1
/* The author disclaims copyright to this source code.
3
* This is an SQLite module implementing full-text search.
7
#if !defined(__APPLE__)
18
#include "tokenizer.h"
20
#include "sqlite3ext.h"
21
SQLITE_EXTENSION_INIT1
23
/* utility functions */
25
/* We encode variable-length integers in little-endian order using seven bits
26
* per byte as follows:
29
** A = 0xxxxxxx 7 bits of data and one flag bit
30
** B = 1xxxxxxx 7 bits of data and one flag bit
38
/* We may need up to VARINT_MAX bytes to store an encoded 64-bit integer. */
41
/* Write a 64-bit variable-length integer to memory starting at p[0].
42
* The length of data written will be between 1 and VARINT_MAX bytes.
43
* The number of bytes written is returned. */
44
static int putVarint(char *p, sqlite_int64 v){
45
unsigned char *q = (unsigned char *) p;
48
*q++ = (unsigned char) ((vu & 0x7f) | 0x80);
51
q[-1] &= 0x7f; /* turn off high bit in final byte */
52
assert( q - (unsigned char *)p <= VARINT_MAX );
53
return (int) (q - (unsigned char *)p);
56
/* Read a 64-bit variable-length integer from memory starting at p[0].
57
* Return the number of bytes read, or 0 on error.
58
* The value is stored in *v. */
59
static int getVarint(const char *p, sqlite_int64 *v){
60
const unsigned char *q = (const unsigned char *) p;
61
sqlite_uint64 x = 0, y = 1;
62
while( (*q & 0x80) == 0x80 ){
63
x += y * (*q++ & 0x7f);
65
if( q - (unsigned char *)p >= VARINT_MAX ){ /* bad data */
71
*v = (sqlite_int64) x;
72
return (int) (q - (unsigned char *)p);
75
static int getVarint32(const char *p, int *pi){
77
int ret = getVarint(p, &i);
83
/*** Document lists ***
85
* A document list holds a sorted list of varint-encoded document IDs.
87
* A doclist with type DL_POSITIONS_OFFSETS is stored like this:
92
* varint position; (delta from previous position plus 1, or 0 for end)
93
* varint startOffset; (delta from previous startOffset)
94
* varint endOffset; (delta from startOffset)
98
* Here, array { X } means zero or more occurrences of X, adjacent in memory.
100
* A doclist with type DL_POSITIONS is like the above, but holds only docids
101
* and positions without offset information.
103
* A doclist with type DL_DOCIDS is like the above, but holds only docids
104
* without positions or offset information.
106
* On disk, every document list has positions and offsets, so we don't bother
107
* to serialize a doclist's type.
109
* We don't yet delta-encode document IDs; doing so will probably be a
112
* NOTE(shess) I've thought of a slightly (1%) better offset encoding.
113
* After the first offset, estimate the next offset by using the
114
* current token position and the previous token position and offset,
115
* offset to handle some variance. So the estimate would be
116
* (iPosition*w->iStartOffset/w->iPosition-64), which is delta-encoded
117
* as normal. Offsets more than 64 chars from the estimate are
118
* encoded as the delta to the previous start offset + 128. An
119
* additional tiny increment can be gained by using the end offset of
120
* the previous token to make the estimate a tiny bit more precise.
123
typedef enum DocListType {
124
DL_DOCIDS, /* docids only */
125
DL_POSITIONS, /* docids + positions */
126
DL_POSITIONS_OFFSETS /* docids + positions + offsets */
129
typedef struct DocList {
133
int iLastPos; /* the last position written */
134
int iLastOffset; /* the last start offset written */
137
/* Initialize a new DocList to hold the given data. */
138
static void docListInit(DocList *d, DocListType iType,
139
const char *pData, int nData){
142
d->pData = malloc(nData);
143
memcpy(d->pData, pData, nData);
152
/* Create a new dynamically-allocated DocList. */
153
static DocList *docListNew(DocListType iType){
154
DocList *d = (DocList *) malloc(sizeof(DocList));
155
docListInit(d, iType, 0, 0);
159
static void docListDestroy(DocList *d){
162
memset(d, 0x55, sizeof(*d));
166
static void docListDelete(DocList *d){
171
static char *docListEnd(DocList *d){
172
return d->pData + d->nData;
175
/* Append a varint to a DocList's data. */
176
static void appendVarint(DocList *d, sqlite_int64 i){
178
int n = putVarint(c, i);
179
d->pData = realloc(d->pData, d->nData + n);
180
memcpy(d->pData + d->nData, c, n);
184
static void docListAddDocid(DocList *d, sqlite_int64 iDocid){
185
appendVarint(d, iDocid);
189
/* Add a position to the last position list in a doclist. */
190
static void docListAddPos(DocList *d, int iPos){
191
assert( d->iType>=DL_POSITIONS );
192
appendVarint(d, iPos-d->iLastPos+1);
196
static void docListAddPosOffset(DocList *d, int iPos,
197
int iStartOffset, int iEndOffset){
198
assert( d->iType==DL_POSITIONS_OFFSETS );
199
docListAddPos(d, iPos);
200
appendVarint(d, iStartOffset-d->iLastOffset);
201
d->iLastOffset = iStartOffset;
202
appendVarint(d, iEndOffset-iStartOffset);
205
/* Terminate the last position list in the given doclist. */
206
static void docListAddEndPos(DocList *d){
210
typedef struct DocListReader {
213
int iLastPos; /* the last position read */
216
static void readerInit(DocListReader *r, DocList *pDoclist){
217
r->pDoclist = pDoclist;
218
if( pDoclist!=NULL ){
219
r->p = pDoclist->pData;
224
static int readerAtEnd(DocListReader *pReader){
225
return pReader->p >= docListEnd(pReader->pDoclist);
228
/* Peek at the next docid without advancing the read pointer. */
229
static sqlite_int64 peekDocid(DocListReader *pReader){
231
assert( !readerAtEnd(pReader) );
232
getVarint(pReader->p, &ret);
236
/* Read the next docid. */
237
static sqlite_int64 readDocid(DocListReader *pReader){
239
assert( !readerAtEnd(pReader) );
240
pReader->p += getVarint(pReader->p, &ret);
241
pReader->iLastPos = 0;
245
/* Read the next position from a position list.
246
* Returns the position, or -1 at the end of the list. */
247
static int readPosition(DocListReader *pReader){
249
int iType = pReader->pDoclist->iType;
250
assert( iType>=DL_POSITIONS );
251
assert( !readerAtEnd(pReader) );
253
pReader->p += getVarint32(pReader->p, &i);
255
pReader->iLastPos = -1;
258
pReader->iLastPos += ((int) i)-1;
259
if( iType>=DL_POSITIONS_OFFSETS ){
260
/* Skip over offsets, ignoring them for now. */
262
pReader->p += getVarint32(pReader->p, &iStart);
263
pReader->p += getVarint32(pReader->p, &iEnd);
265
return pReader->iLastPos;
268
/* Skip past the end of a position list. */
269
static void skipPositionList(DocListReader *pReader){
270
while( readPosition(pReader)!=-1 )
274
/* Skip over a docid, including its position list if the doclist has
276
static void skipDocument(DocListReader *pReader){
278
if( pReader->pDoclist->iType >= DL_POSITIONS ){
279
skipPositionList(pReader);
283
static sqlite_int64 firstDocid(DocList *d){
286
return readDocid(&r);
289
/* Doclist multi-tool. Pass pUpdate==NULL to delete the indicated docid;
290
* otherwise pUpdate, which must contain only the single docid [iDocid], is
291
* inserted (if not present) or updated (if already present). */
292
static int docListUpdate(DocList *d, sqlite_int64 iDocid, DocList *pUpdate){
294
DocListReader reader;
298
assert( d->iType==pUpdate->iType);
299
assert( iDocid==firstDocid(pUpdate) );
302
readerInit(&reader, d);
303
while( !readerAtEnd(&reader) && peekDocid(&reader)<iDocid ){
304
skipDocument(&reader);
308
/* Delete if there is a matching element. */
309
if( !readerAtEnd(&reader) && iDocid==peekDocid(&reader) ){
310
skipDocument(&reader);
311
memmove(p, reader.p, docListEnd(d) - reader.p);
312
d->nData -= (reader.p - p);
316
/* Insert if indicated. */
318
int iDoclist = p-d->pData;
319
docListAddEndPos(pUpdate);
321
d->pData = realloc(d->pData, d->nData+pUpdate->nData);
322
p = d->pData + iDoclist;
324
memmove(p+pUpdate->nData, p, docListEnd(d) - p);
325
memcpy(p, pUpdate->pData, pUpdate->nData);
326
d->nData += pUpdate->nData;
333
/* Split the second half of doclist d into a separate doclist d2. Returns 1
334
* if successful, or 0 if d contains a single document and hence can't be
336
static int docListSplit(DocList *d, DocList *d2){
337
const char *pSplitPoint = d->pData + d->nData / 2;
338
DocListReader reader;
340
readerInit(&reader, d);
341
while( reader.p<pSplitPoint ){
342
skipDocument(&reader);
344
if( readerAtEnd(&reader) ) return 0;
345
docListInit(d2, d->iType, reader.p, docListEnd(d) - reader.p);
346
d->nData = reader.p - d->pData;
347
d->pData = realloc(d->pData, d->nData);
351
/* A DocListMerge computes the AND of an in-memory DocList [in] and a chunked
352
* on-disk doclist, resulting in another in-memory DocList [out]. [in]
353
* and [out] may or may not store position information according to the
354
* caller's wishes. The on-disk doclist always comes with positions.
356
* The caller must read each chunk of the on-disk doclist in succession and
357
* pass it to mergeBlock().
359
* If [in] has positions, then the merge output contains only documents with
360
* matching positions in the two input doclists. If [in] does not have
361
* positions, then the merge output contains all documents common to the two
364
* If [in] is NULL, then the on-disk doclist is copied to [out] directly.
366
* A merge is performed using an integer [iOffset] provided by the caller.
367
* [iOffset] is subtracted from each position in the on-disk doclist for the
368
* purpose of position comparison; this is helpful in implementing phrase
371
* A DocListMerge is not yet able to propagate offsets through query
372
* processing; we should add that capability soon.
374
typedef struct DocListMerge {
380
static void mergeInit(DocListMerge *m,
381
DocList *pIn, int iOffset, DocList *pOut){
382
readerInit(&m->in, pIn);
384
m->iOffset = iOffset;
386
/* can't handle offsets yet */
387
assert( pIn==NULL || pIn->iType <= DL_POSITIONS );
388
assert( pOut->iType <= DL_POSITIONS );
391
/* A helper function for mergeBlock(), below. Merge the position lists
392
* pointed to by m->in and pBlockReader.
393
* If the merge matches, write [iDocid] to m->pOut; if m->pOut
394
* has positions then write all matching positions as well. */
395
static void mergePosList(DocListMerge *m, sqlite_int64 iDocid,
396
DocListReader *pBlockReader){
397
int block_pos = readPosition(pBlockReader);
398
int in_pos = readPosition(&m->in);
400
while( block_pos!=-1 || in_pos!=-1 ){
401
if( block_pos-m->iOffset==in_pos ){
403
docListAddDocid(m->pOut, iDocid);
406
if( m->pOut->iType >= DL_POSITIONS ){
407
docListAddPos(m->pOut, in_pos);
409
block_pos = readPosition(pBlockReader);
410
in_pos = readPosition(&m->in);
411
} else if( in_pos==-1 || (block_pos!=-1 && block_pos-m->iOffset<in_pos) ){
412
block_pos = readPosition(pBlockReader);
414
in_pos = readPosition(&m->in);
417
if( m->pOut->iType >= DL_POSITIONS && match ){
418
docListAddEndPos(m->pOut);
422
/* Merge one block of an on-disk doclist into a DocListMerge. */
423
static void mergeBlock(DocListMerge *m, DocList *pBlock){
424
DocListReader blockReader;
425
assert( pBlock->iType >= DL_POSITIONS );
426
readerInit(&blockReader, pBlock);
427
while( !readerAtEnd(&blockReader) ){
428
sqlite_int64 iDocid = readDocid(&blockReader);
429
if( m->in.pDoclist!=NULL ){
431
if( readerAtEnd(&m->in) ) return; /* nothing more to merge */
432
if( peekDocid(&m->in)>=iDocid ) break;
433
skipDocument(&m->in);
435
if( peekDocid(&m->in)>iDocid ){ /* [pIn] has no match with iDocid */
436
skipPositionList(&blockReader); /* skip this docid in the block */
441
/* We have a document match. */
442
if( m->in.pDoclist==NULL || m->in.pDoclist->iType < DL_POSITIONS ){
443
/* We don't need to do a poslist merge. */
444
docListAddDocid(m->pOut, iDocid);
445
if( m->pOut->iType >= DL_POSITIONS ){
446
/* Copy all positions to the output doclist. */
448
int pos = readPosition(&blockReader);
450
docListAddPos(m->pOut, pos);
452
docListAddEndPos(m->pOut);
453
} else skipPositionList(&blockReader);
456
mergePosList(m, iDocid, &blockReader);
460
static char *string_dup_n(const char *s, int n){
461
char *str = malloc(n + 1);
467
/* Duplicate a string; the caller must free() the returned string.
468
* (We don't use strdup() since it's not part of the standard C library and
469
* may not be available everywhere.) */
470
static char *string_dup(const char *s){
471
return string_dup_n(s, strlen(s));
474
/* Format a string, replacing each occurrence of the % character with
475
* zName. This may be more convenient than sqlite_mprintf()
476
* when one string is used repeatedly in a format string.
477
* The caller must free() the returned string. */
478
static char *string_format(const char *zFormat, const char *zName){
481
size_t nName = strlen(zName);
485
/* first compute length needed */
486
for(p = zFormat ; *p ; ++p){
487
len += (*p=='%' ? nName : 1);
489
len += 1; /* for null terminator */
491
r = result = malloc(len);
492
for(p = zFormat; *p; ++p){
494
memcpy(r, zName, nName);
501
assert( r == result + len );
505
static int sql_exec(sqlite3 *db, const char *zName, const char *zFormat){
506
char *zCommand = string_format(zFormat, zName);
507
int rc = sqlite3_exec(db, zCommand, NULL, 0, NULL);
512
static int sql_prepare(sqlite3 *db, const char *zName, sqlite3_stmt **ppStmt,
513
const char *zFormat){
514
char *zCommand = string_format(zFormat, zName);
515
int rc = sqlite3_prepare(db, zCommand, -1, ppStmt, NULL);
520
/* end utility functions */
522
#define QUERY_GENERIC 0
523
#define QUERY_FULLTEXT 1
525
#define CHUNK_MAX 1024
527
typedef enum fulltext_statement {
533
TERM_CHUNK_SELECT_STMT,
538
MAX_STMT /* Always at end! */
539
} fulltext_statement;
541
/* These must exactly match the enum above. */
542
/* TODO(adam): Is there some risk that a statement (in particular,
543
** pTermSelectStmt) will be used in two cursors at once, e.g. if a
544
** query joins a virtual table to itself? If so perhaps we should
545
** move some of these to the cursor object.
547
static const char *fulltext_zStatement[MAX_STMT] = {
548
/* CONTENT_INSERT */ "insert into %_content (rowid, content) values (?, ?)",
549
/* CONTENT_SELECT */ "select content from %_content where rowid = ?",
550
/* CONTENT_DELETE */ "delete from %_content where rowid = ?",
553
"select rowid, doclist from %_term where term = ? and first = ?",
554
/* TERM_CHUNK_SELECT */
555
"select max(first) from %_term where term = ? and first <= ?",
557
"insert into %_term (term, first, doclist) values (?, ?, ?)",
558
/* TERM_UPDATE */ "update %_term set doclist = ? where rowid = ?",
559
/* TERM_DELETE */ "delete from %_term where rowid = ?",
562
typedef struct fulltext_vtab {
565
const char *zName; /* virtual table name */
566
sqlite3_tokenizer *pTokenizer; /* tokenizer for inserts and queries */
568
/* Precompiled statements which we keep as long as the table is
571
sqlite3_stmt *pFulltextStatements[MAX_STMT];
574
typedef struct fulltext_cursor {
575
sqlite3_vtab_cursor base;
576
int iCursorType; /* QUERY_GENERIC or QUERY_FULLTEXT */
582
/* The following is used only when iCursorType == QUERY_FULLTEXT. */
583
DocListReader result;
586
static struct fulltext_vtab *cursor_vtab(fulltext_cursor *c){
587
return (fulltext_vtab *) c->base.pVtab;
590
static sqlite3_module fulltextModule; /* forward declaration */
592
/* Puts a freshly-prepared statement determined by iStmt in *ppStmt.
593
** If the indicated statement has never been prepared, it is prepared
594
** and cached, otherwise the cached version is reset.
596
static int sql_get_statement(fulltext_vtab *v, fulltext_statement iStmt,
597
sqlite3_stmt **ppStmt){
598
assert( iStmt<MAX_STMT );
599
if( v->pFulltextStatements[iStmt]==NULL ){
600
int rc = sql_prepare(v->db, v->zName, &v->pFulltextStatements[iStmt],
601
fulltext_zStatement[iStmt]);
602
if( rc!=SQLITE_OK ) return rc;
604
int rc = sqlite3_reset(v->pFulltextStatements[iStmt]);
605
if( rc!=SQLITE_OK ) return rc;
608
*ppStmt = v->pFulltextStatements[iStmt];
612
/* Step the indicated statement, handling errors SQLITE_BUSY (by
613
** retrying) and SQLITE_SCHEMA (by re-preparing and transferring
614
** bindings to the new statement).
615
** TODO(adam): We should extend this function so that it can work with
616
** statements declared locally, not only globally cached statements.
618
static int sql_step_statement(fulltext_vtab *v, fulltext_statement iStmt,
619
sqlite3_stmt **ppStmt){
621
sqlite3_stmt *s = *ppStmt;
622
assert( iStmt<MAX_STMT );
623
assert( s==v->pFulltextStatements[iStmt] );
625
while( (rc=sqlite3_step(s))!=SQLITE_DONE && rc!=SQLITE_ROW ){
626
sqlite3_stmt *pNewStmt;
628
if( rc==SQLITE_BUSY ) continue;
629
if( rc!=SQLITE_ERROR ) return rc;
631
rc = sqlite3_reset(s);
632
if( rc!=SQLITE_SCHEMA ) return SQLITE_ERROR;
634
v->pFulltextStatements[iStmt] = NULL; /* Still in s */
635
rc = sql_get_statement(v, iStmt, &pNewStmt);
636
if( rc!=SQLITE_OK ) goto err;
639
rc = sqlite3_transfer_bindings(s, pNewStmt);
640
if( rc!=SQLITE_OK ) goto err;
642
rc = sqlite3_finalize(s);
643
if( rc!=SQLITE_OK ) return rc;
653
/* Like sql_step_statement(), but convert SQLITE_DONE to SQLITE_OK.
654
** Useful for statements like UPDATE, where we expect no results.
656
static int sql_single_step_statement(fulltext_vtab *v,
657
fulltext_statement iStmt,
658
sqlite3_stmt **ppStmt){
659
int rc = sql_step_statement(v, iStmt, ppStmt);
660
return (rc==SQLITE_DONE) ? SQLITE_OK : rc;
663
/* insert into %_content (rowid, content) values ([rowid], [zContent]) */
664
static int content_insert(fulltext_vtab *v, sqlite3_value *rowid,
665
const char *zContent, int nContent){
667
int rc = sql_get_statement(v, CONTENT_INSERT_STMT, &s);
668
if( rc!=SQLITE_OK ) return rc;
670
rc = sqlite3_bind_value(s, 1, rowid);
671
if( rc!=SQLITE_OK ) return rc;
673
rc = sqlite3_bind_text(s, 2, zContent, nContent, SQLITE_STATIC);
674
if( rc!=SQLITE_OK ) return rc;
676
return sql_single_step_statement(v, CONTENT_INSERT_STMT, &s);
679
/* select content from %_content where rowid = [iRow]
680
* The caller must delete the returned string. */
681
static int content_select(fulltext_vtab *v, sqlite_int64 iRow,
684
int rc = sql_get_statement(v, CONTENT_SELECT_STMT, &s);
685
if( rc!=SQLITE_OK ) return rc;
687
rc = sqlite3_bind_int64(s, 1, iRow);
688
if( rc!=SQLITE_OK ) return rc;
690
rc = sql_step_statement(v, CONTENT_SELECT_STMT, &s);
691
if( rc!=SQLITE_ROW ) return rc;
693
*pzContent = string_dup((const char *)sqlite3_column_text(s, 0));
695
/* We expect only one row. We must execute another sqlite3_step()
696
* to complete the iteration; otherwise the table will remain locked. */
697
rc = sqlite3_step(s);
698
if( rc==SQLITE_DONE ) return SQLITE_OK;
704
/* delete from %_content where rowid = [iRow ] */
705
static int content_delete(fulltext_vtab *v, sqlite_int64 iRow){
707
int rc = sql_get_statement(v, CONTENT_DELETE_STMT, &s);
708
if( rc!=SQLITE_OK ) return rc;
710
rc = sqlite3_bind_int64(s, 1, iRow);
711
if( rc!=SQLITE_OK ) return rc;
713
return sql_single_step_statement(v, CONTENT_DELETE_STMT, &s);
716
/* select rowid, doclist from %_term where term = [zTerm] and first = [iFirst]
717
* If found, returns SQLITE_OK; the caller must free the returned doclist.
718
* If no rows found, returns SQLITE_ERROR. */
719
static int term_select(fulltext_vtab *v, const char *zTerm, int nTerm,
724
int rc = sql_get_statement(v, TERM_SELECT_STMT, &s);
725
if( rc!=SQLITE_OK ) return rc;
727
rc = sqlite3_bind_text(s, 1, zTerm, nTerm, SQLITE_TRANSIENT);
728
if( rc!=SQLITE_OK ) return rc;
730
rc = sqlite3_bind_int64(s, 2, iFirst);
731
if( rc!=SQLITE_OK ) return rc;
733
rc = sql_step_statement(v, TERM_SELECT_STMT, &s);
734
if( rc!=SQLITE_ROW ) return rc==SQLITE_DONE ? SQLITE_ERROR : rc;
736
*rowid = sqlite3_column_int64(s, 0);
737
docListInit(out, DL_POSITIONS_OFFSETS,
738
sqlite3_column_blob(s, 1), sqlite3_column_bytes(s, 1));
740
/* We expect only one row. We must execute another sqlite3_step()
741
* to complete the iteration; otherwise the table will remain locked. */
742
rc = sqlite3_step(s);
743
return rc==SQLITE_DONE ? SQLITE_OK : rc;
746
/* select max(first) from %_term where term = [zTerm] and first <= [iFirst]
747
* If found, returns SQLITE_ROW and result in *piResult; if the query returns
748
* NULL (meaning no row found) returns SQLITE_DONE.
750
static int term_chunk_select(fulltext_vtab *v, const char *zTerm, int nTerm,
751
sqlite_int64 iFirst, sqlite_int64 *piResult){
753
int rc = sql_get_statement(v, TERM_CHUNK_SELECT_STMT, &s);
754
if( rc!=SQLITE_OK ) return rc;
756
rc = sqlite3_bind_text(s, 1, zTerm, nTerm, SQLITE_STATIC);
757
if( rc!=SQLITE_OK ) return rc;
759
rc = sqlite3_bind_int64(s, 2, iFirst);
760
if( rc!=SQLITE_OK ) return rc;
762
rc = sql_step_statement(v, TERM_CHUNK_SELECT_STMT, &s);
763
if( rc!=SQLITE_ROW ) return rc==SQLITE_DONE ? SQLITE_ERROR : rc;
765
switch( sqlite3_column_type(s, 0) ){
770
*piResult = sqlite3_column_int64(s, 0);
775
/* We expect only one row. We must execute another sqlite3_step()
776
* to complete the iteration; otherwise the table will remain locked. */
777
if( sqlite3_step(s) != SQLITE_DONE ) return SQLITE_ERROR;
781
/* insert into %_term (term, first, doclist)
782
values ([zTerm], [iFirst], [doclist]) */
783
static int term_insert(fulltext_vtab *v, const char *zTerm, int nTerm,
784
sqlite_int64 iFirst, DocList *doclist){
786
int rc = sql_get_statement(v, TERM_INSERT_STMT, &s);
787
if( rc!=SQLITE_OK ) return rc;
789
rc = sqlite3_bind_text(s, 1, zTerm, nTerm, SQLITE_STATIC);
790
if( rc!=SQLITE_OK ) return rc;
792
rc = sqlite3_bind_int64(s, 2, iFirst);
793
if( rc!=SQLITE_OK ) return rc;
795
rc = sqlite3_bind_blob(s, 3, doclist->pData, doclist->nData, SQLITE_STATIC);
796
if( rc!=SQLITE_OK ) return rc;
798
return sql_single_step_statement(v, TERM_INSERT_STMT, &s);
801
/* update %_term set doclist = [doclist] where rowid = [rowid] */
802
static int term_update(fulltext_vtab *v, sqlite_int64 rowid,
805
int rc = sql_get_statement(v, TERM_UPDATE_STMT, &s);
806
if( rc!=SQLITE_OK ) return rc;
808
rc = sqlite3_bind_blob(s, 1, doclist->pData, doclist->nData,
810
if( rc!=SQLITE_OK ) return rc;
812
rc = sqlite3_bind_int64(s, 2, rowid);
813
if( rc!=SQLITE_OK ) return rc;
815
return sql_single_step_statement(v, TERM_UPDATE_STMT, &s);
818
static int term_delete(fulltext_vtab *v, sqlite_int64 rowid){
820
int rc = sql_get_statement(v, TERM_DELETE_STMT, &s);
821
if( rc!=SQLITE_OK ) return rc;
823
rc = sqlite3_bind_int64(s, 1, rowid);
824
if( rc!=SQLITE_OK ) return rc;
826
return sql_single_step_statement(v, TERM_DELETE_STMT, &s);
829
static void fulltext_vtab_destroy(fulltext_vtab *v){
832
for( iStmt=0; iStmt<MAX_STMT; iStmt++ ){
833
if( v->pFulltextStatements[iStmt]!=NULL ){
834
sqlite3_finalize(v->pFulltextStatements[iStmt]);
835
v->pFulltextStatements[iStmt] = NULL;
839
if( v->pTokenizer!=NULL ){
840
v->pTokenizer->pModule->xDestroy(v->pTokenizer);
841
v->pTokenizer = NULL;
844
free((void *) v->zName);
848
/* Current interface:
849
** argv[0] - module name
850
** argv[1] - database name
851
** argv[2] - table name
852
** argv[3] - tokenizer name (optional, a sensible default is provided)
853
** argv[4..] - passed to tokenizer (optional based on tokenizer)
855
static int fulltextConnect(sqlite3 *db, void *pAux, int argc, char **argv,
856
sqlite3_vtab **ppVTab){
859
sqlite3_tokenizer_module *m = NULL;
862
v = (fulltext_vtab *) malloc(sizeof(fulltext_vtab));
863
/* sqlite will initialize v->base */
865
v->zName = string_dup(argv[2]);
866
v->pTokenizer = NULL;
869
get_simple_tokenizer_module(&m);
871
/* TODO(shess) For now, add new tokenizers as else if clauses. */
872
if( !strcmp(argv[3], "simple") ){
873
get_simple_tokenizer_module(&m);
875
assert( "unrecognized tokenizer"==NULL );
879
/* TODO(shess) Since tokenization impacts the index, the parameters
880
** to the tokenizer need to be identical when a persistent virtual
881
** table is re-created. One solution would be a meta-table to track
882
** such information in the database. Then we could verify that the
883
** information is identical on subsequent creates.
885
/* TODO(shess) Why isn't argv already (const char **)? */
886
rc = m->xCreate(argc-3, (const char **) (argv+3), &v->pTokenizer);
887
if( rc!=SQLITE_OK ) return rc;
888
v->pTokenizer->pModule = m;
890
/* TODO: verify the existence of backing tables foo_content, foo_term */
892
rc = sqlite3_declare_vtab(db, "create table x(content text)");
893
if( rc!=SQLITE_OK ) return rc;
895
memset(v->pFulltextStatements, 0, sizeof(v->pFulltextStatements));
901
static int fulltextCreate(sqlite3 *db, void *pAux, int argc, char **argv,
902
sqlite3_vtab **ppVTab){
906
/* The %_content table holds the text of each full-text item, with
907
** the rowid used as the docid.
909
** The %_term table maps each term to a document list blob
910
** containing elements sorted by ascending docid, each element
913
** docid varint-encoded
914
** token count varint-encoded
915
** "count" token elements (poslist):
916
** position varint-encoded as delta from previous position
917
** start offset varint-encoded as delta from previous start offset
918
** end offset varint-encoded as delta from start offset
920
** Additionally, doclist blobs can be chunked into multiple rows,
921
** using "first" to order the blobs. "first" is simply the first
922
** docid in the blob.
925
** NOTE(shess) That last sentence is incorrect in the face of
926
** deletion, which can leave a doclist that doesn't contain the
927
** first from that row. I _believe_ this does not matter to the
928
** operation of the system, but it might be reasonable to update
929
** appropriately in case this assumption becomes more important.
931
rc = sql_exec(db, argv[2],
932
"create table %_content(content text);"
933
"create table %_term(term text, first integer, doclist blob);"
934
"create index %_index on %_term(term, first)");
935
if( rc!=SQLITE_OK ) return rc;
937
return fulltextConnect(db, pAux, argc, argv, ppVTab);
940
/* Decide how to handle an SQL query.
941
* At the moment, MATCH queries can include implicit boolean ANDs; we
942
* haven't implemented phrase searches or OR yet. */
943
static int fulltextBestIndex(sqlite3_vtab *pVTab, sqlite3_index_info *pInfo){
946
for(i=0; i<pInfo->nConstraint; ++i){
947
const struct sqlite3_index_constraint *pConstraint;
948
pConstraint = &pInfo->aConstraint[i];
949
if( pConstraint->iColumn==0 &&
950
pConstraint->op==SQLITE_INDEX_CONSTRAINT_MATCH &&
951
pConstraint->usable ){ /* a full-text search */
952
pInfo->aConstraintUsage[i].argvIndex = 1;
953
pInfo->aConstraintUsage[i].omit = 1;
954
pInfo->idxNum = QUERY_FULLTEXT;
955
pInfo->estimatedCost = 1.0; /* an arbitrary value for now */
959
pInfo->idxNum = QUERY_GENERIC;
963
static int fulltextDisconnect(sqlite3_vtab *pVTab){
964
fulltext_vtab_destroy((fulltext_vtab *)pVTab);
968
static int fulltextDestroy(sqlite3_vtab *pVTab){
969
fulltext_vtab *v = (fulltext_vtab *)pVTab;
971
int rc = sql_exec(v->db, v->zName,
972
"drop table %_content; drop table %_term");
973
if( rc!=SQLITE_OK ) return rc;
975
fulltext_vtab_destroy((fulltext_vtab *)pVTab);
979
static int fulltextOpen(sqlite3_vtab *pVTab, sqlite3_vtab_cursor **ppCursor){
982
c = (fulltext_cursor *) calloc(sizeof(fulltext_cursor), 1);
983
/* sqlite will initialize c->base */
984
*ppCursor = &c->base;
989
static int fulltextClose(sqlite3_vtab_cursor *pCursor){
990
fulltext_cursor *c = (fulltext_cursor *) pCursor;
991
sqlite3_finalize(c->pStmt);
992
if( c->result.pDoclist!=NULL ){
993
docListDelete(c->result.pDoclist);
999
static int fulltextNext(sqlite3_vtab_cursor *pCursor){
1000
fulltext_cursor *c = (fulltext_cursor *) pCursor;
1001
sqlite_int64 iDocid;
1004
switch( c->iCursorType ){
1006
/* TODO(shess) Handle SQLITE_SCHEMA AND SQLITE_BUSY. */
1007
rc = sqlite3_step(c->pStmt);
1019
case QUERY_FULLTEXT:
1020
rc = sqlite3_reset(c->pStmt);
1021
if( rc!=SQLITE_OK ) return rc;
1023
if( readerAtEnd(&c->result)){
1027
iDocid = readDocid(&c->result);
1028
rc = sqlite3_bind_int64(c->pStmt, 1, iDocid);
1029
if( rc!=SQLITE_OK ) return rc;
1030
/* TODO(shess) Handle SQLITE_SCHEMA AND SQLITE_BUSY. */
1031
rc = sqlite3_step(c->pStmt);
1032
if( rc==SQLITE_ROW ){ /* the case we expect */
1036
/* an error occurred; abort */
1037
return rc==SQLITE_DONE ? SQLITE_ERROR : rc;
1040
return SQLITE_ERROR; /* not reached */
1044
static int term_select_doclist(fulltext_vtab *v, const char *pTerm, int nTerm,
1045
sqlite3_stmt **ppStmt){
1048
rc = sqlite3_reset(*ppStmt);
1050
rc = sql_prepare(v->db, v->zName, ppStmt,
1051
"select doclist from %_term where term = ? order by first");
1053
if( rc!=SQLITE_OK ) return rc;
1055
rc = sqlite3_bind_text(*ppStmt, 1, pTerm, nTerm, SQLITE_TRANSIENT);
1056
if( rc!=SQLITE_OK ) return rc;
1058
return sqlite3_step(*ppStmt); /* TODO(adamd): handle schema error */
1061
/* Read the posting list for [zTerm]; AND it with the doclist [in] to
1062
* produce the doclist [out], using the given offset [iOffset] for phrase
1064
* (*pSelect) is used to hold an SQLite statement used inside this function;
1065
* the caller should initialize *pSelect to NULL before the first call.
1067
static int query_merge(fulltext_vtab *v, sqlite3_stmt **pSelect,
1069
DocList *pIn, int iOffset, DocList *out){
1073
if( pIn!=NULL && !pIn->nData ){
1074
/* If [pIn] is already empty, there's no point in reading the
1075
* posting list to AND it in; return immediately. */
1079
rc = term_select_doclist(v, zTerm, -1, pSelect);
1080
if( rc!=SQLITE_ROW && rc!=SQLITE_DONE ) return rc;
1082
mergeInit(&merge, pIn, iOffset, out);
1083
while( rc==SQLITE_ROW ){
1085
docListInit(&block, DL_POSITIONS_OFFSETS,
1086
sqlite3_column_blob(*pSelect, 0),
1087
sqlite3_column_bytes(*pSelect, 0));
1088
mergeBlock(&merge, &block);
1089
docListDestroy(&block);
1091
rc = sqlite3_step(*pSelect);
1092
if( rc!=SQLITE_ROW && rc!=SQLITE_DONE ){
1100
typedef struct QueryTerm {
1101
int is_phrase; /* true if this term begins a new phrase */
1107
* As an example, parsing the query ["four score" years "new nation"] will
1108
* yield a Query with 5 terms:
1109
* "four", is_phrase = 1
1110
* "score", is_phrase = 0
1111
* "years", is_phrase = 1
1112
* "new", is_phrase = 1
1113
* "nation", is_phrase = 0
1115
typedef struct Query {
1120
static void query_add(Query *q, int is_phrase, const char *zTerm){
1123
q->pTerm = realloc(q->pTerm, q->nTerms * sizeof(q->pTerm[0]));
1124
t = &q->pTerm[q->nTerms - 1];
1125
t->is_phrase = is_phrase;
1129
static void query_free(Query *q){
1131
for(i = 0; i < q->nTerms; ++i){
1132
free((void *) q->pTerm[i].zTerm);
1137
static int tokenize_segment(sqlite3_tokenizer *pTokenizer,
1138
const char *zQuery, int in_phrase,
1140
sqlite3_tokenizer_module *pModule = pTokenizer->pModule;
1141
sqlite3_tokenizer_cursor *pCursor;
1144
int rc = pModule->xOpen(pTokenizer, zQuery, -1, &pCursor);
1145
if( rc!=SQLITE_OK ) return rc;
1146
pCursor->pTokenizer = pTokenizer;
1150
int nToken, iStartOffset, iEndOffset, dummy_pos;
1152
rc = pModule->xNext(pCursor,
1154
&iStartOffset, &iEndOffset,
1156
if( rc!=SQLITE_OK ) break;
1157
query_add(pQuery, !in_phrase || is_first, string_dup_n(zToken, nToken));
1161
return pModule->xClose(pCursor);
1164
/* Parse a query string, yielding a Query object. */
1165
static int parse_query(fulltext_vtab *v, const char *zQuery, Query *pQuery){
1166
char *zQuery1 = string_dup(zQuery);
1170
pQuery->pTerm = NULL;
1182
tokenize_segment(v->pTokenizer, s, in_phrase, pQuery);
1185
in_phrase = !in_phrase;
1192
/* Perform a full-text query; return a list of documents in [pResult]. */
1193
static int fulltext_query(fulltext_vtab *v, const char *zQuery,
1196
int phrase_start = -1;
1198
sqlite3_stmt *pSelect = NULL;
1201
int rc = parse_query(v, zQuery, &q);
1202
if( rc!=SQLITE_OK ) return rc;
1205
for(i = 0 ; i < q.nTerms ; ++i){
1206
/* In each merge step, we need to generate positions whenever we're
1207
* processing a phrase which hasn't ended yet. */
1208
int need_positions = i<q.nTerms-1 && !q.pTerm[i+1].is_phrase;
1209
DocList *next = docListNew(need_positions ? DL_POSITIONS : DL_DOCIDS);
1210
if( q.pTerm[i].is_phrase ){
1213
rc = query_merge(v, &pSelect, q.pTerm[i].zTerm, d, i - phrase_start, next);
1214
if( rc!=SQLITE_OK ) break;
1221
sqlite3_finalize(pSelect);
1227
static int fulltextFilter(sqlite3_vtab_cursor *pCursor,
1228
int idxNum, const char *idxStr,
1229
int argc, sqlite3_value **argv){
1230
fulltext_cursor *c = (fulltext_cursor *) pCursor;
1231
fulltext_vtab *v = cursor_vtab(c);
1233
const char *zStatement;
1235
c->iCursorType = idxNum;
1238
zStatement = "select rowid, content from %_content";
1241
case QUERY_FULLTEXT: /* full-text search */
1243
const char *zQuery = (const char *)sqlite3_value_text(argv[0]);
1246
rc = fulltext_query(v, zQuery, &pResult);
1247
if( rc!=SQLITE_OK ) return rc;
1248
readerInit(&c->result, pResult);
1249
zStatement = "select rowid, content from %_content where rowid = ?";
1257
rc = sql_prepare(v->db, v->zName, &c->pStmt, zStatement);
1258
if( rc!=SQLITE_OK ) return rc;
1260
return fulltextNext(pCursor);
1263
static int fulltextEof(sqlite3_vtab_cursor *pCursor){
1264
fulltext_cursor *c = (fulltext_cursor *) pCursor;
1268
static int fulltextColumn(sqlite3_vtab_cursor *pCursor,
1269
sqlite3_context *pContext, int idxCol){
1270
fulltext_cursor *c = (fulltext_cursor *) pCursor;
1273
assert( idxCol==0 );
1274
s = (const char *) sqlite3_column_text(c->pStmt, 1);
1275
sqlite3_result_text(pContext, s, -1, SQLITE_TRANSIENT);
1280
static int fulltextRowid(sqlite3_vtab_cursor *pCursor, sqlite_int64 *pRowid){
1281
fulltext_cursor *c = (fulltext_cursor *) pCursor;
1283
*pRowid = sqlite3_column_int64(c->pStmt, 0);
1287
/* Build a hash table containing all terms in zText. */
1288
static int build_terms(Hash *terms, sqlite3_tokenizer *pTokenizer,
1289
const char *zText, sqlite_int64 iDocid){
1290
sqlite3_tokenizer_cursor *pCursor;
1293
int iStartOffset, iEndOffset, iPosition;
1295
int rc = pTokenizer->pModule->xOpen(pTokenizer, zText, -1, &pCursor);
1296
if( rc!=SQLITE_OK ) return rc;
1298
pCursor->pTokenizer = pTokenizer;
1299
HashInit(terms, HASH_STRING, 1);
1300
while( SQLITE_OK==pTokenizer->pModule->xNext(pCursor,
1301
&pToken, &nTokenBytes,
1302
&iStartOffset, &iEndOffset,
1306
/* Positions can't be negative; we use -1 as a terminator internally. */
1312
p = HashFind(terms, pToken, nTokenBytes);
1314
p = docListNew(DL_POSITIONS_OFFSETS);
1315
docListAddDocid(p, iDocid);
1316
HashInsert(terms, pToken, nTokenBytes, p);
1318
docListAddPosOffset(p, iPosition, iStartOffset, iEndOffset);
1322
/* TODO(shess) Check return? Should this be able to cause errors at
1323
** this point? Actually, same question about sqlite3_finalize(),
1324
** though one could argue that failure there means that the data is
1325
** not durable. *ponder*
1327
pTokenizer->pModule->xClose(pCursor);
1330
/* Update the %_terms table to map the term [zTerm] to the given rowid. */
1331
static int index_insert_term(fulltext_vtab *v, const char *zTerm, int nTerm,
1332
sqlite_int64 iDocid, DocList *p){
1333
sqlite_int64 iFirst;
1334
sqlite_int64 iIndexRow;
1337
int rc = term_chunk_select(v, zTerm, nTerm, iDocid, &iFirst);
1338
if( rc==SQLITE_DONE ){
1339
docListInit(&doclist, DL_POSITIONS_OFFSETS, 0, 0);
1340
if( docListUpdate(&doclist, iDocid, p) ){
1341
rc = term_insert(v, zTerm, nTerm, iDocid, &doclist);
1342
docListDestroy(&doclist);
1347
if( rc!=SQLITE_ROW ) return SQLITE_ERROR;
1349
/* This word is in the index; add this document ID to its blob. */
1351
rc = term_select(v, zTerm, nTerm, iFirst, &iIndexRow, &doclist);
1352
if( rc!=SQLITE_OK ) return rc;
1354
if( docListUpdate(&doclist, iDocid, p) ){
1355
/* If the blob is too big, split it in half. */
1356
if( doclist.nData>CHUNK_MAX ){
1358
if( docListSplit(&doclist, &half) ){
1359
rc = term_insert(v, zTerm, nTerm, firstDocid(&half), &half);
1360
docListDestroy(&half);
1361
if( rc!=SQLITE_OK ) goto err;
1364
rc = term_update(v, iIndexRow, &doclist);
1368
docListDestroy(&doclist);
1372
/* Insert a row into the full-text index; set *piRowid to be the ID of the
1374
static int index_insert(fulltext_vtab *v,
1375
sqlite3_value *pRequestRowid, const char *zText,
1376
sqlite_int64 *piRowid){
1377
Hash terms; /* maps term string -> PosList */
1380
int rc = content_insert(v, pRequestRowid, zText, -1);
1381
if( rc!=SQLITE_OK ) return rc;
1382
*piRowid = sqlite3_last_insert_rowid(v->db);
1384
if( !zText ) return SQLITE_OK; /* nothing to index */
1386
rc = build_terms(&terms, v->pTokenizer, zText, *piRowid);
1387
if( rc!=SQLITE_OK ) return rc;
1389
for(e=HashFirst(&terms); e; e=HashNext(e)){
1390
DocList *p = HashData(e);
1391
rc = index_insert_term(v, HashKey(e), HashKeysize(e), *piRowid, p);
1392
if( rc!=SQLITE_OK ) break;
1395
for(e=HashFirst(&terms); e; e=HashNext(e)){
1396
DocList *p = HashData(e);
1403
static int index_delete_term(fulltext_vtab *v, const char *zTerm, int nTerm,
1404
sqlite_int64 iDocid){
1405
sqlite_int64 iFirst;
1406
sqlite_int64 iIndexRow;
1409
int rc = term_chunk_select(v, zTerm, nTerm, iDocid, &iFirst);
1410
if( rc!=SQLITE_ROW ) return SQLITE_ERROR;
1412
rc = term_select(v, zTerm, nTerm, iFirst, &iIndexRow, &doclist);
1413
if( rc!=SQLITE_OK ) return rc;
1415
if( docListUpdate(&doclist, iDocid, NULL) ){
1416
if( doclist.nData>0 ){
1417
rc = term_update(v, iIndexRow, &doclist);
1418
} else { /* empty posting list */
1419
rc = term_delete(v, iIndexRow);
1422
docListDestroy(&doclist);
1426
/* Delete a row from the full-text index. */
1427
static int index_delete(fulltext_vtab *v, sqlite_int64 iRow){
1432
int rc = content_select(v, iRow, &zText);
1433
if( rc!=SQLITE_OK ) return rc;
1435
rc = build_terms(&terms, v->pTokenizer, zText, iRow);
1437
if( rc!=SQLITE_OK ) return rc;
1439
for(e=HashFirst(&terms); e; e=HashNext(e)){
1440
rc = index_delete_term(v, HashKey(e), HashKeysize(e), iRow);
1441
if( rc!=SQLITE_OK ) break;
1443
for(e=HashFirst(&terms); e; e=HashNext(e)){
1444
DocList *p = HashData(e);
1449
return content_delete(v, iRow);
1452
static int fulltextUpdate(sqlite3_vtab *pVtab, int nArg, sqlite3_value **ppArg,
1453
sqlite_int64 *pRowid){
1454
fulltext_vtab *v = (fulltext_vtab *) pVtab;
1457
return index_delete(v, sqlite3_value_int64(ppArg[0]));
1460
if( sqlite3_value_type(ppArg[0]) != SQLITE_NULL ){
1461
return SQLITE_ERROR; /* an update; not yet supported */
1464
assert( nArg==3 ); /* ppArg[1] = rowid, ppArg[2] = content */
1465
return index_insert(v, ppArg[1],
1466
(const char *)sqlite3_value_text(ppArg[2]), pRowid);
1469
static sqlite3_module fulltextModule = {
1486
int fulltext_init(sqlite3 *db){
1487
return sqlite3_create_module(db, "fulltext", &fulltextModule, 0);
1491
int sqlite3_extension_init(sqlite3 *db, char **pzErrMsg,
1492
const sqlite3_api_routines *pApi){
1493
SQLITE_EXTENSION_INIT2(pApi)
1494
return fulltext_init(db);
added
removed
third_party/sqlite_google/ext/fts1/fulltext.c
