts5RowCksum(sqlite4 *, Fts5Info *, Mem *, Mem *, i64 *);
int sqlite4Fts5Open(sqlite4*, Fts5Info*, const char*, int, Fts5Cursor**,char**);
int sqlite4Fts5Valid(Fts5Cursor *);
void sqlite4Fts5Close(Fts5Cursor *);

#endif /* _SQLITEINT_H_ */

** to retrieve the collation sequence set by this opcode is not available
** publicly, only to user functions defined in func.c.
*/
case OP_CollSeq: {
  assert( pOp->p4type==P4_COLLSEQ );
  break;
}

/* Opcode: Mifunction P1
*/
case OP_Mifunction: {
  pc++;
  pOp++;
  /* fall through to OP_Function */
};

/* Opcode: Function P1 P2 P3 P4 P5
**
** Invoke a user function (P4 is a pointer to a Function structure that
** defines the function) with P5 arguments taken from register P2 and
** successors.  The result of the function is stored in register P3.
** Register P3 must not be one of the function inputs.

    ctx.pFunc = ctx.pVdbeFunc->pFunc;
  }

  ctx.s.flags = MEM_Null;
  ctx.s.db = db;
  ctx.s.xDel = 0;
  ctx.s.zMalloc = 0;
  if( pOp[-1].opcode==OP_Mifunction ){
    ctx.pFts = p->apCsr[pOp[-1].p1]->pFts;
    apVal++;
    n--;
  }else{
    ctx.pFts = 0;
  }

  /* The output cell may already have a buffer allocated. Move
  ** the pointer to ctx.s so in case the user-function can use
  ** the already allocated buffer instead of allocating a new one.
  */
  sqlite4VdbeMemMove(&ctx.s, pOut);
  MemSetTypeFlag(&ctx.s, MEM_Null);

struct sqlite4_context {
  FuncDef *pFunc;       /* Pointer to function information.  MUST BE FIRST */
  VdbeFunc *pVdbeFunc;  /* Auxilary data, if created. */
  Mem s;                /* The return value is stored here */
  Mem *pMem;            /* Memory cell used to store aggregate context */
  int isError;          /* Error code returned by the function. */
  CollSeq *pColl;       /* Collating sequence */
  Fts5Cursor *pFts;     /* fts5 cursor for matchinfo functions */
};

/*
** An Explain object accumulates indented output which is helpful
** in describing recursive data structures.
*/
struct Explain {

** Close a VDBE cursor and release all the resources that cursor 
** happens to hold.
*/
void sqlite4VdbeFreeCursor(Vdbe *p, VdbeCursor *pCx){
  if( pCx==0 ){
    return;
  }
  sqlite4Fts5Close(pCx->pFts);
  if( pCx->pKVCur ){
    sqlite4KVCursorClose(pCx->pKVCur);
  }
  if( pCx->pTmpKV ){
    sqlite4KVStoreClose(pCx->pTmpKV);
  }
#ifndef SQLITE4_OMIT_VIRTUALTABLE

      while( pOp<pEnd ){
        if( pOp->p1==pLevel->iTabCur && pOp->opcode==OP_Column ){
          pOp->p1 = pLevel->iIdxCur;
        }
        pOp++;
      }
    }

    if( (pLevel->plan.wsFlags & WHERE_INDEXED)
     && (pLevel->plan.u.pIdx->eIndexType==SQLITE4_INDEX_FTS5)
    ){
      VdbeOp *pOp;
      VdbeOp *pEnd;

      assert( pLevel->iTabCur!=pLevel->iIdxCur );
      pOp = sqlite4VdbeGetOp(v, pWInfo->iTop);
      pEnd = &pOp[sqlite4VdbeCurrentAddr(v) - pWInfo->iTop];

      while( pOp<pEnd ){
        if( pOp->p1==pLevel->iTabCur && pOp->opcode==OP_Mifunction ){
          pOp->p1 = pLevel->iIdxCur;
        }
        pOp++;
      }
    }
  }

  /* Final cleanup
  */
  pParse->nQueryLoop = pWInfo->savedNQueryLoop;
  whereInfoFree(db, pWInfo);
  return;

# tree to be flushed to disk, 
#
populate_db_2
do_execsql_test 3.1 {
  BEGIN;
    INSERT INTO t1 VALUES(10, randstr(910, 910));
}
do_test 3.2 { sqlite4_lsm_config db main autoflush } [expr 1*1024*1024]
do_test 3.3 { sqlite4_lsm_config db main autoflush 4096 } 4096

do_test 3.4 {
  set res [list]
  db eval { SELECT a, length(b) AS l FROM t1 } {
    lappend res $a $l
    # The following commit will flush the in-memory tree to disk.

  CREATE INDEX ft ON t2 USING fts5(tukenizer=simple);
} {1 {unrecognized argument: "tukenizer"}}

do_catchsql_test 2.3 { 
  CREATE INDEX ft ON t2 USING fts5("a b c");
} {1 {unrecognized argument: "a b c"}}

breakpoint
do_catchsql_test 2.4 {
  CREATE INDEX ft ON t2 USING fts5(tokenizer="nosuch");
} {1 {no such tokenizer: "nosuch"}}

finish_test

  2 {a:a}  {1}
  3 {b:a}  {2}
  4 {c:a}  {1 2}
  5 {a:a*} {1}
} {
  do_execsql_test 7.$tn {SELECT docid FROM t7 WHERE t7 MATCH $expr} $res
}

#-------------------------------------------------------------------------
#
do_execsql_test 8.0 {
  CREATE TABLE t8(a PRIMARY KEY, b, c);
  CREATE INDEX i8 ON t8 USING fts5();
  INSERT INTO t8 VALUES('one', 'a b c', 'a a a');
  INSERT INTO t8 VALUES('two', 'd e f', 'b b b');
}

#do_execsql_test 8.1 {
#  SELECT rank(t8) FROM t8 WHERE t8 MATCH 'b a'
#}

do_execsql_test 9.0 {
  CREATE TABLE t9(a PRIMARY KEY, b);
  CREATE INDEX i9 ON t9 USING fts5();
  INSERT INTO t9 VALUES('one', 
    'a b c d e f g h i j k l m n o p q r s t u v w x y z ' ||
    'a b c d e f g h i j k l m n o p q r s t u v w x y z'
  );
}

#do_execsql_test 9.1 {
#  SELECT snippet(t9) FROM t9 WHERE t9 MATCH 'b'
#} 

do_execsql_test 10.1 {
  CREATE TABLE ft(content);
  CREATE INDEX fti ON ft USING fts5();
}
do_execsql_test 10.2 {
  INSERT INTO ft VALUES('a b c d e');
  INSERT INTO ft VALUES('f g h i j');
}
do_execsql_test 10.3 { SELECT rowid FROM ft WHERE ft MATCH 'c' } {1}
do_execsql_test 10.4 { SELECT rowid FROM ft WHERE ft MATCH 'f' } {2}

do_execsql_test 10.5 {
  DELETE FROM ft;
  CREATE TABLE ft2(a, b, c);
  CREATE INDEX fti2 ON ft2 USING fts5();
  INSERT INTO ft2 VALUES('1 2 3 4 5', '6 7 8 9 10', '11 12 13 14 15');
  SELECT snippet(ft2, '[', ']', '...', -1, 3) FROM ft2 WHERE ft2 MATCH '5';
} {{...3 4 [5]}}

finish_test

# 2013 January 10
#
# 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.
#
#*************************************************************************
#
# The tests in this file test the FTS5 snippet() function.
#

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

# If SQLITE4_ENABLE_FTS3 is not defined, omit this file.
source $testdir/fts3_common.tcl

set DO_MALLOC_TEST 0

# Transform the list $L to its "normal" form. So that it can be compared to
# another list with the same set of elements using [string compare].
#
proc normalize {L} {
  set ret [list]
  foreach l $L {lappend ret $l}
  return $ret
}

# Document text used by a few tests. Contains the English names of all
# integers between 1 and 300.
#
set numbers [normalize {
  one two three four five six seven eight nine ten eleven twelve thirteen
  fourteen fifteen sixteen seventeen eighteen nineteen twenty twentyone
  twentytwo twentythree twentyfour twentyfive twentysix twentyseven
  twentyeight twentynine thirty thirtyone thirtytwo thirtythree thirtyfour
  thirtyfive thirtysix thirtyseven thirtyeight thirtynine forty fortyone
  fortytwo fortythree fortyfour fortyfive fortysix fortyseven fortyeight
  fortynine fifty fiftyone fiftytwo fiftythree fiftyfour fiftyfive fiftysix
  fiftyseven fiftyeight fiftynine sixty sixtyone sixtytwo sixtythree sixtyfour
  sixtyfive sixtysix sixtyseven sixtyeight sixtynine seventy seventyone
  seventytwo seventythree seventyfour seventyfive seventysix seventyseven
  seventyeight seventynine eighty eightyone eightytwo eightythree eightyfour
  eightyfive eightysix eightyseven eightyeight eightynine ninety ninetyone
  ninetytwo ninetythree ninetyfour ninetyfive ninetysix ninetyseven
  ninetyeight ninetynine onehundred onehundredone onehundredtwo
  onehundredthree onehundredfour onehundredfive onehundredsix onehundredseven
  onehundredeight onehundrednine onehundredten onehundredeleven
  onehundredtwelve onehundredthirteen onehundredfourteen onehundredfifteen
  onehundredsixteen onehundredseventeen onehundredeighteen onehundrednineteen
  onehundredtwenty onehundredtwentyone onehundredtwentytwo
  onehundredtwentythree onehundredtwentyfour onehundredtwentyfive
  onehundredtwentysix onehundredtwentyseven onehundredtwentyeight
  onehundredtwentynine onehundredthirty onehundredthirtyone
  onehundredthirtytwo onehundredthirtythree onehundredthirtyfour
  onehundredthirtyfive onehundredthirtysix onehundredthirtyseven
  onehundredthirtyeight onehundredthirtynine onehundredforty
  onehundredfortyone onehundredfortytwo onehundredfortythree
  onehundredfortyfour onehundredfortyfive onehundredfortysix
  onehundredfortyseven onehundredfortyeight onehundredfortynine
  onehundredfifty onehundredfiftyone onehundredfiftytwo onehundredfiftythree
  onehundredfiftyfour onehundredfiftyfive onehundredfiftysix
  onehundredfiftyseven onehundredfiftyeight onehundredfiftynine
  onehundredsixty onehundredsixtyone onehundredsixtytwo onehundredsixtythree
  onehundredsixtyfour onehundredsixtyfive onehundredsixtysix
  onehundredsixtyseven onehundredsixtyeight onehundredsixtynine
  onehundredseventy onehundredseventyone onehundredseventytwo
  onehundredseventythree onehundredseventyfour onehundredseventyfive
  onehundredseventysix onehundredseventyseven onehundredseventyeight
  onehundredseventynine onehundredeighty onehundredeightyone
  onehundredeightytwo onehundredeightythree onehundredeightyfour
  onehundredeightyfive onehundredeightysix onehundredeightyseven
  onehundredeightyeight onehundredeightynine onehundredninety
  onehundredninetyone onehundredninetytwo onehundredninetythree
  onehundredninetyfour onehundredninetyfive onehundredninetysix
  onehundredninetyseven onehundredninetyeight onehundredninetynine twohundred
  twohundredone twohundredtwo twohundredthree twohundredfour twohundredfive
  twohundredsix twohundredseven twohundredeight twohundrednine twohundredten
  twohundredeleven twohundredtwelve twohundredthirteen twohundredfourteen
  twohundredfifteen twohundredsixteen twohundredseventeen twohundredeighteen
  twohundrednineteen twohundredtwenty twohundredtwentyone twohundredtwentytwo
  twohundredtwentythree twohundredtwentyfour twohundredtwentyfive
  twohundredtwentysix twohundredtwentyseven twohundredtwentyeight
  twohundredtwentynine twohundredthirty twohundredthirtyone
  twohundredthirtytwo twohundredthirtythree twohundredthirtyfour
  twohundredthirtyfive twohundredthirtysix twohundredthirtyseven
  twohundredthirtyeight twohundredthirtynine twohundredforty
  twohundredfortyone twohundredfortytwo twohundredfortythree
  twohundredfortyfour twohundredfortyfive twohundredfortysix
  twohundredfortyseven twohundredfortyeight twohundredfortynine
  twohundredfifty twohundredfiftyone twohundredfiftytwo twohundredfiftythree
  twohundredfiftyfour twohundredfiftyfive twohundredfiftysix
  twohundredfiftyseven twohundredfiftyeight twohundredfiftynine
  twohundredsixty twohundredsixtyone twohundredsixtytwo twohundredsixtythree
  twohundredsixtyfour twohundredsixtyfive twohundredsixtysix
  twohundredsixtyseven twohundredsixtyeight twohundredsixtynine
  twohundredseventy twohundredseventyone twohundredseventytwo
  twohundredseventythree twohundredseventyfour twohundredseventyfive
  twohundredseventysix twohundredseventyseven twohundredseventyeight
  twohundredseventynine twohundredeighty twohundredeightyone
  twohundredeightytwo twohundredeightythree twohundredeightyfour
  twohundredeightyfive twohundredeightysix twohundredeightyseven
  twohundredeightyeight twohundredeightynine twohundredninety
  twohundredninetyone twohundredninetytwo twohundredninetythree
  twohundredninetyfour twohundredninetyfive twohundredninetysix
  twohundredninetyseven twohundredninetyeight twohundredninetynine
  threehundred
}]

foreach {DO_MALLOC_TEST enc} {
  0 utf8
  1 utf8
  1 utf16
} {
if {$DO_MALLOC_TEST || $enc=="utf16"} continue

  db close
  forcedelete test.db
  sqlite4 db test.db
  sqlite4_db_config_lookaside db 0 0 0
  db eval "PRAGMA encoding = \"$enc\""

  # Set variable $T to the test name prefix for this iteration of the loop.
  #
  set T "fts5snippet-$enc"
  
  ##########################################################################
  # Test the snippet function.
  #
  proc do_snippet_test {name expr iCol nTok args} {
    set res [list]
    foreach a $args { lappend res [string trim $a] }
    do_select_test $name {
      SELECT snippet(ft,'{','}','...',$iCol,$nTok) FROM ft WHERE ft MATCH $expr
    } $res
  }
  do_test $T.3.1 {
    execsql {
      DROP TABLE IF EXISTS ft;
      CREATE TABLE ft(content);
      CREATE INDEX fti ON ft USING fts5();

      INSERT INTO ft VALUES('one two three four five six seven eight nine ten');
    }
  } {}
  do_snippet_test $T.3.2  one    0 5 "{one} two three four five..."
  do_snippet_test $T.3.3  two    0 5 "one {two} three four five..."
  do_snippet_test $T.3.4  three  0 5 "one two {three} four five..."
  do_snippet_test $T.3.5  four   0 5 "...two three {four} five six..."
  do_snippet_test $T.3.6  five   0 5 "...three four {five} six seven..."
  do_snippet_test $T.3.7  six    0 5 "...four five {six} seven eight..."
  do_snippet_test $T.3.8  seven  0 5 "...five six {seven} eight nine..."
  do_snippet_test $T.3.9  eight  0 5 "...six seven {eight} nine ten"
  do_snippet_test $T.3.10 nine   0 5 "...six seven eight {nine} ten"
  do_snippet_test $T.3.11 ten    0 5 "...six seven eight nine {ten}"
  
  do_test $T.4.1 {
    execsql {
      INSERT INTO ft VALUES(
           'one two three four five '
        || 'six seven eight nine ten '
        || 'eleven twelve thirteen fourteen fifteen '
        || 'sixteen seventeen eighteen nineteen twenty '
        || 'one two three four five '
        || 'six seven eight nine ten '
        || 'eleven twelve thirteen fourteen fifteen '
        || 'sixteen seventeen eighteen nineteen twenty'
      );
    }
  } {}
  
  do_snippet_test $T.4.2 {one nine} 0 5 {
     {one} two three...eight {nine} ten
  } {
     {one} two three...eight {nine} ten...
  }
  
  do_snippet_test $T.4.3 {one nine} 0 -5 {
     {one} two three four five...six seven eight {nine} ten
  } {
     {one} two three four five...seven eight {nine} ten eleven...
  }
  do_snippet_test $T.4.3 {one nineteen} 0 -5 {
     ...eighteen {nineteen} twenty {one} two...
  }
  do_snippet_test $T.4.4 {two nineteen} 0 -5 {
     ...eighteen {nineteen} twenty one {two}...
  }
  do_snippet_test $T.4.5 {three nineteen} 0 -5 {
     ...{nineteen} twenty one two {three}...
  }
  
  do_snippet_test $T.4.6 {four nineteen} 0 -5 {
     ...two three {four} five six...seventeen eighteen {nineteen} twenty one...
  }
  do_snippet_test $T.4.7 {four NEAR nineteen} 0 -5 {
     ...seventeen eighteen {nineteen} twenty one...two three {four} five six...
  }
  
  do_snippet_test $T.4.8 {four nineteen} 0 5 {
     ...three {four} five...eighteen {nineteen} twenty...
  }
  do_snippet_test $T.4.9 {four NEAR nineteen} 0 5 {
     ...eighteen {nineteen} twenty...three {four} five...
  }
  do_snippet_test $T.4.10 {four NEAR nineteen} 0 -5 {
     ...seventeen eighteen {nineteen} twenty one...two three {four} five six...
  }
  do_snippet_test $T.4.11 {four NOT (nineteen+twentyone)} 0 5 {
     ...two three {four} five six...
  } {
     ...two three {four} five six...
  }
  do_snippet_test $T.4.12 {four OR nineteen NEAR twentyone} 0 5 {
     ...two three {four} five six...
  } {
     ...two three {four} five six...
  }
  
  do_test $T.5.1 {
    execsql {
      DROP TABLE IF EXISTS ft;
      CREATE TABLE ft(a, b, c);
      CREATE INDEX fti ON ft USING fts5();
      INSERT INTO ft VALUES(
        'one two three four five', 
        'four five six seven eight', 
        'seven eight nine ten eleven'
      );
    }
  } {}
  
  do_snippet_test $T.5.2 {five} -1 3 {...three four {five}}
  do_snippet_test $T.5.3 {five}  0 3 {...three four {five}}
  do_snippet_test $T.5.4 {five}  1 3 {four {five} six...}
  do_snippet_test $T.5.5 {five}  2 3 {seven eight nine...}
  
  do_test $T.5.6 {
    execsql { UPDATE ft SET b = NULL }
  } {}
  
  do_snippet_test $T.5.7  {five} -1 3 {...three four {five}}
  do_snippet_test $T.5.8  {five}  0 3 {...three four {five}}
  do_snippet_test $T.5.9  {five}  1 3 {}
  do_snippet_test $T.5.10 {five}  2 3 {seven eight nine...}
  
  do_snippet_test $T.5.11 {one "seven eight nine"} -1 -3 {
    {one} two three...{seven} {eight} {nine}...
  }

  do_test $T.6.1 {
    execsql {
      DROP TABLE IF EXISTS ft;
      CREATE TABLE ft(x);
      CREATE INDEX fti ON ft USING fts5();
      INSERT INTO ft VALUES($numbers);
    }
  } {}
  do_snippet_test $T.6.2 {
    one fifty onehundred onehundredfifty twohundredfifty threehundred
  } -1 4 {
    {one}...{fifty}...{onehundred}...{onehundredfifty}...
  }
  do_snippet_test $T.6.3 {
    one fifty onehundred onehundredfifty twohundredfifty threehundred
  } -1 -4 {
    {one} two three four...fortyeight fortynine {fifty} fiftyone...ninetyeight ninetynine {onehundred} onehundredone...onehundredfortyeight onehundredfortynine {onehundredfifty} onehundredfiftyone...
  }

  do_test $T.7.1 {
    execsql {
      BEGIN;
        DROP TABLE IF EXISTS ft;
        CREATE TABLE ft(x);
        CREATE INDEX fti ON ft USING fts5();
    }
    set testresults [list]
    for {set i 1} {$i < 150} {incr i} {
      set commas [string repeat , $i]
      execsql {INSERT INTO ft VALUES('one' || $commas || 'two')}
      lappend testresults "{one}$commas{two}"
    }
    execsql COMMIT
  } {}
  eval [list do_snippet_test $T.7.2 {one two} -1 3] $testresults
  
}

finish_test

  simple.test simple2.test
  log3.test 
  lsm1.test lsm2.test
  csr1.test
  ckpt1.test
  mc1.test
  fts5expr1.test fts5query1.test fts5rnd1.test fts5create.test
  fts5snippet.test

  aggerror.test
  attach.test
  autoindex1.test
  badutf.test
  between.test
  bigrow.test