15821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)/* fts1 has a design flaw which can lead to database corruption (see
25821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)** below).  It is recommended not to use it any longer, instead use
35821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)** fts3 (or higher).  If you believe that your use of fts1 is safe,
45821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)** add -DSQLITE_ENABLE_BROKEN_FTS1=1 to your CFLAGS.
55821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)*/
65821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)#if (!defined(SQLITE_CORE) || defined(SQLITE_ENABLE_FTS1)) \
75821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)        && !defined(SQLITE_ENABLE_BROKEN_FTS1)
82a99a7e74a7f215066514fe81d2bfa6639d9edddTorne (Richard Coles)#error fts1 has a design flaw and has been deprecated.
95821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)#endif
105821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)/* The flaw is that fts1 uses the content table's unaliased rowid as
115821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)** the unique docid.  fts1 embeds the rowid in the index it builds,
125821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)** and expects the rowid to not change.  The SQLite VACUUM operation
135821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)** will renumber such rowids, thereby breaking fts1.  If you are using
145821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)** fts1 in a system which has disabled VACUUM, then you can continue
155821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)** to use it safely.  Note that PRAGMA auto_vacuum does NOT disable
165821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)** VACUUM, though systems using auto_vacuum are unlikely to invoke
175821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)** VACUUM.
185821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)**
195821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)** fts1 should be safe even across VACUUM if you only insert documents
205821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)** and never delete.
215821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)*/
225821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
235821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)/* The author disclaims copyright to this source code.
245821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles) *
255821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles) * This is an SQLite module implementing full-text search.
265821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles) */
275821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
285821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)/*
295821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)** The code in this file is only compiled if:
305821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)**
315821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)**     * The FTS1 module is being built as an extension
325821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)**       (in which case SQLITE_CORE is not defined), or
335821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)**
345821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)**     * The FTS1 module is being built into the core of
355821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)**       SQLite (in which case SQLITE_ENABLE_FTS1 is defined).
365821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)*/
375821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)#if !defined(SQLITE_CORE) || defined(SQLITE_ENABLE_FTS1)
385821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
395821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)#if defined(SQLITE_ENABLE_FTS1) && !defined(SQLITE_CORE)
405821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)# define SQLITE_CORE 1
415821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)#endif
425821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
435821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)#include <assert.h>
445821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)#include <stdlib.h>
455821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)#include <stdio.h>
465821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)#include <string.h>
475821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)#include <ctype.h>
485821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
495821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)#include "fts1.h"
505821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)#include "fts1_hash.h"
515821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)#include "fts1_tokenizer.h"
522a99a7e74a7f215066514fe81d2bfa6639d9edddTorne (Richard Coles)#include "sqlite3.h"
535821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)#include "sqlite3ext.h"
545821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)SQLITE_EXTENSION_INIT1
555821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
565821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
575821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)#if 0
5858e6fbe4ee35d65e14b626c557d37565bf8ad179Ben Murdoch# define TRACE(A)  printf A; fflush(stdout)
595821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)#else
605821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)# define TRACE(A)
615821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)#endif
625821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
635821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)/* utility functions */
645821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
655821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)typedef struct StringBuffer {
665821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  int len;      /* length, not including null terminator */
675821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  int alloced;  /* Space allocated for s[] */
685821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  char *s;      /* Content of the string */
695821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)} StringBuffer;
705821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
715821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)static void initStringBuffer(StringBuffer *sb){
725821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  sb->len = 0;
735821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  sb->alloced = 100;
745821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  sb->s = malloc(100);
755821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  sb->s[0] = '\0';
765821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)}
775821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)
785821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)static void nappend(StringBuffer *sb, const char *zFrom, int nFrom){
795821806d5e7f356e8fa4b058a389a808ea183019Torne (Richard Coles)  if( sb->len + nFrom >= sb->alloced ){
80    sb->alloced = sb->len + nFrom + 100;
81    sb->s = realloc(sb->s, sb->alloced+1);
82    if( sb->s==0 ){
83      initStringBuffer(sb);
84      return;
85    }
86  }
87  memcpy(sb->s + sb->len, zFrom, nFrom);
88  sb->len += nFrom;
89  sb->s[sb->len] = 0;
90}
91static void append(StringBuffer *sb, const char *zFrom){
92  nappend(sb, zFrom, strlen(zFrom));
93}
94
95/* We encode variable-length integers in little-endian order using seven bits
96 * per byte as follows:
97**
98** KEY:
99**         A = 0xxxxxxx    7 bits of data and one flag bit
100**         B = 1xxxxxxx    7 bits of data and one flag bit
101**
102**  7 bits - A
103** 14 bits - BA
104** 21 bits - BBA
105** and so on.
106*/
107
108/* We may need up to VARINT_MAX bytes to store an encoded 64-bit integer. */
109#define VARINT_MAX 10
110
111/* Write a 64-bit variable-length integer to memory starting at p[0].
112 * The length of data written will be between 1 and VARINT_MAX bytes.
113 * The number of bytes written is returned. */
114static int putVarint(char *p, sqlite_int64 v){
115  unsigned char *q = (unsigned char *) p;
116  sqlite_uint64 vu = v;
117  do{
118    *q++ = (unsigned char) ((vu & 0x7f) | 0x80);
119    vu >>= 7;
120  }while( vu!=0 );
121  q[-1] &= 0x7f;  /* turn off high bit in final byte */
122  assert( q - (unsigned char *)p <= VARINT_MAX );
123  return (int) (q - (unsigned char *)p);
124}
125
126/* Read a 64-bit variable-length integer from memory starting at p[0].
127 * Return the number of bytes read, or 0 on error.
128 * The value is stored in *v. */
129static int getVarint(const char *p, sqlite_int64 *v){
130  const unsigned char *q = (const unsigned char *) p;
131  sqlite_uint64 x = 0, y = 1;
132  while( (*q & 0x80) == 0x80 ){
133    x += y * (*q++ & 0x7f);
134    y <<= 7;
135    if( q - (unsigned char *)p >= VARINT_MAX ){  /* bad data */
136      assert( 0 );
137      return 0;
138    }
139  }
140  x += y * (*q++);
141  *v = (sqlite_int64) x;
142  return (int) (q - (unsigned char *)p);
143}
144
145static int getVarint32(const char *p, int *pi){
146 sqlite_int64 i;
147 int ret = getVarint(p, &i);
148 *pi = (int) i;
149 assert( *pi==i );
150 return ret;
151}
152
153/*** Document lists ***
154 *
155 * A document list holds a sorted list of varint-encoded document IDs.
156 *
157 * A doclist with type DL_POSITIONS_OFFSETS is stored like this:
158 *
159 * array {
160 *   varint docid;
161 *   array {
162 *     varint position;     (delta from previous position plus POS_BASE)
163 *     varint startOffset;  (delta from previous startOffset)
164 *     varint endOffset;    (delta from startOffset)
165 *   }
166 * }
167 *
168 * Here, array { X } means zero or more occurrences of X, adjacent in memory.
169 *
170 * A position list may hold positions for text in multiple columns.  A position
171 * POS_COLUMN is followed by a varint containing the index of the column for
172 * following positions in the list.  Any positions appearing before any
173 * occurrences of POS_COLUMN are for column 0.
174 *
175 * A doclist with type DL_POSITIONS is like the above, but holds only docids
176 * and positions without offset information.
177 *
178 * A doclist with type DL_DOCIDS is like the above, but holds only docids
179 * without positions or offset information.
180 *
181 * On disk, every document list has positions and offsets, so we don't bother
182 * to serialize a doclist's type.
183 *
184 * We don't yet delta-encode document IDs; doing so will probably be a
185 * modest win.
186 *
187 * NOTE(shess) I've thought of a slightly (1%) better offset encoding.
188 * After the first offset, estimate the next offset by using the
189 * current token position and the previous token position and offset,
190 * offset to handle some variance.  So the estimate would be
191 * (iPosition*w->iStartOffset/w->iPosition-64), which is delta-encoded
192 * as normal.  Offsets more than 64 chars from the estimate are
193 * encoded as the delta to the previous start offset + 128.  An
194 * additional tiny increment can be gained by using the end offset of
195 * the previous token to make the estimate a tiny bit more precise.
196*/
197
198/* It is not safe to call isspace(), tolower(), or isalnum() on
199** hi-bit-set characters.  This is the same solution used in the
200** tokenizer.
201*/
202/* TODO(shess) The snippet-generation code should be using the
203** tokenizer-generated tokens rather than doing its own local
204** tokenization.
205*/
206/* TODO(shess) Is __isascii() a portable version of (c&0x80)==0? */
207static int safe_isspace(char c){
208  return (c&0x80)==0 ? isspace(c) : 0;
209}
210static int safe_tolower(char c){
211  return (c&0x80)==0 ? tolower(c) : c;
212}
213static int safe_isalnum(char c){
214  return (c&0x80)==0 ? isalnum(c) : 0;
215}
216
217typedef enum DocListType {
218  DL_DOCIDS,              /* docids only */
219  DL_POSITIONS,           /* docids + positions */
220  DL_POSITIONS_OFFSETS    /* docids + positions + offsets */
221} DocListType;
222
223/*
224** By default, only positions and not offsets are stored in the doclists.
225** To change this so that offsets are stored too, compile with
226**
227**          -DDL_DEFAULT=DL_POSITIONS_OFFSETS
228**
229*/
230#ifndef DL_DEFAULT
231# define DL_DEFAULT DL_POSITIONS
232#endif
233
234typedef struct DocList {
235  char *pData;
236  int nData;
237  DocListType iType;
238  int iLastColumn;    /* the last column written */
239  int iLastPos;       /* the last position written */
240  int iLastOffset;    /* the last start offset written */
241} DocList;
242
243enum {
244  POS_END = 0,        /* end of this position list */
245  POS_COLUMN,         /* followed by new column number */
246  POS_BASE
247};
248
249/* Initialize a new DocList to hold the given data. */
250static void docListInit(DocList *d, DocListType iType,
251                        const char *pData, int nData){
252  d->nData = nData;
253  if( nData>0 ){
254    d->pData = malloc(nData);
255    memcpy(d->pData, pData, nData);
256  } else {
257    d->pData = NULL;
258  }
259  d->iType = iType;
260  d->iLastColumn = 0;
261  d->iLastPos = d->iLastOffset = 0;
262}
263
264/* Create a new dynamically-allocated DocList. */
265static DocList *docListNew(DocListType iType){
266  DocList *d = (DocList *) malloc(sizeof(DocList));
267  docListInit(d, iType, 0, 0);
268  return d;
269}
270
271static void docListDestroy(DocList *d){
272  free(d->pData);
273#ifndef NDEBUG
274  memset(d, 0x55, sizeof(*d));
275#endif
276}
277
278static void docListDelete(DocList *d){
279  docListDestroy(d);
280  free(d);
281}
282
283static char *docListEnd(DocList *d){
284  return d->pData + d->nData;
285}
286
287/* Append a varint to a DocList's data. */
288static void appendVarint(DocList *d, sqlite_int64 i){
289  char c[VARINT_MAX];
290  int n = putVarint(c, i);
291  d->pData = realloc(d->pData, d->nData + n);
292  memcpy(d->pData + d->nData, c, n);
293  d->nData += n;
294}
295
296static void docListAddDocid(DocList *d, sqlite_int64 iDocid){
297  appendVarint(d, iDocid);
298  if( d->iType>=DL_POSITIONS ){
299    appendVarint(d, POS_END);  /* initially empty position list */
300    d->iLastColumn = 0;
301    d->iLastPos = d->iLastOffset = 0;
302  }
303}
304
305/* helper function for docListAddPos and docListAddPosOffset */
306static void addPos(DocList *d, int iColumn, int iPos){
307  assert( d->nData>0 );
308  --d->nData;  /* remove previous terminator */
309  if( iColumn!=d->iLastColumn ){
310    assert( iColumn>d->iLastColumn );
311    appendVarint(d, POS_COLUMN);
312    appendVarint(d, iColumn);
313    d->iLastColumn = iColumn;
314    d->iLastPos = d->iLastOffset = 0;
315  }
316  assert( iPos>=d->iLastPos );
317  appendVarint(d, iPos-d->iLastPos+POS_BASE);
318  d->iLastPos = iPos;
319}
320
321/* Add a position to the last position list in a doclist. */
322static void docListAddPos(DocList *d, int iColumn, int iPos){
323  assert( d->iType==DL_POSITIONS );
324  addPos(d, iColumn, iPos);
325  appendVarint(d, POS_END);  /* add new terminator */
326}
327
328/*
329** Add a position and starting and ending offsets to a doclist.
330**
331** If the doclist is setup to handle only positions, then insert
332** the position only and ignore the offsets.
333*/
334static void docListAddPosOffset(
335  DocList *d,             /* Doclist under construction */
336  int iColumn,            /* Column the inserted term is part of */
337  int iPos,               /* Position of the inserted term */
338  int iStartOffset,       /* Starting offset of inserted term */
339  int iEndOffset          /* Ending offset of inserted term */
340){
341  assert( d->iType>=DL_POSITIONS );
342  addPos(d, iColumn, iPos);
343  if( d->iType==DL_POSITIONS_OFFSETS ){
344    assert( iStartOffset>=d->iLastOffset );
345    appendVarint(d, iStartOffset-d->iLastOffset);
346    d->iLastOffset = iStartOffset;
347    assert( iEndOffset>=iStartOffset );
348    appendVarint(d, iEndOffset-iStartOffset);
349  }
350  appendVarint(d, POS_END);  /* add new terminator */
351}
352
353/*
354** A DocListReader object is a cursor into a doclist.  Initialize
355** the cursor to the beginning of the doclist by calling readerInit().
356** Then use routines
357**
358**      peekDocid()
359**      readDocid()
360**      readPosition()
361**      skipPositionList()
362**      and so forth...
363**
364** to read information out of the doclist.  When we reach the end
365** of the doclist, atEnd() returns TRUE.
366*/
367typedef struct DocListReader {
368  DocList *pDoclist;  /* The document list we are stepping through */
369  char *p;            /* Pointer to next unread byte in the doclist */
370  int iLastColumn;
371  int iLastPos;  /* the last position read, or -1 when not in a position list */
372} DocListReader;
373
374/*
375** Initialize the DocListReader r to point to the beginning of pDoclist.
376*/
377static void readerInit(DocListReader *r, DocList *pDoclist){
378  r->pDoclist = pDoclist;
379  if( pDoclist!=NULL ){
380    r->p = pDoclist->pData;
381  }
382  r->iLastColumn = -1;
383  r->iLastPos = -1;
384}
385
386/*
387** Return TRUE if we have reached then end of pReader and there is
388** nothing else left to read.
389*/
390static int atEnd(DocListReader *pReader){
391  return pReader->pDoclist==0 || (pReader->p >= docListEnd(pReader->pDoclist));
392}
393
394/* Peek at the next docid without advancing the read pointer.
395*/
396static sqlite_int64 peekDocid(DocListReader *pReader){
397  sqlite_int64 ret;
398  assert( !atEnd(pReader) );
399  assert( pReader->iLastPos==-1 );
400  getVarint(pReader->p, &ret);
401  return ret;
402}
403
404/* Read the next docid.   See also nextDocid().
405*/
406static sqlite_int64 readDocid(DocListReader *pReader){
407  sqlite_int64 ret;
408  assert( !atEnd(pReader) );
409  assert( pReader->iLastPos==-1 );
410  pReader->p += getVarint(pReader->p, &ret);
411  if( pReader->pDoclist->iType>=DL_POSITIONS ){
412    pReader->iLastColumn = 0;
413    pReader->iLastPos = 0;
414  }
415  return ret;
416}
417
418/* Read the next position and column index from a position list.
419 * Returns the position, or -1 at the end of the list. */
420static int readPosition(DocListReader *pReader, int *iColumn){
421  int i;
422  int iType = pReader->pDoclist->iType;
423
424  if( pReader->iLastPos==-1 ){
425    return -1;
426  }
427  assert( !atEnd(pReader) );
428
429  if( iType<DL_POSITIONS ){
430    return -1;
431  }
432  pReader->p += getVarint32(pReader->p, &i);
433  if( i==POS_END ){
434    pReader->iLastColumn = pReader->iLastPos = -1;
435    *iColumn = -1;
436    return -1;
437  }
438  if( i==POS_COLUMN ){
439    pReader->p += getVarint32(pReader->p, &pReader->iLastColumn);
440    pReader->iLastPos = 0;
441    pReader->p += getVarint32(pReader->p, &i);
442    assert( i>=POS_BASE );
443  }
444  pReader->iLastPos += ((int) i)-POS_BASE;
445  if( iType>=DL_POSITIONS_OFFSETS ){
446    /* Skip over offsets, ignoring them for now. */
447    int iStart, iEnd;
448    pReader->p += getVarint32(pReader->p, &iStart);
449    pReader->p += getVarint32(pReader->p, &iEnd);
450  }
451  *iColumn = pReader->iLastColumn;
452  return pReader->iLastPos;
453}
454
455/* Skip past the end of a position list. */
456static void skipPositionList(DocListReader *pReader){
457  DocList *p = pReader->pDoclist;
458  if( p && p->iType>=DL_POSITIONS ){
459    int iColumn;
460    while( readPosition(pReader, &iColumn)!=-1 ){}
461  }
462}
463
464/* Skip over a docid, including its position list if the doclist has
465 * positions. */
466static void skipDocument(DocListReader *pReader){
467  readDocid(pReader);
468  skipPositionList(pReader);
469}
470
471/* Skip past all docids which are less than [iDocid].  Returns 1 if a docid
472 * matching [iDocid] was found.  */
473static int skipToDocid(DocListReader *pReader, sqlite_int64 iDocid){
474  sqlite_int64 d = 0;
475  while( !atEnd(pReader) && (d=peekDocid(pReader))<iDocid ){
476    skipDocument(pReader);
477  }
478  return !atEnd(pReader) && d==iDocid;
479}
480
481/* Return the first document in a document list.
482*/
483static sqlite_int64 firstDocid(DocList *d){
484  DocListReader r;
485  readerInit(&r, d);
486  return readDocid(&r);
487}
488
489#ifdef SQLITE_DEBUG
490/*
491** This routine is used for debugging purpose only.
492**
493** Write the content of a doclist to standard output.
494*/
495static void printDoclist(DocList *p){
496  DocListReader r;
497  const char *zSep = "";
498
499  readerInit(&r, p);
500  while( !atEnd(&r) ){
501    sqlite_int64 docid = readDocid(&r);
502    if( docid==0 ){
503      skipPositionList(&r);
504      continue;
505    }
506    printf("%s%lld", zSep, docid);
507    zSep =  ",";
508    if( p->iType>=DL_POSITIONS ){
509      int iPos, iCol;
510      const char *zDiv = "";
511      printf("(");
512      while( (iPos = readPosition(&r, &iCol))>=0 ){
513        printf("%s%d:%d", zDiv, iCol, iPos);
514        zDiv = ":";
515      }
516      printf(")");
517    }
518  }
519  printf("\n");
520  fflush(stdout);
521}
522#endif /* SQLITE_DEBUG */
523
524/* Trim the given doclist to contain only positions in column
525 * [iRestrictColumn]. */
526static void docListRestrictColumn(DocList *in, int iRestrictColumn){
527  DocListReader r;
528  DocList out;
529
530  assert( in->iType>=DL_POSITIONS );
531  readerInit(&r, in);
532  docListInit(&out, DL_POSITIONS, NULL, 0);
533
534  while( !atEnd(&r) ){
535    sqlite_int64 iDocid = readDocid(&r);
536    int iPos, iColumn;
537
538    docListAddDocid(&out, iDocid);
539    while( (iPos = readPosition(&r, &iColumn)) != -1 ){
540      if( iColumn==iRestrictColumn ){
541        docListAddPos(&out, iColumn, iPos);
542      }
543    }
544  }
545
546  docListDestroy(in);
547  *in = out;
548}
549
550/* Trim the given doclist by discarding any docids without any remaining
551 * positions. */
552static void docListDiscardEmpty(DocList *in) {
553  DocListReader r;
554  DocList out;
555
556  /* TODO: It would be nice to implement this operation in place; that
557   * could save a significant amount of memory in queries with long doclists. */
558  assert( in->iType>=DL_POSITIONS );
559  readerInit(&r, in);
560  docListInit(&out, DL_POSITIONS, NULL, 0);
561
562  while( !atEnd(&r) ){
563    sqlite_int64 iDocid = readDocid(&r);
564    int match = 0;
565    int iPos, iColumn;
566    while( (iPos = readPosition(&r, &iColumn)) != -1 ){
567      if( !match ){
568        docListAddDocid(&out, iDocid);
569        match = 1;
570      }
571      docListAddPos(&out, iColumn, iPos);
572    }
573  }
574
575  docListDestroy(in);
576  *in = out;
577}
578
579/* Helper function for docListUpdate() and docListAccumulate().
580** Splices a doclist element into the doclist represented by r,
581** leaving r pointing after the newly spliced element.
582*/
583static void docListSpliceElement(DocListReader *r, sqlite_int64 iDocid,
584                                 const char *pSource, int nSource){
585  DocList *d = r->pDoclist;
586  char *pTarget;
587  int nTarget, found;
588
589  found = skipToDocid(r, iDocid);
590
591  /* Describe slice in d to place pSource/nSource. */
592  pTarget = r->p;
593  if( found ){
594    skipDocument(r);
595    nTarget = r->p-pTarget;
596  }else{
597    nTarget = 0;
598  }
599
600  /* The sense of the following is that there are three possibilities.
601  ** If nTarget==nSource, we should not move any memory nor realloc.
602  ** If nTarget>nSource, trim target and realloc.
603  ** If nTarget<nSource, realloc then expand target.
604  */
605  if( nTarget>nSource ){
606    memmove(pTarget+nSource, pTarget+nTarget, docListEnd(d)-(pTarget+nTarget));
607  }
608  if( nTarget!=nSource ){
609    int iDoclist = pTarget-d->pData;
610    d->pData = realloc(d->pData, d->nData+nSource-nTarget);
611    pTarget = d->pData+iDoclist;
612  }
613  if( nTarget<nSource ){
614    memmove(pTarget+nSource, pTarget+nTarget, docListEnd(d)-(pTarget+nTarget));
615  }
616
617  memcpy(pTarget, pSource, nSource);
618  d->nData += nSource-nTarget;
619  r->p = pTarget+nSource;
620}
621
622/* Insert/update pUpdate into the doclist. */
623static void docListUpdate(DocList *d, DocList *pUpdate){
624  DocListReader reader;
625
626  assert( d!=NULL && pUpdate!=NULL );
627  assert( d->iType==pUpdate->iType);
628
629  readerInit(&reader, d);
630  docListSpliceElement(&reader, firstDocid(pUpdate),
631                       pUpdate->pData, pUpdate->nData);
632}
633
634/* Propagate elements from pUpdate to pAcc, overwriting elements with
635** matching docids.
636*/
637static void docListAccumulate(DocList *pAcc, DocList *pUpdate){
638  DocListReader accReader, updateReader;
639
640  /* Handle edge cases where one doclist is empty. */
641  assert( pAcc!=NULL );
642  if( pUpdate==NULL || pUpdate->nData==0 ) return;
643  if( pAcc->nData==0 ){
644    pAcc->pData = malloc(pUpdate->nData);
645    memcpy(pAcc->pData, pUpdate->pData, pUpdate->nData);
646    pAcc->nData = pUpdate->nData;
647    return;
648  }
649
650  readerInit(&accReader, pAcc);
651  readerInit(&updateReader, pUpdate);
652
653  while( !atEnd(&updateReader) ){
654    char *pSource = updateReader.p;
655    sqlite_int64 iDocid = readDocid(&updateReader);
656    skipPositionList(&updateReader);
657    docListSpliceElement(&accReader, iDocid, pSource, updateReader.p-pSource);
658  }
659}
660
661/*
662** Read the next docid off of pIn.  Return 0 if we reach the end.
663*
664* TODO: This assumes that docids are never 0, but they may actually be 0 since
665* users can choose docids when inserting into a full-text table.  Fix this.
666*/
667static sqlite_int64 nextDocid(DocListReader *pIn){
668  skipPositionList(pIn);
669  return atEnd(pIn) ? 0 : readDocid(pIn);
670}
671
672/*
673** pLeft and pRight are two DocListReaders that are pointing to
674** positions lists of the same document: iDocid.
675**
676** If there are no instances in pLeft or pRight where the position
677** of pLeft is one less than the position of pRight, then this
678** routine adds nothing to pOut.
679**
680** If there are one or more instances where positions from pLeft
681** are exactly one less than positions from pRight, then add a new
682** document record to pOut.  If pOut wants to hold positions, then
683** include the positions from pRight that are one more than a
684** position in pLeft.  In other words:  pRight.iPos==pLeft.iPos+1.
685**
686** pLeft and pRight are left pointing at the next document record.
687*/
688static void mergePosList(
689  DocListReader *pLeft,    /* Left position list */
690  DocListReader *pRight,   /* Right position list */
691  sqlite_int64 iDocid,     /* The docid from pLeft and pRight */
692  DocList *pOut            /* Write the merged document record here */
693){
694  int iLeftCol, iLeftPos = readPosition(pLeft, &iLeftCol);
695  int iRightCol, iRightPos = readPosition(pRight, &iRightCol);
696  int match = 0;
697
698  /* Loop until we've reached the end of both position lists. */
699  while( iLeftPos!=-1 && iRightPos!=-1 ){
700    if( iLeftCol==iRightCol && iLeftPos+1==iRightPos ){
701      if( !match ){
702        docListAddDocid(pOut, iDocid);
703        match = 1;
704      }
705      if( pOut->iType>=DL_POSITIONS ){
706        docListAddPos(pOut, iRightCol, iRightPos);
707      }
708      iLeftPos = readPosition(pLeft, &iLeftCol);
709      iRightPos = readPosition(pRight, &iRightCol);
710    }else if( iRightCol<iLeftCol ||
711              (iRightCol==iLeftCol && iRightPos<iLeftPos+1) ){
712      iRightPos = readPosition(pRight, &iRightCol);
713    }else{
714      iLeftPos = readPosition(pLeft, &iLeftCol);
715    }
716  }
717  if( iLeftPos>=0 ) skipPositionList(pLeft);
718  if( iRightPos>=0 ) skipPositionList(pRight);
719}
720
721/* We have two doclists:  pLeft and pRight.
722** Write the phrase intersection of these two doclists into pOut.
723**
724** A phrase intersection means that two documents only match
725** if pLeft.iPos+1==pRight.iPos.
726**
727** The output pOut may or may not contain positions.  If pOut
728** does contain positions, they are the positions of pRight.
729*/
730static void docListPhraseMerge(
731  DocList *pLeft,    /* Doclist resulting from the words on the left */
732  DocList *pRight,   /* Doclist for the next word to the right */
733  DocList *pOut      /* Write the combined doclist here */
734){
735  DocListReader left, right;
736  sqlite_int64 docidLeft, docidRight;
737
738  readerInit(&left, pLeft);
739  readerInit(&right, pRight);
740  docidLeft = nextDocid(&left);
741  docidRight = nextDocid(&right);
742
743  while( docidLeft>0 && docidRight>0 ){
744    if( docidLeft<docidRight ){
745      docidLeft = nextDocid(&left);
746    }else if( docidRight<docidLeft ){
747      docidRight = nextDocid(&right);
748    }else{
749      mergePosList(&left, &right, docidLeft, pOut);
750      docidLeft = nextDocid(&left);
751      docidRight = nextDocid(&right);
752    }
753  }
754}
755
756/* We have two doclists:  pLeft and pRight.
757** Write the intersection of these two doclists into pOut.
758** Only docids are matched.  Position information is ignored.
759**
760** The output pOut never holds positions.
761*/
762static void docListAndMerge(
763  DocList *pLeft,    /* Doclist resulting from the words on the left */
764  DocList *pRight,   /* Doclist for the next word to the right */
765  DocList *pOut      /* Write the combined doclist here */
766){
767  DocListReader left, right;
768  sqlite_int64 docidLeft, docidRight;
769
770  assert( pOut->iType<DL_POSITIONS );
771
772  readerInit(&left, pLeft);
773  readerInit(&right, pRight);
774  docidLeft = nextDocid(&left);
775  docidRight = nextDocid(&right);
776
777  while( docidLeft>0 && docidRight>0 ){
778    if( docidLeft<docidRight ){
779      docidLeft = nextDocid(&left);
780    }else if( docidRight<docidLeft ){
781      docidRight = nextDocid(&right);
782    }else{
783      docListAddDocid(pOut, docidLeft);
784      docidLeft = nextDocid(&left);
785      docidRight = nextDocid(&right);
786    }
787  }
788}
789
790/* We have two doclists:  pLeft and pRight.
791** Write the union of these two doclists into pOut.
792** Only docids are matched.  Position information is ignored.
793**
794** The output pOut never holds positions.
795*/
796static void docListOrMerge(
797  DocList *pLeft,    /* Doclist resulting from the words on the left */
798  DocList *pRight,   /* Doclist for the next word to the right */
799  DocList *pOut      /* Write the combined doclist here */
800){
801  DocListReader left, right;
802  sqlite_int64 docidLeft, docidRight, priorLeft;
803
804  readerInit(&left, pLeft);
805  readerInit(&right, pRight);
806  docidLeft = nextDocid(&left);
807  docidRight = nextDocid(&right);
808
809  while( docidLeft>0 && docidRight>0 ){
810    if( docidLeft<=docidRight ){
811      docListAddDocid(pOut, docidLeft);
812    }else{
813      docListAddDocid(pOut, docidRight);
814    }
815    priorLeft = docidLeft;
816    if( docidLeft<=docidRight ){
817      docidLeft = nextDocid(&left);
818    }
819    if( docidRight>0 && docidRight<=priorLeft ){
820      docidRight = nextDocid(&right);
821    }
822  }
823  while( docidLeft>0 ){
824    docListAddDocid(pOut, docidLeft);
825    docidLeft = nextDocid(&left);
826  }
827  while( docidRight>0 ){
828    docListAddDocid(pOut, docidRight);
829    docidRight = nextDocid(&right);
830  }
831}
832
833/* We have two doclists:  pLeft and pRight.
834** Write into pOut all documents that occur in pLeft but not
835** in pRight.
836**
837** Only docids are matched.  Position information is ignored.
838**
839** The output pOut never holds positions.
840*/
841static void docListExceptMerge(
842  DocList *pLeft,    /* Doclist resulting from the words on the left */
843  DocList *pRight,   /* Doclist for the next word to the right */
844  DocList *pOut      /* Write the combined doclist here */
845){
846  DocListReader left, right;
847  sqlite_int64 docidLeft, docidRight, priorLeft;
848
849  readerInit(&left, pLeft);
850  readerInit(&right, pRight);
851  docidLeft = nextDocid(&left);
852  docidRight = nextDocid(&right);
853
854  while( docidLeft>0 && docidRight>0 ){
855    priorLeft = docidLeft;
856    if( docidLeft<docidRight ){
857      docListAddDocid(pOut, docidLeft);
858    }
859    if( docidLeft<=docidRight ){
860      docidLeft = nextDocid(&left);
861    }
862    if( docidRight>0 && docidRight<=priorLeft ){
863      docidRight = nextDocid(&right);
864    }
865  }
866  while( docidLeft>0 ){
867    docListAddDocid(pOut, docidLeft);
868    docidLeft = nextDocid(&left);
869  }
870}
871
872static char *string_dup_n(const char *s, int n){
873  char *str = malloc(n + 1);
874  memcpy(str, s, n);
875  str[n] = '\0';
876  return str;
877}
878
879/* Duplicate a string; the caller must free() the returned string.
880 * (We don't use strdup() since it is not part of the standard C library and
881 * may not be available everywhere.) */
882static char *string_dup(const char *s){
883  return string_dup_n(s, strlen(s));
884}
885
886/* Format a string, replacing each occurrence of the % character with
887 * zDb.zName.  This may be more convenient than sqlite_mprintf()
888 * when one string is used repeatedly in a format string.
889 * The caller must free() the returned string. */
890static char *string_format(const char *zFormat,
891                           const char *zDb, const char *zName){
892  const char *p;
893  size_t len = 0;
894  size_t nDb = strlen(zDb);
895  size_t nName = strlen(zName);
896  size_t nFullTableName = nDb+1+nName;
897  char *result;
898  char *r;
899
900  /* first compute length needed */
901  for(p = zFormat ; *p ; ++p){
902    len += (*p=='%' ? nFullTableName : 1);
903  }
904  len += 1;  /* for null terminator */
905
906  r = result = malloc(len);
907  for(p = zFormat; *p; ++p){
908    if( *p=='%' ){
909      memcpy(r, zDb, nDb);
910      r += nDb;
911      *r++ = '.';
912      memcpy(r, zName, nName);
913      r += nName;
914    } else {
915      *r++ = *p;
916    }
917  }
918  *r++ = '\0';
919  assert( r == result + len );
920  return result;
921}
922
923static int sql_exec(sqlite3 *db, const char *zDb, const char *zName,
924                    const char *zFormat){
925  char *zCommand = string_format(zFormat, zDb, zName);
926  int rc;
927  TRACE(("FTS1 sql: %s\n", zCommand));
928  rc = sqlite3_exec(db, zCommand, NULL, 0, NULL);
929  free(zCommand);
930  return rc;
931}
932
933static int sql_prepare(sqlite3 *db, const char *zDb, const char *zName,
934                       sqlite3_stmt **ppStmt, const char *zFormat){
935  char *zCommand = string_format(zFormat, zDb, zName);
936  int rc;
937  TRACE(("FTS1 prepare: %s\n", zCommand));
938  rc = sqlite3_prepare(db, zCommand, -1, ppStmt, NULL);
939  free(zCommand);
940  return rc;
941}
942
943/* end utility functions */
944
945/* Forward reference */
946typedef struct fulltext_vtab fulltext_vtab;
947
948/* A single term in a query is represented by an instances of
949** the following structure.
950*/
951typedef struct QueryTerm {
952  short int nPhrase; /* How many following terms are part of the same phrase */
953  short int iPhrase; /* This is the i-th term of a phrase. */
954  short int iColumn; /* Column of the index that must match this term */
955  signed char isOr;  /* this term is preceded by "OR" */
956  signed char isNot; /* this term is preceded by "-" */
957  char *pTerm;       /* text of the term.  '\000' terminated.  malloced */
958  int nTerm;         /* Number of bytes in pTerm[] */
959} QueryTerm;
960
961
962/* A query string is parsed into a Query structure.
963 *
964 * We could, in theory, allow query strings to be complicated
965 * nested expressions with precedence determined by parentheses.
966 * But none of the major search engines do this.  (Perhaps the
967 * feeling is that an parenthesized expression is two complex of
968 * an idea for the average user to grasp.)  Taking our lead from
969 * the major search engines, we will allow queries to be a list
970 * of terms (with an implied AND operator) or phrases in double-quotes,
971 * with a single optional "-" before each non-phrase term to designate
972 * negation and an optional OR connector.
973 *
974 * OR binds more tightly than the implied AND, which is what the
975 * major search engines seem to do.  So, for example:
976 *
977 *    [one two OR three]     ==>    one AND (two OR three)
978 *    [one OR two three]     ==>    (one OR two) AND three
979 *
980 * A "-" before a term matches all entries that lack that term.
981 * The "-" must occur immediately before the term with in intervening
982 * space.  This is how the search engines do it.
983 *
984 * A NOT term cannot be the right-hand operand of an OR.  If this
985 * occurs in the query string, the NOT is ignored:
986 *
987 *    [one OR -two]          ==>    one OR two
988 *
989 */
990typedef struct Query {
991  fulltext_vtab *pFts;  /* The full text index */
992  int nTerms;           /* Number of terms in the query */
993  QueryTerm *pTerms;    /* Array of terms.  Space obtained from malloc() */
994  int nextIsOr;         /* Set the isOr flag on the next inserted term */
995  int nextColumn;       /* Next word parsed must be in this column */
996  int dfltColumn;       /* The default column */
997} Query;
998
999
1000/*
1001** An instance of the following structure keeps track of generated
1002** matching-word offset information and snippets.
1003*/
1004typedef struct Snippet {
1005  int nMatch;     /* Total number of matches */
1006  int nAlloc;     /* Space allocated for aMatch[] */
1007  struct snippetMatch { /* One entry for each matching term */
1008    char snStatus;       /* Status flag for use while constructing snippets */
1009    short int iCol;      /* The column that contains the match */
1010    short int iTerm;     /* The index in Query.pTerms[] of the matching term */
1011    short int nByte;     /* Number of bytes in the term */
1012    int iStart;          /* The offset to the first character of the term */
1013  } *aMatch;      /* Points to space obtained from malloc */
1014  char *zOffset;  /* Text rendering of aMatch[] */
1015  int nOffset;    /* strlen(zOffset) */
1016  char *zSnippet; /* Snippet text */
1017  int nSnippet;   /* strlen(zSnippet) */
1018} Snippet;
1019
1020
1021typedef enum QueryType {
1022  QUERY_GENERIC,   /* table scan */
1023  QUERY_ROWID,     /* lookup by rowid */
1024  QUERY_FULLTEXT   /* QUERY_FULLTEXT + [i] is a full-text search for column i*/
1025} QueryType;
1026
1027/* TODO(shess) CHUNK_MAX controls how much data we allow in segment 0
1028** before we start aggregating into larger segments.  Lower CHUNK_MAX
1029** means that for a given input we have more individual segments per
1030** term, which means more rows in the table and a bigger index (due to
1031** both more rows and bigger rowids).  But it also reduces the average
1032** cost of adding new elements to the segment 0 doclist, and it seems
1033** to reduce the number of pages read and written during inserts.  256
1034** was chosen by measuring insertion times for a certain input (first
1035** 10k documents of Enron corpus), though including query performance
1036** in the decision may argue for a larger value.
1037*/
1038#define CHUNK_MAX 256
1039
1040typedef enum fulltext_statement {
1041  CONTENT_INSERT_STMT,
1042  CONTENT_SELECT_STMT,
1043  CONTENT_UPDATE_STMT,
1044  CONTENT_DELETE_STMT,
1045
1046  TERM_SELECT_STMT,
1047  TERM_SELECT_ALL_STMT,
1048  TERM_INSERT_STMT,
1049  TERM_UPDATE_STMT,
1050  TERM_DELETE_STMT,
1051
1052  MAX_STMT                     /* Always at end! */
1053} fulltext_statement;
1054
1055/* These must exactly match the enum above. */
1056/* TODO(adam): Is there some risk that a statement (in particular,
1057** pTermSelectStmt) will be used in two cursors at once, e.g.  if a
1058** query joins a virtual table to itself?  If so perhaps we should
1059** move some of these to the cursor object.
1060*/
1061static const char *const fulltext_zStatement[MAX_STMT] = {
1062  /* CONTENT_INSERT */ NULL,  /* generated in contentInsertStatement() */
1063  /* CONTENT_SELECT */ "select * from %_content where rowid = ?",
1064  /* CONTENT_UPDATE */ NULL,  /* generated in contentUpdateStatement() */
1065  /* CONTENT_DELETE */ "delete from %_content where rowid = ?",
1066
1067  /* TERM_SELECT */
1068  "select rowid, doclist from %_term where term = ? and segment = ?",
1069  /* TERM_SELECT_ALL */
1070  "select doclist from %_term where term = ? order by segment",
1071  /* TERM_INSERT */
1072  "insert into %_term (rowid, term, segment, doclist) values (?, ?, ?, ?)",
1073  /* TERM_UPDATE */ "update %_term set doclist = ? where rowid = ?",
1074  /* TERM_DELETE */ "delete from %_term where rowid = ?",
1075};
1076
1077/*
1078** A connection to a fulltext index is an instance of the following
1079** structure.  The xCreate and xConnect methods create an instance
1080** of this structure and xDestroy and xDisconnect free that instance.
1081** All other methods receive a pointer to the structure as one of their
1082** arguments.
1083*/
1084struct fulltext_vtab {
1085  sqlite3_vtab base;               /* Base class used by SQLite core */
1086  sqlite3 *db;                     /* The database connection */
1087  const char *zDb;                 /* logical database name */
1088  const char *zName;               /* virtual table name */
1089  int nColumn;                     /* number of columns in virtual table */
1090  char **azColumn;                 /* column names.  malloced */
1091  char **azContentColumn;          /* column names in content table; malloced */
1092  sqlite3_tokenizer *pTokenizer;   /* tokenizer for inserts and queries */
1093
1094  /* Precompiled statements which we keep as long as the table is
1095  ** open.
1096  */
1097  sqlite3_stmt *pFulltextStatements[MAX_STMT];
1098};
1099
1100/*
1101** When the core wants to do a query, it create a cursor using a
1102** call to xOpen.  This structure is an instance of a cursor.  It
1103** is destroyed by xClose.
1104*/
1105typedef struct fulltext_cursor {
1106  sqlite3_vtab_cursor base;        /* Base class used by SQLite core */
1107  QueryType iCursorType;           /* Copy of sqlite3_index_info.idxNum */
1108  sqlite3_stmt *pStmt;             /* Prepared statement in use by the cursor */
1109  int eof;                         /* True if at End Of Results */
1110  Query q;                         /* Parsed query string */
1111  Snippet snippet;                 /* Cached snippet for the current row */
1112  int iColumn;                     /* Column being searched */
1113  DocListReader result;  /* used when iCursorType == QUERY_FULLTEXT */
1114} fulltext_cursor;
1115
1116static struct fulltext_vtab *cursor_vtab(fulltext_cursor *c){
1117  return (fulltext_vtab *) c->base.pVtab;
1118}
1119
1120static const sqlite3_module fulltextModule;   /* forward declaration */
1121
1122/* Append a list of strings separated by commas to a StringBuffer. */
1123static void appendList(StringBuffer *sb, int nString, char **azString){
1124  int i;
1125  for(i=0; i<nString; ++i){
1126    if( i>0 ) append(sb, ", ");
1127    append(sb, azString[i]);
1128  }
1129}
1130
1131/* Return a dynamically generated statement of the form
1132 *   insert into %_content (rowid, ...) values (?, ...)
1133 */
1134static const char *contentInsertStatement(fulltext_vtab *v){
1135  StringBuffer sb;
1136  int i;
1137
1138  initStringBuffer(&sb);
1139  append(&sb, "insert into %_content (rowid, ");
1140  appendList(&sb, v->nColumn, v->azContentColumn);
1141  append(&sb, ") values (?");
1142  for(i=0; i<v->nColumn; ++i)
1143    append(&sb, ", ?");
1144  append(&sb, ")");
1145  return sb.s;
1146}
1147
1148/* Return a dynamically generated statement of the form
1149 *   update %_content set [col_0] = ?, [col_1] = ?, ...
1150 *                    where rowid = ?
1151 */
1152static const char *contentUpdateStatement(fulltext_vtab *v){
1153  StringBuffer sb;
1154  int i;
1155
1156  initStringBuffer(&sb);
1157  append(&sb, "update %_content set ");
1158  for(i=0; i<v->nColumn; ++i) {
1159    if( i>0 ){
1160      append(&sb, ", ");
1161    }
1162    append(&sb, v->azContentColumn[i]);
1163    append(&sb, " = ?");
1164  }
1165  append(&sb, " where rowid = ?");
1166  return sb.s;
1167}
1168
1169/* Puts a freshly-prepared statement determined by iStmt in *ppStmt.
1170** If the indicated statement has never been prepared, it is prepared
1171** and cached, otherwise the cached version is reset.
1172*/
1173static int sql_get_statement(fulltext_vtab *v, fulltext_statement iStmt,
1174                             sqlite3_stmt **ppStmt){
1175  assert( iStmt<MAX_STMT );
1176  if( v->pFulltextStatements[iStmt]==NULL ){
1177    const char *zStmt;
1178    int rc;
1179    switch( iStmt ){
1180      case CONTENT_INSERT_STMT:
1181        zStmt = contentInsertStatement(v); break;
1182      case CONTENT_UPDATE_STMT:
1183        zStmt = contentUpdateStatement(v); break;
1184      default:
1185        zStmt = fulltext_zStatement[iStmt];
1186    }
1187    rc = sql_prepare(v->db, v->zDb, v->zName, &v->pFulltextStatements[iStmt],
1188                         zStmt);
1189    if( zStmt != fulltext_zStatement[iStmt]) free((void *) zStmt);
1190    if( rc!=SQLITE_OK ) return rc;
1191  } else {
1192    int rc = sqlite3_reset(v->pFulltextStatements[iStmt]);
1193    if( rc!=SQLITE_OK ) return rc;
1194  }
1195
1196  *ppStmt = v->pFulltextStatements[iStmt];
1197  return SQLITE_OK;
1198}
1199
1200/* Step the indicated statement, handling errors SQLITE_BUSY (by
1201** retrying) and SQLITE_SCHEMA (by re-preparing and transferring
1202** bindings to the new statement).
1203** TODO(adam): We should extend this function so that it can work with
1204** statements declared locally, not only globally cached statements.
1205*/
1206static int sql_step_statement(fulltext_vtab *v, fulltext_statement iStmt,
1207                              sqlite3_stmt **ppStmt){
1208  int rc;
1209  sqlite3_stmt *s = *ppStmt;
1210  assert( iStmt<MAX_STMT );
1211  assert( s==v->pFulltextStatements[iStmt] );
1212
1213  while( (rc=sqlite3_step(s))!=SQLITE_DONE && rc!=SQLITE_ROW ){
1214    if( rc==SQLITE_BUSY ) continue;
1215    if( rc!=SQLITE_ERROR ) return rc;
1216
1217    /* If an SQLITE_SCHEMA error has occurred, then finalizing this
1218     * statement is going to delete the fulltext_vtab structure. If
1219     * the statement just executed is in the pFulltextStatements[]
1220     * array, it will be finalized twice. So remove it before
1221     * calling sqlite3_finalize().
1222     */
1223    v->pFulltextStatements[iStmt] = NULL;
1224    rc = sqlite3_finalize(s);
1225    break;
1226  }
1227  return rc;
1228
1229 err:
1230  sqlite3_finalize(s);
1231  return rc;
1232}
1233
1234/* Like sql_step_statement(), but convert SQLITE_DONE to SQLITE_OK.
1235** Useful for statements like UPDATE, where we expect no results.
1236*/
1237static int sql_single_step_statement(fulltext_vtab *v,
1238                                     fulltext_statement iStmt,
1239                                     sqlite3_stmt **ppStmt){
1240  int rc = sql_step_statement(v, iStmt, ppStmt);
1241  return (rc==SQLITE_DONE) ? SQLITE_OK : rc;
1242}
1243
1244/* insert into %_content (rowid, ...) values ([rowid], [pValues]) */
1245static int content_insert(fulltext_vtab *v, sqlite3_value *rowid,
1246                          sqlite3_value **pValues){
1247  sqlite3_stmt *s;
1248  int i;
1249  int rc = sql_get_statement(v, CONTENT_INSERT_STMT, &s);
1250  if( rc!=SQLITE_OK ) return rc;
1251
1252  rc = sqlite3_bind_value(s, 1, rowid);
1253  if( rc!=SQLITE_OK ) return rc;
1254
1255  for(i=0; i<v->nColumn; ++i){
1256    rc = sqlite3_bind_value(s, 2+i, pValues[i]);
1257    if( rc!=SQLITE_OK ) return rc;
1258  }
1259
1260  return sql_single_step_statement(v, CONTENT_INSERT_STMT, &s);
1261}
1262
1263/* update %_content set col0 = pValues[0], col1 = pValues[1], ...
1264 *                  where rowid = [iRowid] */
1265static int content_update(fulltext_vtab *v, sqlite3_value **pValues,
1266                          sqlite_int64 iRowid){
1267  sqlite3_stmt *s;
1268  int i;
1269  int rc = sql_get_statement(v, CONTENT_UPDATE_STMT, &s);
1270  if( rc!=SQLITE_OK ) return rc;
1271
1272  for(i=0; i<v->nColumn; ++i){
1273    rc = sqlite3_bind_value(s, 1+i, pValues[i]);
1274    if( rc!=SQLITE_OK ) return rc;
1275  }
1276
1277  rc = sqlite3_bind_int64(s, 1+v->nColumn, iRowid);
1278  if( rc!=SQLITE_OK ) return rc;
1279
1280  return sql_single_step_statement(v, CONTENT_UPDATE_STMT, &s);
1281}
1282
1283static void freeStringArray(int nString, const char **pString){
1284  int i;
1285
1286  for (i=0 ; i < nString ; ++i) {
1287    if( pString[i]!=NULL ) free((void *) pString[i]);
1288  }
1289  free((void *) pString);
1290}
1291
1292/* select * from %_content where rowid = [iRow]
1293 * The caller must delete the returned array and all strings in it.
1294 * null fields will be NULL in the returned array.
1295 *
1296 * TODO: Perhaps we should return pointer/length strings here for consistency
1297 * with other code which uses pointer/length. */
1298static int content_select(fulltext_vtab *v, sqlite_int64 iRow,
1299                          const char ***pValues){
1300  sqlite3_stmt *s;
1301  const char **values;
1302  int i;
1303  int rc;
1304
1305  *pValues = NULL;
1306
1307  rc = sql_get_statement(v, CONTENT_SELECT_STMT, &s);
1308  if( rc!=SQLITE_OK ) return rc;
1309
1310  rc = sqlite3_bind_int64(s, 1, iRow);
1311  if( rc!=SQLITE_OK ) return rc;
1312
1313  rc = sql_step_statement(v, CONTENT_SELECT_STMT, &s);
1314  if( rc!=SQLITE_ROW ) return rc;
1315
1316  values = (const char **) malloc(v->nColumn * sizeof(const char *));
1317  for(i=0; i<v->nColumn; ++i){
1318    if( sqlite3_column_type(s, i)==SQLITE_NULL ){
1319      values[i] = NULL;
1320    }else{
1321      values[i] = string_dup((char*)sqlite3_column_text(s, i));
1322    }
1323  }
1324
1325  /* We expect only one row.  We must execute another sqlite3_step()
1326   * to complete the iteration; otherwise the table will remain locked. */
1327  rc = sqlite3_step(s);
1328  if( rc==SQLITE_DONE ){
1329    *pValues = values;
1330    return SQLITE_OK;
1331  }
1332
1333  freeStringArray(v->nColumn, values);
1334  return rc;
1335}
1336
1337/* delete from %_content where rowid = [iRow ] */
1338static int content_delete(fulltext_vtab *v, sqlite_int64 iRow){
1339  sqlite3_stmt *s;
1340  int rc = sql_get_statement(v, CONTENT_DELETE_STMT, &s);
1341  if( rc!=SQLITE_OK ) return rc;
1342
1343  rc = sqlite3_bind_int64(s, 1, iRow);
1344  if( rc!=SQLITE_OK ) return rc;
1345
1346  return sql_single_step_statement(v, CONTENT_DELETE_STMT, &s);
1347}
1348
1349/* select rowid, doclist from %_term
1350 *  where term = [pTerm] and segment = [iSegment]
1351 * If found, returns SQLITE_ROW; the caller must free the
1352 * returned doclist.  If no rows found, returns SQLITE_DONE. */
1353static int term_select(fulltext_vtab *v, const char *pTerm, int nTerm,
1354                       int iSegment,
1355                       sqlite_int64 *rowid, DocList *out){
1356  sqlite3_stmt *s;
1357  int rc = sql_get_statement(v, TERM_SELECT_STMT, &s);
1358  if( rc!=SQLITE_OK ) return rc;
1359
1360  rc = sqlite3_bind_text(s, 1, pTerm, nTerm, SQLITE_STATIC);
1361  if( rc!=SQLITE_OK ) return rc;
1362
1363  rc = sqlite3_bind_int(s, 2, iSegment);
1364  if( rc!=SQLITE_OK ) return rc;
1365
1366  rc = sql_step_statement(v, TERM_SELECT_STMT, &s);
1367  if( rc!=SQLITE_ROW ) return rc;
1368
1369  *rowid = sqlite3_column_int64(s, 0);
1370  docListInit(out, DL_DEFAULT,
1371              sqlite3_column_blob(s, 1), sqlite3_column_bytes(s, 1));
1372
1373  /* We expect only one row.  We must execute another sqlite3_step()
1374   * to complete the iteration; otherwise the table will remain locked. */
1375  rc = sqlite3_step(s);
1376  return rc==SQLITE_DONE ? SQLITE_ROW : rc;
1377}
1378
1379/* Load the segment doclists for term pTerm and merge them in
1380** appropriate order into out.  Returns SQLITE_OK if successful.  If
1381** there are no segments for pTerm, successfully returns an empty
1382** doclist in out.
1383**
1384** Each document consists of 1 or more "columns".  The number of
1385** columns is v->nColumn.  If iColumn==v->nColumn, then return
1386** position information about all columns.  If iColumn<v->nColumn,
1387** then only return position information about the iColumn-th column
1388** (where the first column is 0).
1389*/
1390static int term_select_all(
1391  fulltext_vtab *v,     /* The fulltext index we are querying against */
1392  int iColumn,          /* If <nColumn, only look at the iColumn-th column */
1393  const char *pTerm,    /* The term whose posting lists we want */
1394  int nTerm,            /* Number of bytes in pTerm */
1395  DocList *out          /* Write the resulting doclist here */
1396){
1397  DocList doclist;
1398  sqlite3_stmt *s;
1399  int rc = sql_get_statement(v, TERM_SELECT_ALL_STMT, &s);
1400  if( rc!=SQLITE_OK ) return rc;
1401
1402  rc = sqlite3_bind_text(s, 1, pTerm, nTerm, SQLITE_STATIC);
1403  if( rc!=SQLITE_OK ) return rc;
1404
1405  docListInit(&doclist, DL_DEFAULT, 0, 0);
1406
1407  /* TODO(shess) Handle schema and busy errors. */
1408  while( (rc=sql_step_statement(v, TERM_SELECT_ALL_STMT, &s))==SQLITE_ROW ){
1409    DocList old;
1410
1411    /* TODO(shess) If we processed doclists from oldest to newest, we
1412    ** could skip the malloc() involved with the following call.  For
1413    ** now, I'd rather keep this logic similar to index_insert_term().
1414    ** We could additionally drop elements when we see deletes, but
1415    ** that would require a distinct version of docListAccumulate().
1416    */
1417    docListInit(&old, DL_DEFAULT,
1418                sqlite3_column_blob(s, 0), sqlite3_column_bytes(s, 0));
1419
1420    if( iColumn<v->nColumn ){   /* querying a single column */
1421      docListRestrictColumn(&old, iColumn);
1422    }
1423
1424    /* doclist contains the newer data, so write it over old.  Then
1425    ** steal accumulated result for doclist.
1426    */
1427    docListAccumulate(&old, &doclist);
1428    docListDestroy(&doclist);
1429    doclist = old;
1430  }
1431  if( rc!=SQLITE_DONE ){
1432    docListDestroy(&doclist);
1433    return rc;
1434  }
1435
1436  docListDiscardEmpty(&doclist);
1437  *out = doclist;
1438  return SQLITE_OK;
1439}
1440
1441/* insert into %_term (rowid, term, segment, doclist)
1442               values ([piRowid], [pTerm], [iSegment], [doclist])
1443** Lets sqlite select rowid if piRowid is NULL, else uses *piRowid.
1444**
1445** NOTE(shess) piRowid is IN, with values of "space of int64" plus
1446** null, it is not used to pass data back to the caller.
1447*/
1448static int term_insert(fulltext_vtab *v, sqlite_int64 *piRowid,
1449                       const char *pTerm, int nTerm,
1450                       int iSegment, DocList *doclist){
1451  sqlite3_stmt *s;
1452  int rc = sql_get_statement(v, TERM_INSERT_STMT, &s);
1453  if( rc!=SQLITE_OK ) return rc;
1454
1455  if( piRowid==NULL ){
1456    rc = sqlite3_bind_null(s, 1);
1457  }else{
1458    rc = sqlite3_bind_int64(s, 1, *piRowid);
1459  }
1460  if( rc!=SQLITE_OK ) return rc;
1461
1462  rc = sqlite3_bind_text(s, 2, pTerm, nTerm, SQLITE_STATIC);
1463  if( rc!=SQLITE_OK ) return rc;
1464
1465  rc = sqlite3_bind_int(s, 3, iSegment);
1466  if( rc!=SQLITE_OK ) return rc;
1467
1468  rc = sqlite3_bind_blob(s, 4, doclist->pData, doclist->nData, SQLITE_STATIC);
1469  if( rc!=SQLITE_OK ) return rc;
1470
1471  return sql_single_step_statement(v, TERM_INSERT_STMT, &s);
1472}
1473
1474/* update %_term set doclist = [doclist] where rowid = [rowid] */
1475static int term_update(fulltext_vtab *v, sqlite_int64 rowid,
1476                       DocList *doclist){
1477  sqlite3_stmt *s;
1478  int rc = sql_get_statement(v, TERM_UPDATE_STMT, &s);
1479  if( rc!=SQLITE_OK ) return rc;
1480
1481  rc = sqlite3_bind_blob(s, 1, doclist->pData, doclist->nData, SQLITE_STATIC);
1482  if( rc!=SQLITE_OK ) return rc;
1483
1484  rc = sqlite3_bind_int64(s, 2, rowid);
1485  if( rc!=SQLITE_OK ) return rc;
1486
1487  return sql_single_step_statement(v, TERM_UPDATE_STMT, &s);
1488}
1489
1490static int term_delete(fulltext_vtab *v, sqlite_int64 rowid){
1491  sqlite3_stmt *s;
1492  int rc = sql_get_statement(v, TERM_DELETE_STMT, &s);
1493  if( rc!=SQLITE_OK ) return rc;
1494
1495  rc = sqlite3_bind_int64(s, 1, rowid);
1496  if( rc!=SQLITE_OK ) return rc;
1497
1498  return sql_single_step_statement(v, TERM_DELETE_STMT, &s);
1499}
1500
1501/*
1502** Free the memory used to contain a fulltext_vtab structure.
1503*/
1504static void fulltext_vtab_destroy(fulltext_vtab *v){
1505  int iStmt, i;
1506
1507  TRACE(("FTS1 Destroy %p\n", v));
1508  for( iStmt=0; iStmt<MAX_STMT; iStmt++ ){
1509    if( v->pFulltextStatements[iStmt]!=NULL ){
1510      sqlite3_finalize(v->pFulltextStatements[iStmt]);
1511      v->pFulltextStatements[iStmt] = NULL;
1512    }
1513  }
1514
1515  if( v->pTokenizer!=NULL ){
1516    v->pTokenizer->pModule->xDestroy(v->pTokenizer);
1517    v->pTokenizer = NULL;
1518  }
1519
1520  free(v->azColumn);
1521  for(i = 0; i < v->nColumn; ++i) {
1522    sqlite3_free(v->azContentColumn[i]);
1523  }
1524  free(v->azContentColumn);
1525  free(v);
1526}
1527
1528/*
1529** Token types for parsing the arguments to xConnect or xCreate.
1530*/
1531#define TOKEN_EOF         0    /* End of file */
1532#define TOKEN_SPACE       1    /* Any kind of whitespace */
1533#define TOKEN_ID          2    /* An identifier */
1534#define TOKEN_STRING      3    /* A string literal */
1535#define TOKEN_PUNCT       4    /* A single punctuation character */
1536
1537/*
1538** If X is a character that can be used in an identifier then
1539** IdChar(X) will be true.  Otherwise it is false.
1540**
1541** For ASCII, any character with the high-order bit set is
1542** allowed in an identifier.  For 7-bit characters,
1543** sqlite3IsIdChar[X] must be 1.
1544**
1545** Ticket #1066.  the SQL standard does not allow '$' in the
1546** middle of identfiers.  But many SQL implementations do.
1547** SQLite will allow '$' in identifiers for compatibility.
1548** But the feature is undocumented.
1549*/
1550static const char isIdChar[] = {
1551/* x0 x1 x2 x3 x4 x5 x6 x7 x8 x9 xA xB xC xD xE xF */
1552    0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,  /* 2x */
1553    1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0,  /* 3x */
1554    0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,  /* 4x */
1555    1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 1,  /* 5x */
1556    0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,  /* 6x */
1557    1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0,  /* 7x */
1558};
1559#define IdChar(C)  (((c=C)&0x80)!=0 || (c>0x1f && isIdChar[c-0x20]))
1560
1561
1562/*
1563** Return the length of the token that begins at z[0].
1564** Store the token type in *tokenType before returning.
1565*/
1566static int getToken(const char *z, int *tokenType){
1567  int i, c;
1568  switch( *z ){
1569    case 0: {
1570      *tokenType = TOKEN_EOF;
1571      return 0;
1572    }
1573    case ' ': case '\t': case '\n': case '\f': case '\r': {
1574      for(i=1; safe_isspace(z[i]); i++){}
1575      *tokenType = TOKEN_SPACE;
1576      return i;
1577    }
1578    case '`':
1579    case '\'':
1580    case '"': {
1581      int delim = z[0];
1582      for(i=1; (c=z[i])!=0; i++){
1583        if( c==delim ){
1584          if( z[i+1]==delim ){
1585            i++;
1586          }else{
1587            break;
1588          }
1589        }
1590      }
1591      *tokenType = TOKEN_STRING;
1592      return i + (c!=0);
1593    }
1594    case '[': {
1595      for(i=1, c=z[0]; c!=']' && (c=z[i])!=0; i++){}
1596      *tokenType = TOKEN_ID;
1597      return i;
1598    }
1599    default: {
1600      if( !IdChar(*z) ){
1601        break;
1602      }
1603      for(i=1; IdChar(z[i]); i++){}
1604      *tokenType = TOKEN_ID;
1605      return i;
1606    }
1607  }
1608  *tokenType = TOKEN_PUNCT;
1609  return 1;
1610}
1611
1612/*
1613** A token extracted from a string is an instance of the following
1614** structure.
1615*/
1616typedef struct Token {
1617  const char *z;       /* Pointer to token text.  Not '\000' terminated */
1618  short int n;         /* Length of the token text in bytes. */
1619} Token;
1620
1621/*
1622** Given a input string (which is really one of the argv[] parameters
1623** passed into xConnect or xCreate) split the string up into tokens.
1624** Return an array of pointers to '\000' terminated strings, one string
1625** for each non-whitespace token.
1626**
1627** The returned array is terminated by a single NULL pointer.
1628**
1629** Space to hold the returned array is obtained from a single
1630** malloc and should be freed by passing the return value to free().
1631** The individual strings within the token list are all a part of
1632** the single memory allocation and will all be freed at once.
1633*/
1634static char **tokenizeString(const char *z, int *pnToken){
1635  int nToken = 0;
1636  Token *aToken = malloc( strlen(z) * sizeof(aToken[0]) );
1637  int n = 1;
1638  int e, i;
1639  int totalSize = 0;
1640  char **azToken;
1641  char *zCopy;
1642  while( n>0 ){
1643    n = getToken(z, &e);
1644    if( e!=TOKEN_SPACE ){
1645      aToken[nToken].z = z;
1646      aToken[nToken].n = n;
1647      nToken++;
1648      totalSize += n+1;
1649    }
1650    z += n;
1651  }
1652  azToken = (char**)malloc( nToken*sizeof(char*) + totalSize );
1653  zCopy = (char*)&azToken[nToken];
1654  nToken--;
1655  for(i=0; i<nToken; i++){
1656    azToken[i] = zCopy;
1657    n = aToken[i].n;
1658    memcpy(zCopy, aToken[i].z, n);
1659    zCopy[n] = 0;
1660    zCopy += n+1;
1661  }
1662  azToken[nToken] = 0;
1663  free(aToken);
1664  *pnToken = nToken;
1665  return azToken;
1666}
1667
1668/*
1669** Convert an SQL-style quoted string into a normal string by removing
1670** the quote characters.  The conversion is done in-place.  If the
1671** input does not begin with a quote character, then this routine
1672** is a no-op.
1673**
1674** Examples:
1675**
1676**     "abc"   becomes   abc
1677**     'xyz'   becomes   xyz
1678**     [pqr]   becomes   pqr
1679**     `mno`   becomes   mno
1680*/
1681static void dequoteString(char *z){
1682  int quote;
1683  int i, j;
1684  if( z==0 ) return;
1685  quote = z[0];
1686  switch( quote ){
1687    case '\'':  break;
1688    case '"':   break;
1689    case '`':   break;                /* For MySQL compatibility */
1690    case '[':   quote = ']';  break;  /* For MS SqlServer compatibility */
1691    default:    return;
1692  }
1693  for(i=1, j=0; z[i]; i++){
1694    if( z[i]==quote ){
1695      if( z[i+1]==quote ){
1696        z[j++] = quote;
1697        i++;
1698      }else{
1699        z[j++] = 0;
1700        break;
1701      }
1702    }else{
1703      z[j++] = z[i];
1704    }
1705  }
1706}
1707
1708/*
1709** The input azIn is a NULL-terminated list of tokens.  Remove the first
1710** token and all punctuation tokens.  Remove the quotes from
1711** around string literal tokens.
1712**
1713** Example:
1714**
1715**     input:      tokenize chinese ( 'simplifed' , 'mixed' )
1716**     output:     chinese simplifed mixed
1717**
1718** Another example:
1719**
1720**     input:      delimiters ( '[' , ']' , '...' )
1721**     output:     [ ] ...
1722*/
1723static void tokenListToIdList(char **azIn){
1724  int i, j;
1725  if( azIn ){
1726    for(i=0, j=-1; azIn[i]; i++){
1727      if( safe_isalnum(azIn[i][0]) || azIn[i][1] ){
1728        dequoteString(azIn[i]);
1729        if( j>=0 ){
1730          azIn[j] = azIn[i];
1731        }
1732        j++;
1733      }
1734    }
1735    azIn[j] = 0;
1736  }
1737}
1738
1739
1740/*
1741** Find the first alphanumeric token in the string zIn.  Null-terminate
1742** this token.  Remove any quotation marks.  And return a pointer to
1743** the result.
1744*/
1745static char *firstToken(char *zIn, char **pzTail){
1746  int n, ttype;
1747  while(1){
1748    n = getToken(zIn, &ttype);
1749    if( ttype==TOKEN_SPACE ){
1750      zIn += n;
1751    }else if( ttype==TOKEN_EOF ){
1752      *pzTail = zIn;
1753      return 0;
1754    }else{
1755      zIn[n] = 0;
1756      *pzTail = &zIn[1];
1757      dequoteString(zIn);
1758      return zIn;
1759    }
1760  }
1761  /*NOTREACHED*/
1762}
1763
1764/* Return true if...
1765**
1766**   *  s begins with the string t, ignoring case
1767**   *  s is longer than t
1768**   *  The first character of s beyond t is not a alphanumeric
1769**
1770** Ignore leading space in *s.
1771**
1772** To put it another way, return true if the first token of
1773** s[] is t[].
1774*/
1775static int startsWith(const char *s, const char *t){
1776  while( safe_isspace(*s) ){ s++; }
1777  while( *t ){
1778    if( safe_tolower(*s++)!=safe_tolower(*t++) ) return 0;
1779  }
1780  return *s!='_' && !safe_isalnum(*s);
1781}
1782
1783/*
1784** An instance of this structure defines the "spec" of a
1785** full text index.  This structure is populated by parseSpec
1786** and use by fulltextConnect and fulltextCreate.
1787*/
1788typedef struct TableSpec {
1789  const char *zDb;         /* Logical database name */
1790  const char *zName;       /* Name of the full-text index */
1791  int nColumn;             /* Number of columns to be indexed */
1792  char **azColumn;         /* Original names of columns to be indexed */
1793  char **azContentColumn;  /* Column names for %_content */
1794  char **azTokenizer;      /* Name of tokenizer and its arguments */
1795} TableSpec;
1796
1797/*
1798** Reclaim all of the memory used by a TableSpec
1799*/
1800static void clearTableSpec(TableSpec *p) {
1801  free(p->azColumn);
1802  free(p->azContentColumn);
1803  free(p->azTokenizer);
1804}
1805
1806/* Parse a CREATE VIRTUAL TABLE statement, which looks like this:
1807 *
1808 * CREATE VIRTUAL TABLE email
1809 *        USING fts1(subject, body, tokenize mytokenizer(myarg))
1810 *
1811 * We return parsed information in a TableSpec structure.
1812 *
1813 */
1814static int parseSpec(TableSpec *pSpec, int argc, const char *const*argv,
1815                     char**pzErr){
1816  int i, n;
1817  char *z, *zDummy;
1818  char **azArg;
1819  const char *zTokenizer = 0;    /* argv[] entry describing the tokenizer */
1820
1821  assert( argc>=3 );
1822  /* Current interface:
1823  ** argv[0] - module name
1824  ** argv[1] - database name
1825  ** argv[2] - table name
1826  ** argv[3..] - columns, optionally followed by tokenizer specification
1827  **             and snippet delimiters specification.
1828  */
1829
1830  /* Make a copy of the complete argv[][] array in a single allocation.
1831  ** The argv[][] array is read-only and transient.  We can write to the
1832  ** copy in order to modify things and the copy is persistent.
1833  */
1834  memset(pSpec, 0, sizeof(*pSpec));
1835  for(i=n=0; i<argc; i++){
1836    n += strlen(argv[i]) + 1;
1837  }
1838  azArg = malloc( sizeof(char*)*argc + n );
1839  if( azArg==0 ){
1840    return SQLITE_NOMEM;
1841  }
1842  z = (char*)&azArg[argc];
1843  for(i=0; i<argc; i++){
1844    azArg[i] = z;
1845    strcpy(z, argv[i]);
1846    z += strlen(z)+1;
1847  }
1848
1849  /* Identify the column names and the tokenizer and delimiter arguments
1850  ** in the argv[][] array.
1851  */
1852  pSpec->zDb = azArg[1];
1853  pSpec->zName = azArg[2];
1854  pSpec->nColumn = 0;
1855  pSpec->azColumn = azArg;
1856  zTokenizer = "tokenize simple";
1857  for(i=3; i<argc; ++i){
1858    if( startsWith(azArg[i],"tokenize") ){
1859      zTokenizer = azArg[i];
1860    }else{
1861      z = azArg[pSpec->nColumn] = firstToken(azArg[i], &zDummy);
1862      pSpec->nColumn++;
1863    }
1864  }
1865  if( pSpec->nColumn==0 ){
1866    azArg[0] = "content";
1867    pSpec->nColumn = 1;
1868  }
1869
1870  /*
1871  ** Construct the list of content column names.
1872  **
1873  ** Each content column name will be of the form cNNAAAA
1874  ** where NN is the column number and AAAA is the sanitized
1875  ** column name.  "sanitized" means that special characters are
1876  ** converted to "_".  The cNN prefix guarantees that all column
1877  ** names are unique.
1878  **
1879  ** The AAAA suffix is not strictly necessary.  It is included
1880  ** for the convenience of people who might examine the generated
1881  ** %_content table and wonder what the columns are used for.
1882  */
1883  pSpec->azContentColumn = malloc( pSpec->nColumn * sizeof(char *) );
1884  if( pSpec->azContentColumn==0 ){
1885    clearTableSpec(pSpec);
1886    return SQLITE_NOMEM;
1887  }
1888  for(i=0; i<pSpec->nColumn; i++){
1889    char *p;
1890    pSpec->azContentColumn[i] = sqlite3_mprintf("c%d%s", i, azArg[i]);
1891    for (p = pSpec->azContentColumn[i]; *p ; ++p) {
1892      if( !safe_isalnum(*p) ) *p = '_';
1893    }
1894  }
1895
1896  /*
1897  ** Parse the tokenizer specification string.
1898  */
1899  pSpec->azTokenizer = tokenizeString(zTokenizer, &n);
1900  tokenListToIdList(pSpec->azTokenizer);
1901
1902  return SQLITE_OK;
1903}
1904
1905/*
1906** Generate a CREATE TABLE statement that describes the schema of
1907** the virtual table.  Return a pointer to this schema string.
1908**
1909** Space is obtained from sqlite3_mprintf() and should be freed
1910** using sqlite3_free().
1911*/
1912static char *fulltextSchema(
1913  int nColumn,                  /* Number of columns */
1914  const char *const* azColumn,  /* List of columns */
1915  const char *zTableName        /* Name of the table */
1916){
1917  int i;
1918  char *zSchema, *zNext;
1919  const char *zSep = "(";
1920  zSchema = sqlite3_mprintf("CREATE TABLE x");
1921  for(i=0; i<nColumn; i++){
1922    zNext = sqlite3_mprintf("%s%s%Q", zSchema, zSep, azColumn[i]);
1923    sqlite3_free(zSchema);
1924    zSchema = zNext;
1925    zSep = ",";
1926  }
1927  zNext = sqlite3_mprintf("%s,%Q)", zSchema, zTableName);
1928  sqlite3_free(zSchema);
1929  return zNext;
1930}
1931
1932/*
1933** Build a new sqlite3_vtab structure that will describe the
1934** fulltext index defined by spec.
1935*/
1936static int constructVtab(
1937  sqlite3 *db,              /* The SQLite database connection */
1938  TableSpec *spec,          /* Parsed spec information from parseSpec() */
1939  sqlite3_vtab **ppVTab,    /* Write the resulting vtab structure here */
1940  char **pzErr              /* Write any error message here */
1941){
1942  int rc;
1943  int n;
1944  fulltext_vtab *v = 0;
1945  const sqlite3_tokenizer_module *m = NULL;
1946  char *schema;
1947
1948  v = (fulltext_vtab *) malloc(sizeof(fulltext_vtab));
1949  if( v==0 ) return SQLITE_NOMEM;
1950  memset(v, 0, sizeof(*v));
1951  /* sqlite will initialize v->base */
1952  v->db = db;
1953  v->zDb = spec->zDb;       /* Freed when azColumn is freed */
1954  v->zName = spec->zName;   /* Freed when azColumn is freed */
1955  v->nColumn = spec->nColumn;
1956  v->azContentColumn = spec->azContentColumn;
1957  spec->azContentColumn = 0;
1958  v->azColumn = spec->azColumn;
1959  spec->azColumn = 0;
1960
1961  if( spec->azTokenizer==0 ){
1962    return SQLITE_NOMEM;
1963  }
1964  /* TODO(shess) For now, add new tokenizers as else if clauses. */
1965  if( spec->azTokenizer[0]==0 || startsWith(spec->azTokenizer[0], "simple") ){
1966    sqlite3Fts1SimpleTokenizerModule(&m);
1967  }else if( startsWith(spec->azTokenizer[0], "porter") ){
1968    sqlite3Fts1PorterTokenizerModule(&m);
1969  }else{
1970    *pzErr = sqlite3_mprintf("unknown tokenizer: %s", spec->azTokenizer[0]);
1971    rc = SQLITE_ERROR;
1972    goto err;
1973  }
1974  for(n=0; spec->azTokenizer[n]; n++){}
1975  if( n ){
1976    rc = m->xCreate(n-1, (const char*const*)&spec->azTokenizer[1],
1977                    &v->pTokenizer);
1978  }else{
1979    rc = m->xCreate(0, 0, &v->pTokenizer);
1980  }
1981  if( rc!=SQLITE_OK ) goto err;
1982  v->pTokenizer->pModule = m;
1983
1984  /* TODO: verify the existence of backing tables foo_content, foo_term */
1985
1986  schema = fulltextSchema(v->nColumn, (const char*const*)v->azColumn,
1987                          spec->zName);
1988  rc = sqlite3_declare_vtab(db, schema);
1989  sqlite3_free(schema);
1990  if( rc!=SQLITE_OK ) goto err;
1991
1992  memset(v->pFulltextStatements, 0, sizeof(v->pFulltextStatements));
1993
1994  *ppVTab = &v->base;
1995  TRACE(("FTS1 Connect %p\n", v));
1996
1997  return rc;
1998
1999err:
2000  fulltext_vtab_destroy(v);
2001  return rc;
2002}
2003
2004static int fulltextConnect(
2005  sqlite3 *db,
2006  void *pAux,
2007  int argc, const char *const*argv,
2008  sqlite3_vtab **ppVTab,
2009  char **pzErr
2010){
2011  TableSpec spec;
2012  int rc = parseSpec(&spec, argc, argv, pzErr);
2013  if( rc!=SQLITE_OK ) return rc;
2014
2015  rc = constructVtab(db, &spec, ppVTab, pzErr);
2016  clearTableSpec(&spec);
2017  return rc;
2018}
2019
2020  /* The %_content table holds the text of each document, with
2021  ** the rowid used as the docid.
2022  **
2023  ** The %_term table maps each term to a document list blob
2024  ** containing elements sorted by ascending docid, each element
2025  ** encoded as:
2026  **
2027  **   docid varint-encoded
2028  **   token elements:
2029  **     position+1 varint-encoded as delta from previous position
2030  **     start offset varint-encoded as delta from previous start offset
2031  **     end offset varint-encoded as delta from start offset
2032  **
2033  ** The sentinel position of 0 indicates the end of the token list.
2034  **
2035  ** Additionally, doclist blobs are chunked into multiple segments,
2036  ** using segment to order the segments.  New elements are added to
2037  ** the segment at segment 0, until it exceeds CHUNK_MAX.  Then
2038  ** segment 0 is deleted, and the doclist is inserted at segment 1.
2039  ** If there is already a doclist at segment 1, the segment 0 doclist
2040  ** is merged with it, the segment 1 doclist is deleted, and the
2041  ** merged doclist is inserted at segment 2, repeating those
2042  ** operations until an insert succeeds.
2043  **
2044  ** Since this structure doesn't allow us to update elements in place
2045  ** in case of deletion or update, these are simply written to
2046  ** segment 0 (with an empty token list in case of deletion), with
2047  ** docListAccumulate() taking care to retain lower-segment
2048  ** information in preference to higher-segment information.
2049  */
2050  /* TODO(shess) Provide a VACUUM type operation which both removes
2051  ** deleted elements which are no longer necessary, and duplicated
2052  ** elements.  I suspect this will probably not be necessary in
2053  ** practice, though.
2054  */
2055static int fulltextCreate(sqlite3 *db, void *pAux,
2056                          int argc, const char * const *argv,
2057                          sqlite3_vtab **ppVTab, char **pzErr){
2058  int rc;
2059  TableSpec spec;
2060  StringBuffer schema;
2061  TRACE(("FTS1 Create\n"));
2062
2063  rc = parseSpec(&spec, argc, argv, pzErr);
2064  if( rc!=SQLITE_OK ) return rc;
2065
2066  initStringBuffer(&schema);
2067  append(&schema, "CREATE TABLE %_content(");
2068  appendList(&schema, spec.nColumn, spec.azContentColumn);
2069  append(&schema, ")");
2070  rc = sql_exec(db, spec.zDb, spec.zName, schema.s);
2071  free(schema.s);
2072  if( rc!=SQLITE_OK ) goto out;
2073
2074  rc = sql_exec(db, spec.zDb, spec.zName,
2075    "create table %_term(term text, segment integer, doclist blob, "
2076                        "primary key(term, segment));");
2077  if( rc!=SQLITE_OK ) goto out;
2078
2079  rc = constructVtab(db, &spec, ppVTab, pzErr);
2080
2081out:
2082  clearTableSpec(&spec);
2083  return rc;
2084}
2085
2086/* Decide how to handle an SQL query. */
2087static int fulltextBestIndex(sqlite3_vtab *pVTab, sqlite3_index_info *pInfo){
2088  int i;
2089  TRACE(("FTS1 BestIndex\n"));
2090
2091  for(i=0; i<pInfo->nConstraint; ++i){
2092    const struct sqlite3_index_constraint *pConstraint;
2093    pConstraint = &pInfo->aConstraint[i];
2094    if( pConstraint->usable ) {
2095      if( pConstraint->iColumn==-1 &&
2096          pConstraint->op==SQLITE_INDEX_CONSTRAINT_EQ ){
2097        pInfo->idxNum = QUERY_ROWID;      /* lookup by rowid */
2098        TRACE(("FTS1 QUERY_ROWID\n"));
2099      } else if( pConstraint->iColumn>=0 &&
2100                 pConstraint->op==SQLITE_INDEX_CONSTRAINT_MATCH ){
2101        /* full-text search */
2102        pInfo->idxNum = QUERY_FULLTEXT + pConstraint->iColumn;
2103        TRACE(("FTS1 QUERY_FULLTEXT %d\n", pConstraint->iColumn));
2104      } else continue;
2105
2106      pInfo->aConstraintUsage[i].argvIndex = 1;
2107      pInfo->aConstraintUsage[i].omit = 1;
2108
2109      /* An arbitrary value for now.
2110       * TODO: Perhaps rowid matches should be considered cheaper than
2111       * full-text searches. */
2112      pInfo->estimatedCost = 1.0;
2113
2114      return SQLITE_OK;
2115    }
2116  }
2117  pInfo->idxNum = QUERY_GENERIC;
2118  return SQLITE_OK;
2119}
2120
2121static int fulltextDisconnect(sqlite3_vtab *pVTab){
2122  TRACE(("FTS1 Disconnect %p\n", pVTab));
2123  fulltext_vtab_destroy((fulltext_vtab *)pVTab);
2124  return SQLITE_OK;
2125}
2126
2127static int fulltextDestroy(sqlite3_vtab *pVTab){
2128  fulltext_vtab *v = (fulltext_vtab *)pVTab;
2129  int rc;
2130
2131  TRACE(("FTS1 Destroy %p\n", pVTab));
2132  rc = sql_exec(v->db, v->zDb, v->zName,
2133                "drop table if exists %_content;"
2134                "drop table if exists %_term;"
2135                );
2136  if( rc!=SQLITE_OK ) return rc;
2137
2138  fulltext_vtab_destroy((fulltext_vtab *)pVTab);
2139  return SQLITE_OK;
2140}
2141
2142static int fulltextOpen(sqlite3_vtab *pVTab, sqlite3_vtab_cursor **ppCursor){
2143  fulltext_cursor *c;
2144
2145  c = (fulltext_cursor *) calloc(sizeof(fulltext_cursor), 1);
2146  /* sqlite will initialize c->base */
2147  *ppCursor = &c->base;
2148  TRACE(("FTS1 Open %p: %p\n", pVTab, c));
2149
2150  return SQLITE_OK;
2151}
2152
2153
2154/* Free all of the dynamically allocated memory held by *q
2155*/
2156static void queryClear(Query *q){
2157  int i;
2158  for(i = 0; i < q->nTerms; ++i){
2159    free(q->pTerms[i].pTerm);
2160  }
2161  free(q->pTerms);
2162  memset(q, 0, sizeof(*q));
2163}
2164
2165/* Free all of the dynamically allocated memory held by the
2166** Snippet
2167*/
2168static void snippetClear(Snippet *p){
2169  free(p->aMatch);
2170  free(p->zOffset);
2171  free(p->zSnippet);
2172  memset(p, 0, sizeof(*p));
2173}
2174/*
2175** Append a single entry to the p->aMatch[] log.
2176*/
2177static void snippetAppendMatch(
2178  Snippet *p,               /* Append the entry to this snippet */
2179  int iCol, int iTerm,      /* The column and query term */
2180  int iStart, int nByte     /* Offset and size of the match */
2181){
2182  int i;
2183  struct snippetMatch *pMatch;
2184  if( p->nMatch+1>=p->nAlloc ){
2185    p->nAlloc = p->nAlloc*2 + 10;
2186    p->aMatch = realloc(p->aMatch, p->nAlloc*sizeof(p->aMatch[0]) );
2187    if( p->aMatch==0 ){
2188      p->nMatch = 0;
2189      p->nAlloc = 0;
2190      return;
2191    }
2192  }
2193  i = p->nMatch++;
2194  pMatch = &p->aMatch[i];
2195  pMatch->iCol = iCol;
2196  pMatch->iTerm = iTerm;
2197  pMatch->iStart = iStart;
2198  pMatch->nByte = nByte;
2199}
2200
2201/*
2202** Sizing information for the circular buffer used in snippetOffsetsOfColumn()
2203*/
2204#define FTS1_ROTOR_SZ   (32)
2205#define FTS1_ROTOR_MASK (FTS1_ROTOR_SZ-1)
2206
2207/*
2208** Add entries to pSnippet->aMatch[] for every match that occurs against
2209** document zDoc[0..nDoc-1] which is stored in column iColumn.
2210*/
2211static void snippetOffsetsOfColumn(
2212  Query *pQuery,
2213  Snippet *pSnippet,
2214  int iColumn,
2215  const char *zDoc,
2216  int nDoc
2217){
2218  const sqlite3_tokenizer_module *pTModule;  /* The tokenizer module */
2219  sqlite3_tokenizer *pTokenizer;             /* The specific tokenizer */
2220  sqlite3_tokenizer_cursor *pTCursor;        /* Tokenizer cursor */
2221  fulltext_vtab *pVtab;                /* The full text index */
2222  int nColumn;                         /* Number of columns in the index */
2223  const QueryTerm *aTerm;              /* Query string terms */
2224  int nTerm;                           /* Number of query string terms */
2225  int i, j;                            /* Loop counters */
2226  int rc;                              /* Return code */
2227  unsigned int match, prevMatch;       /* Phrase search bitmasks */
2228  const char *zToken;                  /* Next token from the tokenizer */
2229  int nToken;                          /* Size of zToken */
2230  int iBegin, iEnd, iPos;              /* Offsets of beginning and end */
2231
2232  /* The following variables keep a circular buffer of the last
2233  ** few tokens */
2234  unsigned int iRotor = 0;             /* Index of current token */
2235  int iRotorBegin[FTS1_ROTOR_SZ];      /* Beginning offset of token */
2236  int iRotorLen[FTS1_ROTOR_SZ];        /* Length of token */
2237
2238  pVtab = pQuery->pFts;
2239  nColumn = pVtab->nColumn;
2240  pTokenizer = pVtab->pTokenizer;
2241  pTModule = pTokenizer->pModule;
2242  rc = pTModule->xOpen(pTokenizer, zDoc, nDoc, &pTCursor);
2243  if( rc ) return;
2244  pTCursor->pTokenizer = pTokenizer;
2245  aTerm = pQuery->pTerms;
2246  nTerm = pQuery->nTerms;
2247  if( nTerm>=FTS1_ROTOR_SZ ){
2248    nTerm = FTS1_ROTOR_SZ - 1;
2249  }
2250  prevMatch = 0;
2251  while(1){
2252    rc = pTModule->xNext(pTCursor, &zToken, &nToken, &iBegin, &iEnd, &iPos);
2253    if( rc ) break;
2254    iRotorBegin[iRotor&FTS1_ROTOR_MASK] = iBegin;
2255    iRotorLen[iRotor&FTS1_ROTOR_MASK] = iEnd-iBegin;
2256    match = 0;
2257    for(i=0; i<nTerm; i++){
2258      int iCol;
2259      iCol = aTerm[i].iColumn;
2260      if( iCol>=0 && iCol<nColumn && iCol!=iColumn ) continue;
2261      if( aTerm[i].nTerm!=nToken ) continue;
2262      if( memcmp(aTerm[i].pTerm, zToken, nToken) ) continue;
2263      if( aTerm[i].iPhrase>1 && (prevMatch & (1<<i))==0 ) continue;
2264      match |= 1<<i;
2265      if( i==nTerm-1 || aTerm[i+1].iPhrase==1 ){
2266        for(j=aTerm[i].iPhrase-1; j>=0; j--){
2267          int k = (iRotor-j) & FTS1_ROTOR_MASK;
2268          snippetAppendMatch(pSnippet, iColumn, i-j,
2269                iRotorBegin[k], iRotorLen[k]);
2270        }
2271      }
2272    }
2273    prevMatch = match<<1;
2274    iRotor++;
2275  }
2276  pTModule->xClose(pTCursor);
2277}
2278
2279
2280/*
2281** Compute all offsets for the current row of the query.
2282** If the offsets have already been computed, this routine is a no-op.
2283*/
2284static void snippetAllOffsets(fulltext_cursor *p){
2285  int nColumn;
2286  int iColumn, i;
2287  int iFirst, iLast;
2288  fulltext_vtab *pFts;
2289
2290  if( p->snippet.nMatch ) return;
2291  if( p->q.nTerms==0 ) return;
2292  pFts = p->q.pFts;
2293  nColumn = pFts->nColumn;
2294  iColumn = p->iCursorType - QUERY_FULLTEXT;
2295  if( iColumn<0 || iColumn>=nColumn ){
2296    iFirst = 0;
2297    iLast = nColumn-1;
2298  }else{
2299    iFirst = iColumn;
2300    iLast = iColumn;
2301  }
2302  for(i=iFirst; i<=iLast; i++){
2303    const char *zDoc;
2304    int nDoc;
2305    zDoc = (const char*)sqlite3_column_text(p->pStmt, i+1);
2306    nDoc = sqlite3_column_bytes(p->pStmt, i+1);
2307    snippetOffsetsOfColumn(&p->q, &p->snippet, i, zDoc, nDoc);
2308  }
2309}
2310
2311/*
2312** Convert the information in the aMatch[] array of the snippet
2313** into the string zOffset[0..nOffset-1].
2314*/
2315static void snippetOffsetText(Snippet *p){
2316  int i;
2317  int cnt = 0;
2318  StringBuffer sb;
2319  char zBuf[200];
2320  if( p->zOffset ) return;
2321  initStringBuffer(&sb);
2322  for(i=0; i<p->nMatch; i++){
2323    struct snippetMatch *pMatch = &p->aMatch[i];
2324    zBuf[0] = ' ';
2325    sqlite3_snprintf(sizeof(zBuf)-1, &zBuf[cnt>0], "%d %d %d %d",
2326        pMatch->iCol, pMatch->iTerm, pMatch->iStart, pMatch->nByte);
2327    append(&sb, zBuf);
2328    cnt++;
2329  }
2330  p->zOffset = sb.s;
2331  p->nOffset = sb.len;
2332}
2333
2334/*
2335** zDoc[0..nDoc-1] is phrase of text.  aMatch[0..nMatch-1] are a set
2336** of matching words some of which might be in zDoc.  zDoc is column
2337** number iCol.
2338**
2339** iBreak is suggested spot in zDoc where we could begin or end an
2340** excerpt.  Return a value similar to iBreak but possibly adjusted
2341** to be a little left or right so that the break point is better.
2342*/
2343static int wordBoundary(
2344  int iBreak,                   /* The suggested break point */
2345  const char *zDoc,             /* Document text */
2346  int nDoc,                     /* Number of bytes in zDoc[] */
2347  struct snippetMatch *aMatch,  /* Matching words */
2348  int nMatch,                   /* Number of entries in aMatch[] */
2349  int iCol                      /* The column number for zDoc[] */
2350){
2351  int i;
2352  if( iBreak<=10 ){
2353    return 0;
2354  }
2355  if( iBreak>=nDoc-10 ){
2356    return nDoc;
2357  }
2358  for(i=0; i<nMatch && aMatch[i].iCol<iCol; i++){}
2359  while( i<nMatch && aMatch[i].iStart+aMatch[i].nByte<iBreak ){ i++; }
2360  if( i<nMatch ){
2361    if( aMatch[i].iStart<iBreak+10 ){
2362      return aMatch[i].iStart;
2363    }
2364    if( i>0 && aMatch[i-1].iStart+aMatch[i-1].nByte>=iBreak ){
2365      return aMatch[i-1].iStart;
2366    }
2367  }
2368  for(i=1; i<=10; i++){
2369    if( safe_isspace(zDoc[iBreak-i]) ){
2370      return iBreak - i + 1;
2371    }
2372    if( safe_isspace(zDoc[iBreak+i]) ){
2373      return iBreak + i + 1;
2374    }
2375  }
2376  return iBreak;
2377}
2378
2379/*
2380** If the StringBuffer does not end in white space, add a single
2381** space character to the end.
2382*/
2383static void appendWhiteSpace(StringBuffer *p){
2384  if( p->len==0 ) return;
2385  if( safe_isspace(p->s[p->len-1]) ) return;
2386  append(p, " ");
2387}
2388
2389/*
2390** Remove white space from teh end of the StringBuffer
2391*/
2392static void trimWhiteSpace(StringBuffer *p){
2393  while( p->len>0 && safe_isspace(p->s[p->len-1]) ){
2394    p->len--;
2395  }
2396}
2397
2398
2399
2400/*
2401** Allowed values for Snippet.aMatch[].snStatus
2402*/
2403#define SNIPPET_IGNORE  0   /* It is ok to omit this match from the snippet */
2404#define SNIPPET_DESIRED 1   /* We want to include this match in the snippet */
2405
2406/*
2407** Generate the text of a snippet.
2408*/
2409static void snippetText(
2410  fulltext_cursor *pCursor,   /* The cursor we need the snippet for */
2411  const char *zStartMark,     /* Markup to appear before each match */
2412  const char *zEndMark,       /* Markup to appear after each match */
2413  const char *zEllipsis       /* Ellipsis mark */
2414){
2415  int i, j;
2416  struct snippetMatch *aMatch;
2417  int nMatch;
2418  int nDesired;
2419  StringBuffer sb;
2420  int tailCol;
2421  int tailOffset;
2422  int iCol;
2423  int nDoc;
2424  const char *zDoc;
2425  int iStart, iEnd;
2426  int tailEllipsis = 0;
2427  int iMatch;
2428
2429
2430  free(pCursor->snippet.zSnippet);
2431  pCursor->snippet.zSnippet = 0;
2432  aMatch = pCursor->snippet.aMatch;
2433  nMatch = pCursor->snippet.nMatch;
2434  initStringBuffer(&sb);
2435
2436  for(i=0; i<nMatch; i++){
2437    aMatch[i].snStatus = SNIPPET_IGNORE;
2438  }
2439  nDesired = 0;
2440  for(i=0; i<pCursor->q.nTerms; i++){
2441    for(j=0; j<nMatch; j++){
2442      if( aMatch[j].iTerm==i ){
2443        aMatch[j].snStatus = SNIPPET_DESIRED;
2444        nDesired++;
2445        break;
2446      }
2447    }
2448  }
2449
2450  iMatch = 0;
2451  tailCol = -1;
2452  tailOffset = 0;
2453  for(i=0; i<nMatch && nDesired>0; i++){
2454    if( aMatch[i].snStatus!=SNIPPET_DESIRED ) continue;
2455    nDesired--;
2456    iCol = aMatch[i].iCol;
2457    zDoc = (const char*)sqlite3_column_text(pCursor->pStmt, iCol+1);
2458    nDoc = sqlite3_column_bytes(pCursor->pStmt, iCol+1);
2459    iStart = aMatch[i].iStart - 40;
2460    iStart = wordBoundary(iStart, zDoc, nDoc, aMatch, nMatch, iCol);
2461    if( iStart<=10 ){
2462      iStart = 0;
2463    }
2464    if( iCol==tailCol && iStart<=tailOffset+20 ){
2465      iStart = tailOffset;
2466    }
2467    if( (iCol!=tailCol && tailCol>=0) || iStart!=tailOffset ){
2468      trimWhiteSpace(&sb);
2469      appendWhiteSpace(&sb);
2470      append(&sb, zEllipsis);
2471      appendWhiteSpace(&sb);
2472    }
2473    iEnd = aMatch[i].iStart + aMatch[i].nByte + 40;
2474    iEnd = wordBoundary(iEnd, zDoc, nDoc, aMatch, nMatch, iCol);
2475    if( iEnd>=nDoc-10 ){
2476      iEnd = nDoc;
2477      tailEllipsis = 0;
2478    }else{
2479      tailEllipsis = 1;
2480    }
2481    while( iMatch<nMatch && aMatch[iMatch].iCol<iCol ){ iMatch++; }
2482    while( iStart<iEnd ){
2483      while( iMatch<nMatch && aMatch[iMatch].iStart<iStart
2484             && aMatch[iMatch].iCol<=iCol ){
2485        iMatch++;
2486      }
2487      if( iMatch<nMatch && aMatch[iMatch].iStart<iEnd
2488             && aMatch[iMatch].iCol==iCol ){
2489        nappend(&sb, &zDoc[iStart], aMatch[iMatch].iStart - iStart);
2490        iStart = aMatch[iMatch].iStart;
2491        append(&sb, zStartMark);
2492        nappend(&sb, &zDoc[iStart], aMatch[iMatch].nByte);
2493        append(&sb, zEndMark);
2494        iStart += aMatch[iMatch].nByte;
2495        for(j=iMatch+1; j<nMatch; j++){
2496          if( aMatch[j].iTerm==aMatch[iMatch].iTerm
2497              && aMatch[j].snStatus==SNIPPET_DESIRED ){
2498            nDesired--;
2499            aMatch[j].snStatus = SNIPPET_IGNORE;
2500          }
2501        }
2502      }else{
2503        nappend(&sb, &zDoc[iStart], iEnd - iStart);
2504        iStart = iEnd;
2505      }
2506    }
2507    tailCol = iCol;
2508    tailOffset = iEnd;
2509  }
2510  trimWhiteSpace(&sb);
2511  if( tailEllipsis ){
2512    appendWhiteSpace(&sb);
2513    append(&sb, zEllipsis);
2514  }
2515  pCursor->snippet.zSnippet = sb.s;
2516  pCursor->snippet.nSnippet = sb.len;
2517}
2518
2519
2520/*
2521** Close the cursor.  For additional information see the documentation
2522** on the xClose method of the virtual table interface.
2523*/
2524static int fulltextClose(sqlite3_vtab_cursor *pCursor){
2525  fulltext_cursor *c = (fulltext_cursor *) pCursor;
2526  TRACE(("FTS1 Close %p\n", c));
2527  sqlite3_finalize(c->pStmt);
2528  queryClear(&c->q);
2529  snippetClear(&c->snippet);
2530  if( c->result.pDoclist!=NULL ){
2531    docListDelete(c->result.pDoclist);
2532  }
2533  free(c);
2534  return SQLITE_OK;
2535}
2536
2537static int fulltextNext(sqlite3_vtab_cursor *pCursor){
2538  fulltext_cursor *c = (fulltext_cursor *) pCursor;
2539  sqlite_int64 iDocid;
2540  int rc;
2541
2542  TRACE(("FTS1 Next %p\n", pCursor));
2543  snippetClear(&c->snippet);
2544  if( c->iCursorType < QUERY_FULLTEXT ){
2545    /* TODO(shess) Handle SQLITE_SCHEMA AND SQLITE_BUSY. */
2546    rc = sqlite3_step(c->pStmt);
2547    switch( rc ){
2548      case SQLITE_ROW:
2549        c->eof = 0;
2550        return SQLITE_OK;
2551      case SQLITE_DONE:
2552        c->eof = 1;
2553        return SQLITE_OK;
2554      default:
2555        c->eof = 1;
2556        return rc;
2557    }
2558  } else {  /* full-text query */
2559    rc = sqlite3_reset(c->pStmt);
2560    if( rc!=SQLITE_OK ) return rc;
2561
2562    iDocid = nextDocid(&c->result);
2563    if( iDocid==0 ){
2564      c->eof = 1;
2565      return SQLITE_OK;
2566    }
2567    rc = sqlite3_bind_int64(c->pStmt, 1, iDocid);
2568    if( rc!=SQLITE_OK ) return rc;
2569    /* TODO(shess) Handle SQLITE_SCHEMA AND SQLITE_BUSY. */
2570    rc = sqlite3_step(c->pStmt);
2571    if( rc==SQLITE_ROW ){   /* the case we expect */
2572      c->eof = 0;
2573      return SQLITE_OK;
2574    }
2575    /* an error occurred; abort */
2576    return rc==SQLITE_DONE ? SQLITE_ERROR : rc;
2577  }
2578}
2579
2580
2581/* Return a DocList corresponding to the query term *pTerm.  If *pTerm
2582** is the first term of a phrase query, go ahead and evaluate the phrase
2583** query and return the doclist for the entire phrase query.
2584**
2585** The result is stored in pTerm->doclist.
2586*/
2587static int docListOfTerm(
2588  fulltext_vtab *v,     /* The full text index */
2589  int iColumn,          /* column to restrict to.  No restrition if >=nColumn */
2590  QueryTerm *pQTerm,    /* Term we are looking for, or 1st term of a phrase */
2591  DocList **ppResult    /* Write the result here */
2592){
2593  DocList *pLeft, *pRight, *pNew;
2594  int i, rc;
2595
2596  pLeft = docListNew(DL_POSITIONS);
2597  rc = term_select_all(v, iColumn, pQTerm->pTerm, pQTerm->nTerm, pLeft);
2598  if( rc ){
2599    docListDelete(pLeft);
2600    return rc;
2601  }
2602  for(i=1; i<=pQTerm->nPhrase; i++){
2603    pRight = docListNew(DL_POSITIONS);
2604    rc = term_select_all(v, iColumn, pQTerm[i].pTerm, pQTerm[i].nTerm, pRight);
2605    if( rc ){
2606      docListDelete(pLeft);
2607      return rc;
2608    }
2609    pNew = docListNew(i<pQTerm->nPhrase ? DL_POSITIONS : DL_DOCIDS);
2610    docListPhraseMerge(pLeft, pRight, pNew);
2611    docListDelete(pLeft);
2612    docListDelete(pRight);
2613    pLeft = pNew;
2614  }
2615  *ppResult = pLeft;
2616  return SQLITE_OK;
2617}
2618
2619/* Add a new term pTerm[0..nTerm-1] to the query *q.
2620*/
2621static void queryAdd(Query *q, const char *pTerm, int nTerm){
2622  QueryTerm *t;
2623  ++q->nTerms;
2624  q->pTerms = realloc(q->pTerms, q->nTerms * sizeof(q->pTerms[0]));
2625  if( q->pTerms==0 ){
2626    q->nTerms = 0;
2627    return;
2628  }
2629  t = &q->pTerms[q->nTerms - 1];
2630  memset(t, 0, sizeof(*t));
2631  t->pTerm = malloc(nTerm+1);
2632  memcpy(t->pTerm, pTerm, nTerm);
2633  t->pTerm[nTerm] = 0;
2634  t->nTerm = nTerm;
2635  t->isOr = q->nextIsOr;
2636  q->nextIsOr = 0;
2637  t->iColumn = q->nextColumn;
2638  q->nextColumn = q->dfltColumn;
2639}
2640
2641/*
2642** Check to see if the string zToken[0...nToken-1] matches any
2643** column name in the virtual table.   If it does,
2644** return the zero-indexed column number.  If not, return -1.
2645*/
2646static int checkColumnSpecifier(
2647  fulltext_vtab *pVtab,    /* The virtual table */
2648  const char *zToken,      /* Text of the token */
2649  int nToken               /* Number of characters in the token */
2650){
2651  int i;
2652  for(i=0; i<pVtab->nColumn; i++){
2653    if( memcmp(pVtab->azColumn[i], zToken, nToken)==0
2654        && pVtab->azColumn[i][nToken]==0 ){
2655      return i;
2656    }
2657  }
2658  return -1;
2659}
2660
2661/*
2662** Parse the text at pSegment[0..nSegment-1].  Add additional terms
2663** to the query being assemblied in pQuery.
2664**
2665** inPhrase is true if pSegment[0..nSegement-1] is contained within
2666** double-quotes.  If inPhrase is true, then the first term
2667** is marked with the number of terms in the phrase less one and
2668** OR and "-" syntax is ignored.  If inPhrase is false, then every
2669** term found is marked with nPhrase=0 and OR and "-" syntax is significant.
2670*/
2671static int tokenizeSegment(
2672  sqlite3_tokenizer *pTokenizer,          /* The tokenizer to use */
2673  const char *pSegment, int nSegment,     /* Query expression being parsed */
2674  int inPhrase,                           /* True if within "..." */
2675  Query *pQuery                           /* Append results here */
2676){
2677  const sqlite3_tokenizer_module *pModule = pTokenizer->pModule;
2678  sqlite3_tokenizer_cursor *pCursor;
2679  int firstIndex = pQuery->nTerms;
2680  int iCol;
2681  int nTerm = 1;
2682
2683  int rc = pModule->xOpen(pTokenizer, pSegment, nSegment, &pCursor);
2684  if( rc!=SQLITE_OK ) return rc;
2685  pCursor->pTokenizer = pTokenizer;
2686
2687  while( 1 ){
2688    const char *pToken;
2689    int nToken, iBegin, iEnd, iPos;
2690
2691    rc = pModule->xNext(pCursor,
2692                        &pToken, &nToken,
2693                        &iBegin, &iEnd, &iPos);
2694    if( rc!=SQLITE_OK ) break;
2695    if( !inPhrase &&
2696        pSegment[iEnd]==':' &&
2697         (iCol = checkColumnSpecifier(pQuery->pFts, pToken, nToken))>=0 ){
2698      pQuery->nextColumn = iCol;
2699      continue;
2700    }
2701    if( !inPhrase && pQuery->nTerms>0 && nToken==2
2702         && pSegment[iBegin]=='O' && pSegment[iBegin+1]=='R' ){
2703      pQuery->nextIsOr = 1;
2704      continue;
2705    }
2706    queryAdd(pQuery, pToken, nToken);
2707    if( !inPhrase && iBegin>0 && pSegment[iBegin-1]=='-' ){
2708      pQuery->pTerms[pQuery->nTerms-1].isNot = 1;
2709    }
2710    pQuery->pTerms[pQuery->nTerms-1].iPhrase = nTerm;
2711    if( inPhrase ){
2712      nTerm++;
2713    }
2714  }
2715
2716  if( inPhrase && pQuery->nTerms>firstIndex ){
2717    pQuery->pTerms[firstIndex].nPhrase = pQuery->nTerms - firstIndex - 1;
2718  }
2719
2720  return pModule->xClose(pCursor);
2721}
2722
2723/* Parse a query string, yielding a Query object pQuery.
2724**
2725** The calling function will need to queryClear() to clean up
2726** the dynamically allocated memory held by pQuery.
2727*/
2728static int parseQuery(
2729  fulltext_vtab *v,        /* The fulltext index */
2730  const char *zInput,      /* Input text of the query string */
2731  int nInput,              /* Size of the input text */
2732  int dfltColumn,          /* Default column of the index to match against */
2733  Query *pQuery            /* Write the parse results here. */
2734){
2735  int iInput, inPhrase = 0;
2736
2737  if( zInput==0 ) nInput = 0;
2738  if( nInput<0 ) nInput = strlen(zInput);
2739  pQuery->nTerms = 0;
2740  pQuery->pTerms = NULL;
2741  pQuery->nextIsOr = 0;
2742  pQuery->nextColumn = dfltColumn;
2743  pQuery->dfltColumn = dfltColumn;
2744  pQuery->pFts = v;
2745
2746  for(iInput=0; iInput<nInput; ++iInput){
2747    int i;
2748    for(i=iInput; i<nInput && zInput[i]!='"'; ++i){}
2749    if( i>iInput ){
2750      tokenizeSegment(v->pTokenizer, zInput+iInput, i-iInput, inPhrase,
2751                       pQuery);
2752    }
2753    iInput = i;
2754    if( i<nInput ){
2755      assert( zInput[i]=='"' );
2756      inPhrase = !inPhrase;
2757    }
2758  }
2759
2760  if( inPhrase ){
2761    /* unmatched quote */
2762    queryClear(pQuery);
2763    return SQLITE_ERROR;
2764  }
2765  return SQLITE_OK;
2766}
2767
2768/* Perform a full-text query using the search expression in
2769** zInput[0..nInput-1].  Return a list of matching documents
2770** in pResult.
2771**
2772** Queries must match column iColumn.  Or if iColumn>=nColumn
2773** they are allowed to match against any column.
2774*/
2775static int fulltextQuery(
2776  fulltext_vtab *v,      /* The full text index */
2777  int iColumn,           /* Match against this column by default */
2778  const char *zInput,    /* The query string */
2779  int nInput,            /* Number of bytes in zInput[] */
2780  DocList **pResult,     /* Write the result doclist here */
2781  Query *pQuery          /* Put parsed query string here */
2782){
2783  int i, iNext, rc;
2784  DocList *pLeft = NULL;
2785  DocList *pRight, *pNew, *pOr;
2786  int nNot = 0;
2787  QueryTerm *aTerm;
2788
2789  rc = parseQuery(v, zInput, nInput, iColumn, pQuery);
2790  if( rc!=SQLITE_OK ) return rc;
2791
2792  /* Merge AND terms. */
2793  aTerm = pQuery->pTerms;
2794  for(i = 0; i<pQuery->nTerms; i=iNext){
2795    if( aTerm[i].isNot ){
2796      /* Handle all NOT terms in a separate pass */
2797      nNot++;
2798      iNext = i + aTerm[i].nPhrase+1;
2799      continue;
2800    }
2801    iNext = i + aTerm[i].nPhrase + 1;
2802    rc = docListOfTerm(v, aTerm[i].iColumn, &aTerm[i], &pRight);
2803    if( rc ){
2804      queryClear(pQuery);
2805      return rc;
2806    }
2807    while( iNext<pQuery->nTerms && aTerm[iNext].isOr ){
2808      rc = docListOfTerm(v, aTerm[iNext].iColumn, &aTerm[iNext], &pOr);
2809      iNext += aTerm[iNext].nPhrase + 1;
2810      if( rc ){
2811        queryClear(pQuery);
2812        return rc;
2813      }
2814      pNew = docListNew(DL_DOCIDS);
2815      docListOrMerge(pRight, pOr, pNew);
2816      docListDelete(pRight);
2817      docListDelete(pOr);
2818      pRight = pNew;
2819    }
2820    if( pLeft==0 ){
2821      pLeft = pRight;
2822    }else{
2823      pNew = docListNew(DL_DOCIDS);
2824      docListAndMerge(pLeft, pRight, pNew);
2825      docListDelete(pRight);
2826      docListDelete(pLeft);
2827      pLeft = pNew;
2828    }
2829  }
2830
2831  if( nNot && pLeft==0 ){
2832    /* We do not yet know how to handle a query of only NOT terms */
2833    return SQLITE_ERROR;
2834  }
2835
2836  /* Do the EXCEPT terms */
2837  for(i=0; i<pQuery->nTerms;  i += aTerm[i].nPhrase + 1){
2838    if( !aTerm[i].isNot ) continue;
2839    rc = docListOfTerm(v, aTerm[i].iColumn, &aTerm[i], &pRight);
2840    if( rc ){
2841      queryClear(pQuery);
2842      docListDelete(pLeft);
2843      return rc;
2844    }
2845    pNew = docListNew(DL_DOCIDS);
2846    docListExceptMerge(pLeft, pRight, pNew);
2847    docListDelete(pRight);
2848    docListDelete(pLeft);
2849    pLeft = pNew;
2850  }
2851
2852  *pResult = pLeft;
2853  return rc;
2854}
2855
2856/*
2857** This is the xFilter interface for the virtual table.  See
2858** the virtual table xFilter method documentation for additional
2859** information.
2860**
2861** If idxNum==QUERY_GENERIC then do a full table scan against
2862** the %_content table.
2863**
2864** If idxNum==QUERY_ROWID then do a rowid lookup for a single entry
2865** in the %_content table.
2866**
2867** If idxNum>=QUERY_FULLTEXT then use the full text index.  The
2868** column on the left-hand side of the MATCH operator is column
2869** number idxNum-QUERY_FULLTEXT, 0 indexed.  argv[0] is the right-hand
2870** side of the MATCH operator.
2871*/
2872/* TODO(shess) Upgrade the cursor initialization and destruction to
2873** account for fulltextFilter() being called multiple times on the
2874** same cursor.  The current solution is very fragile.  Apply fix to
2875** fts2 as appropriate.
2876*/
2877static int fulltextFilter(
2878  sqlite3_vtab_cursor *pCursor,     /* The cursor used for this query */
2879  int idxNum, const char *idxStr,   /* Which indexing scheme to use */
2880  int argc, sqlite3_value **argv    /* Arguments for the indexing scheme */
2881){
2882  fulltext_cursor *c = (fulltext_cursor *) pCursor;
2883  fulltext_vtab *v = cursor_vtab(c);
2884  int rc;
2885  char *zSql;
2886
2887  TRACE(("FTS1 Filter %p\n",pCursor));
2888
2889  zSql = sqlite3_mprintf("select rowid, * from %%_content %s",
2890                          idxNum==QUERY_GENERIC ? "" : "where rowid=?");
2891  sqlite3_finalize(c->pStmt);
2892  rc = sql_prepare(v->db, v->zDb, v->zName, &c->pStmt, zSql);
2893  sqlite3_free(zSql);
2894  if( rc!=SQLITE_OK ) return rc;
2895
2896  c->iCursorType = idxNum;
2897  switch( idxNum ){
2898    case QUERY_GENERIC:
2899      break;
2900
2901    case QUERY_ROWID:
2902      rc = sqlite3_bind_int64(c->pStmt, 1, sqlite3_value_int64(argv[0]));
2903      if( rc!=SQLITE_OK ) return rc;
2904      break;
2905
2906    default:   /* full-text search */
2907    {
2908      const char *zQuery = (const char *)sqlite3_value_text(argv[0]);
2909      DocList *pResult;
2910      assert( idxNum<=QUERY_FULLTEXT+v->nColumn);
2911      assert( argc==1 );
2912      queryClear(&c->q);
2913      rc = fulltextQuery(v, idxNum-QUERY_FULLTEXT, zQuery, -1, &pResult, &c->q);
2914      if( rc!=SQLITE_OK ) return rc;
2915      if( c->result.pDoclist!=NULL ) docListDelete(c->result.pDoclist);
2916      readerInit(&c->result, pResult);
2917      break;
2918    }
2919  }
2920
2921  return fulltextNext(pCursor);
2922}
2923
2924/* This is the xEof method of the virtual table.  The SQLite core
2925** calls this routine to find out if it has reached the end of
2926** a query's results set.
2927*/
2928static int fulltextEof(sqlite3_vtab_cursor *pCursor){
2929  fulltext_cursor *c = (fulltext_cursor *) pCursor;
2930  return c->eof;
2931}
2932
2933/* This is the xColumn method of the virtual table.  The SQLite
2934** core calls this method during a query when it needs the value
2935** of a column from the virtual table.  This method needs to use
2936** one of the sqlite3_result_*() routines to store the requested
2937** value back in the pContext.
2938*/
2939static int fulltextColumn(sqlite3_vtab_cursor *pCursor,
2940                          sqlite3_context *pContext, int idxCol){
2941  fulltext_cursor *c = (fulltext_cursor *) pCursor;
2942  fulltext_vtab *v = cursor_vtab(c);
2943
2944  if( idxCol<v->nColumn ){
2945    sqlite3_value *pVal = sqlite3_column_value(c->pStmt, idxCol+1);
2946    sqlite3_result_value(pContext, pVal);
2947  }else if( idxCol==v->nColumn ){
2948    /* The extra column whose name is the same as the table.
2949    ** Return a blob which is a pointer to the cursor
2950    */
2951    sqlite3_result_blob(pContext, &c, sizeof(c), SQLITE_TRANSIENT);
2952  }
2953  return SQLITE_OK;
2954}
2955
2956/* This is the xRowid method.  The SQLite core calls this routine to
2957** retrive the rowid for the current row of the result set.  The
2958** rowid should be written to *pRowid.
2959*/
2960static int fulltextRowid(sqlite3_vtab_cursor *pCursor, sqlite_int64 *pRowid){
2961  fulltext_cursor *c = (fulltext_cursor *) pCursor;
2962
2963  *pRowid = sqlite3_column_int64(c->pStmt, 0);
2964  return SQLITE_OK;
2965}
2966
2967/* Add all terms in [zText] to the given hash table.  If [iColumn] > 0,
2968 * we also store positions and offsets in the hash table using the given
2969 * column number. */
2970static int buildTerms(fulltext_vtab *v, fts1Hash *terms, sqlite_int64 iDocid,
2971                      const char *zText, int iColumn){
2972  sqlite3_tokenizer *pTokenizer = v->pTokenizer;
2973  sqlite3_tokenizer_cursor *pCursor;
2974  const char *pToken;
2975  int nTokenBytes;
2976  int iStartOffset, iEndOffset, iPosition;
2977  int rc;
2978
2979  rc = pTokenizer->pModule->xOpen(pTokenizer, zText, -1, &pCursor);
2980  if( rc!=SQLITE_OK ) return rc;
2981
2982  pCursor->pTokenizer = pTokenizer;
2983  while( SQLITE_OK==pTokenizer->pModule->xNext(pCursor,
2984                                               &pToken, &nTokenBytes,
2985                                               &iStartOffset, &iEndOffset,
2986                                               &iPosition) ){
2987    DocList *p;
2988
2989    /* Positions can't be negative; we use -1 as a terminator internally. */
2990    if( iPosition<0 ){
2991      pTokenizer->pModule->xClose(pCursor);
2992      return SQLITE_ERROR;
2993    }
2994
2995    p = fts1HashFind(terms, pToken, nTokenBytes);
2996    if( p==NULL ){
2997      p = docListNew(DL_DEFAULT);
2998      docListAddDocid(p, iDocid);
2999      fts1HashInsert(terms, pToken, nTokenBytes, p);
3000    }
3001    if( iColumn>=0 ){
3002      docListAddPosOffset(p, iColumn, iPosition, iStartOffset, iEndOffset);
3003    }
3004  }
3005
3006  /* TODO(shess) Check return?  Should this be able to cause errors at
3007  ** this point?  Actually, same question about sqlite3_finalize(),
3008  ** though one could argue that failure there means that the data is
3009  ** not durable.  *ponder*
3010  */
3011  pTokenizer->pModule->xClose(pCursor);
3012  return rc;
3013}
3014
3015/* Update the %_terms table to map the term [pTerm] to the given rowid. */
3016static int index_insert_term(fulltext_vtab *v, const char *pTerm, int nTerm,
3017                             DocList *d){
3018  sqlite_int64 iIndexRow;
3019  DocList doclist;
3020  int iSegment = 0, rc;
3021
3022  rc = term_select(v, pTerm, nTerm, iSegment, &iIndexRow, &doclist);
3023  if( rc==SQLITE_DONE ){
3024    docListInit(&doclist, DL_DEFAULT, 0, 0);
3025    docListUpdate(&doclist, d);
3026    /* TODO(shess) Consider length(doclist)>CHUNK_MAX? */
3027    rc = term_insert(v, NULL, pTerm, nTerm, iSegment, &doclist);
3028    goto err;
3029  }
3030  if( rc!=SQLITE_ROW ) return SQLITE_ERROR;
3031
3032  docListUpdate(&doclist, d);
3033  if( doclist.nData<=CHUNK_MAX ){
3034    rc = term_update(v, iIndexRow, &doclist);
3035    goto err;
3036  }
3037
3038  /* Doclist doesn't fit, delete what's there, and accumulate
3039  ** forward.
3040  */
3041  rc = term_delete(v, iIndexRow);
3042  if( rc!=SQLITE_OK ) goto err;
3043
3044  /* Try to insert the doclist into a higher segment bucket.  On
3045  ** failure, accumulate existing doclist with the doclist from that
3046  ** bucket, and put results in the next bucket.
3047  */
3048  iSegment++;
3049  while( (rc=term_insert(v, &iIndexRow, pTerm, nTerm, iSegment,
3050                         &doclist))!=SQLITE_OK ){
3051    sqlite_int64 iSegmentRow;
3052    DocList old;
3053    int rc2;
3054
3055    /* Retain old error in case the term_insert() error was really an
3056    ** error rather than a bounced insert.
3057    */
3058    rc2 = term_select(v, pTerm, nTerm, iSegment, &iSegmentRow, &old);
3059    if( rc2!=SQLITE_ROW ) goto err;
3060
3061    rc = term_delete(v, iSegmentRow);
3062    if( rc!=SQLITE_OK ) goto err;
3063
3064    /* Reusing lowest-number deleted row keeps the index smaller. */
3065    if( iSegmentRow<iIndexRow ) iIndexRow = iSegmentRow;
3066
3067    /* doclist contains the newer data, so accumulate it over old.
3068    ** Then steal accumulated data for doclist.
3069    */
3070    docListAccumulate(&old, &doclist);
3071    docListDestroy(&doclist);
3072    doclist = old;
3073
3074    iSegment++;
3075  }
3076
3077 err:
3078  docListDestroy(&doclist);
3079  return rc;
3080}
3081
3082/* Add doclists for all terms in [pValues] to the hash table [terms]. */
3083static int insertTerms(fulltext_vtab *v, fts1Hash *terms, sqlite_int64 iRowid,
3084                sqlite3_value **pValues){
3085  int i;
3086  for(i = 0; i < v->nColumn ; ++i){
3087    char *zText = (char*)sqlite3_value_text(pValues[i]);
3088    int rc = buildTerms(v, terms, iRowid, zText, i);
3089    if( rc!=SQLITE_OK ) return rc;
3090  }
3091  return SQLITE_OK;
3092}
3093
3094/* Add empty doclists for all terms in the given row's content to the hash
3095 * table [pTerms]. */
3096static int deleteTerms(fulltext_vtab *v, fts1Hash *pTerms, sqlite_int64 iRowid){
3097  const char **pValues;
3098  int i;
3099
3100  int rc = content_select(v, iRowid, &pValues);
3101  if( rc!=SQLITE_OK ) return rc;
3102
3103  for(i = 0 ; i < v->nColumn; ++i) {
3104    rc = buildTerms(v, pTerms, iRowid, pValues[i], -1);
3105    if( rc!=SQLITE_OK ) break;
3106  }
3107
3108  freeStringArray(v->nColumn, pValues);
3109  return SQLITE_OK;
3110}
3111
3112/* Insert a row into the %_content table; set *piRowid to be the ID of the
3113 * new row.  Fill [pTerms] with new doclists for the %_term table. */
3114static int index_insert(fulltext_vtab *v, sqlite3_value *pRequestRowid,
3115                        sqlite3_value **pValues,
3116                        sqlite_int64 *piRowid, fts1Hash *pTerms){
3117  int rc;
3118
3119  rc = content_insert(v, pRequestRowid, pValues);  /* execute an SQL INSERT */
3120  if( rc!=SQLITE_OK ) return rc;
3121  *piRowid = sqlite3_last_insert_rowid(v->db);
3122  return insertTerms(v, pTerms, *piRowid, pValues);
3123}
3124
3125/* Delete a row from the %_content table; fill [pTerms] with empty doclists
3126 * to be written to the %_term table. */
3127static int index_delete(fulltext_vtab *v, sqlite_int64 iRow, fts1Hash *pTerms){
3128  int rc = deleteTerms(v, pTerms, iRow);
3129  if( rc!=SQLITE_OK ) return rc;
3130  return content_delete(v, iRow);  /* execute an SQL DELETE */
3131}
3132
3133/* Update a row in the %_content table; fill [pTerms] with new doclists for the
3134 * %_term table. */
3135static int index_update(fulltext_vtab *v, sqlite_int64 iRow,
3136                        sqlite3_value **pValues, fts1Hash *pTerms){
3137  /* Generate an empty doclist for each term that previously appeared in this
3138   * row. */
3139  int rc = deleteTerms(v, pTerms, iRow);
3140  if( rc!=SQLITE_OK ) return rc;
3141
3142  rc = content_update(v, pValues, iRow);  /* execute an SQL UPDATE */
3143  if( rc!=SQLITE_OK ) return rc;
3144
3145  /* Now add positions for terms which appear in the updated row. */
3146  return insertTerms(v, pTerms, iRow, pValues);
3147}
3148
3149/* This function implements the xUpdate callback; it is the top-level entry
3150 * point for inserting, deleting or updating a row in a full-text table. */
3151static int fulltextUpdate(sqlite3_vtab *pVtab, int nArg, sqlite3_value **ppArg,
3152                   sqlite_int64 *pRowid){
3153  fulltext_vtab *v = (fulltext_vtab *) pVtab;
3154  fts1Hash terms;   /* maps term string -> PosList */
3155  int rc;
3156  fts1HashElem *e;
3157
3158  TRACE(("FTS1 Update %p\n", pVtab));
3159
3160  fts1HashInit(&terms, FTS1_HASH_STRING, 1);
3161
3162  if( nArg<2 ){
3163    rc = index_delete(v, sqlite3_value_int64(ppArg[0]), &terms);
3164  } else if( sqlite3_value_type(ppArg[0]) != SQLITE_NULL ){
3165    /* An update:
3166     * ppArg[0] = old rowid
3167     * ppArg[1] = new rowid
3168     * ppArg[2..2+v->nColumn-1] = values
3169     * ppArg[2+v->nColumn] = value for magic column (we ignore this)
3170     */
3171    sqlite_int64 rowid = sqlite3_value_int64(ppArg[0]);
3172    if( sqlite3_value_type(ppArg[1]) != SQLITE_INTEGER ||
3173      sqlite3_value_int64(ppArg[1]) != rowid ){
3174      rc = SQLITE_ERROR;  /* we don't allow changing the rowid */
3175    } else {
3176      assert( nArg==2+v->nColumn+1);
3177      rc = index_update(v, rowid, &ppArg[2], &terms);
3178    }
3179  } else {
3180    /* An insert:
3181     * ppArg[1] = requested rowid
3182     * ppArg[2..2+v->nColumn-1] = values
3183     * ppArg[2+v->nColumn] = value for magic column (we ignore this)
3184     */
3185    assert( nArg==2+v->nColumn+1);
3186    rc = index_insert(v, ppArg[1], &ppArg[2], pRowid, &terms);
3187  }
3188
3189  if( rc==SQLITE_OK ){
3190    /* Write updated doclists to disk. */
3191    for(e=fts1HashFirst(&terms); e; e=fts1HashNext(e)){
3192      DocList *p = fts1HashData(e);
3193      rc = index_insert_term(v, fts1HashKey(e), fts1HashKeysize(e), p);
3194      if( rc!=SQLITE_OK ) break;
3195    }
3196  }
3197
3198  /* clean up */
3199  for(e=fts1HashFirst(&terms); e; e=fts1HashNext(e)){
3200    DocList *p = fts1HashData(e);
3201    docListDelete(p);
3202  }
3203  fts1HashClear(&terms);
3204
3205  return rc;
3206}
3207
3208/*
3209** Implementation of the snippet() function for FTS1
3210*/
3211static void snippetFunc(
3212  sqlite3_context *pContext,
3213  int argc,
3214  sqlite3_value **argv
3215){
3216  fulltext_cursor *pCursor;
3217  if( argc<1 ) return;
3218  if( sqlite3_value_type(argv[0])!=SQLITE_BLOB ||
3219      sqlite3_value_bytes(argv[0])!=sizeof(pCursor) ){
3220    sqlite3_result_error(pContext, "illegal first argument to html_snippet",-1);
3221  }else{
3222    const char *zStart = "<b>";
3223    const char *zEnd = "</b>";
3224    const char *zEllipsis = "<b>...</b>";
3225    memcpy(&pCursor, sqlite3_value_blob(argv[0]), sizeof(pCursor));
3226    if( argc>=2 ){
3227      zStart = (const char*)sqlite3_value_text(argv[1]);
3228      if( argc>=3 ){
3229        zEnd = (const char*)sqlite3_value_text(argv[2]);
3230        if( argc>=4 ){
3231          zEllipsis = (const char*)sqlite3_value_text(argv[3]);
3232        }
3233      }
3234    }
3235    snippetAllOffsets(pCursor);
3236    snippetText(pCursor, zStart, zEnd, zEllipsis);
3237    sqlite3_result_text(pContext, pCursor->snippet.zSnippet,
3238                        pCursor->snippet.nSnippet, SQLITE_STATIC);
3239  }
3240}
3241
3242/*
3243** Implementation of the offsets() function for FTS1
3244*/
3245static void snippetOffsetsFunc(
3246  sqlite3_context *pContext,
3247  int argc,
3248  sqlite3_value **argv
3249){
3250  fulltext_cursor *pCursor;
3251  if( argc<1 ) return;
3252  if( sqlite3_value_type(argv[0])!=SQLITE_BLOB ||
3253      sqlite3_value_bytes(argv[0])!=sizeof(pCursor) ){
3254    sqlite3_result_error(pContext, "illegal first argument to offsets",-1);
3255  }else{
3256    memcpy(&pCursor, sqlite3_value_blob(argv[0]), sizeof(pCursor));
3257    snippetAllOffsets(pCursor);
3258    snippetOffsetText(&pCursor->snippet);
3259    sqlite3_result_text(pContext,
3260                        pCursor->snippet.zOffset, pCursor->snippet.nOffset,
3261                        SQLITE_STATIC);
3262  }
3263}
3264
3265/*
3266** This routine implements the xFindFunction method for the FTS1
3267** virtual table.
3268*/
3269static int fulltextFindFunction(
3270  sqlite3_vtab *pVtab,
3271  int nArg,
3272  const char *zName,
3273  void (**pxFunc)(sqlite3_context*,int,sqlite3_value**),
3274  void **ppArg
3275){
3276  if( strcmp(zName,"snippet")==0 ){
3277    *pxFunc = snippetFunc;
3278    return 1;
3279  }else if( strcmp(zName,"offsets")==0 ){
3280    *pxFunc = snippetOffsetsFunc;
3281    return 1;
3282  }
3283  return 0;
3284}
3285
3286/*
3287** Rename an fts1 table.
3288*/
3289static int fulltextRename(
3290  sqlite3_vtab *pVtab,
3291  const char *zName
3292){
3293  fulltext_vtab *p = (fulltext_vtab *)pVtab;
3294  int rc = SQLITE_NOMEM;
3295  char *zSql = sqlite3_mprintf(
3296    "ALTER TABLE %Q.'%q_content'  RENAME TO '%q_content';"
3297    "ALTER TABLE %Q.'%q_term' RENAME TO '%q_term';"
3298    , p->zDb, p->zName, zName
3299    , p->zDb, p->zName, zName
3300  );
3301  if( zSql ){
3302    rc = sqlite3_exec(p->db, zSql, 0, 0, 0);
3303    sqlite3_free(zSql);
3304  }
3305  return rc;
3306}
3307
3308static const sqlite3_module fulltextModule = {
3309  /* iVersion      */ 0,
3310  /* xCreate       */ fulltextCreate,
3311  /* xConnect      */ fulltextConnect,
3312  /* xBestIndex    */ fulltextBestIndex,
3313  /* xDisconnect   */ fulltextDisconnect,
3314  /* xDestroy      */ fulltextDestroy,
3315  /* xOpen         */ fulltextOpen,
3316  /* xClose        */ fulltextClose,
3317  /* xFilter       */ fulltextFilter,
3318  /* xNext         */ fulltextNext,
3319  /* xEof          */ fulltextEof,
3320  /* xColumn       */ fulltextColumn,
3321  /* xRowid        */ fulltextRowid,
3322  /* xUpdate       */ fulltextUpdate,
3323  /* xBegin        */ 0,
3324  /* xSync         */ 0,
3325  /* xCommit       */ 0,
3326  /* xRollback     */ 0,
3327  /* xFindFunction */ fulltextFindFunction,
3328  /* xRename       */ fulltextRename,
3329};
3330
3331int sqlite3Fts1Init(sqlite3 *db){
3332  sqlite3_overload_function(db, "snippet", -1);
3333  sqlite3_overload_function(db, "offsets", -1);
3334  return sqlite3_create_module(db, "fts1", &fulltextModule, 0);
3335}
3336
3337#if !SQLITE_CORE
3338int sqlite3_extension_init(sqlite3 *db, char **pzErrMsg,
3339                           const sqlite3_api_routines *pApi){
3340  SQLITE_EXTENSION_INIT2(pApi)
3341  return sqlite3Fts1Init(db);
3342}
3343#endif
3344
3345#endif /* !defined(SQLITE_CORE) || defined(SQLITE_ENABLE_FTS1) */
3346