SQLite

  }
  return pRet;
}

typedef struct TokenCtx TokenCtx;
struct TokenCtx {
  Fts5ExprPhrase *pPhrase;
  int rc;
};

/*
** Callback for tokenizing terms used by ParseTerm().
*/
static int fts5ParseTokenize(
  void *pContext,                 /* Pointer to Fts5InsertCtx object */
  int tflags,                     /* Mask of FTS5_TOKEN_* flags */
  const char *pToken,             /* Buffer containing token */
  int nToken,                     /* Size of token in bytes */
  int iUnused1,                   /* Start offset of token */
  int iUnused2                    /* End offset of token */
){
  int rc = SQLITE_OK;
  const int SZALLOC = 8;
  TokenCtx *pCtx = (TokenCtx*)pContext;
  Fts5ExprPhrase *pPhrase = pCtx->pPhrase;

  /* If an error has already occurred, this is a no-op */
  if( pCtx->rc!=SQLITE_OK ) return pCtx->rc;

  assert( pPhrase==0 || pPhrase->nTerm>0 );
  if( pPhrase && (tflags & FTS5_TOKEN_COLOCATED) ){
    Fts5ExprTerm *pSyn;
    int nByte = sizeof(Fts5ExprTerm) + nToken+1;
    pSyn = (Fts5ExprTerm*)sqlite3_malloc(nByte);
    if( pSyn==0 ) return SQLITE_NOMEM;
    memset(pSyn, 0, nByte);
    pSyn->zTerm = (char*)&pSyn[1];
    memcpy(pSyn->zTerm, pToken, nToken);
    pSyn->pSynonym = pPhrase->aTerm[pPhrase->nTerm-1].pSynonym;
    pPhrase->aTerm[pPhrase->nTerm-1].pSynonym = pSyn;
    if( pSyn==0 ){
      rc = SQLITE_NOMEM;
    }else{
      memset(pSyn, 0, nByte);
      pSyn->zTerm = (char*)&pSyn[1];
      memcpy(pSyn->zTerm, pToken, nToken);
      pSyn->pSynonym = pPhrase->aTerm[pPhrase->nTerm-1].pSynonym;
      pPhrase->aTerm[pPhrase->nTerm-1].pSynonym = pSyn;
    }
  }else{
    Fts5ExprTerm *pTerm;
    if( pPhrase==0 || (pPhrase->nTerm % SZALLOC)==0 ){
      Fts5ExprPhrase *pNew;
      int nNew = SZALLOC + (pPhrase ? pPhrase->nTerm : 0);

      pNew = (Fts5ExprPhrase*)sqlite3_realloc(pPhrase, 
          sizeof(Fts5ExprPhrase) + sizeof(Fts5ExprTerm) * nNew
      );
      if( pNew==0 ) return SQLITE_NOMEM;
      if( pPhrase==0 ) memset(pNew, 0, sizeof(Fts5ExprPhrase));
      pCtx->pPhrase = pPhrase = pNew;
      pNew->nTerm = nNew - SZALLOC;
    }

    pTerm = &pPhrase->aTerm[pPhrase->nTerm++];
    memset(pTerm, 0, sizeof(Fts5ExprTerm));
    pTerm->zTerm = sqlite3Fts5Strndup(&rc, pToken, nToken);
  }
      if( pNew==0 ){
        rc = SQLITE_NOMEM;
      }else{
        if( pPhrase==0 ) memset(pNew, 0, sizeof(Fts5ExprPhrase));
        pCtx->pPhrase = pPhrase = pNew;
        pNew->nTerm = nNew - SZALLOC;
      }
    }

    if( rc==SQLITE_OK ){
      pTerm = &pPhrase->aTerm[pPhrase->nTerm++];
      memset(pTerm, 0, sizeof(Fts5ExprTerm));
      pTerm->zTerm = sqlite3Fts5Strndup(&rc, pToken, nToken);
    }
  }

  pCtx->rc = rc;
  return rc;
}


/*
** Free the phrase object passed as the only argument.
*/

    int flags = FTS5_TOKENIZE_QUERY | (bPrefix ? FTS5_TOKENIZE_QUERY : 0);
    int n;
    sqlite3Fts5Dequote(z);
    n = strlen(z);
    rc = sqlite3Fts5Tokenize(pConfig, flags, z, n, &sCtx, fts5ParseTokenize);
  }
  sqlite3_free(z);
  if( rc ){
  if( rc || (rc = sCtx.rc) ){
    pParse->rc = rc;
    fts5ExprPhraseFree(sCtx.pPhrase);
    sCtx.pPhrase = 0;
  }else if( sCtx.pPhrase ){

    if( pAppend==0 ){
      if( (pParse->nPhrase % 8)==0 ){

  int i;                          /* Used to iterate through phrase terms */

  Fts5Expr *pNew = 0;             /* Expression to return via *ppNew */
  Fts5ExprPhrase **apPhrase;      /* pNew->apPhrase */
  Fts5ExprNode *pNode;            /* pNew->pRoot */
  Fts5ExprNearset *pNear;         /* pNew->pRoot->pNear */

  TokenCtx sCtx = {0};            /* Context object for fts5ParseTokenize */
  TokenCtx sCtx = {0,0};          /* Context object for fts5ParseTokenize */


  pOrig = pExpr->apExprPhrase[iPhrase];

  pNew = (Fts5Expr*)sqlite3Fts5MallocZero(&rc, sizeof(Fts5Expr));
  if( rc==SQLITE_OK ){
    pNew->apExprPhrase = (Fts5ExprPhrase**)sqlite3Fts5MallocZero(&rc, 

  Fts5Context *pFts,              /* First arg to pass to pApi functions */
  sqlite3_context *pCtx,          /* Context for returning result/error */
  int nVal,                       /* Number of values in apVal[] array */
  sqlite3_value **apVal           /* Array of trailing arguments */
){
  const char *zArg;
  Fts5MatchinfoCtx *p;
  int rc;
  int rc = SQLITE_OK;

  if( nVal>0 ){
    zArg = (const char*)sqlite3_value_text(apVal[0]);
  }else{
    zArg = "pcx";
  }

  p = (Fts5MatchinfoCtx*)pApi->xGetAuxdata(pFts, 0);
  if( p==0 || sqlite3_stricmp(zArg, p->zArg) ){
    p = fts5MatchinfoNew(pApi, pFts, pCtx, zArg);
    if( p==0 ){
      rc = SQLITE_NOMEM;
    }else{
    pApi->xSetAuxdata(pFts, p, sqlite3_free);
      rc = pApi->xSetAuxdata(pFts, p, sqlite3_free);
    if( p==0 ) return;
  }

  rc = fts5MatchinfoIter(pApi, pFts, p, fts5MatchinfoLocalCb);
    }
  }

  if( rc==SQLITE_OK ){
    rc = fts5MatchinfoIter(pApi, pFts, p, fts5MatchinfoLocalCb);
  }
  if( rc!=SQLITE_OK ){
    sqlite3_result_error_code(pCtx, rc);
  }else{
    /* No errors has occured, so return a copy of the array of integers. */
    int nByte = p->nRet * sizeof(u32);
    sqlite3_result_blob(pCtx, (void*)p->aRet, nByte, SQLITE_TRANSIENT);
  }

proc OR {args} {
  sort_poslist [concat {*}$args]
}
proc NOT {a b} {
  if {[llength $b]>0} { return [list] }
  return $a
}

#-------------------------------------------------------------------------
# This command is similar to [split], except that it also provides the
# start and end offsets of each token. For example:
#
#   [fts5_tokenize_split "abc d ef"] -> {abc 0 3 d 4 5 ef 6 8}
#

proc gobble_whitespace {textvar} {
  upvar $textvar t
  regexp {([ ]*)(.*)} $t -> space t
  return [string length $space]
}

proc gobble_text {textvar wordvar} {
  upvar $textvar t
  upvar $wordvar w
  regexp {([^ ]*)(.*)} $t -> w t
  return [string length $w]
}

proc fts5_tokenize_split {text} {
  set token ""
  set ret [list]
  set iOff [gobble_whitespace text]
  while {[set nToken [gobble_text text word]]} {
    lappend ret $word $iOff [expr $iOff+$nToken]
    incr iOff $nToken
    incr iOff [gobble_whitespace text]
  }

  set ret
}

} -body {
  db eval { 
    CREATE VIRTUAL TABLE yu USING fts5(x, tokenize="unicode61 separators abc");
  }
} -test {
  faultsim_test_result {0 {}}
}

#-------------------------------------------------------------------------
# OOM while running a query that includes synonyms and matchinfo().
#
proc mit {blob} {
  set scan(littleEndian) i*
  set scan(bigEndian) I*
  binary scan $blob $scan($::tcl_platform(byteOrder)) r
  return $r
}
proc tcl_tokenize {tflags text} {
  foreach {w iStart iEnd} [fts5_tokenize_split $text] {
    sqlite3_fts5_token $w $iStart $iEnd
    if {$tflags=="query" && [string length $w]==1} {
      for {set i 2} {$i < 7} {incr i} {
        sqlite3_fts5_token -colo [string repeat $w $i] $iStart $iEnd
      }
    }
  }
}
proc tcl_create {args} { return "tcl_tokenize" }
reset_db
sqlite3_fts5_create_tokenizer db tcl tcl_create
db func mit mit
sqlite3_fts5_register_matchinfo db
do_test 5.0 {
  execsql { CREATE VIRTUAL TABLE t1 USING fts5(a, tokenize=tcl) }
  foreach {rowid text} {
    1 {aaaa cc b aaaaa cc aa} 
    2 {aa aa bb a bbb}
    3 {bb aaaaa aaaaa b aaaa aaaaa}
    4 {aa a b aaaa aa}
    5 {aa b ccc aaaaa cc}
    6 {aa aaaaa bbbb cc aaa}
    7 {aaaaa aa aa ccccc bb}
    8 {ccc bbbbb ccccc bbb c}
    9 {cccccc bbbb a aaa cccc c}
  } {
    execsql { INSERT INTO t1(rowid, a) VALUES($rowid, $text) }
  }
} {}

set res [list {*}{
  1 {3 24 8 2 12 6}
  5 {2 24 8 2 12 6}
  6 {3 24 8 1 12 6}
  7 {3 24 8 1 12 6}
  9 {2 24 8 3 12 6}
}]
do_execsql_test 5.1 {
  SELECT rowid, mit(matchinfo(t1, 'x')) FROM t1 WHERE t1 MATCH 'a AND c'
} $res

faultsim_save_and_close
do_faultsim_test 5.2 -faults oom* -prep {
  faultsim_restore_and_reopen
  sqlite3_fts5_create_tokenizer db tcl tcl_create
  sqlite3_fts5_register_matchinfo db
  db func mit mit
} -body {
  db eval { 
    SELECT rowid, mit(matchinfo(t1, 'x')) FROM t1 WHERE t1 MATCH 'a AND c'
  }
} -test {
  faultsim_test_result [list 0 $::res]
}

finish_test

  foreach s $S {
    set o [list]
    foreach x $S {if {$x!=$s} {lappend o $x}}
    set ::syn($s) $o
  }
}

proc gobble_whitespace {textvar} {
  upvar $textvar t
  regexp {([ ]*)(.*)} $t -> space t
  return [string length $space]
}

proc gobble_text {textvar wordvar} {
  upvar $textvar t
  upvar $wordvar w
  regexp {([^ ]*)(.*)} $t -> w t
  return [string length $w]
}

proc do_tokenize_split {text} {
  set token ""
  set ret [list]
  set iOff [gobble_whitespace text]
  while {[set nToken [gobble_text text word]]} {
    lappend ret $word $iOff [expr $iOff+$nToken]
    incr iOff $nToken
    incr iOff [gobble_whitespace text]
  }

  set ret
}

proc tcl_tokenize {tflags text} {
  foreach {w iStart iEnd} [do_tokenize_split $text] {
  foreach {w iStart iEnd} [fts5_tokenize_split $text] {
    sqlite3_fts5_token $w $iStart $iEnd
  }
}

proc tcl_create {args} {
  return "tcl_tokenize"
}

#-------------------------------------------------------------------------
# Test a tokenizer that supports synonyms by adding extra entries to the
# FTS index.
#

proc tcl_tokenize {tflags text} {
  foreach {w iStart iEnd} [do_tokenize_split $text] {
  foreach {w iStart iEnd} [fts5_tokenize_split $text] {
    sqlite3_fts5_token $w $iStart $iEnd
    if {$tflags=="document" && [info exists ::syn($w)]} {
      foreach s $::syn($w) {
        sqlite3_fts5_token -colo $s $iStart $iEnd
      }
    }
  }

#   3.2.*: A tokenizer that reports two identical tokens at the same position.
#          This is allowed.
#
reset_db
sqlite3_fts5_create_tokenizer db tcl tcl_create
proc tcl_tokenize {tflags text} {
  set bColo 1
  foreach {w iStart iEnd} [do_tokenize_split $text] {
  foreach {w iStart iEnd} [fts5_tokenize_split $text] {
    if {$bColo} {
      sqlite3_fts5_token -colo $w $iStart $iEnd
      set bColo 0
    } {
      sqlite3_fts5_token $w $iStart $iEnd
    }
  }

}

do_execsql_test 3.1.1 {
  INSERT INTO ft(ft) VALUES('integrity-check');
} {}

proc tcl_tokenize {tflags text} {
  foreach {w iStart iEnd} [do_tokenize_split $text] {
  foreach {w iStart iEnd} [fts5_tokenize_split $text] {
    sqlite3_fts5_token $w $iStart $iEnd
  }
}

do_execsql_test 3.1.2 {
  SELECT rowid FROM ft WHERE ft MATCH 'one two three'
} {1}

reset_db
sqlite3_fts5_create_tokenizer db tcl tcl_create
proc tcl_tokenize {tflags text} {
  foreach {w iStart iEnd} [do_tokenize_split $text] {
  foreach {w iStart iEnd} [fts5_tokenize_split $text] {
    sqlite3_fts5_token $w $iStart $iEnd
    sqlite3_fts5_token -colo $w $iStart $iEnd
  }
}
do_execsql_test 3.2.0 {
  CREATE VIRTUAL TABLE ft USING fts5(x, tokenize = tcl);
  INSERT INTO ft VALUES('one one two three');

#-------------------------------------------------------------------------
# Check that expressions with synonyms can be parsed and executed.
#
reset_db
sqlite3_fts5_create_tokenizer db tcl tcl_create
proc tcl_tokenize {tflags text} {
  foreach {w iStart iEnd} [do_tokenize_split $text] {
  foreach {w iStart iEnd} [fts5_tokenize_split $text] {
    sqlite3_fts5_token $w $iStart $iEnd
    if {$tflags=="query" && [info exists ::syn($w)]} {
      foreach s $::syn($w) {
        sqlite3_fts5_token -colo $s $iStart $iEnd
      }
    }
  }

#-------------------------------------------------------------------------
# Test terms with more than 4 synonyms.
#
reset_db
sqlite3_fts5_create_tokenizer db tcl tcl_create
proc tcl_tokenize {tflags text} {
  foreach {w iStart iEnd} [do_tokenize_split $text] {
  foreach {w iStart iEnd} [fts5_tokenize_split $text] {
    sqlite3_fts5_token $w $iStart $iEnd
    if {$tflags=="query" && [string length $w]==1} {
      for {set i 2} {$i<=10} {incr i} {
        sqlite3_fts5_token -colo [string repeat $w $i] $iStart $iEnd
      }
    }
  }