Snapshot of upstream SQLite 3.37.2
[sqlcipher.git] / ext / fts3 / fts3.c
blob074123d65899e2b8f8ed028aba10670e91bcde89
1 /*
2 ** 2006 Oct 10
3 **
4 ** The author disclaims copyright to this source code. In place of
5 ** a legal notice, here is a blessing:
6 **
7 ** May you do good and not evil.
8 ** May you find forgiveness for yourself and forgive others.
9 ** May you share freely, never taking more than you give.
11 ******************************************************************************
13 ** This is an SQLite module implementing full-text search.
17 ** The code in this file is only compiled if:
19 ** * The FTS3 module is being built as an extension
20 ** (in which case SQLITE_CORE is not defined), or
22 ** * The FTS3 module is being built into the core of
23 ** SQLite (in which case SQLITE_ENABLE_FTS3 is defined).
26 /* The full-text index is stored in a series of b+tree (-like)
27 ** structures called segments which map terms to doclists. The
28 ** structures are like b+trees in layout, but are constructed from the
29 ** bottom up in optimal fashion and are not updatable. Since trees
30 ** are built from the bottom up, things will be described from the
31 ** bottom up.
34 **** Varints ****
35 ** The basic unit of encoding is a variable-length integer called a
36 ** varint. We encode variable-length integers in little-endian order
37 ** using seven bits * per byte as follows:
39 ** KEY:
40 ** A = 0xxxxxxx 7 bits of data and one flag bit
41 ** B = 1xxxxxxx 7 bits of data and one flag bit
43 ** 7 bits - A
44 ** 14 bits - BA
45 ** 21 bits - BBA
46 ** and so on.
48 ** This is similar in concept to how sqlite encodes "varints" but
49 ** the encoding is not the same. SQLite varints are big-endian
50 ** are are limited to 9 bytes in length whereas FTS3 varints are
51 ** little-endian and can be up to 10 bytes in length (in theory).
53 ** Example encodings:
55 ** 1: 0x01
56 ** 127: 0x7f
57 ** 128: 0x81 0x00
60 **** Document lists ****
61 ** A doclist (document list) holds a docid-sorted list of hits for a
62 ** given term. Doclists hold docids and associated token positions.
63 ** A docid is the unique integer identifier for a single document.
64 ** A position is the index of a word within the document. The first
65 ** word of the document has a position of 0.
67 ** FTS3 used to optionally store character offsets using a compile-time
68 ** option. But that functionality is no longer supported.
70 ** A doclist is stored like this:
72 ** array {
73 ** varint docid; (delta from previous doclist)
74 ** array { (position list for column 0)
75 ** varint position; (2 more than the delta from previous position)
76 ** }
77 ** array {
78 ** varint POS_COLUMN; (marks start of position list for new column)
79 ** varint column; (index of new column)
80 ** array {
81 ** varint position; (2 more than the delta from previous position)
82 ** }
83 ** }
84 ** varint POS_END; (marks end of positions for this document.
85 ** }
87 ** Here, array { X } means zero or more occurrences of X, adjacent in
88 ** memory. A "position" is an index of a token in the token stream
89 ** generated by the tokenizer. Note that POS_END and POS_COLUMN occur
90 ** in the same logical place as the position element, and act as sentinals
91 ** ending a position list array. POS_END is 0. POS_COLUMN is 1.
92 ** The positions numbers are not stored literally but rather as two more
93 ** than the difference from the prior position, or the just the position plus
94 ** 2 for the first position. Example:
96 ** label: A B C D E F G H I J K
97 ** value: 123 5 9 1 1 14 35 0 234 72 0
99 ** The 123 value is the first docid. For column zero in this document
100 ** there are two matches at positions 3 and 10 (5-2 and 9-2+3). The 1
101 ** at D signals the start of a new column; the 1 at E indicates that the
102 ** new column is column number 1. There are two positions at 12 and 45
103 ** (14-2 and 35-2+12). The 0 at H indicate the end-of-document. The
104 ** 234 at I is the delta to next docid (357). It has one position 70
105 ** (72-2) and then terminates with the 0 at K.
107 ** A "position-list" is the list of positions for multiple columns for
108 ** a single docid. A "column-list" is the set of positions for a single
109 ** column. Hence, a position-list consists of one or more column-lists,
110 ** a document record consists of a docid followed by a position-list and
111 ** a doclist consists of one or more document records.
113 ** A bare doclist omits the position information, becoming an
114 ** array of varint-encoded docids.
116 **** Segment leaf nodes ****
117 ** Segment leaf nodes store terms and doclists, ordered by term. Leaf
118 ** nodes are written using LeafWriter, and read using LeafReader (to
119 ** iterate through a single leaf node's data) and LeavesReader (to
120 ** iterate through a segment's entire leaf layer). Leaf nodes have
121 ** the format:
123 ** varint iHeight; (height from leaf level, always 0)
124 ** varint nTerm; (length of first term)
125 ** char pTerm[nTerm]; (content of first term)
126 ** varint nDoclist; (length of term's associated doclist)
127 ** char pDoclist[nDoclist]; (content of doclist)
128 ** array {
129 ** (further terms are delta-encoded)
130 ** varint nPrefix; (length of prefix shared with previous term)
131 ** varint nSuffix; (length of unshared suffix)
132 ** char pTermSuffix[nSuffix];(unshared suffix of next term)
133 ** varint nDoclist; (length of term's associated doclist)
134 ** char pDoclist[nDoclist]; (content of doclist)
135 ** }
137 ** Here, array { X } means zero or more occurrences of X, adjacent in
138 ** memory.
140 ** Leaf nodes are broken into blocks which are stored contiguously in
141 ** the %_segments table in sorted order. This means that when the end
142 ** of a node is reached, the next term is in the node with the next
143 ** greater node id.
145 ** New data is spilled to a new leaf node when the current node
146 ** exceeds LEAF_MAX bytes (default 2048). New data which itself is
147 ** larger than STANDALONE_MIN (default 1024) is placed in a standalone
148 ** node (a leaf node with a single term and doclist). The goal of
149 ** these settings is to pack together groups of small doclists while
150 ** making it efficient to directly access large doclists. The
151 ** assumption is that large doclists represent terms which are more
152 ** likely to be query targets.
154 ** TODO(shess) It may be useful for blocking decisions to be more
155 ** dynamic. For instance, it may make more sense to have a 2.5k leaf
156 ** node rather than splitting into 2k and .5k nodes. My intuition is
157 ** that this might extend through 2x or 4x the pagesize.
160 **** Segment interior nodes ****
161 ** Segment interior nodes store blockids for subtree nodes and terms
162 ** to describe what data is stored by the each subtree. Interior
163 ** nodes are written using InteriorWriter, and read using
164 ** InteriorReader. InteriorWriters are created as needed when
165 ** SegmentWriter creates new leaf nodes, or when an interior node
166 ** itself grows too big and must be split. The format of interior
167 ** nodes:
169 ** varint iHeight; (height from leaf level, always >0)
170 ** varint iBlockid; (block id of node's leftmost subtree)
171 ** optional {
172 ** varint nTerm; (length of first term)
173 ** char pTerm[nTerm]; (content of first term)
174 ** array {
175 ** (further terms are delta-encoded)
176 ** varint nPrefix; (length of shared prefix with previous term)
177 ** varint nSuffix; (length of unshared suffix)
178 ** char pTermSuffix[nSuffix]; (unshared suffix of next term)
179 ** }
180 ** }
182 ** Here, optional { X } means an optional element, while array { X }
183 ** means zero or more occurrences of X, adjacent in memory.
185 ** An interior node encodes n terms separating n+1 subtrees. The
186 ** subtree blocks are contiguous, so only the first subtree's blockid
187 ** is encoded. The subtree at iBlockid will contain all terms less
188 ** than the first term encoded (or all terms if no term is encoded).
189 ** Otherwise, for terms greater than or equal to pTerm[i] but less
190 ** than pTerm[i+1], the subtree for that term will be rooted at
191 ** iBlockid+i. Interior nodes only store enough term data to
192 ** distinguish adjacent children (if the rightmost term of the left
193 ** child is "something", and the leftmost term of the right child is
194 ** "wicked", only "w" is stored).
196 ** New data is spilled to a new interior node at the same height when
197 ** the current node exceeds INTERIOR_MAX bytes (default 2048).
198 ** INTERIOR_MIN_TERMS (default 7) keeps large terms from monopolizing
199 ** interior nodes and making the tree too skinny. The interior nodes
200 ** at a given height are naturally tracked by interior nodes at
201 ** height+1, and so on.
204 **** Segment directory ****
205 ** The segment directory in table %_segdir stores meta-information for
206 ** merging and deleting segments, and also the root node of the
207 ** segment's tree.
209 ** The root node is the top node of the segment's tree after encoding
210 ** the entire segment, restricted to ROOT_MAX bytes (default 1024).
211 ** This could be either a leaf node or an interior node. If the top
212 ** node requires more than ROOT_MAX bytes, it is flushed to %_segments
213 ** and a new root interior node is generated (which should always fit
214 ** within ROOT_MAX because it only needs space for 2 varints, the
215 ** height and the blockid of the previous root).
217 ** The meta-information in the segment directory is:
218 ** level - segment level (see below)
219 ** idx - index within level
220 ** - (level,idx uniquely identify a segment)
221 ** start_block - first leaf node
222 ** leaves_end_block - last leaf node
223 ** end_block - last block (including interior nodes)
224 ** root - contents of root node
226 ** If the root node is a leaf node, then start_block,
227 ** leaves_end_block, and end_block are all 0.
230 **** Segment merging ****
231 ** To amortize update costs, segments are grouped into levels and
232 ** merged in batches. Each increase in level represents exponentially
233 ** more documents.
235 ** New documents (actually, document updates) are tokenized and
236 ** written individually (using LeafWriter) to a level 0 segment, with
237 ** incrementing idx. When idx reaches MERGE_COUNT (default 16), all
238 ** level 0 segments are merged into a single level 1 segment. Level 1
239 ** is populated like level 0, and eventually MERGE_COUNT level 1
240 ** segments are merged to a single level 2 segment (representing
241 ** MERGE_COUNT^2 updates), and so on.
243 ** A segment merge traverses all segments at a given level in
244 ** parallel, performing a straightforward sorted merge. Since segment
245 ** leaf nodes are written in to the %_segments table in order, this
246 ** merge traverses the underlying sqlite disk structures efficiently.
247 ** After the merge, all segment blocks from the merged level are
248 ** deleted.
250 ** MERGE_COUNT controls how often we merge segments. 16 seems to be
251 ** somewhat of a sweet spot for insertion performance. 32 and 64 show
252 ** very similar performance numbers to 16 on insertion, though they're
253 ** a tiny bit slower (perhaps due to more overhead in merge-time
254 ** sorting). 8 is about 20% slower than 16, 4 about 50% slower than
255 ** 16, 2 about 66% slower than 16.
257 ** At query time, high MERGE_COUNT increases the number of segments
258 ** which need to be scanned and merged. For instance, with 100k docs
259 ** inserted:
261 ** MERGE_COUNT segments
262 ** 16 25
263 ** 8 12
264 ** 4 10
265 ** 2 6
267 ** This appears to have only a moderate impact on queries for very
268 ** frequent terms (which are somewhat dominated by segment merge
269 ** costs), and infrequent and non-existent terms still seem to be fast
270 ** even with many segments.
272 ** TODO(shess) That said, it would be nice to have a better query-side
273 ** argument for MERGE_COUNT of 16. Also, it is possible/likely that
274 ** optimizations to things like doclist merging will swing the sweet
275 ** spot around.
279 **** Handling of deletions and updates ****
280 ** Since we're using a segmented structure, with no docid-oriented
281 ** index into the term index, we clearly cannot simply update the term
282 ** index when a document is deleted or updated. For deletions, we
283 ** write an empty doclist (varint(docid) varint(POS_END)), for updates
284 ** we simply write the new doclist. Segment merges overwrite older
285 ** data for a particular docid with newer data, so deletes or updates
286 ** will eventually overtake the earlier data and knock it out. The
287 ** query logic likewise merges doclists so that newer data knocks out
288 ** older data.
291 #include "fts3Int.h"
292 #if !defined(SQLITE_CORE) || defined(SQLITE_ENABLE_FTS3)
294 #if defined(SQLITE_ENABLE_FTS3) && !defined(SQLITE_CORE)
295 # define SQLITE_CORE 1
296 #endif
298 #include <assert.h>
299 #include <stdlib.h>
300 #include <stddef.h>
301 #include <stdio.h>
302 #include <string.h>
303 #include <stdarg.h>
305 #include "fts3.h"
306 #ifndef SQLITE_CORE
307 # include "sqlite3ext.h"
308 SQLITE_EXTENSION_INIT1
309 #endif
311 static int fts3EvalNext(Fts3Cursor *pCsr);
312 static int fts3EvalStart(Fts3Cursor *pCsr);
313 static int fts3TermSegReaderCursor(
314 Fts3Cursor *, const char *, int, int, Fts3MultiSegReader **);
317 ** This variable is set to false when running tests for which the on disk
318 ** structures should not be corrupt. Otherwise, true. If it is false, extra
319 ** assert() conditions in the fts3 code are activated - conditions that are
320 ** only true if it is guaranteed that the fts3 database is not corrupt.
322 #ifdef SQLITE_DEBUG
323 int sqlite3_fts3_may_be_corrupt = 1;
324 #endif
327 ** Write a 64-bit variable-length integer to memory starting at p[0].
328 ** The length of data written will be between 1 and FTS3_VARINT_MAX bytes.
329 ** The number of bytes written is returned.
331 int sqlite3Fts3PutVarint(char *p, sqlite_int64 v){
332 unsigned char *q = (unsigned char *) p;
333 sqlite_uint64 vu = v;
335 *q++ = (unsigned char) ((vu & 0x7f) | 0x80);
336 vu >>= 7;
337 }while( vu!=0 );
338 q[-1] &= 0x7f; /* turn off high bit in final byte */
339 assert( q - (unsigned char *)p <= FTS3_VARINT_MAX );
340 return (int) (q - (unsigned char *)p);
343 #define GETVARINT_STEP(v, ptr, shift, mask1, mask2, var, ret) \
344 v = (v & mask1) | ( (*(const unsigned char*)(ptr++)) << shift ); \
345 if( (v & mask2)==0 ){ var = v; return ret; }
346 #define GETVARINT_INIT(v, ptr, shift, mask1, mask2, var, ret) \
347 v = (*ptr++); \
348 if( (v & mask2)==0 ){ var = v; return ret; }
350 int sqlite3Fts3GetVarintU(const char *pBuf, sqlite_uint64 *v){
351 const unsigned char *p = (const unsigned char*)pBuf;
352 const unsigned char *pStart = p;
353 u32 a;
354 u64 b;
355 int shift;
357 GETVARINT_INIT(a, p, 0, 0x00, 0x80, *v, 1);
358 GETVARINT_STEP(a, p, 7, 0x7F, 0x4000, *v, 2);
359 GETVARINT_STEP(a, p, 14, 0x3FFF, 0x200000, *v, 3);
360 GETVARINT_STEP(a, p, 21, 0x1FFFFF, 0x10000000, *v, 4);
361 b = (a & 0x0FFFFFFF );
363 for(shift=28; shift<=63; shift+=7){
364 u64 c = *p++;
365 b += (c&0x7F) << shift;
366 if( (c & 0x80)==0 ) break;
368 *v = b;
369 return (int)(p - pStart);
373 ** Read a 64-bit variable-length integer from memory starting at p[0].
374 ** Return the number of bytes read, or 0 on error.
375 ** The value is stored in *v.
377 int sqlite3Fts3GetVarint(const char *pBuf, sqlite_int64 *v){
378 return sqlite3Fts3GetVarintU(pBuf, (sqlite3_uint64*)v);
382 ** Read a 64-bit variable-length integer from memory starting at p[0] and
383 ** not extending past pEnd[-1].
384 ** Return the number of bytes read, or 0 on error.
385 ** The value is stored in *v.
387 int sqlite3Fts3GetVarintBounded(
388 const char *pBuf,
389 const char *pEnd,
390 sqlite_int64 *v
392 const unsigned char *p = (const unsigned char*)pBuf;
393 const unsigned char *pStart = p;
394 const unsigned char *pX = (const unsigned char*)pEnd;
395 u64 b = 0;
396 int shift;
397 for(shift=0; shift<=63; shift+=7){
398 u64 c = p<pX ? *p : 0;
399 p++;
400 b += (c&0x7F) << shift;
401 if( (c & 0x80)==0 ) break;
403 *v = b;
404 return (int)(p - pStart);
408 ** Similar to sqlite3Fts3GetVarint(), except that the output is truncated to
409 ** a non-negative 32-bit integer before it is returned.
411 int sqlite3Fts3GetVarint32(const char *p, int *pi){
412 const unsigned char *ptr = (const unsigned char*)p;
413 u32 a;
415 #ifndef fts3GetVarint32
416 GETVARINT_INIT(a, ptr, 0, 0x00, 0x80, *pi, 1);
417 #else
418 a = (*ptr++);
419 assert( a & 0x80 );
420 #endif
422 GETVARINT_STEP(a, ptr, 7, 0x7F, 0x4000, *pi, 2);
423 GETVARINT_STEP(a, ptr, 14, 0x3FFF, 0x200000, *pi, 3);
424 GETVARINT_STEP(a, ptr, 21, 0x1FFFFF, 0x10000000, *pi, 4);
425 a = (a & 0x0FFFFFFF );
426 *pi = (int)(a | ((u32)(*ptr & 0x07) << 28));
427 assert( 0==(a & 0x80000000) );
428 assert( *pi>=0 );
429 return 5;
433 ** Return the number of bytes required to encode v as a varint
435 int sqlite3Fts3VarintLen(sqlite3_uint64 v){
436 int i = 0;
438 i++;
439 v >>= 7;
440 }while( v!=0 );
441 return i;
445 ** Convert an SQL-style quoted string into a normal string by removing
446 ** the quote characters. The conversion is done in-place. If the
447 ** input does not begin with a quote character, then this routine
448 ** is a no-op.
450 ** Examples:
452 ** "abc" becomes abc
453 ** 'xyz' becomes xyz
454 ** [pqr] becomes pqr
455 ** `mno` becomes mno
458 void sqlite3Fts3Dequote(char *z){
459 char quote; /* Quote character (if any ) */
461 quote = z[0];
462 if( quote=='[' || quote=='\'' || quote=='"' || quote=='`' ){
463 int iIn = 1; /* Index of next byte to read from input */
464 int iOut = 0; /* Index of next byte to write to output */
466 /* If the first byte was a '[', then the close-quote character is a ']' */
467 if( quote=='[' ) quote = ']';
469 while( z[iIn] ){
470 if( z[iIn]==quote ){
471 if( z[iIn+1]!=quote ) break;
472 z[iOut++] = quote;
473 iIn += 2;
474 }else{
475 z[iOut++] = z[iIn++];
478 z[iOut] = '\0';
483 ** Read a single varint from the doclist at *pp and advance *pp to point
484 ** to the first byte past the end of the varint. Add the value of the varint
485 ** to *pVal.
487 static void fts3GetDeltaVarint(char **pp, sqlite3_int64 *pVal){
488 sqlite3_int64 iVal;
489 *pp += sqlite3Fts3GetVarint(*pp, &iVal);
490 *pVal += iVal;
494 ** When this function is called, *pp points to the first byte following a
495 ** varint that is part of a doclist (or position-list, or any other list
496 ** of varints). This function moves *pp to point to the start of that varint,
497 ** and sets *pVal by the varint value.
499 ** Argument pStart points to the first byte of the doclist that the
500 ** varint is part of.
502 static void fts3GetReverseVarint(
503 char **pp,
504 char *pStart,
505 sqlite3_int64 *pVal
507 sqlite3_int64 iVal;
508 char *p;
510 /* Pointer p now points at the first byte past the varint we are
511 ** interested in. So, unless the doclist is corrupt, the 0x80 bit is
512 ** clear on character p[-1]. */
513 for(p = (*pp)-2; p>=pStart && *p&0x80; p--);
514 p++;
515 *pp = p;
517 sqlite3Fts3GetVarint(p, &iVal);
518 *pVal = iVal;
522 ** The xDisconnect() virtual table method.
524 static int fts3DisconnectMethod(sqlite3_vtab *pVtab){
525 Fts3Table *p = (Fts3Table *)pVtab;
526 int i;
528 assert( p->nPendingData==0 );
529 assert( p->pSegments==0 );
531 /* Free any prepared statements held */
532 sqlite3_finalize(p->pSeekStmt);
533 for(i=0; i<SizeofArray(p->aStmt); i++){
534 sqlite3_finalize(p->aStmt[i]);
536 sqlite3_free(p->zSegmentsTbl);
537 sqlite3_free(p->zReadExprlist);
538 sqlite3_free(p->zWriteExprlist);
539 sqlite3_free(p->zContentTbl);
540 sqlite3_free(p->zLanguageid);
542 /* Invoke the tokenizer destructor to free the tokenizer. */
543 p->pTokenizer->pModule->xDestroy(p->pTokenizer);
545 sqlite3_free(p);
546 return SQLITE_OK;
550 ** Write an error message into *pzErr
552 void sqlite3Fts3ErrMsg(char **pzErr, const char *zFormat, ...){
553 va_list ap;
554 sqlite3_free(*pzErr);
555 va_start(ap, zFormat);
556 *pzErr = sqlite3_vmprintf(zFormat, ap);
557 va_end(ap);
561 ** Construct one or more SQL statements from the format string given
562 ** and then evaluate those statements. The success code is written
563 ** into *pRc.
565 ** If *pRc is initially non-zero then this routine is a no-op.
567 static void fts3DbExec(
568 int *pRc, /* Success code */
569 sqlite3 *db, /* Database in which to run SQL */
570 const char *zFormat, /* Format string for SQL */
571 ... /* Arguments to the format string */
573 va_list ap;
574 char *zSql;
575 if( *pRc ) return;
576 va_start(ap, zFormat);
577 zSql = sqlite3_vmprintf(zFormat, ap);
578 va_end(ap);
579 if( zSql==0 ){
580 *pRc = SQLITE_NOMEM;
581 }else{
582 *pRc = sqlite3_exec(db, zSql, 0, 0, 0);
583 sqlite3_free(zSql);
588 ** The xDestroy() virtual table method.
590 static int fts3DestroyMethod(sqlite3_vtab *pVtab){
591 Fts3Table *p = (Fts3Table *)pVtab;
592 int rc = SQLITE_OK; /* Return code */
593 const char *zDb = p->zDb; /* Name of database (e.g. "main", "temp") */
594 sqlite3 *db = p->db; /* Database handle */
596 /* Drop the shadow tables */
597 fts3DbExec(&rc, db,
598 "DROP TABLE IF EXISTS %Q.'%q_segments';"
599 "DROP TABLE IF EXISTS %Q.'%q_segdir';"
600 "DROP TABLE IF EXISTS %Q.'%q_docsize';"
601 "DROP TABLE IF EXISTS %Q.'%q_stat';"
602 "%s DROP TABLE IF EXISTS %Q.'%q_content';",
603 zDb, p->zName,
604 zDb, p->zName,
605 zDb, p->zName,
606 zDb, p->zName,
607 (p->zContentTbl ? "--" : ""), zDb,p->zName
610 /* If everything has worked, invoke fts3DisconnectMethod() to free the
611 ** memory associated with the Fts3Table structure and return SQLITE_OK.
612 ** Otherwise, return an SQLite error code.
614 return (rc==SQLITE_OK ? fts3DisconnectMethod(pVtab) : rc);
619 ** Invoke sqlite3_declare_vtab() to declare the schema for the FTS3 table
620 ** passed as the first argument. This is done as part of the xConnect()
621 ** and xCreate() methods.
623 ** If *pRc is non-zero when this function is called, it is a no-op.
624 ** Otherwise, if an error occurs, an SQLite error code is stored in *pRc
625 ** before returning.
627 static void fts3DeclareVtab(int *pRc, Fts3Table *p){
628 if( *pRc==SQLITE_OK ){
629 int i; /* Iterator variable */
630 int rc; /* Return code */
631 char *zSql; /* SQL statement passed to declare_vtab() */
632 char *zCols; /* List of user defined columns */
633 const char *zLanguageid;
635 zLanguageid = (p->zLanguageid ? p->zLanguageid : "__langid");
636 sqlite3_vtab_config(p->db, SQLITE_VTAB_CONSTRAINT_SUPPORT, 1);
638 /* Create a list of user columns for the virtual table */
639 zCols = sqlite3_mprintf("%Q, ", p->azColumn[0]);
640 for(i=1; zCols && i<p->nColumn; i++){
641 zCols = sqlite3_mprintf("%z%Q, ", zCols, p->azColumn[i]);
644 /* Create the whole "CREATE TABLE" statement to pass to SQLite */
645 zSql = sqlite3_mprintf(
646 "CREATE TABLE x(%s %Q HIDDEN, docid HIDDEN, %Q HIDDEN)",
647 zCols, p->zName, zLanguageid
649 if( !zCols || !zSql ){
650 rc = SQLITE_NOMEM;
651 }else{
652 rc = sqlite3_declare_vtab(p->db, zSql);
655 sqlite3_free(zSql);
656 sqlite3_free(zCols);
657 *pRc = rc;
662 ** Create the %_stat table if it does not already exist.
664 void sqlite3Fts3CreateStatTable(int *pRc, Fts3Table *p){
665 fts3DbExec(pRc, p->db,
666 "CREATE TABLE IF NOT EXISTS %Q.'%q_stat'"
667 "(id INTEGER PRIMARY KEY, value BLOB);",
668 p->zDb, p->zName
670 if( (*pRc)==SQLITE_OK ) p->bHasStat = 1;
674 ** Create the backing store tables (%_content, %_segments and %_segdir)
675 ** required by the FTS3 table passed as the only argument. This is done
676 ** as part of the vtab xCreate() method.
678 ** If the p->bHasDocsize boolean is true (indicating that this is an
679 ** FTS4 table, not an FTS3 table) then also create the %_docsize and
680 ** %_stat tables required by FTS4.
682 static int fts3CreateTables(Fts3Table *p){
683 int rc = SQLITE_OK; /* Return code */
684 int i; /* Iterator variable */
685 sqlite3 *db = p->db; /* The database connection */
687 if( p->zContentTbl==0 ){
688 const char *zLanguageid = p->zLanguageid;
689 char *zContentCols; /* Columns of %_content table */
691 /* Create a list of user columns for the content table */
692 zContentCols = sqlite3_mprintf("docid INTEGER PRIMARY KEY");
693 for(i=0; zContentCols && i<p->nColumn; i++){
694 char *z = p->azColumn[i];
695 zContentCols = sqlite3_mprintf("%z, 'c%d%q'", zContentCols, i, z);
697 if( zLanguageid && zContentCols ){
698 zContentCols = sqlite3_mprintf("%z, langid", zContentCols, zLanguageid);
700 if( zContentCols==0 ) rc = SQLITE_NOMEM;
702 /* Create the content table */
703 fts3DbExec(&rc, db,
704 "CREATE TABLE %Q.'%q_content'(%s)",
705 p->zDb, p->zName, zContentCols
707 sqlite3_free(zContentCols);
710 /* Create other tables */
711 fts3DbExec(&rc, db,
712 "CREATE TABLE %Q.'%q_segments'(blockid INTEGER PRIMARY KEY, block BLOB);",
713 p->zDb, p->zName
715 fts3DbExec(&rc, db,
716 "CREATE TABLE %Q.'%q_segdir'("
717 "level INTEGER,"
718 "idx INTEGER,"
719 "start_block INTEGER,"
720 "leaves_end_block INTEGER,"
721 "end_block INTEGER,"
722 "root BLOB,"
723 "PRIMARY KEY(level, idx)"
724 ");",
725 p->zDb, p->zName
727 if( p->bHasDocsize ){
728 fts3DbExec(&rc, db,
729 "CREATE TABLE %Q.'%q_docsize'(docid INTEGER PRIMARY KEY, size BLOB);",
730 p->zDb, p->zName
733 assert( p->bHasStat==p->bFts4 );
734 if( p->bHasStat ){
735 sqlite3Fts3CreateStatTable(&rc, p);
737 return rc;
741 ** Store the current database page-size in bytes in p->nPgsz.
743 ** If *pRc is non-zero when this function is called, it is a no-op.
744 ** Otherwise, if an error occurs, an SQLite error code is stored in *pRc
745 ** before returning.
747 static void fts3DatabasePageSize(int *pRc, Fts3Table *p){
748 if( *pRc==SQLITE_OK ){
749 int rc; /* Return code */
750 char *zSql; /* SQL text "PRAGMA %Q.page_size" */
751 sqlite3_stmt *pStmt; /* Compiled "PRAGMA %Q.page_size" statement */
753 zSql = sqlite3_mprintf("PRAGMA %Q.page_size", p->zDb);
754 if( !zSql ){
755 rc = SQLITE_NOMEM;
756 }else{
757 rc = sqlite3_prepare(p->db, zSql, -1, &pStmt, 0);
758 if( rc==SQLITE_OK ){
759 sqlite3_step(pStmt);
760 p->nPgsz = sqlite3_column_int(pStmt, 0);
761 rc = sqlite3_finalize(pStmt);
762 }else if( rc==SQLITE_AUTH ){
763 p->nPgsz = 1024;
764 rc = SQLITE_OK;
767 assert( p->nPgsz>0 || rc!=SQLITE_OK );
768 sqlite3_free(zSql);
769 *pRc = rc;
774 ** "Special" FTS4 arguments are column specifications of the following form:
776 ** <key> = <value>
778 ** There may not be whitespace surrounding the "=" character. The <value>
779 ** term may be quoted, but the <key> may not.
781 static int fts3IsSpecialColumn(
782 const char *z,
783 int *pnKey,
784 char **pzValue
786 char *zValue;
787 const char *zCsr = z;
789 while( *zCsr!='=' ){
790 if( *zCsr=='\0' ) return 0;
791 zCsr++;
794 *pnKey = (int)(zCsr-z);
795 zValue = sqlite3_mprintf("%s", &zCsr[1]);
796 if( zValue ){
797 sqlite3Fts3Dequote(zValue);
799 *pzValue = zValue;
800 return 1;
804 ** Append the output of a printf() style formatting to an existing string.
806 static void fts3Appendf(
807 int *pRc, /* IN/OUT: Error code */
808 char **pz, /* IN/OUT: Pointer to string buffer */
809 const char *zFormat, /* Printf format string to append */
810 ... /* Arguments for printf format string */
812 if( *pRc==SQLITE_OK ){
813 va_list ap;
814 char *z;
815 va_start(ap, zFormat);
816 z = sqlite3_vmprintf(zFormat, ap);
817 va_end(ap);
818 if( z && *pz ){
819 char *z2 = sqlite3_mprintf("%s%s", *pz, z);
820 sqlite3_free(z);
821 z = z2;
823 if( z==0 ) *pRc = SQLITE_NOMEM;
824 sqlite3_free(*pz);
825 *pz = z;
830 ** Return a copy of input string zInput enclosed in double-quotes (") and
831 ** with all double quote characters escaped. For example:
833 ** fts3QuoteId("un \"zip\"") -> "un \"\"zip\"\""
835 ** The pointer returned points to memory obtained from sqlite3_malloc(). It
836 ** is the callers responsibility to call sqlite3_free() to release this
837 ** memory.
839 static char *fts3QuoteId(char const *zInput){
840 sqlite3_int64 nRet;
841 char *zRet;
842 nRet = 2 + (int)strlen(zInput)*2 + 1;
843 zRet = sqlite3_malloc64(nRet);
844 if( zRet ){
845 int i;
846 char *z = zRet;
847 *(z++) = '"';
848 for(i=0; zInput[i]; i++){
849 if( zInput[i]=='"' ) *(z++) = '"';
850 *(z++) = zInput[i];
852 *(z++) = '"';
853 *(z++) = '\0';
855 return zRet;
859 ** Return a list of comma separated SQL expressions and a FROM clause that
860 ** could be used in a SELECT statement such as the following:
862 ** SELECT <list of expressions> FROM %_content AS x ...
864 ** to return the docid, followed by each column of text data in order
865 ** from left to write. If parameter zFunc is not NULL, then instead of
866 ** being returned directly each column of text data is passed to an SQL
867 ** function named zFunc first. For example, if zFunc is "unzip" and the
868 ** table has the three user-defined columns "a", "b", and "c", the following
869 ** string is returned:
871 ** "docid, unzip(x.'a'), unzip(x.'b'), unzip(x.'c') FROM %_content AS x"
873 ** The pointer returned points to a buffer allocated by sqlite3_malloc(). It
874 ** is the responsibility of the caller to eventually free it.
876 ** If *pRc is not SQLITE_OK when this function is called, it is a no-op (and
877 ** a NULL pointer is returned). Otherwise, if an OOM error is encountered
878 ** by this function, NULL is returned and *pRc is set to SQLITE_NOMEM. If
879 ** no error occurs, *pRc is left unmodified.
881 static char *fts3ReadExprList(Fts3Table *p, const char *zFunc, int *pRc){
882 char *zRet = 0;
883 char *zFree = 0;
884 char *zFunction;
885 int i;
887 if( p->zContentTbl==0 ){
888 if( !zFunc ){
889 zFunction = "";
890 }else{
891 zFree = zFunction = fts3QuoteId(zFunc);
893 fts3Appendf(pRc, &zRet, "docid");
894 for(i=0; i<p->nColumn; i++){
895 fts3Appendf(pRc, &zRet, ",%s(x.'c%d%q')", zFunction, i, p->azColumn[i]);
897 if( p->zLanguageid ){
898 fts3Appendf(pRc, &zRet, ", x.%Q", "langid");
900 sqlite3_free(zFree);
901 }else{
902 fts3Appendf(pRc, &zRet, "rowid");
903 for(i=0; i<p->nColumn; i++){
904 fts3Appendf(pRc, &zRet, ", x.'%q'", p->azColumn[i]);
906 if( p->zLanguageid ){
907 fts3Appendf(pRc, &zRet, ", x.%Q", p->zLanguageid);
910 fts3Appendf(pRc, &zRet, " FROM '%q'.'%q%s' AS x",
911 p->zDb,
912 (p->zContentTbl ? p->zContentTbl : p->zName),
913 (p->zContentTbl ? "" : "_content")
915 return zRet;
919 ** Return a list of N comma separated question marks, where N is the number
920 ** of columns in the %_content table (one for the docid plus one for each
921 ** user-defined text column).
923 ** If argument zFunc is not NULL, then all but the first question mark
924 ** is preceded by zFunc and an open bracket, and followed by a closed
925 ** bracket. For example, if zFunc is "zip" and the FTS3 table has three
926 ** user-defined text columns, the following string is returned:
928 ** "?, zip(?), zip(?), zip(?)"
930 ** The pointer returned points to a buffer allocated by sqlite3_malloc(). It
931 ** is the responsibility of the caller to eventually free it.
933 ** If *pRc is not SQLITE_OK when this function is called, it is a no-op (and
934 ** a NULL pointer is returned). Otherwise, if an OOM error is encountered
935 ** by this function, NULL is returned and *pRc is set to SQLITE_NOMEM. If
936 ** no error occurs, *pRc is left unmodified.
938 static char *fts3WriteExprList(Fts3Table *p, const char *zFunc, int *pRc){
939 char *zRet = 0;
940 char *zFree = 0;
941 char *zFunction;
942 int i;
944 if( !zFunc ){
945 zFunction = "";
946 }else{
947 zFree = zFunction = fts3QuoteId(zFunc);
949 fts3Appendf(pRc, &zRet, "?");
950 for(i=0; i<p->nColumn; i++){
951 fts3Appendf(pRc, &zRet, ",%s(?)", zFunction);
953 if( p->zLanguageid ){
954 fts3Appendf(pRc, &zRet, ", ?");
956 sqlite3_free(zFree);
957 return zRet;
961 ** Buffer z contains a positive integer value encoded as utf-8 text.
962 ** Decode this value and store it in *pnOut, returning the number of bytes
963 ** consumed. If an overflow error occurs return a negative value.
965 int sqlite3Fts3ReadInt(const char *z, int *pnOut){
966 u64 iVal = 0;
967 int i;
968 for(i=0; z[i]>='0' && z[i]<='9'; i++){
969 iVal = iVal*10 + (z[i] - '0');
970 if( iVal>0x7FFFFFFF ) return -1;
972 *pnOut = (int)iVal;
973 return i;
977 ** This function interprets the string at (*pp) as a non-negative integer
978 ** value. It reads the integer and sets *pnOut to the value read, then
979 ** sets *pp to point to the byte immediately following the last byte of
980 ** the integer value.
982 ** Only decimal digits ('0'..'9') may be part of an integer value.
984 ** If *pp does not being with a decimal digit SQLITE_ERROR is returned and
985 ** the output value undefined. Otherwise SQLITE_OK is returned.
987 ** This function is used when parsing the "prefix=" FTS4 parameter.
989 static int fts3GobbleInt(const char **pp, int *pnOut){
990 const int MAX_NPREFIX = 10000000;
991 int nInt = 0; /* Output value */
992 int nByte;
993 nByte = sqlite3Fts3ReadInt(*pp, &nInt);
994 if( nInt>MAX_NPREFIX ){
995 nInt = 0;
997 if( nByte==0 ){
998 return SQLITE_ERROR;
1000 *pnOut = nInt;
1001 *pp += nByte;
1002 return SQLITE_OK;
1006 ** This function is called to allocate an array of Fts3Index structures
1007 ** representing the indexes maintained by the current FTS table. FTS tables
1008 ** always maintain the main "terms" index, but may also maintain one or
1009 ** more "prefix" indexes, depending on the value of the "prefix=" parameter
1010 ** (if any) specified as part of the CREATE VIRTUAL TABLE statement.
1012 ** Argument zParam is passed the value of the "prefix=" option if one was
1013 ** specified, or NULL otherwise.
1015 ** If no error occurs, SQLITE_OK is returned and *apIndex set to point to
1016 ** the allocated array. *pnIndex is set to the number of elements in the
1017 ** array. If an error does occur, an SQLite error code is returned.
1019 ** Regardless of whether or not an error is returned, it is the responsibility
1020 ** of the caller to call sqlite3_free() on the output array to free it.
1022 static int fts3PrefixParameter(
1023 const char *zParam, /* ABC in prefix=ABC parameter to parse */
1024 int *pnIndex, /* OUT: size of *apIndex[] array */
1025 struct Fts3Index **apIndex /* OUT: Array of indexes for this table */
1027 struct Fts3Index *aIndex; /* Allocated array */
1028 int nIndex = 1; /* Number of entries in array */
1030 if( zParam && zParam[0] ){
1031 const char *p;
1032 nIndex++;
1033 for(p=zParam; *p; p++){
1034 if( *p==',' ) nIndex++;
1038 aIndex = sqlite3_malloc64(sizeof(struct Fts3Index) * nIndex);
1039 *apIndex = aIndex;
1040 if( !aIndex ){
1041 return SQLITE_NOMEM;
1044 memset(aIndex, 0, sizeof(struct Fts3Index) * nIndex);
1045 if( zParam ){
1046 const char *p = zParam;
1047 int i;
1048 for(i=1; i<nIndex; i++){
1049 int nPrefix = 0;
1050 if( fts3GobbleInt(&p, &nPrefix) ) return SQLITE_ERROR;
1051 assert( nPrefix>=0 );
1052 if( nPrefix==0 ){
1053 nIndex--;
1054 i--;
1055 }else{
1056 aIndex[i].nPrefix = nPrefix;
1058 p++;
1062 *pnIndex = nIndex;
1063 return SQLITE_OK;
1067 ** This function is called when initializing an FTS4 table that uses the
1068 ** content=xxx option. It determines the number of and names of the columns
1069 ** of the new FTS4 table.
1071 ** The third argument passed to this function is the value passed to the
1072 ** config=xxx option (i.e. "xxx"). This function queries the database for
1073 ** a table of that name. If found, the output variables are populated
1074 ** as follows:
1076 ** *pnCol: Set to the number of columns table xxx has,
1078 ** *pnStr: Set to the total amount of space required to store a copy
1079 ** of each columns name, including the nul-terminator.
1081 ** *pazCol: Set to point to an array of *pnCol strings. Each string is
1082 ** the name of the corresponding column in table xxx. The array
1083 ** and its contents are allocated using a single allocation. It
1084 ** is the responsibility of the caller to free this allocation
1085 ** by eventually passing the *pazCol value to sqlite3_free().
1087 ** If the table cannot be found, an error code is returned and the output
1088 ** variables are undefined. Or, if an OOM is encountered, SQLITE_NOMEM is
1089 ** returned (and the output variables are undefined).
1091 static int fts3ContentColumns(
1092 sqlite3 *db, /* Database handle */
1093 const char *zDb, /* Name of db (i.e. "main", "temp" etc.) */
1094 const char *zTbl, /* Name of content table */
1095 const char ***pazCol, /* OUT: Malloc'd array of column names */
1096 int *pnCol, /* OUT: Size of array *pazCol */
1097 int *pnStr, /* OUT: Bytes of string content */
1098 char **pzErr /* OUT: error message */
1100 int rc = SQLITE_OK; /* Return code */
1101 char *zSql; /* "SELECT *" statement on zTbl */
1102 sqlite3_stmt *pStmt = 0; /* Compiled version of zSql */
1104 zSql = sqlite3_mprintf("SELECT * FROM %Q.%Q", zDb, zTbl);
1105 if( !zSql ){
1106 rc = SQLITE_NOMEM;
1107 }else{
1108 rc = sqlite3_prepare(db, zSql, -1, &pStmt, 0);
1109 if( rc!=SQLITE_OK ){
1110 sqlite3Fts3ErrMsg(pzErr, "%s", sqlite3_errmsg(db));
1113 sqlite3_free(zSql);
1115 if( rc==SQLITE_OK ){
1116 const char **azCol; /* Output array */
1117 sqlite3_int64 nStr = 0; /* Size of all column names (incl. 0x00) */
1118 int nCol; /* Number of table columns */
1119 int i; /* Used to iterate through columns */
1121 /* Loop through the returned columns. Set nStr to the number of bytes of
1122 ** space required to store a copy of each column name, including the
1123 ** nul-terminator byte. */
1124 nCol = sqlite3_column_count(pStmt);
1125 for(i=0; i<nCol; i++){
1126 const char *zCol = sqlite3_column_name(pStmt, i);
1127 nStr += strlen(zCol) + 1;
1130 /* Allocate and populate the array to return. */
1131 azCol = (const char **)sqlite3_malloc64(sizeof(char *) * nCol + nStr);
1132 if( azCol==0 ){
1133 rc = SQLITE_NOMEM;
1134 }else{
1135 char *p = (char *)&azCol[nCol];
1136 for(i=0; i<nCol; i++){
1137 const char *zCol = sqlite3_column_name(pStmt, i);
1138 int n = (int)strlen(zCol)+1;
1139 memcpy(p, zCol, n);
1140 azCol[i] = p;
1141 p += n;
1144 sqlite3_finalize(pStmt);
1146 /* Set the output variables. */
1147 *pnCol = nCol;
1148 *pnStr = nStr;
1149 *pazCol = azCol;
1152 return rc;
1156 ** This function is the implementation of both the xConnect and xCreate
1157 ** methods of the FTS3 virtual table.
1159 ** The argv[] array contains the following:
1161 ** argv[0] -> module name ("fts3" or "fts4")
1162 ** argv[1] -> database name
1163 ** argv[2] -> table name
1164 ** argv[...] -> "column name" and other module argument fields.
1166 static int fts3InitVtab(
1167 int isCreate, /* True for xCreate, false for xConnect */
1168 sqlite3 *db, /* The SQLite database connection */
1169 void *pAux, /* Hash table containing tokenizers */
1170 int argc, /* Number of elements in argv array */
1171 const char * const *argv, /* xCreate/xConnect argument array */
1172 sqlite3_vtab **ppVTab, /* Write the resulting vtab structure here */
1173 char **pzErr /* Write any error message here */
1175 Fts3Hash *pHash = (Fts3Hash *)pAux;
1176 Fts3Table *p = 0; /* Pointer to allocated vtab */
1177 int rc = SQLITE_OK; /* Return code */
1178 int i; /* Iterator variable */
1179 sqlite3_int64 nByte; /* Size of allocation used for *p */
1180 int iCol; /* Column index */
1181 int nString = 0; /* Bytes required to hold all column names */
1182 int nCol = 0; /* Number of columns in the FTS table */
1183 char *zCsr; /* Space for holding column names */
1184 int nDb; /* Bytes required to hold database name */
1185 int nName; /* Bytes required to hold table name */
1186 int isFts4 = (argv[0][3]=='4'); /* True for FTS4, false for FTS3 */
1187 const char **aCol; /* Array of column names */
1188 sqlite3_tokenizer *pTokenizer = 0; /* Tokenizer for this table */
1190 int nIndex = 0; /* Size of aIndex[] array */
1191 struct Fts3Index *aIndex = 0; /* Array of indexes for this table */
1193 /* The results of parsing supported FTS4 key=value options: */
1194 int bNoDocsize = 0; /* True to omit %_docsize table */
1195 int bDescIdx = 0; /* True to store descending indexes */
1196 char *zPrefix = 0; /* Prefix parameter value (or NULL) */
1197 char *zCompress = 0; /* compress=? parameter (or NULL) */
1198 char *zUncompress = 0; /* uncompress=? parameter (or NULL) */
1199 char *zContent = 0; /* content=? parameter (or NULL) */
1200 char *zLanguageid = 0; /* languageid=? parameter (or NULL) */
1201 char **azNotindexed = 0; /* The set of notindexed= columns */
1202 int nNotindexed = 0; /* Size of azNotindexed[] array */
1204 assert( strlen(argv[0])==4 );
1205 assert( (sqlite3_strnicmp(argv[0], "fts4", 4)==0 && isFts4)
1206 || (sqlite3_strnicmp(argv[0], "fts3", 4)==0 && !isFts4)
1209 nDb = (int)strlen(argv[1]) + 1;
1210 nName = (int)strlen(argv[2]) + 1;
1212 nByte = sizeof(const char *) * (argc-2);
1213 aCol = (const char **)sqlite3_malloc64(nByte);
1214 if( aCol ){
1215 memset((void*)aCol, 0, nByte);
1216 azNotindexed = (char **)sqlite3_malloc64(nByte);
1218 if( azNotindexed ){
1219 memset(azNotindexed, 0, nByte);
1221 if( !aCol || !azNotindexed ){
1222 rc = SQLITE_NOMEM;
1223 goto fts3_init_out;
1226 /* Loop through all of the arguments passed by the user to the FTS3/4
1227 ** module (i.e. all the column names and special arguments). This loop
1228 ** does the following:
1230 ** + Figures out the number of columns the FTSX table will have, and
1231 ** the number of bytes of space that must be allocated to store copies
1232 ** of the column names.
1234 ** + If there is a tokenizer specification included in the arguments,
1235 ** initializes the tokenizer pTokenizer.
1237 for(i=3; rc==SQLITE_OK && i<argc; i++){
1238 char const *z = argv[i];
1239 int nKey;
1240 char *zVal;
1242 /* Check if this is a tokenizer specification */
1243 if( !pTokenizer
1244 && strlen(z)>8
1245 && 0==sqlite3_strnicmp(z, "tokenize", 8)
1246 && 0==sqlite3Fts3IsIdChar(z[8])
1248 rc = sqlite3Fts3InitTokenizer(pHash, &z[9], &pTokenizer, pzErr);
1251 /* Check if it is an FTS4 special argument. */
1252 else if( isFts4 && fts3IsSpecialColumn(z, &nKey, &zVal) ){
1253 struct Fts4Option {
1254 const char *zOpt;
1255 int nOpt;
1256 } aFts4Opt[] = {
1257 { "matchinfo", 9 }, /* 0 -> MATCHINFO */
1258 { "prefix", 6 }, /* 1 -> PREFIX */
1259 { "compress", 8 }, /* 2 -> COMPRESS */
1260 { "uncompress", 10 }, /* 3 -> UNCOMPRESS */
1261 { "order", 5 }, /* 4 -> ORDER */
1262 { "content", 7 }, /* 5 -> CONTENT */
1263 { "languageid", 10 }, /* 6 -> LANGUAGEID */
1264 { "notindexed", 10 } /* 7 -> NOTINDEXED */
1267 int iOpt;
1268 if( !zVal ){
1269 rc = SQLITE_NOMEM;
1270 }else{
1271 for(iOpt=0; iOpt<SizeofArray(aFts4Opt); iOpt++){
1272 struct Fts4Option *pOp = &aFts4Opt[iOpt];
1273 if( nKey==pOp->nOpt && !sqlite3_strnicmp(z, pOp->zOpt, pOp->nOpt) ){
1274 break;
1277 switch( iOpt ){
1278 case 0: /* MATCHINFO */
1279 if( strlen(zVal)!=4 || sqlite3_strnicmp(zVal, "fts3", 4) ){
1280 sqlite3Fts3ErrMsg(pzErr, "unrecognized matchinfo: %s", zVal);
1281 rc = SQLITE_ERROR;
1283 bNoDocsize = 1;
1284 break;
1286 case 1: /* PREFIX */
1287 sqlite3_free(zPrefix);
1288 zPrefix = zVal;
1289 zVal = 0;
1290 break;
1292 case 2: /* COMPRESS */
1293 sqlite3_free(zCompress);
1294 zCompress = zVal;
1295 zVal = 0;
1296 break;
1298 case 3: /* UNCOMPRESS */
1299 sqlite3_free(zUncompress);
1300 zUncompress = zVal;
1301 zVal = 0;
1302 break;
1304 case 4: /* ORDER */
1305 if( (strlen(zVal)!=3 || sqlite3_strnicmp(zVal, "asc", 3))
1306 && (strlen(zVal)!=4 || sqlite3_strnicmp(zVal, "desc", 4))
1308 sqlite3Fts3ErrMsg(pzErr, "unrecognized order: %s", zVal);
1309 rc = SQLITE_ERROR;
1311 bDescIdx = (zVal[0]=='d' || zVal[0]=='D');
1312 break;
1314 case 5: /* CONTENT */
1315 sqlite3_free(zContent);
1316 zContent = zVal;
1317 zVal = 0;
1318 break;
1320 case 6: /* LANGUAGEID */
1321 assert( iOpt==6 );
1322 sqlite3_free(zLanguageid);
1323 zLanguageid = zVal;
1324 zVal = 0;
1325 break;
1327 case 7: /* NOTINDEXED */
1328 azNotindexed[nNotindexed++] = zVal;
1329 zVal = 0;
1330 break;
1332 default:
1333 assert( iOpt==SizeofArray(aFts4Opt) );
1334 sqlite3Fts3ErrMsg(pzErr, "unrecognized parameter: %s", z);
1335 rc = SQLITE_ERROR;
1336 break;
1338 sqlite3_free(zVal);
1342 /* Otherwise, the argument is a column name. */
1343 else {
1344 nString += (int)(strlen(z) + 1);
1345 aCol[nCol++] = z;
1349 /* If a content=xxx option was specified, the following:
1351 ** 1. Ignore any compress= and uncompress= options.
1353 ** 2. If no column names were specified as part of the CREATE VIRTUAL
1354 ** TABLE statement, use all columns from the content table.
1356 if( rc==SQLITE_OK && zContent ){
1357 sqlite3_free(zCompress);
1358 sqlite3_free(zUncompress);
1359 zCompress = 0;
1360 zUncompress = 0;
1361 if( nCol==0 ){
1362 sqlite3_free((void*)aCol);
1363 aCol = 0;
1364 rc = fts3ContentColumns(db, argv[1], zContent,&aCol,&nCol,&nString,pzErr);
1366 /* If a languageid= option was specified, remove the language id
1367 ** column from the aCol[] array. */
1368 if( rc==SQLITE_OK && zLanguageid ){
1369 int j;
1370 for(j=0; j<nCol; j++){
1371 if( sqlite3_stricmp(zLanguageid, aCol[j])==0 ){
1372 int k;
1373 for(k=j; k<nCol; k++) aCol[k] = aCol[k+1];
1374 nCol--;
1375 break;
1381 if( rc!=SQLITE_OK ) goto fts3_init_out;
1383 if( nCol==0 ){
1384 assert( nString==0 );
1385 aCol[0] = "content";
1386 nString = 8;
1387 nCol = 1;
1390 if( pTokenizer==0 ){
1391 rc = sqlite3Fts3InitTokenizer(pHash, "simple", &pTokenizer, pzErr);
1392 if( rc!=SQLITE_OK ) goto fts3_init_out;
1394 assert( pTokenizer );
1396 rc = fts3PrefixParameter(zPrefix, &nIndex, &aIndex);
1397 if( rc==SQLITE_ERROR ){
1398 assert( zPrefix );
1399 sqlite3Fts3ErrMsg(pzErr, "error parsing prefix parameter: %s", zPrefix);
1401 if( rc!=SQLITE_OK ) goto fts3_init_out;
1403 /* Allocate and populate the Fts3Table structure. */
1404 nByte = sizeof(Fts3Table) + /* Fts3Table */
1405 nCol * sizeof(char *) + /* azColumn */
1406 nIndex * sizeof(struct Fts3Index) + /* aIndex */
1407 nCol * sizeof(u8) + /* abNotindexed */
1408 nName + /* zName */
1409 nDb + /* zDb */
1410 nString; /* Space for azColumn strings */
1411 p = (Fts3Table*)sqlite3_malloc64(nByte);
1412 if( p==0 ){
1413 rc = SQLITE_NOMEM;
1414 goto fts3_init_out;
1416 memset(p, 0, nByte);
1417 p->db = db;
1418 p->nColumn = nCol;
1419 p->nPendingData = 0;
1420 p->azColumn = (char **)&p[1];
1421 p->pTokenizer = pTokenizer;
1422 p->nMaxPendingData = FTS3_MAX_PENDING_DATA;
1423 p->bHasDocsize = (isFts4 && bNoDocsize==0);
1424 p->bHasStat = (u8)isFts4;
1425 p->bFts4 = (u8)isFts4;
1426 p->bDescIdx = (u8)bDescIdx;
1427 p->nAutoincrmerge = 0xff; /* 0xff means setting unknown */
1428 p->zContentTbl = zContent;
1429 p->zLanguageid = zLanguageid;
1430 zContent = 0;
1431 zLanguageid = 0;
1432 TESTONLY( p->inTransaction = -1 );
1433 TESTONLY( p->mxSavepoint = -1 );
1435 p->aIndex = (struct Fts3Index *)&p->azColumn[nCol];
1436 memcpy(p->aIndex, aIndex, sizeof(struct Fts3Index) * nIndex);
1437 p->nIndex = nIndex;
1438 for(i=0; i<nIndex; i++){
1439 fts3HashInit(&p->aIndex[i].hPending, FTS3_HASH_STRING, 1);
1441 p->abNotindexed = (u8 *)&p->aIndex[nIndex];
1443 /* Fill in the zName and zDb fields of the vtab structure. */
1444 zCsr = (char *)&p->abNotindexed[nCol];
1445 p->zName = zCsr;
1446 memcpy(zCsr, argv[2], nName);
1447 zCsr += nName;
1448 p->zDb = zCsr;
1449 memcpy(zCsr, argv[1], nDb);
1450 zCsr += nDb;
1452 /* Fill in the azColumn array */
1453 for(iCol=0; iCol<nCol; iCol++){
1454 char *z;
1455 int n = 0;
1456 z = (char *)sqlite3Fts3NextToken(aCol[iCol], &n);
1457 if( n>0 ){
1458 memcpy(zCsr, z, n);
1460 zCsr[n] = '\0';
1461 sqlite3Fts3Dequote(zCsr);
1462 p->azColumn[iCol] = zCsr;
1463 zCsr += n+1;
1464 assert( zCsr <= &((char *)p)[nByte] );
1467 /* Fill in the abNotindexed array */
1468 for(iCol=0; iCol<nCol; iCol++){
1469 int n = (int)strlen(p->azColumn[iCol]);
1470 for(i=0; i<nNotindexed; i++){
1471 char *zNot = azNotindexed[i];
1472 if( zNot && n==(int)strlen(zNot)
1473 && 0==sqlite3_strnicmp(p->azColumn[iCol], zNot, n)
1475 p->abNotindexed[iCol] = 1;
1476 sqlite3_free(zNot);
1477 azNotindexed[i] = 0;
1481 for(i=0; i<nNotindexed; i++){
1482 if( azNotindexed[i] ){
1483 sqlite3Fts3ErrMsg(pzErr, "no such column: %s", azNotindexed[i]);
1484 rc = SQLITE_ERROR;
1488 if( rc==SQLITE_OK && (zCompress==0)!=(zUncompress==0) ){
1489 char const *zMiss = (zCompress==0 ? "compress" : "uncompress");
1490 rc = SQLITE_ERROR;
1491 sqlite3Fts3ErrMsg(pzErr, "missing %s parameter in fts4 constructor", zMiss);
1493 p->zReadExprlist = fts3ReadExprList(p, zUncompress, &rc);
1494 p->zWriteExprlist = fts3WriteExprList(p, zCompress, &rc);
1495 if( rc!=SQLITE_OK ) goto fts3_init_out;
1497 /* If this is an xCreate call, create the underlying tables in the
1498 ** database. TODO: For xConnect(), it could verify that said tables exist.
1500 if( isCreate ){
1501 rc = fts3CreateTables(p);
1504 /* Check to see if a legacy fts3 table has been "upgraded" by the
1505 ** addition of a %_stat table so that it can use incremental merge.
1507 if( !isFts4 && !isCreate ){
1508 p->bHasStat = 2;
1511 /* Figure out the page-size for the database. This is required in order to
1512 ** estimate the cost of loading large doclists from the database. */
1513 fts3DatabasePageSize(&rc, p);
1514 p->nNodeSize = p->nPgsz-35;
1516 #if defined(SQLITE_DEBUG)||defined(SQLITE_TEST)
1517 p->nMergeCount = FTS3_MERGE_COUNT;
1518 #endif
1520 /* Declare the table schema to SQLite. */
1521 fts3DeclareVtab(&rc, p);
1523 fts3_init_out:
1524 sqlite3_free(zPrefix);
1525 sqlite3_free(aIndex);
1526 sqlite3_free(zCompress);
1527 sqlite3_free(zUncompress);
1528 sqlite3_free(zContent);
1529 sqlite3_free(zLanguageid);
1530 for(i=0; i<nNotindexed; i++) sqlite3_free(azNotindexed[i]);
1531 sqlite3_free((void *)aCol);
1532 sqlite3_free((void *)azNotindexed);
1533 if( rc!=SQLITE_OK ){
1534 if( p ){
1535 fts3DisconnectMethod((sqlite3_vtab *)p);
1536 }else if( pTokenizer ){
1537 pTokenizer->pModule->xDestroy(pTokenizer);
1539 }else{
1540 assert( p->pSegments==0 );
1541 *ppVTab = &p->base;
1543 return rc;
1547 ** The xConnect() and xCreate() methods for the virtual table. All the
1548 ** work is done in function fts3InitVtab().
1550 static int fts3ConnectMethod(
1551 sqlite3 *db, /* Database connection */
1552 void *pAux, /* Pointer to tokenizer hash table */
1553 int argc, /* Number of elements in argv array */
1554 const char * const *argv, /* xCreate/xConnect argument array */
1555 sqlite3_vtab **ppVtab, /* OUT: New sqlite3_vtab object */
1556 char **pzErr /* OUT: sqlite3_malloc'd error message */
1558 return fts3InitVtab(0, db, pAux, argc, argv, ppVtab, pzErr);
1560 static int fts3CreateMethod(
1561 sqlite3 *db, /* Database connection */
1562 void *pAux, /* Pointer to tokenizer hash table */
1563 int argc, /* Number of elements in argv array */
1564 const char * const *argv, /* xCreate/xConnect argument array */
1565 sqlite3_vtab **ppVtab, /* OUT: New sqlite3_vtab object */
1566 char **pzErr /* OUT: sqlite3_malloc'd error message */
1568 return fts3InitVtab(1, db, pAux, argc, argv, ppVtab, pzErr);
1572 ** Set the pIdxInfo->estimatedRows variable to nRow. Unless this
1573 ** extension is currently being used by a version of SQLite too old to
1574 ** support estimatedRows. In that case this function is a no-op.
1576 static void fts3SetEstimatedRows(sqlite3_index_info *pIdxInfo, i64 nRow){
1577 #if SQLITE_VERSION_NUMBER>=3008002
1578 if( sqlite3_libversion_number()>=3008002 ){
1579 pIdxInfo->estimatedRows = nRow;
1581 #endif
1585 ** Set the SQLITE_INDEX_SCAN_UNIQUE flag in pIdxInfo->flags. Unless this
1586 ** extension is currently being used by a version of SQLite too old to
1587 ** support index-info flags. In that case this function is a no-op.
1589 static void fts3SetUniqueFlag(sqlite3_index_info *pIdxInfo){
1590 #if SQLITE_VERSION_NUMBER>=3008012
1591 if( sqlite3_libversion_number()>=3008012 ){
1592 pIdxInfo->idxFlags |= SQLITE_INDEX_SCAN_UNIQUE;
1594 #endif
1598 ** Implementation of the xBestIndex method for FTS3 tables. There
1599 ** are three possible strategies, in order of preference:
1601 ** 1. Direct lookup by rowid or docid.
1602 ** 2. Full-text search using a MATCH operator on a non-docid column.
1603 ** 3. Linear scan of %_content table.
1605 static int fts3BestIndexMethod(sqlite3_vtab *pVTab, sqlite3_index_info *pInfo){
1606 Fts3Table *p = (Fts3Table *)pVTab;
1607 int i; /* Iterator variable */
1608 int iCons = -1; /* Index of constraint to use */
1610 int iLangidCons = -1; /* Index of langid=x constraint, if present */
1611 int iDocidGe = -1; /* Index of docid>=x constraint, if present */
1612 int iDocidLe = -1; /* Index of docid<=x constraint, if present */
1613 int iIdx;
1615 if( p->bLock ){
1616 return SQLITE_ERROR;
1619 /* By default use a full table scan. This is an expensive option,
1620 ** so search through the constraints to see if a more efficient
1621 ** strategy is possible.
1623 pInfo->idxNum = FTS3_FULLSCAN_SEARCH;
1624 pInfo->estimatedCost = 5000000;
1625 for(i=0; i<pInfo->nConstraint; i++){
1626 int bDocid; /* True if this constraint is on docid */
1627 struct sqlite3_index_constraint *pCons = &pInfo->aConstraint[i];
1628 if( pCons->usable==0 ){
1629 if( pCons->op==SQLITE_INDEX_CONSTRAINT_MATCH ){
1630 /* There exists an unusable MATCH constraint. This means that if
1631 ** the planner does elect to use the results of this call as part
1632 ** of the overall query plan the user will see an "unable to use
1633 ** function MATCH in the requested context" error. To discourage
1634 ** this, return a very high cost here. */
1635 pInfo->idxNum = FTS3_FULLSCAN_SEARCH;
1636 pInfo->estimatedCost = 1e50;
1637 fts3SetEstimatedRows(pInfo, ((sqlite3_int64)1) << 50);
1638 return SQLITE_OK;
1640 continue;
1643 bDocid = (pCons->iColumn<0 || pCons->iColumn==p->nColumn+1);
1645 /* A direct lookup on the rowid or docid column. Assign a cost of 1.0. */
1646 if( iCons<0 && pCons->op==SQLITE_INDEX_CONSTRAINT_EQ && bDocid ){
1647 pInfo->idxNum = FTS3_DOCID_SEARCH;
1648 pInfo->estimatedCost = 1.0;
1649 iCons = i;
1652 /* A MATCH constraint. Use a full-text search.
1654 ** If there is more than one MATCH constraint available, use the first
1655 ** one encountered. If there is both a MATCH constraint and a direct
1656 ** rowid/docid lookup, prefer the MATCH strategy. This is done even
1657 ** though the rowid/docid lookup is faster than a MATCH query, selecting
1658 ** it would lead to an "unable to use function MATCH in the requested
1659 ** context" error.
1661 if( pCons->op==SQLITE_INDEX_CONSTRAINT_MATCH
1662 && pCons->iColumn>=0 && pCons->iColumn<=p->nColumn
1664 pInfo->idxNum = FTS3_FULLTEXT_SEARCH + pCons->iColumn;
1665 pInfo->estimatedCost = 2.0;
1666 iCons = i;
1669 /* Equality constraint on the langid column */
1670 if( pCons->op==SQLITE_INDEX_CONSTRAINT_EQ
1671 && pCons->iColumn==p->nColumn + 2
1673 iLangidCons = i;
1676 if( bDocid ){
1677 switch( pCons->op ){
1678 case SQLITE_INDEX_CONSTRAINT_GE:
1679 case SQLITE_INDEX_CONSTRAINT_GT:
1680 iDocidGe = i;
1681 break;
1683 case SQLITE_INDEX_CONSTRAINT_LE:
1684 case SQLITE_INDEX_CONSTRAINT_LT:
1685 iDocidLe = i;
1686 break;
1691 /* If using a docid=? or rowid=? strategy, set the UNIQUE flag. */
1692 if( pInfo->idxNum==FTS3_DOCID_SEARCH ) fts3SetUniqueFlag(pInfo);
1694 iIdx = 1;
1695 if( iCons>=0 ){
1696 pInfo->aConstraintUsage[iCons].argvIndex = iIdx++;
1697 pInfo->aConstraintUsage[iCons].omit = 1;
1699 if( iLangidCons>=0 ){
1700 pInfo->idxNum |= FTS3_HAVE_LANGID;
1701 pInfo->aConstraintUsage[iLangidCons].argvIndex = iIdx++;
1703 if( iDocidGe>=0 ){
1704 pInfo->idxNum |= FTS3_HAVE_DOCID_GE;
1705 pInfo->aConstraintUsage[iDocidGe].argvIndex = iIdx++;
1707 if( iDocidLe>=0 ){
1708 pInfo->idxNum |= FTS3_HAVE_DOCID_LE;
1709 pInfo->aConstraintUsage[iDocidLe].argvIndex = iIdx++;
1712 /* Regardless of the strategy selected, FTS can deliver rows in rowid (or
1713 ** docid) order. Both ascending and descending are possible.
1715 if( pInfo->nOrderBy==1 ){
1716 struct sqlite3_index_orderby *pOrder = &pInfo->aOrderBy[0];
1717 if( pOrder->iColumn<0 || pOrder->iColumn==p->nColumn+1 ){
1718 if( pOrder->desc ){
1719 pInfo->idxStr = "DESC";
1720 }else{
1721 pInfo->idxStr = "ASC";
1723 pInfo->orderByConsumed = 1;
1727 assert( p->pSegments==0 );
1728 return SQLITE_OK;
1732 ** Implementation of xOpen method.
1734 static int fts3OpenMethod(sqlite3_vtab *pVTab, sqlite3_vtab_cursor **ppCsr){
1735 sqlite3_vtab_cursor *pCsr; /* Allocated cursor */
1737 UNUSED_PARAMETER(pVTab);
1739 /* Allocate a buffer large enough for an Fts3Cursor structure. If the
1740 ** allocation succeeds, zero it and return SQLITE_OK. Otherwise,
1741 ** if the allocation fails, return SQLITE_NOMEM.
1743 *ppCsr = pCsr = (sqlite3_vtab_cursor *)sqlite3_malloc(sizeof(Fts3Cursor));
1744 if( !pCsr ){
1745 return SQLITE_NOMEM;
1747 memset(pCsr, 0, sizeof(Fts3Cursor));
1748 return SQLITE_OK;
1752 ** Finalize the statement handle at pCsr->pStmt.
1754 ** Or, if that statement handle is one created by fts3CursorSeekStmt(),
1755 ** and the Fts3Table.pSeekStmt slot is currently NULL, save the statement
1756 ** pointer there instead of finalizing it.
1758 static void fts3CursorFinalizeStmt(Fts3Cursor *pCsr){
1759 if( pCsr->bSeekStmt ){
1760 Fts3Table *p = (Fts3Table *)pCsr->base.pVtab;
1761 if( p->pSeekStmt==0 ){
1762 p->pSeekStmt = pCsr->pStmt;
1763 sqlite3_reset(pCsr->pStmt);
1764 pCsr->pStmt = 0;
1766 pCsr->bSeekStmt = 0;
1768 sqlite3_finalize(pCsr->pStmt);
1772 ** Free all resources currently held by the cursor passed as the only
1773 ** argument.
1775 static void fts3ClearCursor(Fts3Cursor *pCsr){
1776 fts3CursorFinalizeStmt(pCsr);
1777 sqlite3Fts3FreeDeferredTokens(pCsr);
1778 sqlite3_free(pCsr->aDoclist);
1779 sqlite3Fts3MIBufferFree(pCsr->pMIBuffer);
1780 sqlite3Fts3ExprFree(pCsr->pExpr);
1781 memset(&(&pCsr->base)[1], 0, sizeof(Fts3Cursor)-sizeof(sqlite3_vtab_cursor));
1785 ** Close the cursor. For additional information see the documentation
1786 ** on the xClose method of the virtual table interface.
1788 static int fts3CloseMethod(sqlite3_vtab_cursor *pCursor){
1789 Fts3Cursor *pCsr = (Fts3Cursor *)pCursor;
1790 assert( ((Fts3Table *)pCsr->base.pVtab)->pSegments==0 );
1791 fts3ClearCursor(pCsr);
1792 assert( ((Fts3Table *)pCsr->base.pVtab)->pSegments==0 );
1793 sqlite3_free(pCsr);
1794 return SQLITE_OK;
1798 ** If pCsr->pStmt has not been prepared (i.e. if pCsr->pStmt==0), then
1799 ** compose and prepare an SQL statement of the form:
1801 ** "SELECT <columns> FROM %_content WHERE rowid = ?"
1803 ** (or the equivalent for a content=xxx table) and set pCsr->pStmt to
1804 ** it. If an error occurs, return an SQLite error code.
1806 static int fts3CursorSeekStmt(Fts3Cursor *pCsr){
1807 int rc = SQLITE_OK;
1808 if( pCsr->pStmt==0 ){
1809 Fts3Table *p = (Fts3Table *)pCsr->base.pVtab;
1810 char *zSql;
1811 if( p->pSeekStmt ){
1812 pCsr->pStmt = p->pSeekStmt;
1813 p->pSeekStmt = 0;
1814 }else{
1815 zSql = sqlite3_mprintf("SELECT %s WHERE rowid = ?", p->zReadExprlist);
1816 if( !zSql ) return SQLITE_NOMEM;
1817 p->bLock++;
1818 rc = sqlite3_prepare_v3(
1819 p->db, zSql,-1,SQLITE_PREPARE_PERSISTENT,&pCsr->pStmt,0
1821 p->bLock--;
1822 sqlite3_free(zSql);
1824 if( rc==SQLITE_OK ) pCsr->bSeekStmt = 1;
1826 return rc;
1830 ** Position the pCsr->pStmt statement so that it is on the row
1831 ** of the %_content table that contains the last match. Return
1832 ** SQLITE_OK on success.
1834 static int fts3CursorSeek(sqlite3_context *pContext, Fts3Cursor *pCsr){
1835 int rc = SQLITE_OK;
1836 if( pCsr->isRequireSeek ){
1837 rc = fts3CursorSeekStmt(pCsr);
1838 if( rc==SQLITE_OK ){
1839 Fts3Table *pTab = (Fts3Table*)pCsr->base.pVtab;
1840 pTab->bLock++;
1841 sqlite3_bind_int64(pCsr->pStmt, 1, pCsr->iPrevId);
1842 pCsr->isRequireSeek = 0;
1843 if( SQLITE_ROW==sqlite3_step(pCsr->pStmt) ){
1844 pTab->bLock--;
1845 return SQLITE_OK;
1846 }else{
1847 pTab->bLock--;
1848 rc = sqlite3_reset(pCsr->pStmt);
1849 if( rc==SQLITE_OK && ((Fts3Table *)pCsr->base.pVtab)->zContentTbl==0 ){
1850 /* If no row was found and no error has occurred, then the %_content
1851 ** table is missing a row that is present in the full-text index.
1852 ** The data structures are corrupt. */
1853 rc = FTS_CORRUPT_VTAB;
1854 pCsr->isEof = 1;
1860 if( rc!=SQLITE_OK && pContext ){
1861 sqlite3_result_error_code(pContext, rc);
1863 return rc;
1867 ** This function is used to process a single interior node when searching
1868 ** a b-tree for a term or term prefix. The node data is passed to this
1869 ** function via the zNode/nNode parameters. The term to search for is
1870 ** passed in zTerm/nTerm.
1872 ** If piFirst is not NULL, then this function sets *piFirst to the blockid
1873 ** of the child node that heads the sub-tree that may contain the term.
1875 ** If piLast is not NULL, then *piLast is set to the right-most child node
1876 ** that heads a sub-tree that may contain a term for which zTerm/nTerm is
1877 ** a prefix.
1879 ** If an OOM error occurs, SQLITE_NOMEM is returned. Otherwise, SQLITE_OK.
1881 static int fts3ScanInteriorNode(
1882 const char *zTerm, /* Term to select leaves for */
1883 int nTerm, /* Size of term zTerm in bytes */
1884 const char *zNode, /* Buffer containing segment interior node */
1885 int nNode, /* Size of buffer at zNode */
1886 sqlite3_int64 *piFirst, /* OUT: Selected child node */
1887 sqlite3_int64 *piLast /* OUT: Selected child node */
1889 int rc = SQLITE_OK; /* Return code */
1890 const char *zCsr = zNode; /* Cursor to iterate through node */
1891 const char *zEnd = &zCsr[nNode];/* End of interior node buffer */
1892 char *zBuffer = 0; /* Buffer to load terms into */
1893 i64 nAlloc = 0; /* Size of allocated buffer */
1894 int isFirstTerm = 1; /* True when processing first term on page */
1895 u64 iChild; /* Block id of child node to descend to */
1896 int nBuffer = 0; /* Total term size */
1898 /* Skip over the 'height' varint that occurs at the start of every
1899 ** interior node. Then load the blockid of the left-child of the b-tree
1900 ** node into variable iChild.
1902 ** Even if the data structure on disk is corrupted, this (reading two
1903 ** varints from the buffer) does not risk an overread. If zNode is a
1904 ** root node, then the buffer comes from a SELECT statement. SQLite does
1905 ** not make this guarantee explicitly, but in practice there are always
1906 ** either more than 20 bytes of allocated space following the nNode bytes of
1907 ** contents, or two zero bytes. Or, if the node is read from the %_segments
1908 ** table, then there are always 20 bytes of zeroed padding following the
1909 ** nNode bytes of content (see sqlite3Fts3ReadBlock() for details).
1911 zCsr += sqlite3Fts3GetVarintU(zCsr, &iChild);
1912 zCsr += sqlite3Fts3GetVarintU(zCsr, &iChild);
1913 if( zCsr>zEnd ){
1914 return FTS_CORRUPT_VTAB;
1917 while( zCsr<zEnd && (piFirst || piLast) ){
1918 int cmp; /* memcmp() result */
1919 int nSuffix; /* Size of term suffix */
1920 int nPrefix = 0; /* Size of term prefix */
1922 /* Load the next term on the node into zBuffer. Use realloc() to expand
1923 ** the size of zBuffer if required. */
1924 if( !isFirstTerm ){
1925 zCsr += fts3GetVarint32(zCsr, &nPrefix);
1926 if( nPrefix>nBuffer ){
1927 rc = FTS_CORRUPT_VTAB;
1928 goto finish_scan;
1931 isFirstTerm = 0;
1932 zCsr += fts3GetVarint32(zCsr, &nSuffix);
1934 assert( nPrefix>=0 && nSuffix>=0 );
1935 if( nPrefix>zCsr-zNode || nSuffix>zEnd-zCsr || nSuffix==0 ){
1936 rc = FTS_CORRUPT_VTAB;
1937 goto finish_scan;
1939 if( (i64)nPrefix+nSuffix>nAlloc ){
1940 char *zNew;
1941 nAlloc = ((i64)nPrefix+nSuffix) * 2;
1942 zNew = (char *)sqlite3_realloc64(zBuffer, nAlloc);
1943 if( !zNew ){
1944 rc = SQLITE_NOMEM;
1945 goto finish_scan;
1947 zBuffer = zNew;
1949 assert( zBuffer );
1950 memcpy(&zBuffer[nPrefix], zCsr, nSuffix);
1951 nBuffer = nPrefix + nSuffix;
1952 zCsr += nSuffix;
1954 /* Compare the term we are searching for with the term just loaded from
1955 ** the interior node. If the specified term is greater than or equal
1956 ** to the term from the interior node, then all terms on the sub-tree
1957 ** headed by node iChild are smaller than zTerm. No need to search
1958 ** iChild.
1960 ** If the interior node term is larger than the specified term, then
1961 ** the tree headed by iChild may contain the specified term.
1963 cmp = memcmp(zTerm, zBuffer, (nBuffer>nTerm ? nTerm : nBuffer));
1964 if( piFirst && (cmp<0 || (cmp==0 && nBuffer>nTerm)) ){
1965 *piFirst = (i64)iChild;
1966 piFirst = 0;
1969 if( piLast && cmp<0 ){
1970 *piLast = (i64)iChild;
1971 piLast = 0;
1974 iChild++;
1977 if( piFirst ) *piFirst = (i64)iChild;
1978 if( piLast ) *piLast = (i64)iChild;
1980 finish_scan:
1981 sqlite3_free(zBuffer);
1982 return rc;
1987 ** The buffer pointed to by argument zNode (size nNode bytes) contains an
1988 ** interior node of a b-tree segment. The zTerm buffer (size nTerm bytes)
1989 ** contains a term. This function searches the sub-tree headed by the zNode
1990 ** node for the range of leaf nodes that may contain the specified term
1991 ** or terms for which the specified term is a prefix.
1993 ** If piLeaf is not NULL, then *piLeaf is set to the blockid of the
1994 ** left-most leaf node in the tree that may contain the specified term.
1995 ** If piLeaf2 is not NULL, then *piLeaf2 is set to the blockid of the
1996 ** right-most leaf node that may contain a term for which the specified
1997 ** term is a prefix.
1999 ** It is possible that the range of returned leaf nodes does not contain
2000 ** the specified term or any terms for which it is a prefix. However, if the
2001 ** segment does contain any such terms, they are stored within the identified
2002 ** range. Because this function only inspects interior segment nodes (and
2003 ** never loads leaf nodes into memory), it is not possible to be sure.
2005 ** If an error occurs, an error code other than SQLITE_OK is returned.
2007 static int fts3SelectLeaf(
2008 Fts3Table *p, /* Virtual table handle */
2009 const char *zTerm, /* Term to select leaves for */
2010 int nTerm, /* Size of term zTerm in bytes */
2011 const char *zNode, /* Buffer containing segment interior node */
2012 int nNode, /* Size of buffer at zNode */
2013 sqlite3_int64 *piLeaf, /* Selected leaf node */
2014 sqlite3_int64 *piLeaf2 /* Selected leaf node */
2016 int rc = SQLITE_OK; /* Return code */
2017 int iHeight; /* Height of this node in tree */
2019 assert( piLeaf || piLeaf2 );
2021 fts3GetVarint32(zNode, &iHeight);
2022 rc = fts3ScanInteriorNode(zTerm, nTerm, zNode, nNode, piLeaf, piLeaf2);
2023 assert_fts3_nc( !piLeaf2 || !piLeaf || rc!=SQLITE_OK || (*piLeaf<=*piLeaf2) );
2025 if( rc==SQLITE_OK && iHeight>1 ){
2026 char *zBlob = 0; /* Blob read from %_segments table */
2027 int nBlob = 0; /* Size of zBlob in bytes */
2029 if( piLeaf && piLeaf2 && (*piLeaf!=*piLeaf2) ){
2030 rc = sqlite3Fts3ReadBlock(p, *piLeaf, &zBlob, &nBlob, 0);
2031 if( rc==SQLITE_OK ){
2032 rc = fts3SelectLeaf(p, zTerm, nTerm, zBlob, nBlob, piLeaf, 0);
2034 sqlite3_free(zBlob);
2035 piLeaf = 0;
2036 zBlob = 0;
2039 if( rc==SQLITE_OK ){
2040 rc = sqlite3Fts3ReadBlock(p, piLeaf?*piLeaf:*piLeaf2, &zBlob, &nBlob, 0);
2042 if( rc==SQLITE_OK ){
2043 int iNewHeight = 0;
2044 fts3GetVarint32(zBlob, &iNewHeight);
2045 if( iNewHeight>=iHeight ){
2046 rc = FTS_CORRUPT_VTAB;
2047 }else{
2048 rc = fts3SelectLeaf(p, zTerm, nTerm, zBlob, nBlob, piLeaf, piLeaf2);
2051 sqlite3_free(zBlob);
2054 return rc;
2058 ** This function is used to create delta-encoded serialized lists of FTS3
2059 ** varints. Each call to this function appends a single varint to a list.
2061 static void fts3PutDeltaVarint(
2062 char **pp, /* IN/OUT: Output pointer */
2063 sqlite3_int64 *piPrev, /* IN/OUT: Previous value written to list */
2064 sqlite3_int64 iVal /* Write this value to the list */
2066 assert_fts3_nc( iVal-*piPrev > 0 || (*piPrev==0 && iVal==0) );
2067 *pp += sqlite3Fts3PutVarint(*pp, iVal-*piPrev);
2068 *piPrev = iVal;
2072 ** When this function is called, *ppPoslist is assumed to point to the
2073 ** start of a position-list. After it returns, *ppPoslist points to the
2074 ** first byte after the position-list.
2076 ** A position list is list of positions (delta encoded) and columns for
2077 ** a single document record of a doclist. So, in other words, this
2078 ** routine advances *ppPoslist so that it points to the next docid in
2079 ** the doclist, or to the first byte past the end of the doclist.
2081 ** If pp is not NULL, then the contents of the position list are copied
2082 ** to *pp. *pp is set to point to the first byte past the last byte copied
2083 ** before this function returns.
2085 static void fts3PoslistCopy(char **pp, char **ppPoslist){
2086 char *pEnd = *ppPoslist;
2087 char c = 0;
2089 /* The end of a position list is marked by a zero encoded as an FTS3
2090 ** varint. A single POS_END (0) byte. Except, if the 0 byte is preceded by
2091 ** a byte with the 0x80 bit set, then it is not a varint 0, but the tail
2092 ** of some other, multi-byte, value.
2094 ** The following while-loop moves pEnd to point to the first byte that is not
2095 ** immediately preceded by a byte with the 0x80 bit set. Then increments
2096 ** pEnd once more so that it points to the byte immediately following the
2097 ** last byte in the position-list.
2099 while( *pEnd | c ){
2100 c = *pEnd++ & 0x80;
2101 testcase( c!=0 && (*pEnd)==0 );
2103 pEnd++; /* Advance past the POS_END terminator byte */
2105 if( pp ){
2106 int n = (int)(pEnd - *ppPoslist);
2107 char *p = *pp;
2108 memcpy(p, *ppPoslist, n);
2109 p += n;
2110 *pp = p;
2112 *ppPoslist = pEnd;
2116 ** When this function is called, *ppPoslist is assumed to point to the
2117 ** start of a column-list. After it returns, *ppPoslist points to the
2118 ** to the terminator (POS_COLUMN or POS_END) byte of the column-list.
2120 ** A column-list is list of delta-encoded positions for a single column
2121 ** within a single document within a doclist.
2123 ** The column-list is terminated either by a POS_COLUMN varint (1) or
2124 ** a POS_END varint (0). This routine leaves *ppPoslist pointing to
2125 ** the POS_COLUMN or POS_END that terminates the column-list.
2127 ** If pp is not NULL, then the contents of the column-list are copied
2128 ** to *pp. *pp is set to point to the first byte past the last byte copied
2129 ** before this function returns. The POS_COLUMN or POS_END terminator
2130 ** is not copied into *pp.
2132 static void fts3ColumnlistCopy(char **pp, char **ppPoslist){
2133 char *pEnd = *ppPoslist;
2134 char c = 0;
2136 /* A column-list is terminated by either a 0x01 or 0x00 byte that is
2137 ** not part of a multi-byte varint.
2139 while( 0xFE & (*pEnd | c) ){
2140 c = *pEnd++ & 0x80;
2141 testcase( c!=0 && ((*pEnd)&0xfe)==0 );
2143 if( pp ){
2144 int n = (int)(pEnd - *ppPoslist);
2145 char *p = *pp;
2146 memcpy(p, *ppPoslist, n);
2147 p += n;
2148 *pp = p;
2150 *ppPoslist = pEnd;
2154 ** Value used to signify the end of an position-list. This must be
2155 ** as large or larger than any value that might appear on the
2156 ** position-list, even a position list that has been corrupted.
2158 #define POSITION_LIST_END LARGEST_INT64
2161 ** This function is used to help parse position-lists. When this function is
2162 ** called, *pp may point to the start of the next varint in the position-list
2163 ** being parsed, or it may point to 1 byte past the end of the position-list
2164 ** (in which case **pp will be a terminator bytes POS_END (0) or
2165 ** (1)).
2167 ** If *pp points past the end of the current position-list, set *pi to
2168 ** POSITION_LIST_END and return. Otherwise, read the next varint from *pp,
2169 ** increment the current value of *pi by the value read, and set *pp to
2170 ** point to the next value before returning.
2172 ** Before calling this routine *pi must be initialized to the value of
2173 ** the previous position, or zero if we are reading the first position
2174 ** in the position-list. Because positions are delta-encoded, the value
2175 ** of the previous position is needed in order to compute the value of
2176 ** the next position.
2178 static void fts3ReadNextPos(
2179 char **pp, /* IN/OUT: Pointer into position-list buffer */
2180 sqlite3_int64 *pi /* IN/OUT: Value read from position-list */
2182 if( (**pp)&0xFE ){
2183 int iVal;
2184 *pp += fts3GetVarint32((*pp), &iVal);
2185 *pi += iVal;
2186 *pi -= 2;
2187 }else{
2188 *pi = POSITION_LIST_END;
2193 ** If parameter iCol is not 0, write an POS_COLUMN (1) byte followed by
2194 ** the value of iCol encoded as a varint to *pp. This will start a new
2195 ** column list.
2197 ** Set *pp to point to the byte just after the last byte written before
2198 ** returning (do not modify it if iCol==0). Return the total number of bytes
2199 ** written (0 if iCol==0).
2201 static int fts3PutColNumber(char **pp, int iCol){
2202 int n = 0; /* Number of bytes written */
2203 if( iCol ){
2204 char *p = *pp; /* Output pointer */
2205 n = 1 + sqlite3Fts3PutVarint(&p[1], iCol);
2206 *p = 0x01;
2207 *pp = &p[n];
2209 return n;
2213 ** Compute the union of two position lists. The output written
2214 ** into *pp contains all positions of both *pp1 and *pp2 in sorted
2215 ** order and with any duplicates removed. All pointers are
2216 ** updated appropriately. The caller is responsible for insuring
2217 ** that there is enough space in *pp to hold the complete output.
2219 static int fts3PoslistMerge(
2220 char **pp, /* Output buffer */
2221 char **pp1, /* Left input list */
2222 char **pp2 /* Right input list */
2224 char *p = *pp;
2225 char *p1 = *pp1;
2226 char *p2 = *pp2;
2228 while( *p1 || *p2 ){
2229 int iCol1; /* The current column index in pp1 */
2230 int iCol2; /* The current column index in pp2 */
2232 if( *p1==POS_COLUMN ){
2233 fts3GetVarint32(&p1[1], &iCol1);
2234 if( iCol1==0 ) return FTS_CORRUPT_VTAB;
2236 else if( *p1==POS_END ) iCol1 = 0x7fffffff;
2237 else iCol1 = 0;
2239 if( *p2==POS_COLUMN ){
2240 fts3GetVarint32(&p2[1], &iCol2);
2241 if( iCol2==0 ) return FTS_CORRUPT_VTAB;
2243 else if( *p2==POS_END ) iCol2 = 0x7fffffff;
2244 else iCol2 = 0;
2246 if( iCol1==iCol2 ){
2247 sqlite3_int64 i1 = 0; /* Last position from pp1 */
2248 sqlite3_int64 i2 = 0; /* Last position from pp2 */
2249 sqlite3_int64 iPrev = 0;
2250 int n = fts3PutColNumber(&p, iCol1);
2251 p1 += n;
2252 p2 += n;
2254 /* At this point, both p1 and p2 point to the start of column-lists
2255 ** for the same column (the column with index iCol1 and iCol2).
2256 ** A column-list is a list of non-negative delta-encoded varints, each
2257 ** incremented by 2 before being stored. Each list is terminated by a
2258 ** POS_END (0) or POS_COLUMN (1). The following block merges the two lists
2259 ** and writes the results to buffer p. p is left pointing to the byte
2260 ** after the list written. No terminator (POS_END or POS_COLUMN) is
2261 ** written to the output.
2263 fts3GetDeltaVarint(&p1, &i1);
2264 fts3GetDeltaVarint(&p2, &i2);
2265 if( i1<2 || i2<2 ){
2266 break;
2268 do {
2269 fts3PutDeltaVarint(&p, &iPrev, (i1<i2) ? i1 : i2);
2270 iPrev -= 2;
2271 if( i1==i2 ){
2272 fts3ReadNextPos(&p1, &i1);
2273 fts3ReadNextPos(&p2, &i2);
2274 }else if( i1<i2 ){
2275 fts3ReadNextPos(&p1, &i1);
2276 }else{
2277 fts3ReadNextPos(&p2, &i2);
2279 }while( i1!=POSITION_LIST_END || i2!=POSITION_LIST_END );
2280 }else if( iCol1<iCol2 ){
2281 p1 += fts3PutColNumber(&p, iCol1);
2282 fts3ColumnlistCopy(&p, &p1);
2283 }else{
2284 p2 += fts3PutColNumber(&p, iCol2);
2285 fts3ColumnlistCopy(&p, &p2);
2289 *p++ = POS_END;
2290 *pp = p;
2291 *pp1 = p1 + 1;
2292 *pp2 = p2 + 1;
2293 return SQLITE_OK;
2297 ** This function is used to merge two position lists into one. When it is
2298 ** called, *pp1 and *pp2 must both point to position lists. A position-list is
2299 ** the part of a doclist that follows each document id. For example, if a row
2300 ** contains:
2302 ** 'a b c'|'x y z'|'a b b a'
2304 ** Then the position list for this row for token 'b' would consist of:
2306 ** 0x02 0x01 0x02 0x03 0x03 0x00
2308 ** When this function returns, both *pp1 and *pp2 are left pointing to the
2309 ** byte following the 0x00 terminator of their respective position lists.
2311 ** If isSaveLeft is 0, an entry is added to the output position list for
2312 ** each position in *pp2 for which there exists one or more positions in
2313 ** *pp1 so that (pos(*pp2)>pos(*pp1) && pos(*pp2)-pos(*pp1)<=nToken). i.e.
2314 ** when the *pp1 token appears before the *pp2 token, but not more than nToken
2315 ** slots before it.
2317 ** e.g. nToken==1 searches for adjacent positions.
2319 static int fts3PoslistPhraseMerge(
2320 char **pp, /* IN/OUT: Preallocated output buffer */
2321 int nToken, /* Maximum difference in token positions */
2322 int isSaveLeft, /* Save the left position */
2323 int isExact, /* If *pp1 is exactly nTokens before *pp2 */
2324 char **pp1, /* IN/OUT: Left input list */
2325 char **pp2 /* IN/OUT: Right input list */
2327 char *p = *pp;
2328 char *p1 = *pp1;
2329 char *p2 = *pp2;
2330 int iCol1 = 0;
2331 int iCol2 = 0;
2333 /* Never set both isSaveLeft and isExact for the same invocation. */
2334 assert( isSaveLeft==0 || isExact==0 );
2336 assert_fts3_nc( p!=0 && *p1!=0 && *p2!=0 );
2337 if( *p1==POS_COLUMN ){
2338 p1++;
2339 p1 += fts3GetVarint32(p1, &iCol1);
2341 if( *p2==POS_COLUMN ){
2342 p2++;
2343 p2 += fts3GetVarint32(p2, &iCol2);
2346 while( 1 ){
2347 if( iCol1==iCol2 ){
2348 char *pSave = p;
2349 sqlite3_int64 iPrev = 0;
2350 sqlite3_int64 iPos1 = 0;
2351 sqlite3_int64 iPos2 = 0;
2353 if( iCol1 ){
2354 *p++ = POS_COLUMN;
2355 p += sqlite3Fts3PutVarint(p, iCol1);
2358 fts3GetDeltaVarint(&p1, &iPos1); iPos1 -= 2;
2359 fts3GetDeltaVarint(&p2, &iPos2); iPos2 -= 2;
2360 if( iPos1<0 || iPos2<0 ) break;
2362 while( 1 ){
2363 if( iPos2==iPos1+nToken
2364 || (isExact==0 && iPos2>iPos1 && iPos2<=iPos1+nToken)
2366 sqlite3_int64 iSave;
2367 iSave = isSaveLeft ? iPos1 : iPos2;
2368 fts3PutDeltaVarint(&p, &iPrev, iSave+2); iPrev -= 2;
2369 pSave = 0;
2370 assert( p );
2372 if( (!isSaveLeft && iPos2<=(iPos1+nToken)) || iPos2<=iPos1 ){
2373 if( (*p2&0xFE)==0 ) break;
2374 fts3GetDeltaVarint(&p2, &iPos2); iPos2 -= 2;
2375 }else{
2376 if( (*p1&0xFE)==0 ) break;
2377 fts3GetDeltaVarint(&p1, &iPos1); iPos1 -= 2;
2381 if( pSave ){
2382 assert( pp && p );
2383 p = pSave;
2386 fts3ColumnlistCopy(0, &p1);
2387 fts3ColumnlistCopy(0, &p2);
2388 assert( (*p1&0xFE)==0 && (*p2&0xFE)==0 );
2389 if( 0==*p1 || 0==*p2 ) break;
2391 p1++;
2392 p1 += fts3GetVarint32(p1, &iCol1);
2393 p2++;
2394 p2 += fts3GetVarint32(p2, &iCol2);
2397 /* Advance pointer p1 or p2 (whichever corresponds to the smaller of
2398 ** iCol1 and iCol2) so that it points to either the 0x00 that marks the
2399 ** end of the position list, or the 0x01 that precedes the next
2400 ** column-number in the position list.
2402 else if( iCol1<iCol2 ){
2403 fts3ColumnlistCopy(0, &p1);
2404 if( 0==*p1 ) break;
2405 p1++;
2406 p1 += fts3GetVarint32(p1, &iCol1);
2407 }else{
2408 fts3ColumnlistCopy(0, &p2);
2409 if( 0==*p2 ) break;
2410 p2++;
2411 p2 += fts3GetVarint32(p2, &iCol2);
2415 fts3PoslistCopy(0, &p2);
2416 fts3PoslistCopy(0, &p1);
2417 *pp1 = p1;
2418 *pp2 = p2;
2419 if( *pp==p ){
2420 return 0;
2422 *p++ = 0x00;
2423 *pp = p;
2424 return 1;
2428 ** Merge two position-lists as required by the NEAR operator. The argument
2429 ** position lists correspond to the left and right phrases of an expression
2430 ** like:
2432 ** "phrase 1" NEAR "phrase number 2"
2434 ** Position list *pp1 corresponds to the left-hand side of the NEAR
2435 ** expression and *pp2 to the right. As usual, the indexes in the position
2436 ** lists are the offsets of the last token in each phrase (tokens "1" and "2"
2437 ** in the example above).
2439 ** The output position list - written to *pp - is a copy of *pp2 with those
2440 ** entries that are not sufficiently NEAR entries in *pp1 removed.
2442 static int fts3PoslistNearMerge(
2443 char **pp, /* Output buffer */
2444 char *aTmp, /* Temporary buffer space */
2445 int nRight, /* Maximum difference in token positions */
2446 int nLeft, /* Maximum difference in token positions */
2447 char **pp1, /* IN/OUT: Left input list */
2448 char **pp2 /* IN/OUT: Right input list */
2450 char *p1 = *pp1;
2451 char *p2 = *pp2;
2453 char *pTmp1 = aTmp;
2454 char *pTmp2;
2455 char *aTmp2;
2456 int res = 1;
2458 fts3PoslistPhraseMerge(&pTmp1, nRight, 0, 0, pp1, pp2);
2459 aTmp2 = pTmp2 = pTmp1;
2460 *pp1 = p1;
2461 *pp2 = p2;
2462 fts3PoslistPhraseMerge(&pTmp2, nLeft, 1, 0, pp2, pp1);
2463 if( pTmp1!=aTmp && pTmp2!=aTmp2 ){
2464 fts3PoslistMerge(pp, &aTmp, &aTmp2);
2465 }else if( pTmp1!=aTmp ){
2466 fts3PoslistCopy(pp, &aTmp);
2467 }else if( pTmp2!=aTmp2 ){
2468 fts3PoslistCopy(pp, &aTmp2);
2469 }else{
2470 res = 0;
2473 return res;
2477 ** An instance of this function is used to merge together the (potentially
2478 ** large number of) doclists for each term that matches a prefix query.
2479 ** See function fts3TermSelectMerge() for details.
2481 typedef struct TermSelect TermSelect;
2482 struct TermSelect {
2483 char *aaOutput[16]; /* Malloc'd output buffers */
2484 int anOutput[16]; /* Size each output buffer in bytes */
2488 ** This function is used to read a single varint from a buffer. Parameter
2489 ** pEnd points 1 byte past the end of the buffer. When this function is
2490 ** called, if *pp points to pEnd or greater, then the end of the buffer
2491 ** has been reached. In this case *pp is set to 0 and the function returns.
2493 ** If *pp does not point to or past pEnd, then a single varint is read
2494 ** from *pp. *pp is then set to point 1 byte past the end of the read varint.
2496 ** If bDescIdx is false, the value read is added to *pVal before returning.
2497 ** If it is true, the value read is subtracted from *pVal before this
2498 ** function returns.
2500 static void fts3GetDeltaVarint3(
2501 char **pp, /* IN/OUT: Point to read varint from */
2502 char *pEnd, /* End of buffer */
2503 int bDescIdx, /* True if docids are descending */
2504 sqlite3_int64 *pVal /* IN/OUT: Integer value */
2506 if( *pp>=pEnd ){
2507 *pp = 0;
2508 }else{
2509 u64 iVal;
2510 *pp += sqlite3Fts3GetVarintU(*pp, &iVal);
2511 if( bDescIdx ){
2512 *pVal = (i64)((u64)*pVal - iVal);
2513 }else{
2514 *pVal = (i64)((u64)*pVal + iVal);
2520 ** This function is used to write a single varint to a buffer. The varint
2521 ** is written to *pp. Before returning, *pp is set to point 1 byte past the
2522 ** end of the value written.
2524 ** If *pbFirst is zero when this function is called, the value written to
2525 ** the buffer is that of parameter iVal.
2527 ** If *pbFirst is non-zero when this function is called, then the value
2528 ** written is either (iVal-*piPrev) (if bDescIdx is zero) or (*piPrev-iVal)
2529 ** (if bDescIdx is non-zero).
2531 ** Before returning, this function always sets *pbFirst to 1 and *piPrev
2532 ** to the value of parameter iVal.
2534 static void fts3PutDeltaVarint3(
2535 char **pp, /* IN/OUT: Output pointer */
2536 int bDescIdx, /* True for descending docids */
2537 sqlite3_int64 *piPrev, /* IN/OUT: Previous value written to list */
2538 int *pbFirst, /* IN/OUT: True after first int written */
2539 sqlite3_int64 iVal /* Write this value to the list */
2541 sqlite3_uint64 iWrite;
2542 if( bDescIdx==0 || *pbFirst==0 ){
2543 assert_fts3_nc( *pbFirst==0 || iVal>=*piPrev );
2544 iWrite = (u64)iVal - (u64)*piPrev;
2545 }else{
2546 assert_fts3_nc( *piPrev>=iVal );
2547 iWrite = (u64)*piPrev - (u64)iVal;
2549 assert( *pbFirst || *piPrev==0 );
2550 assert_fts3_nc( *pbFirst==0 || iWrite>0 );
2551 *pp += sqlite3Fts3PutVarint(*pp, iWrite);
2552 *piPrev = iVal;
2553 *pbFirst = 1;
2558 ** This macro is used by various functions that merge doclists. The two
2559 ** arguments are 64-bit docid values. If the value of the stack variable
2560 ** bDescDoclist is 0 when this macro is invoked, then it returns (i1-i2).
2561 ** Otherwise, (i2-i1).
2563 ** Using this makes it easier to write code that can merge doclists that are
2564 ** sorted in either ascending or descending order.
2566 /* #define DOCID_CMP(i1, i2) ((bDescDoclist?-1:1) * (i64)((u64)i1-i2)) */
2567 #define DOCID_CMP(i1, i2) ((bDescDoclist?-1:1) * (i1>i2?1:((i1==i2)?0:-1)))
2570 ** This function does an "OR" merge of two doclists (output contains all
2571 ** positions contained in either argument doclist). If the docids in the
2572 ** input doclists are sorted in ascending order, parameter bDescDoclist
2573 ** should be false. If they are sorted in ascending order, it should be
2574 ** passed a non-zero value.
2576 ** If no error occurs, *paOut is set to point at an sqlite3_malloc'd buffer
2577 ** containing the output doclist and SQLITE_OK is returned. In this case
2578 ** *pnOut is set to the number of bytes in the output doclist.
2580 ** If an error occurs, an SQLite error code is returned. The output values
2581 ** are undefined in this case.
2583 static int fts3DoclistOrMerge(
2584 int bDescDoclist, /* True if arguments are desc */
2585 char *a1, int n1, /* First doclist */
2586 char *a2, int n2, /* Second doclist */
2587 char **paOut, int *pnOut /* OUT: Malloc'd doclist */
2589 int rc = SQLITE_OK;
2590 sqlite3_int64 i1 = 0;
2591 sqlite3_int64 i2 = 0;
2592 sqlite3_int64 iPrev = 0;
2593 char *pEnd1 = &a1[n1];
2594 char *pEnd2 = &a2[n2];
2595 char *p1 = a1;
2596 char *p2 = a2;
2597 char *p;
2598 char *aOut;
2599 int bFirstOut = 0;
2601 *paOut = 0;
2602 *pnOut = 0;
2604 /* Allocate space for the output. Both the input and output doclists
2605 ** are delta encoded. If they are in ascending order (bDescDoclist==0),
2606 ** then the first docid in each list is simply encoded as a varint. For
2607 ** each subsequent docid, the varint stored is the difference between the
2608 ** current and previous docid (a positive number - since the list is in
2609 ** ascending order).
2611 ** The first docid written to the output is therefore encoded using the
2612 ** same number of bytes as it is in whichever of the input lists it is
2613 ** read from. And each subsequent docid read from the same input list
2614 ** consumes either the same or less bytes as it did in the input (since
2615 ** the difference between it and the previous value in the output must
2616 ** be a positive value less than or equal to the delta value read from
2617 ** the input list). The same argument applies to all but the first docid
2618 ** read from the 'other' list. And to the contents of all position lists
2619 ** that will be copied and merged from the input to the output.
2621 ** However, if the first docid copied to the output is a negative number,
2622 ** then the encoding of the first docid from the 'other' input list may
2623 ** be larger in the output than it was in the input (since the delta value
2624 ** may be a larger positive integer than the actual docid).
2626 ** The space required to store the output is therefore the sum of the
2627 ** sizes of the two inputs, plus enough space for exactly one of the input
2628 ** docids to grow.
2630 ** A symetric argument may be made if the doclists are in descending
2631 ** order.
2633 aOut = sqlite3_malloc64((i64)n1+n2+FTS3_VARINT_MAX-1+FTS3_BUFFER_PADDING);
2634 if( !aOut ) return SQLITE_NOMEM;
2636 p = aOut;
2637 fts3GetDeltaVarint3(&p1, pEnd1, 0, &i1);
2638 fts3GetDeltaVarint3(&p2, pEnd2, 0, &i2);
2639 while( p1 || p2 ){
2640 sqlite3_int64 iDiff = DOCID_CMP(i1, i2);
2642 if( p2 && p1 && iDiff==0 ){
2643 fts3PutDeltaVarint3(&p, bDescDoclist, &iPrev, &bFirstOut, i1);
2644 rc = fts3PoslistMerge(&p, &p1, &p2);
2645 if( rc ) break;
2646 fts3GetDeltaVarint3(&p1, pEnd1, bDescDoclist, &i1);
2647 fts3GetDeltaVarint3(&p2, pEnd2, bDescDoclist, &i2);
2648 }else if( !p2 || (p1 && iDiff<0) ){
2649 fts3PutDeltaVarint3(&p, bDescDoclist, &iPrev, &bFirstOut, i1);
2650 fts3PoslistCopy(&p, &p1);
2651 fts3GetDeltaVarint3(&p1, pEnd1, bDescDoclist, &i1);
2652 }else{
2653 fts3PutDeltaVarint3(&p, bDescDoclist, &iPrev, &bFirstOut, i2);
2654 fts3PoslistCopy(&p, &p2);
2655 fts3GetDeltaVarint3(&p2, pEnd2, bDescDoclist, &i2);
2658 assert( (p-aOut)<=((p1?(p1-a1):n1)+(p2?(p2-a2):n2)+FTS3_VARINT_MAX-1) );
2661 if( rc!=SQLITE_OK ){
2662 sqlite3_free(aOut);
2663 p = aOut = 0;
2664 }else{
2665 assert( (p-aOut)<=n1+n2+FTS3_VARINT_MAX-1 );
2666 memset(&aOut[(p-aOut)], 0, FTS3_BUFFER_PADDING);
2668 *paOut = aOut;
2669 *pnOut = (int)(p-aOut);
2670 return rc;
2674 ** This function does a "phrase" merge of two doclists. In a phrase merge,
2675 ** the output contains a copy of each position from the right-hand input
2676 ** doclist for which there is a position in the left-hand input doclist
2677 ** exactly nDist tokens before it.
2679 ** If the docids in the input doclists are sorted in ascending order,
2680 ** parameter bDescDoclist should be false. If they are sorted in ascending
2681 ** order, it should be passed a non-zero value.
2683 ** The right-hand input doclist is overwritten by this function.
2685 static int fts3DoclistPhraseMerge(
2686 int bDescDoclist, /* True if arguments are desc */
2687 int nDist, /* Distance from left to right (1=adjacent) */
2688 char *aLeft, int nLeft, /* Left doclist */
2689 char **paRight, int *pnRight /* IN/OUT: Right/output doclist */
2691 sqlite3_int64 i1 = 0;
2692 sqlite3_int64 i2 = 0;
2693 sqlite3_int64 iPrev = 0;
2694 char *aRight = *paRight;
2695 char *pEnd1 = &aLeft[nLeft];
2696 char *pEnd2 = &aRight[*pnRight];
2697 char *p1 = aLeft;
2698 char *p2 = aRight;
2699 char *p;
2700 int bFirstOut = 0;
2701 char *aOut;
2703 assert( nDist>0 );
2704 if( bDescDoclist ){
2705 aOut = sqlite3_malloc64((sqlite3_int64)*pnRight + FTS3_VARINT_MAX);
2706 if( aOut==0 ) return SQLITE_NOMEM;
2707 }else{
2708 aOut = aRight;
2710 p = aOut;
2712 fts3GetDeltaVarint3(&p1, pEnd1, 0, &i1);
2713 fts3GetDeltaVarint3(&p2, pEnd2, 0, &i2);
2715 while( p1 && p2 ){
2716 sqlite3_int64 iDiff = DOCID_CMP(i1, i2);
2717 if( iDiff==0 ){
2718 char *pSave = p;
2719 sqlite3_int64 iPrevSave = iPrev;
2720 int bFirstOutSave = bFirstOut;
2722 fts3PutDeltaVarint3(&p, bDescDoclist, &iPrev, &bFirstOut, i1);
2723 if( 0==fts3PoslistPhraseMerge(&p, nDist, 0, 1, &p1, &p2) ){
2724 p = pSave;
2725 iPrev = iPrevSave;
2726 bFirstOut = bFirstOutSave;
2728 fts3GetDeltaVarint3(&p1, pEnd1, bDescDoclist, &i1);
2729 fts3GetDeltaVarint3(&p2, pEnd2, bDescDoclist, &i2);
2730 }else if( iDiff<0 ){
2731 fts3PoslistCopy(0, &p1);
2732 fts3GetDeltaVarint3(&p1, pEnd1, bDescDoclist, &i1);
2733 }else{
2734 fts3PoslistCopy(0, &p2);
2735 fts3GetDeltaVarint3(&p2, pEnd2, bDescDoclist, &i2);
2739 *pnRight = (int)(p - aOut);
2740 if( bDescDoclist ){
2741 sqlite3_free(aRight);
2742 *paRight = aOut;
2745 return SQLITE_OK;
2749 ** Argument pList points to a position list nList bytes in size. This
2750 ** function checks to see if the position list contains any entries for
2751 ** a token in position 0 (of any column). If so, it writes argument iDelta
2752 ** to the output buffer pOut, followed by a position list consisting only
2753 ** of the entries from pList at position 0, and terminated by an 0x00 byte.
2754 ** The value returned is the number of bytes written to pOut (if any).
2756 int sqlite3Fts3FirstFilter(
2757 sqlite3_int64 iDelta, /* Varint that may be written to pOut */
2758 char *pList, /* Position list (no 0x00 term) */
2759 int nList, /* Size of pList in bytes */
2760 char *pOut /* Write output here */
2762 int nOut = 0;
2763 int bWritten = 0; /* True once iDelta has been written */
2764 char *p = pList;
2765 char *pEnd = &pList[nList];
2767 if( *p!=0x01 ){
2768 if( *p==0x02 ){
2769 nOut += sqlite3Fts3PutVarint(&pOut[nOut], iDelta);
2770 pOut[nOut++] = 0x02;
2771 bWritten = 1;
2773 fts3ColumnlistCopy(0, &p);
2776 while( p<pEnd ){
2777 sqlite3_int64 iCol;
2778 p++;
2779 p += sqlite3Fts3GetVarint(p, &iCol);
2780 if( *p==0x02 ){
2781 if( bWritten==0 ){
2782 nOut += sqlite3Fts3PutVarint(&pOut[nOut], iDelta);
2783 bWritten = 1;
2785 pOut[nOut++] = 0x01;
2786 nOut += sqlite3Fts3PutVarint(&pOut[nOut], iCol);
2787 pOut[nOut++] = 0x02;
2789 fts3ColumnlistCopy(0, &p);
2791 if( bWritten ){
2792 pOut[nOut++] = 0x00;
2795 return nOut;
2800 ** Merge all doclists in the TermSelect.aaOutput[] array into a single
2801 ** doclist stored in TermSelect.aaOutput[0]. If successful, delete all
2802 ** other doclists (except the aaOutput[0] one) and return SQLITE_OK.
2804 ** If an OOM error occurs, return SQLITE_NOMEM. In this case it is
2805 ** the responsibility of the caller to free any doclists left in the
2806 ** TermSelect.aaOutput[] array.
2808 static int fts3TermSelectFinishMerge(Fts3Table *p, TermSelect *pTS){
2809 char *aOut = 0;
2810 int nOut = 0;
2811 int i;
2813 /* Loop through the doclists in the aaOutput[] array. Merge them all
2814 ** into a single doclist.
2816 for(i=0; i<SizeofArray(pTS->aaOutput); i++){
2817 if( pTS->aaOutput[i] ){
2818 if( !aOut ){
2819 aOut = pTS->aaOutput[i];
2820 nOut = pTS->anOutput[i];
2821 pTS->aaOutput[i] = 0;
2822 }else{
2823 int nNew;
2824 char *aNew;
2826 int rc = fts3DoclistOrMerge(p->bDescIdx,
2827 pTS->aaOutput[i], pTS->anOutput[i], aOut, nOut, &aNew, &nNew
2829 if( rc!=SQLITE_OK ){
2830 sqlite3_free(aOut);
2831 return rc;
2834 sqlite3_free(pTS->aaOutput[i]);
2835 sqlite3_free(aOut);
2836 pTS->aaOutput[i] = 0;
2837 aOut = aNew;
2838 nOut = nNew;
2843 pTS->aaOutput[0] = aOut;
2844 pTS->anOutput[0] = nOut;
2845 return SQLITE_OK;
2849 ** Merge the doclist aDoclist/nDoclist into the TermSelect object passed
2850 ** as the first argument. The merge is an "OR" merge (see function
2851 ** fts3DoclistOrMerge() for details).
2853 ** This function is called with the doclist for each term that matches
2854 ** a queried prefix. It merges all these doclists into one, the doclist
2855 ** for the specified prefix. Since there can be a very large number of
2856 ** doclists to merge, the merging is done pair-wise using the TermSelect
2857 ** object.
2859 ** This function returns SQLITE_OK if the merge is successful, or an
2860 ** SQLite error code (SQLITE_NOMEM) if an error occurs.
2862 static int fts3TermSelectMerge(
2863 Fts3Table *p, /* FTS table handle */
2864 TermSelect *pTS, /* TermSelect object to merge into */
2865 char *aDoclist, /* Pointer to doclist */
2866 int nDoclist /* Size of aDoclist in bytes */
2868 if( pTS->aaOutput[0]==0 ){
2869 /* If this is the first term selected, copy the doclist to the output
2870 ** buffer using memcpy().
2872 ** Add FTS3_VARINT_MAX bytes of unused space to the end of the
2873 ** allocation. This is so as to ensure that the buffer is big enough
2874 ** to hold the current doclist AND'd with any other doclist. If the
2875 ** doclists are stored in order=ASC order, this padding would not be
2876 ** required (since the size of [doclistA AND doclistB] is always less
2877 ** than or equal to the size of [doclistA] in that case). But this is
2878 ** not true for order=DESC. For example, a doclist containing (1, -1)
2879 ** may be smaller than (-1), as in the first example the -1 may be stored
2880 ** as a single-byte delta, whereas in the second it must be stored as a
2881 ** FTS3_VARINT_MAX byte varint.
2883 ** Similar padding is added in the fts3DoclistOrMerge() function.
2885 pTS->aaOutput[0] = sqlite3_malloc(nDoclist + FTS3_VARINT_MAX + 1);
2886 pTS->anOutput[0] = nDoclist;
2887 if( pTS->aaOutput[0] ){
2888 memcpy(pTS->aaOutput[0], aDoclist, nDoclist);
2889 memset(&pTS->aaOutput[0][nDoclist], 0, FTS3_VARINT_MAX);
2890 }else{
2891 return SQLITE_NOMEM;
2893 }else{
2894 char *aMerge = aDoclist;
2895 int nMerge = nDoclist;
2896 int iOut;
2898 for(iOut=0; iOut<SizeofArray(pTS->aaOutput); iOut++){
2899 if( pTS->aaOutput[iOut]==0 ){
2900 assert( iOut>0 );
2901 pTS->aaOutput[iOut] = aMerge;
2902 pTS->anOutput[iOut] = nMerge;
2903 break;
2904 }else{
2905 char *aNew;
2906 int nNew;
2908 int rc = fts3DoclistOrMerge(p->bDescIdx, aMerge, nMerge,
2909 pTS->aaOutput[iOut], pTS->anOutput[iOut], &aNew, &nNew
2911 if( rc!=SQLITE_OK ){
2912 if( aMerge!=aDoclist ) sqlite3_free(aMerge);
2913 return rc;
2916 if( aMerge!=aDoclist ) sqlite3_free(aMerge);
2917 sqlite3_free(pTS->aaOutput[iOut]);
2918 pTS->aaOutput[iOut] = 0;
2920 aMerge = aNew;
2921 nMerge = nNew;
2922 if( (iOut+1)==SizeofArray(pTS->aaOutput) ){
2923 pTS->aaOutput[iOut] = aMerge;
2924 pTS->anOutput[iOut] = nMerge;
2929 return SQLITE_OK;
2933 ** Append SegReader object pNew to the end of the pCsr->apSegment[] array.
2935 static int fts3SegReaderCursorAppend(
2936 Fts3MultiSegReader *pCsr,
2937 Fts3SegReader *pNew
2939 if( (pCsr->nSegment%16)==0 ){
2940 Fts3SegReader **apNew;
2941 sqlite3_int64 nByte = (pCsr->nSegment + 16)*sizeof(Fts3SegReader*);
2942 apNew = (Fts3SegReader **)sqlite3_realloc64(pCsr->apSegment, nByte);
2943 if( !apNew ){
2944 sqlite3Fts3SegReaderFree(pNew);
2945 return SQLITE_NOMEM;
2947 pCsr->apSegment = apNew;
2949 pCsr->apSegment[pCsr->nSegment++] = pNew;
2950 return SQLITE_OK;
2954 ** Add seg-reader objects to the Fts3MultiSegReader object passed as the
2955 ** 8th argument.
2957 ** This function returns SQLITE_OK if successful, or an SQLite error code
2958 ** otherwise.
2960 static int fts3SegReaderCursor(
2961 Fts3Table *p, /* FTS3 table handle */
2962 int iLangid, /* Language id */
2963 int iIndex, /* Index to search (from 0 to p->nIndex-1) */
2964 int iLevel, /* Level of segments to scan */
2965 const char *zTerm, /* Term to query for */
2966 int nTerm, /* Size of zTerm in bytes */
2967 int isPrefix, /* True for a prefix search */
2968 int isScan, /* True to scan from zTerm to EOF */
2969 Fts3MultiSegReader *pCsr /* Cursor object to populate */
2971 int rc = SQLITE_OK; /* Error code */
2972 sqlite3_stmt *pStmt = 0; /* Statement to iterate through segments */
2973 int rc2; /* Result of sqlite3_reset() */
2975 /* If iLevel is less than 0 and this is not a scan, include a seg-reader
2976 ** for the pending-terms. If this is a scan, then this call must be being
2977 ** made by an fts4aux module, not an FTS table. In this case calling
2978 ** Fts3SegReaderPending might segfault, as the data structures used by
2979 ** fts4aux are not completely populated. So it's easiest to filter these
2980 ** calls out here. */
2981 if( iLevel<0 && p->aIndex && p->iPrevLangid==iLangid ){
2982 Fts3SegReader *pSeg = 0;
2983 rc = sqlite3Fts3SegReaderPending(p, iIndex, zTerm, nTerm, isPrefix||isScan, &pSeg);
2984 if( rc==SQLITE_OK && pSeg ){
2985 rc = fts3SegReaderCursorAppend(pCsr, pSeg);
2989 if( iLevel!=FTS3_SEGCURSOR_PENDING ){
2990 if( rc==SQLITE_OK ){
2991 rc = sqlite3Fts3AllSegdirs(p, iLangid, iIndex, iLevel, &pStmt);
2994 while( rc==SQLITE_OK && SQLITE_ROW==(rc = sqlite3_step(pStmt)) ){
2995 Fts3SegReader *pSeg = 0;
2997 /* Read the values returned by the SELECT into local variables. */
2998 sqlite3_int64 iStartBlock = sqlite3_column_int64(pStmt, 1);
2999 sqlite3_int64 iLeavesEndBlock = sqlite3_column_int64(pStmt, 2);
3000 sqlite3_int64 iEndBlock = sqlite3_column_int64(pStmt, 3);
3001 int nRoot = sqlite3_column_bytes(pStmt, 4);
3002 char const *zRoot = sqlite3_column_blob(pStmt, 4);
3004 /* If zTerm is not NULL, and this segment is not stored entirely on its
3005 ** root node, the range of leaves scanned can be reduced. Do this. */
3006 if( iStartBlock && zTerm && zRoot ){
3007 sqlite3_int64 *pi = (isPrefix ? &iLeavesEndBlock : 0);
3008 rc = fts3SelectLeaf(p, zTerm, nTerm, zRoot, nRoot, &iStartBlock, pi);
3009 if( rc!=SQLITE_OK ) goto finished;
3010 if( isPrefix==0 && isScan==0 ) iLeavesEndBlock = iStartBlock;
3013 rc = sqlite3Fts3SegReaderNew(pCsr->nSegment+1,
3014 (isPrefix==0 && isScan==0),
3015 iStartBlock, iLeavesEndBlock,
3016 iEndBlock, zRoot, nRoot, &pSeg
3018 if( rc!=SQLITE_OK ) goto finished;
3019 rc = fts3SegReaderCursorAppend(pCsr, pSeg);
3023 finished:
3024 rc2 = sqlite3_reset(pStmt);
3025 if( rc==SQLITE_DONE ) rc = rc2;
3027 return rc;
3031 ** Set up a cursor object for iterating through a full-text index or a
3032 ** single level therein.
3034 int sqlite3Fts3SegReaderCursor(
3035 Fts3Table *p, /* FTS3 table handle */
3036 int iLangid, /* Language-id to search */
3037 int iIndex, /* Index to search (from 0 to p->nIndex-1) */
3038 int iLevel, /* Level of segments to scan */
3039 const char *zTerm, /* Term to query for */
3040 int nTerm, /* Size of zTerm in bytes */
3041 int isPrefix, /* True for a prefix search */
3042 int isScan, /* True to scan from zTerm to EOF */
3043 Fts3MultiSegReader *pCsr /* Cursor object to populate */
3045 assert( iIndex>=0 && iIndex<p->nIndex );
3046 assert( iLevel==FTS3_SEGCURSOR_ALL
3047 || iLevel==FTS3_SEGCURSOR_PENDING
3048 || iLevel>=0
3050 assert( iLevel<FTS3_SEGDIR_MAXLEVEL );
3051 assert( FTS3_SEGCURSOR_ALL<0 && FTS3_SEGCURSOR_PENDING<0 );
3052 assert( isPrefix==0 || isScan==0 );
3054 memset(pCsr, 0, sizeof(Fts3MultiSegReader));
3055 return fts3SegReaderCursor(
3056 p, iLangid, iIndex, iLevel, zTerm, nTerm, isPrefix, isScan, pCsr
3061 ** In addition to its current configuration, have the Fts3MultiSegReader
3062 ** passed as the 4th argument also scan the doclist for term zTerm/nTerm.
3064 ** SQLITE_OK is returned if no error occurs, otherwise an SQLite error code.
3066 static int fts3SegReaderCursorAddZero(
3067 Fts3Table *p, /* FTS virtual table handle */
3068 int iLangid,
3069 const char *zTerm, /* Term to scan doclist of */
3070 int nTerm, /* Number of bytes in zTerm */
3071 Fts3MultiSegReader *pCsr /* Fts3MultiSegReader to modify */
3073 return fts3SegReaderCursor(p,
3074 iLangid, 0, FTS3_SEGCURSOR_ALL, zTerm, nTerm, 0, 0,pCsr
3079 ** Open an Fts3MultiSegReader to scan the doclist for term zTerm/nTerm. Or,
3080 ** if isPrefix is true, to scan the doclist for all terms for which
3081 ** zTerm/nTerm is a prefix. If successful, return SQLITE_OK and write
3082 ** a pointer to the new Fts3MultiSegReader to *ppSegcsr. Otherwise, return
3083 ** an SQLite error code.
3085 ** It is the responsibility of the caller to free this object by eventually
3086 ** passing it to fts3SegReaderCursorFree()
3088 ** SQLITE_OK is returned if no error occurs, otherwise an SQLite error code.
3089 ** Output parameter *ppSegcsr is set to 0 if an error occurs.
3091 static int fts3TermSegReaderCursor(
3092 Fts3Cursor *pCsr, /* Virtual table cursor handle */
3093 const char *zTerm, /* Term to query for */
3094 int nTerm, /* Size of zTerm in bytes */
3095 int isPrefix, /* True for a prefix search */
3096 Fts3MultiSegReader **ppSegcsr /* OUT: Allocated seg-reader cursor */
3098 Fts3MultiSegReader *pSegcsr; /* Object to allocate and return */
3099 int rc = SQLITE_NOMEM; /* Return code */
3101 pSegcsr = sqlite3_malloc(sizeof(Fts3MultiSegReader));
3102 if( pSegcsr ){
3103 int i;
3104 int bFound = 0; /* True once an index has been found */
3105 Fts3Table *p = (Fts3Table *)pCsr->base.pVtab;
3107 if( isPrefix ){
3108 for(i=1; bFound==0 && i<p->nIndex; i++){
3109 if( p->aIndex[i].nPrefix==nTerm ){
3110 bFound = 1;
3111 rc = sqlite3Fts3SegReaderCursor(p, pCsr->iLangid,
3112 i, FTS3_SEGCURSOR_ALL, zTerm, nTerm, 0, 0, pSegcsr
3114 pSegcsr->bLookup = 1;
3118 for(i=1; bFound==0 && i<p->nIndex; i++){
3119 if( p->aIndex[i].nPrefix==nTerm+1 ){
3120 bFound = 1;
3121 rc = sqlite3Fts3SegReaderCursor(p, pCsr->iLangid,
3122 i, FTS3_SEGCURSOR_ALL, zTerm, nTerm, 1, 0, pSegcsr
3124 if( rc==SQLITE_OK ){
3125 rc = fts3SegReaderCursorAddZero(
3126 p, pCsr->iLangid, zTerm, nTerm, pSegcsr
3133 if( bFound==0 ){
3134 rc = sqlite3Fts3SegReaderCursor(p, pCsr->iLangid,
3135 0, FTS3_SEGCURSOR_ALL, zTerm, nTerm, isPrefix, 0, pSegcsr
3137 pSegcsr->bLookup = !isPrefix;
3141 *ppSegcsr = pSegcsr;
3142 return rc;
3146 ** Free an Fts3MultiSegReader allocated by fts3TermSegReaderCursor().
3148 static void fts3SegReaderCursorFree(Fts3MultiSegReader *pSegcsr){
3149 sqlite3Fts3SegReaderFinish(pSegcsr);
3150 sqlite3_free(pSegcsr);
3154 ** This function retrieves the doclist for the specified term (or term
3155 ** prefix) from the database.
3157 static int fts3TermSelect(
3158 Fts3Table *p, /* Virtual table handle */
3159 Fts3PhraseToken *pTok, /* Token to query for */
3160 int iColumn, /* Column to query (or -ve for all columns) */
3161 int *pnOut, /* OUT: Size of buffer at *ppOut */
3162 char **ppOut /* OUT: Malloced result buffer */
3164 int rc; /* Return code */
3165 Fts3MultiSegReader *pSegcsr; /* Seg-reader cursor for this term */
3166 TermSelect tsc; /* Object for pair-wise doclist merging */
3167 Fts3SegFilter filter; /* Segment term filter configuration */
3169 pSegcsr = pTok->pSegcsr;
3170 memset(&tsc, 0, sizeof(TermSelect));
3172 filter.flags = FTS3_SEGMENT_IGNORE_EMPTY | FTS3_SEGMENT_REQUIRE_POS
3173 | (pTok->isPrefix ? FTS3_SEGMENT_PREFIX : 0)
3174 | (pTok->bFirst ? FTS3_SEGMENT_FIRST : 0)
3175 | (iColumn<p->nColumn ? FTS3_SEGMENT_COLUMN_FILTER : 0);
3176 filter.iCol = iColumn;
3177 filter.zTerm = pTok->z;
3178 filter.nTerm = pTok->n;
3180 rc = sqlite3Fts3SegReaderStart(p, pSegcsr, &filter);
3181 while( SQLITE_OK==rc
3182 && SQLITE_ROW==(rc = sqlite3Fts3SegReaderStep(p, pSegcsr))
3184 rc = fts3TermSelectMerge(p, &tsc, pSegcsr->aDoclist, pSegcsr->nDoclist);
3187 if( rc==SQLITE_OK ){
3188 rc = fts3TermSelectFinishMerge(p, &tsc);
3190 if( rc==SQLITE_OK ){
3191 *ppOut = tsc.aaOutput[0];
3192 *pnOut = tsc.anOutput[0];
3193 }else{
3194 int i;
3195 for(i=0; i<SizeofArray(tsc.aaOutput); i++){
3196 sqlite3_free(tsc.aaOutput[i]);
3200 fts3SegReaderCursorFree(pSegcsr);
3201 pTok->pSegcsr = 0;
3202 return rc;
3206 ** This function counts the total number of docids in the doclist stored
3207 ** in buffer aList[], size nList bytes.
3209 ** If the isPoslist argument is true, then it is assumed that the doclist
3210 ** contains a position-list following each docid. Otherwise, it is assumed
3211 ** that the doclist is simply a list of docids stored as delta encoded
3212 ** varints.
3214 static int fts3DoclistCountDocids(char *aList, int nList){
3215 int nDoc = 0; /* Return value */
3216 if( aList ){
3217 char *aEnd = &aList[nList]; /* Pointer to one byte after EOF */
3218 char *p = aList; /* Cursor */
3219 while( p<aEnd ){
3220 nDoc++;
3221 while( (*p++)&0x80 ); /* Skip docid varint */
3222 fts3PoslistCopy(0, &p); /* Skip over position list */
3226 return nDoc;
3230 ** Advance the cursor to the next row in the %_content table that
3231 ** matches the search criteria. For a MATCH search, this will be
3232 ** the next row that matches. For a full-table scan, this will be
3233 ** simply the next row in the %_content table. For a docid lookup,
3234 ** this routine simply sets the EOF flag.
3236 ** Return SQLITE_OK if nothing goes wrong. SQLITE_OK is returned
3237 ** even if we reach end-of-file. The fts3EofMethod() will be called
3238 ** subsequently to determine whether or not an EOF was hit.
3240 static int fts3NextMethod(sqlite3_vtab_cursor *pCursor){
3241 int rc;
3242 Fts3Cursor *pCsr = (Fts3Cursor *)pCursor;
3243 if( pCsr->eSearch==FTS3_DOCID_SEARCH || pCsr->eSearch==FTS3_FULLSCAN_SEARCH ){
3244 Fts3Table *pTab = (Fts3Table*)pCursor->pVtab;
3245 pTab->bLock++;
3246 if( SQLITE_ROW!=sqlite3_step(pCsr->pStmt) ){
3247 pCsr->isEof = 1;
3248 rc = sqlite3_reset(pCsr->pStmt);
3249 }else{
3250 pCsr->iPrevId = sqlite3_column_int64(pCsr->pStmt, 0);
3251 rc = SQLITE_OK;
3253 pTab->bLock--;
3254 }else{
3255 rc = fts3EvalNext((Fts3Cursor *)pCursor);
3257 assert( ((Fts3Table *)pCsr->base.pVtab)->pSegments==0 );
3258 return rc;
3262 ** If the numeric type of argument pVal is "integer", then return it
3263 ** converted to a 64-bit signed integer. Otherwise, return a copy of
3264 ** the second parameter, iDefault.
3266 static sqlite3_int64 fts3DocidRange(sqlite3_value *pVal, i64 iDefault){
3267 if( pVal ){
3268 int eType = sqlite3_value_numeric_type(pVal);
3269 if( eType==SQLITE_INTEGER ){
3270 return sqlite3_value_int64(pVal);
3273 return iDefault;
3277 ** This is the xFilter interface for the virtual table. See
3278 ** the virtual table xFilter method documentation for additional
3279 ** information.
3281 ** If idxNum==FTS3_FULLSCAN_SEARCH then do a full table scan against
3282 ** the %_content table.
3284 ** If idxNum==FTS3_DOCID_SEARCH then do a docid lookup for a single entry
3285 ** in the %_content table.
3287 ** If idxNum>=FTS3_FULLTEXT_SEARCH then use the full text index. The
3288 ** column on the left-hand side of the MATCH operator is column
3289 ** number idxNum-FTS3_FULLTEXT_SEARCH, 0 indexed. argv[0] is the right-hand
3290 ** side of the MATCH operator.
3292 static int fts3FilterMethod(
3293 sqlite3_vtab_cursor *pCursor, /* The cursor used for this query */
3294 int idxNum, /* Strategy index */
3295 const char *idxStr, /* Unused */
3296 int nVal, /* Number of elements in apVal */
3297 sqlite3_value **apVal /* Arguments for the indexing scheme */
3299 int rc = SQLITE_OK;
3300 char *zSql; /* SQL statement used to access %_content */
3301 int eSearch;
3302 Fts3Table *p = (Fts3Table *)pCursor->pVtab;
3303 Fts3Cursor *pCsr = (Fts3Cursor *)pCursor;
3305 sqlite3_value *pCons = 0; /* The MATCH or rowid constraint, if any */
3306 sqlite3_value *pLangid = 0; /* The "langid = ?" constraint, if any */
3307 sqlite3_value *pDocidGe = 0; /* The "docid >= ?" constraint, if any */
3308 sqlite3_value *pDocidLe = 0; /* The "docid <= ?" constraint, if any */
3309 int iIdx;
3311 UNUSED_PARAMETER(idxStr);
3312 UNUSED_PARAMETER(nVal);
3314 if( p->bLock ){
3315 return SQLITE_ERROR;
3318 eSearch = (idxNum & 0x0000FFFF);
3319 assert( eSearch>=0 && eSearch<=(FTS3_FULLTEXT_SEARCH+p->nColumn) );
3320 assert( p->pSegments==0 );
3322 /* Collect arguments into local variables */
3323 iIdx = 0;
3324 if( eSearch!=FTS3_FULLSCAN_SEARCH ) pCons = apVal[iIdx++];
3325 if( idxNum & FTS3_HAVE_LANGID ) pLangid = apVal[iIdx++];
3326 if( idxNum & FTS3_HAVE_DOCID_GE ) pDocidGe = apVal[iIdx++];
3327 if( idxNum & FTS3_HAVE_DOCID_LE ) pDocidLe = apVal[iIdx++];
3328 assert( iIdx==nVal );
3330 /* In case the cursor has been used before, clear it now. */
3331 fts3ClearCursor(pCsr);
3333 /* Set the lower and upper bounds on docids to return */
3334 pCsr->iMinDocid = fts3DocidRange(pDocidGe, SMALLEST_INT64);
3335 pCsr->iMaxDocid = fts3DocidRange(pDocidLe, LARGEST_INT64);
3337 if( idxStr ){
3338 pCsr->bDesc = (idxStr[0]=='D');
3339 }else{
3340 pCsr->bDesc = p->bDescIdx;
3342 pCsr->eSearch = (i16)eSearch;
3344 if( eSearch!=FTS3_DOCID_SEARCH && eSearch!=FTS3_FULLSCAN_SEARCH ){
3345 int iCol = eSearch-FTS3_FULLTEXT_SEARCH;
3346 const char *zQuery = (const char *)sqlite3_value_text(pCons);
3348 if( zQuery==0 && sqlite3_value_type(pCons)!=SQLITE_NULL ){
3349 return SQLITE_NOMEM;
3352 pCsr->iLangid = 0;
3353 if( pLangid ) pCsr->iLangid = sqlite3_value_int(pLangid);
3355 assert( p->base.zErrMsg==0 );
3356 rc = sqlite3Fts3ExprParse(p->pTokenizer, pCsr->iLangid,
3357 p->azColumn, p->bFts4, p->nColumn, iCol, zQuery, -1, &pCsr->pExpr,
3358 &p->base.zErrMsg
3360 if( rc!=SQLITE_OK ){
3361 return rc;
3364 rc = fts3EvalStart(pCsr);
3365 sqlite3Fts3SegmentsClose(p);
3366 if( rc!=SQLITE_OK ) return rc;
3367 pCsr->pNextId = pCsr->aDoclist;
3368 pCsr->iPrevId = 0;
3371 /* Compile a SELECT statement for this cursor. For a full-table-scan, the
3372 ** statement loops through all rows of the %_content table. For a
3373 ** full-text query or docid lookup, the statement retrieves a single
3374 ** row by docid.
3376 if( eSearch==FTS3_FULLSCAN_SEARCH ){
3377 if( pDocidGe || pDocidLe ){
3378 zSql = sqlite3_mprintf(
3379 "SELECT %s WHERE rowid BETWEEN %lld AND %lld ORDER BY rowid %s",
3380 p->zReadExprlist, pCsr->iMinDocid, pCsr->iMaxDocid,
3381 (pCsr->bDesc ? "DESC" : "ASC")
3383 }else{
3384 zSql = sqlite3_mprintf("SELECT %s ORDER BY rowid %s",
3385 p->zReadExprlist, (pCsr->bDesc ? "DESC" : "ASC")
3388 if( zSql ){
3389 p->bLock++;
3390 rc = sqlite3_prepare_v3(
3391 p->db,zSql,-1,SQLITE_PREPARE_PERSISTENT,&pCsr->pStmt,0
3393 p->bLock--;
3394 sqlite3_free(zSql);
3395 }else{
3396 rc = SQLITE_NOMEM;
3398 }else if( eSearch==FTS3_DOCID_SEARCH ){
3399 rc = fts3CursorSeekStmt(pCsr);
3400 if( rc==SQLITE_OK ){
3401 rc = sqlite3_bind_value(pCsr->pStmt, 1, pCons);
3404 if( rc!=SQLITE_OK ) return rc;
3406 return fts3NextMethod(pCursor);
3410 ** This is the xEof method of the virtual table. SQLite calls this
3411 ** routine to find out if it has reached the end of a result set.
3413 static int fts3EofMethod(sqlite3_vtab_cursor *pCursor){
3414 Fts3Cursor *pCsr = (Fts3Cursor*)pCursor;
3415 if( pCsr->isEof ){
3416 fts3ClearCursor(pCsr);
3417 pCsr->isEof = 1;
3419 return pCsr->isEof;
3423 ** This is the xRowid method. The SQLite core calls this routine to
3424 ** retrieve the rowid for the current row of the result set. fts3
3425 ** exposes %_content.docid as the rowid for the virtual table. The
3426 ** rowid should be written to *pRowid.
3428 static int fts3RowidMethod(sqlite3_vtab_cursor *pCursor, sqlite_int64 *pRowid){
3429 Fts3Cursor *pCsr = (Fts3Cursor *) pCursor;
3430 *pRowid = pCsr->iPrevId;
3431 return SQLITE_OK;
3435 ** This is the xColumn method, called by SQLite to request a value from
3436 ** the row that the supplied cursor currently points to.
3438 ** If:
3440 ** (iCol < p->nColumn) -> The value of the iCol'th user column.
3441 ** (iCol == p->nColumn) -> Magic column with the same name as the table.
3442 ** (iCol == p->nColumn+1) -> Docid column
3443 ** (iCol == p->nColumn+2) -> Langid column
3445 static int fts3ColumnMethod(
3446 sqlite3_vtab_cursor *pCursor, /* Cursor to retrieve value from */
3447 sqlite3_context *pCtx, /* Context for sqlite3_result_xxx() calls */
3448 int iCol /* Index of column to read value from */
3450 int rc = SQLITE_OK; /* Return Code */
3451 Fts3Cursor *pCsr = (Fts3Cursor *) pCursor;
3452 Fts3Table *p = (Fts3Table *)pCursor->pVtab;
3454 /* The column value supplied by SQLite must be in range. */
3455 assert( iCol>=0 && iCol<=p->nColumn+2 );
3457 switch( iCol-p->nColumn ){
3458 case 0:
3459 /* The special 'table-name' column */
3460 sqlite3_result_pointer(pCtx, pCsr, "fts3cursor", 0);
3461 break;
3463 case 1:
3464 /* The docid column */
3465 sqlite3_result_int64(pCtx, pCsr->iPrevId);
3466 break;
3468 case 2:
3469 if( pCsr->pExpr ){
3470 sqlite3_result_int64(pCtx, pCsr->iLangid);
3471 break;
3472 }else if( p->zLanguageid==0 ){
3473 sqlite3_result_int(pCtx, 0);
3474 break;
3475 }else{
3476 iCol = p->nColumn;
3477 /* no break */ deliberate_fall_through
3480 default:
3481 /* A user column. Or, if this is a full-table scan, possibly the
3482 ** language-id column. Seek the cursor. */
3483 rc = fts3CursorSeek(0, pCsr);
3484 if( rc==SQLITE_OK && sqlite3_data_count(pCsr->pStmt)-1>iCol ){
3485 sqlite3_result_value(pCtx, sqlite3_column_value(pCsr->pStmt, iCol+1));
3487 break;
3490 assert( ((Fts3Table *)pCsr->base.pVtab)->pSegments==0 );
3491 return rc;
3495 ** This function is the implementation of the xUpdate callback used by
3496 ** FTS3 virtual tables. It is invoked by SQLite each time a row is to be
3497 ** inserted, updated or deleted.
3499 static int fts3UpdateMethod(
3500 sqlite3_vtab *pVtab, /* Virtual table handle */
3501 int nArg, /* Size of argument array */
3502 sqlite3_value **apVal, /* Array of arguments */
3503 sqlite_int64 *pRowid /* OUT: The affected (or effected) rowid */
3505 return sqlite3Fts3UpdateMethod(pVtab, nArg, apVal, pRowid);
3509 ** Implementation of xSync() method. Flush the contents of the pending-terms
3510 ** hash-table to the database.
3512 static int fts3SyncMethod(sqlite3_vtab *pVtab){
3514 /* Following an incremental-merge operation, assuming that the input
3515 ** segments are not completely consumed (the usual case), they are updated
3516 ** in place to remove the entries that have already been merged. This
3517 ** involves updating the leaf block that contains the smallest unmerged
3518 ** entry and each block (if any) between the leaf and the root node. So
3519 ** if the height of the input segment b-trees is N, and input segments
3520 ** are merged eight at a time, updating the input segments at the end
3521 ** of an incremental-merge requires writing (8*(1+N)) blocks. N is usually
3522 ** small - often between 0 and 2. So the overhead of the incremental
3523 ** merge is somewhere between 8 and 24 blocks. To avoid this overhead
3524 ** dwarfing the actual productive work accomplished, the incremental merge
3525 ** is only attempted if it will write at least 64 leaf blocks. Hence
3526 ** nMinMerge.
3528 ** Of course, updating the input segments also involves deleting a bunch
3529 ** of blocks from the segments table. But this is not considered overhead
3530 ** as it would also be required by a crisis-merge that used the same input
3531 ** segments.
3533 const u32 nMinMerge = 64; /* Minimum amount of incr-merge work to do */
3535 Fts3Table *p = (Fts3Table*)pVtab;
3536 int rc;
3537 i64 iLastRowid = sqlite3_last_insert_rowid(p->db);
3539 rc = sqlite3Fts3PendingTermsFlush(p);
3540 if( rc==SQLITE_OK
3541 && p->nLeafAdd>(nMinMerge/16)
3542 && p->nAutoincrmerge && p->nAutoincrmerge!=0xff
3544 int mxLevel = 0; /* Maximum relative level value in db */
3545 int A; /* Incr-merge parameter A */
3547 rc = sqlite3Fts3MaxLevel(p, &mxLevel);
3548 assert( rc==SQLITE_OK || mxLevel==0 );
3549 A = p->nLeafAdd * mxLevel;
3550 A += (A/2);
3551 if( A>(int)nMinMerge ) rc = sqlite3Fts3Incrmerge(p, A, p->nAutoincrmerge);
3553 sqlite3Fts3SegmentsClose(p);
3554 sqlite3_set_last_insert_rowid(p->db, iLastRowid);
3555 return rc;
3559 ** If it is currently unknown whether or not the FTS table has an %_stat
3560 ** table (if p->bHasStat==2), attempt to determine this (set p->bHasStat
3561 ** to 0 or 1). Return SQLITE_OK if successful, or an SQLite error code
3562 ** if an error occurs.
3564 static int fts3SetHasStat(Fts3Table *p){
3565 int rc = SQLITE_OK;
3566 if( p->bHasStat==2 ){
3567 char *zTbl = sqlite3_mprintf("%s_stat", p->zName);
3568 if( zTbl ){
3569 int res = sqlite3_table_column_metadata(p->db, p->zDb, zTbl, 0,0,0,0,0,0);
3570 sqlite3_free(zTbl);
3571 p->bHasStat = (res==SQLITE_OK);
3572 }else{
3573 rc = SQLITE_NOMEM;
3576 return rc;
3580 ** Implementation of xBegin() method.
3582 static int fts3BeginMethod(sqlite3_vtab *pVtab){
3583 Fts3Table *p = (Fts3Table*)pVtab;
3584 int rc;
3585 UNUSED_PARAMETER(pVtab);
3586 assert( p->pSegments==0 );
3587 assert( p->nPendingData==0 );
3588 assert( p->inTransaction!=1 );
3589 p->nLeafAdd = 0;
3590 rc = fts3SetHasStat(p);
3591 #ifdef SQLITE_DEBUG
3592 if( rc==SQLITE_OK ){
3593 p->inTransaction = 1;
3594 p->mxSavepoint = -1;
3596 #endif
3597 return rc;
3601 ** Implementation of xCommit() method. This is a no-op. The contents of
3602 ** the pending-terms hash-table have already been flushed into the database
3603 ** by fts3SyncMethod().
3605 static int fts3CommitMethod(sqlite3_vtab *pVtab){
3606 TESTONLY( Fts3Table *p = (Fts3Table*)pVtab );
3607 UNUSED_PARAMETER(pVtab);
3608 assert( p->nPendingData==0 );
3609 assert( p->inTransaction!=0 );
3610 assert( p->pSegments==0 );
3611 TESTONLY( p->inTransaction = 0 );
3612 TESTONLY( p->mxSavepoint = -1; );
3613 return SQLITE_OK;
3617 ** Implementation of xRollback(). Discard the contents of the pending-terms
3618 ** hash-table. Any changes made to the database are reverted by SQLite.
3620 static int fts3RollbackMethod(sqlite3_vtab *pVtab){
3621 Fts3Table *p = (Fts3Table*)pVtab;
3622 sqlite3Fts3PendingTermsClear(p);
3623 assert( p->inTransaction!=0 );
3624 TESTONLY( p->inTransaction = 0 );
3625 TESTONLY( p->mxSavepoint = -1; );
3626 return SQLITE_OK;
3630 ** When called, *ppPoslist must point to the byte immediately following the
3631 ** end of a position-list. i.e. ( (*ppPoslist)[-1]==POS_END ). This function
3632 ** moves *ppPoslist so that it instead points to the first byte of the
3633 ** same position list.
3635 static void fts3ReversePoslist(char *pStart, char **ppPoslist){
3636 char *p = &(*ppPoslist)[-2];
3637 char c = 0;
3639 /* Skip backwards passed any trailing 0x00 bytes added by NearTrim() */
3640 while( p>pStart && (c=*p--)==0 );
3642 /* Search backwards for a varint with value zero (the end of the previous
3643 ** poslist). This is an 0x00 byte preceded by some byte that does not
3644 ** have the 0x80 bit set. */
3645 while( p>pStart && (*p & 0x80) | c ){
3646 c = *p--;
3648 assert( p==pStart || c==0 );
3650 /* At this point p points to that preceding byte without the 0x80 bit
3651 ** set. So to find the start of the poslist, skip forward 2 bytes then
3652 ** over a varint.
3654 ** Normally. The other case is that p==pStart and the poslist to return
3655 ** is the first in the doclist. In this case do not skip forward 2 bytes.
3656 ** The second part of the if condition (c==0 && *ppPoslist>&p[2])
3657 ** is required for cases where the first byte of a doclist and the
3658 ** doclist is empty. For example, if the first docid is 10, a doclist
3659 ** that begins with:
3661 ** 0x0A 0x00 <next docid delta varint>
3663 if( p>pStart || (c==0 && *ppPoslist>&p[2]) ){ p = &p[2]; }
3664 while( *p++&0x80 );
3665 *ppPoslist = p;
3669 ** Helper function used by the implementation of the overloaded snippet(),
3670 ** offsets() and optimize() SQL functions.
3672 ** If the value passed as the third argument is a blob of size
3673 ** sizeof(Fts3Cursor*), then the blob contents are copied to the
3674 ** output variable *ppCsr and SQLITE_OK is returned. Otherwise, an error
3675 ** message is written to context pContext and SQLITE_ERROR returned. The
3676 ** string passed via zFunc is used as part of the error message.
3678 static int fts3FunctionArg(
3679 sqlite3_context *pContext, /* SQL function call context */
3680 const char *zFunc, /* Function name */
3681 sqlite3_value *pVal, /* argv[0] passed to function */
3682 Fts3Cursor **ppCsr /* OUT: Store cursor handle here */
3684 int rc;
3685 *ppCsr = (Fts3Cursor*)sqlite3_value_pointer(pVal, "fts3cursor");
3686 if( (*ppCsr)!=0 ){
3687 rc = SQLITE_OK;
3688 }else{
3689 char *zErr = sqlite3_mprintf("illegal first argument to %s", zFunc);
3690 sqlite3_result_error(pContext, zErr, -1);
3691 sqlite3_free(zErr);
3692 rc = SQLITE_ERROR;
3694 return rc;
3698 ** Implementation of the snippet() function for FTS3
3700 static void fts3SnippetFunc(
3701 sqlite3_context *pContext, /* SQLite function call context */
3702 int nVal, /* Size of apVal[] array */
3703 sqlite3_value **apVal /* Array of arguments */
3705 Fts3Cursor *pCsr; /* Cursor handle passed through apVal[0] */
3706 const char *zStart = "<b>";
3707 const char *zEnd = "</b>";
3708 const char *zEllipsis = "<b>...</b>";
3709 int iCol = -1;
3710 int nToken = 15; /* Default number of tokens in snippet */
3712 /* There must be at least one argument passed to this function (otherwise
3713 ** the non-overloaded version would have been called instead of this one).
3715 assert( nVal>=1 );
3717 if( nVal>6 ){
3718 sqlite3_result_error(pContext,
3719 "wrong number of arguments to function snippet()", -1);
3720 return;
3722 if( fts3FunctionArg(pContext, "snippet", apVal[0], &pCsr) ) return;
3724 switch( nVal ){
3725 case 6: nToken = sqlite3_value_int(apVal[5]);
3726 /* no break */ deliberate_fall_through
3727 case 5: iCol = sqlite3_value_int(apVal[4]);
3728 /* no break */ deliberate_fall_through
3729 case 4: zEllipsis = (const char*)sqlite3_value_text(apVal[3]);
3730 /* no break */ deliberate_fall_through
3731 case 3: zEnd = (const char*)sqlite3_value_text(apVal[2]);
3732 /* no break */ deliberate_fall_through
3733 case 2: zStart = (const char*)sqlite3_value_text(apVal[1]);
3735 if( !zEllipsis || !zEnd || !zStart ){
3736 sqlite3_result_error_nomem(pContext);
3737 }else if( nToken==0 ){
3738 sqlite3_result_text(pContext, "", -1, SQLITE_STATIC);
3739 }else if( SQLITE_OK==fts3CursorSeek(pContext, pCsr) ){
3740 sqlite3Fts3Snippet(pContext, pCsr, zStart, zEnd, zEllipsis, iCol, nToken);
3745 ** Implementation of the offsets() function for FTS3
3747 static void fts3OffsetsFunc(
3748 sqlite3_context *pContext, /* SQLite function call context */
3749 int nVal, /* Size of argument array */
3750 sqlite3_value **apVal /* Array of arguments */
3752 Fts3Cursor *pCsr; /* Cursor handle passed through apVal[0] */
3754 UNUSED_PARAMETER(nVal);
3756 assert( nVal==1 );
3757 if( fts3FunctionArg(pContext, "offsets", apVal[0], &pCsr) ) return;
3758 assert( pCsr );
3759 if( SQLITE_OK==fts3CursorSeek(pContext, pCsr) ){
3760 sqlite3Fts3Offsets(pContext, pCsr);
3765 ** Implementation of the special optimize() function for FTS3. This
3766 ** function merges all segments in the database to a single segment.
3767 ** Example usage is:
3769 ** SELECT optimize(t) FROM t LIMIT 1;
3771 ** where 't' is the name of an FTS3 table.
3773 static void fts3OptimizeFunc(
3774 sqlite3_context *pContext, /* SQLite function call context */
3775 int nVal, /* Size of argument array */
3776 sqlite3_value **apVal /* Array of arguments */
3778 int rc; /* Return code */
3779 Fts3Table *p; /* Virtual table handle */
3780 Fts3Cursor *pCursor; /* Cursor handle passed through apVal[0] */
3782 UNUSED_PARAMETER(nVal);
3784 assert( nVal==1 );
3785 if( fts3FunctionArg(pContext, "optimize", apVal[0], &pCursor) ) return;
3786 p = (Fts3Table *)pCursor->base.pVtab;
3787 assert( p );
3789 rc = sqlite3Fts3Optimize(p);
3791 switch( rc ){
3792 case SQLITE_OK:
3793 sqlite3_result_text(pContext, "Index optimized", -1, SQLITE_STATIC);
3794 break;
3795 case SQLITE_DONE:
3796 sqlite3_result_text(pContext, "Index already optimal", -1, SQLITE_STATIC);
3797 break;
3798 default:
3799 sqlite3_result_error_code(pContext, rc);
3800 break;
3805 ** Implementation of the matchinfo() function for FTS3
3807 static void fts3MatchinfoFunc(
3808 sqlite3_context *pContext, /* SQLite function call context */
3809 int nVal, /* Size of argument array */
3810 sqlite3_value **apVal /* Array of arguments */
3812 Fts3Cursor *pCsr; /* Cursor handle passed through apVal[0] */
3813 assert( nVal==1 || nVal==2 );
3814 if( SQLITE_OK==fts3FunctionArg(pContext, "matchinfo", apVal[0], &pCsr) ){
3815 const char *zArg = 0;
3816 if( nVal>1 ){
3817 zArg = (const char *)sqlite3_value_text(apVal[1]);
3819 sqlite3Fts3Matchinfo(pContext, pCsr, zArg);
3824 ** This routine implements the xFindFunction method for the FTS3
3825 ** virtual table.
3827 static int fts3FindFunctionMethod(
3828 sqlite3_vtab *pVtab, /* Virtual table handle */
3829 int nArg, /* Number of SQL function arguments */
3830 const char *zName, /* Name of SQL function */
3831 void (**pxFunc)(sqlite3_context*,int,sqlite3_value**), /* OUT: Result */
3832 void **ppArg /* Unused */
3834 struct Overloaded {
3835 const char *zName;
3836 void (*xFunc)(sqlite3_context*,int,sqlite3_value**);
3837 } aOverload[] = {
3838 { "snippet", fts3SnippetFunc },
3839 { "offsets", fts3OffsetsFunc },
3840 { "optimize", fts3OptimizeFunc },
3841 { "matchinfo", fts3MatchinfoFunc },
3843 int i; /* Iterator variable */
3845 UNUSED_PARAMETER(pVtab);
3846 UNUSED_PARAMETER(nArg);
3847 UNUSED_PARAMETER(ppArg);
3849 for(i=0; i<SizeofArray(aOverload); i++){
3850 if( strcmp(zName, aOverload[i].zName)==0 ){
3851 *pxFunc = aOverload[i].xFunc;
3852 return 1;
3856 /* No function of the specified name was found. Return 0. */
3857 return 0;
3861 ** Implementation of FTS3 xRename method. Rename an fts3 table.
3863 static int fts3RenameMethod(
3864 sqlite3_vtab *pVtab, /* Virtual table handle */
3865 const char *zName /* New name of table */
3867 Fts3Table *p = (Fts3Table *)pVtab;
3868 sqlite3 *db = p->db; /* Database connection */
3869 int rc; /* Return Code */
3871 /* At this point it must be known if the %_stat table exists or not.
3872 ** So bHasStat may not be 2. */
3873 rc = fts3SetHasStat(p);
3875 /* As it happens, the pending terms table is always empty here. This is
3876 ** because an "ALTER TABLE RENAME TABLE" statement inside a transaction
3877 ** always opens a savepoint transaction. And the xSavepoint() method
3878 ** flushes the pending terms table. But leave the (no-op) call to
3879 ** PendingTermsFlush() in in case that changes.
3881 assert( p->nPendingData==0 );
3882 if( rc==SQLITE_OK ){
3883 rc = sqlite3Fts3PendingTermsFlush(p);
3886 if( p->zContentTbl==0 ){
3887 fts3DbExec(&rc, db,
3888 "ALTER TABLE %Q.'%q_content' RENAME TO '%q_content';",
3889 p->zDb, p->zName, zName
3893 if( p->bHasDocsize ){
3894 fts3DbExec(&rc, db,
3895 "ALTER TABLE %Q.'%q_docsize' RENAME TO '%q_docsize';",
3896 p->zDb, p->zName, zName
3899 if( p->bHasStat ){
3900 fts3DbExec(&rc, db,
3901 "ALTER TABLE %Q.'%q_stat' RENAME TO '%q_stat';",
3902 p->zDb, p->zName, zName
3905 fts3DbExec(&rc, db,
3906 "ALTER TABLE %Q.'%q_segments' RENAME TO '%q_segments';",
3907 p->zDb, p->zName, zName
3909 fts3DbExec(&rc, db,
3910 "ALTER TABLE %Q.'%q_segdir' RENAME TO '%q_segdir';",
3911 p->zDb, p->zName, zName
3913 return rc;
3917 ** The xSavepoint() method.
3919 ** Flush the contents of the pending-terms table to disk.
3921 static int fts3SavepointMethod(sqlite3_vtab *pVtab, int iSavepoint){
3922 int rc = SQLITE_OK;
3923 UNUSED_PARAMETER(iSavepoint);
3924 assert( ((Fts3Table *)pVtab)->inTransaction );
3925 assert( ((Fts3Table *)pVtab)->mxSavepoint <= iSavepoint );
3926 TESTONLY( ((Fts3Table *)pVtab)->mxSavepoint = iSavepoint );
3927 if( ((Fts3Table *)pVtab)->bIgnoreSavepoint==0 ){
3928 rc = fts3SyncMethod(pVtab);
3930 return rc;
3934 ** The xRelease() method.
3936 ** This is a no-op.
3938 static int fts3ReleaseMethod(sqlite3_vtab *pVtab, int iSavepoint){
3939 TESTONLY( Fts3Table *p = (Fts3Table*)pVtab );
3940 UNUSED_PARAMETER(iSavepoint);
3941 UNUSED_PARAMETER(pVtab);
3942 assert( p->inTransaction );
3943 assert( p->mxSavepoint >= iSavepoint );
3944 TESTONLY( p->mxSavepoint = iSavepoint-1 );
3945 return SQLITE_OK;
3949 ** The xRollbackTo() method.
3951 ** Discard the contents of the pending terms table.
3953 static int fts3RollbackToMethod(sqlite3_vtab *pVtab, int iSavepoint){
3954 Fts3Table *p = (Fts3Table*)pVtab;
3955 UNUSED_PARAMETER(iSavepoint);
3956 assert( p->inTransaction );
3957 TESTONLY( p->mxSavepoint = iSavepoint );
3958 sqlite3Fts3PendingTermsClear(p);
3959 return SQLITE_OK;
3963 ** Return true if zName is the extension on one of the shadow tables used
3964 ** by this module.
3966 static int fts3ShadowName(const char *zName){
3967 static const char *azName[] = {
3968 "content", "docsize", "segdir", "segments", "stat",
3970 unsigned int i;
3971 for(i=0; i<sizeof(azName)/sizeof(azName[0]); i++){
3972 if( sqlite3_stricmp(zName, azName[i])==0 ) return 1;
3974 return 0;
3977 static const sqlite3_module fts3Module = {
3978 /* iVersion */ 3,
3979 /* xCreate */ fts3CreateMethod,
3980 /* xConnect */ fts3ConnectMethod,
3981 /* xBestIndex */ fts3BestIndexMethod,
3982 /* xDisconnect */ fts3DisconnectMethod,
3983 /* xDestroy */ fts3DestroyMethod,
3984 /* xOpen */ fts3OpenMethod,
3985 /* xClose */ fts3CloseMethod,
3986 /* xFilter */ fts3FilterMethod,
3987 /* xNext */ fts3NextMethod,
3988 /* xEof */ fts3EofMethod,
3989 /* xColumn */ fts3ColumnMethod,
3990 /* xRowid */ fts3RowidMethod,
3991 /* xUpdate */ fts3UpdateMethod,
3992 /* xBegin */ fts3BeginMethod,
3993 /* xSync */ fts3SyncMethod,
3994 /* xCommit */ fts3CommitMethod,
3995 /* xRollback */ fts3RollbackMethod,
3996 /* xFindFunction */ fts3FindFunctionMethod,
3997 /* xRename */ fts3RenameMethod,
3998 /* xSavepoint */ fts3SavepointMethod,
3999 /* xRelease */ fts3ReleaseMethod,
4000 /* xRollbackTo */ fts3RollbackToMethod,
4001 /* xShadowName */ fts3ShadowName,
4005 ** This function is registered as the module destructor (called when an
4006 ** FTS3 enabled database connection is closed). It frees the memory
4007 ** allocated for the tokenizer hash table.
4009 static void hashDestroy(void *p){
4010 Fts3Hash *pHash = (Fts3Hash *)p;
4011 sqlite3Fts3HashClear(pHash);
4012 sqlite3_free(pHash);
4016 ** The fts3 built-in tokenizers - "simple", "porter" and "icu"- are
4017 ** implemented in files fts3_tokenizer1.c, fts3_porter.c and fts3_icu.c
4018 ** respectively. The following three forward declarations are for functions
4019 ** declared in these files used to retrieve the respective implementations.
4021 ** Calling sqlite3Fts3SimpleTokenizerModule() sets the value pointed
4022 ** to by the argument to point to the "simple" tokenizer implementation.
4023 ** And so on.
4025 void sqlite3Fts3SimpleTokenizerModule(sqlite3_tokenizer_module const**ppModule);
4026 void sqlite3Fts3PorterTokenizerModule(sqlite3_tokenizer_module const**ppModule);
4027 #ifndef SQLITE_DISABLE_FTS3_UNICODE
4028 void sqlite3Fts3UnicodeTokenizer(sqlite3_tokenizer_module const**ppModule);
4029 #endif
4030 #ifdef SQLITE_ENABLE_ICU
4031 void sqlite3Fts3IcuTokenizerModule(sqlite3_tokenizer_module const**ppModule);
4032 #endif
4035 ** Initialize the fts3 extension. If this extension is built as part
4036 ** of the sqlite library, then this function is called directly by
4037 ** SQLite. If fts3 is built as a dynamically loadable extension, this
4038 ** function is called by the sqlite3_extension_init() entry point.
4040 int sqlite3Fts3Init(sqlite3 *db){
4041 int rc = SQLITE_OK;
4042 Fts3Hash *pHash = 0;
4043 const sqlite3_tokenizer_module *pSimple = 0;
4044 const sqlite3_tokenizer_module *pPorter = 0;
4045 #ifndef SQLITE_DISABLE_FTS3_UNICODE
4046 const sqlite3_tokenizer_module *pUnicode = 0;
4047 #endif
4049 #ifdef SQLITE_ENABLE_ICU
4050 const sqlite3_tokenizer_module *pIcu = 0;
4051 sqlite3Fts3IcuTokenizerModule(&pIcu);
4052 #endif
4054 #ifndef SQLITE_DISABLE_FTS3_UNICODE
4055 sqlite3Fts3UnicodeTokenizer(&pUnicode);
4056 #endif
4058 #ifdef SQLITE_TEST
4059 rc = sqlite3Fts3InitTerm(db);
4060 if( rc!=SQLITE_OK ) return rc;
4061 #endif
4063 rc = sqlite3Fts3InitAux(db);
4064 if( rc!=SQLITE_OK ) return rc;
4066 sqlite3Fts3SimpleTokenizerModule(&pSimple);
4067 sqlite3Fts3PorterTokenizerModule(&pPorter);
4069 /* Allocate and initialize the hash-table used to store tokenizers. */
4070 pHash = sqlite3_malloc(sizeof(Fts3Hash));
4071 if( !pHash ){
4072 rc = SQLITE_NOMEM;
4073 }else{
4074 sqlite3Fts3HashInit(pHash, FTS3_HASH_STRING, 1);
4077 /* Load the built-in tokenizers into the hash table */
4078 if( rc==SQLITE_OK ){
4079 if( sqlite3Fts3HashInsert(pHash, "simple", 7, (void *)pSimple)
4080 || sqlite3Fts3HashInsert(pHash, "porter", 7, (void *)pPorter)
4082 #ifndef SQLITE_DISABLE_FTS3_UNICODE
4083 || sqlite3Fts3HashInsert(pHash, "unicode61", 10, (void *)pUnicode)
4084 #endif
4085 #ifdef SQLITE_ENABLE_ICU
4086 || (pIcu && sqlite3Fts3HashInsert(pHash, "icu", 4, (void *)pIcu))
4087 #endif
4089 rc = SQLITE_NOMEM;
4093 #ifdef SQLITE_TEST
4094 if( rc==SQLITE_OK ){
4095 rc = sqlite3Fts3ExprInitTestInterface(db, pHash);
4097 #endif
4099 /* Create the virtual table wrapper around the hash-table and overload
4100 ** the four scalar functions. If this is successful, register the
4101 ** module with sqlite.
4103 if( SQLITE_OK==rc
4104 && SQLITE_OK==(rc = sqlite3Fts3InitHashTable(db, pHash, "fts3_tokenizer"))
4105 && SQLITE_OK==(rc = sqlite3_overload_function(db, "snippet", -1))
4106 && SQLITE_OK==(rc = sqlite3_overload_function(db, "offsets", 1))
4107 && SQLITE_OK==(rc = sqlite3_overload_function(db, "matchinfo", 1))
4108 && SQLITE_OK==(rc = sqlite3_overload_function(db, "matchinfo", 2))
4109 && SQLITE_OK==(rc = sqlite3_overload_function(db, "optimize", 1))
4111 rc = sqlite3_create_module_v2(
4112 db, "fts3", &fts3Module, (void *)pHash, hashDestroy
4114 if( rc==SQLITE_OK ){
4115 rc = sqlite3_create_module_v2(
4116 db, "fts4", &fts3Module, (void *)pHash, 0
4119 if( rc==SQLITE_OK ){
4120 rc = sqlite3Fts3InitTok(db, (void *)pHash);
4122 return rc;
4126 /* An error has occurred. Delete the hash table and return the error code. */
4127 assert( rc!=SQLITE_OK );
4128 if( pHash ){
4129 sqlite3Fts3HashClear(pHash);
4130 sqlite3_free(pHash);
4132 return rc;
4136 ** Allocate an Fts3MultiSegReader for each token in the expression headed
4137 ** by pExpr.
4139 ** An Fts3SegReader object is a cursor that can seek or scan a range of
4140 ** entries within a single segment b-tree. An Fts3MultiSegReader uses multiple
4141 ** Fts3SegReader objects internally to provide an interface to seek or scan
4142 ** within the union of all segments of a b-tree. Hence the name.
4144 ** If the allocated Fts3MultiSegReader just seeks to a single entry in a
4145 ** segment b-tree (if the term is not a prefix or it is a prefix for which
4146 ** there exists prefix b-tree of the right length) then it may be traversed
4147 ** and merged incrementally. Otherwise, it has to be merged into an in-memory
4148 ** doclist and then traversed.
4150 static void fts3EvalAllocateReaders(
4151 Fts3Cursor *pCsr, /* FTS cursor handle */
4152 Fts3Expr *pExpr, /* Allocate readers for this expression */
4153 int *pnToken, /* OUT: Total number of tokens in phrase. */
4154 int *pnOr, /* OUT: Total number of OR nodes in expr. */
4155 int *pRc /* IN/OUT: Error code */
4157 if( pExpr && SQLITE_OK==*pRc ){
4158 if( pExpr->eType==FTSQUERY_PHRASE ){
4159 int i;
4160 int nToken = pExpr->pPhrase->nToken;
4161 *pnToken += nToken;
4162 for(i=0; i<nToken; i++){
4163 Fts3PhraseToken *pToken = &pExpr->pPhrase->aToken[i];
4164 int rc = fts3TermSegReaderCursor(pCsr,
4165 pToken->z, pToken->n, pToken->isPrefix, &pToken->pSegcsr
4167 if( rc!=SQLITE_OK ){
4168 *pRc = rc;
4169 return;
4172 assert( pExpr->pPhrase->iDoclistToken==0 );
4173 pExpr->pPhrase->iDoclistToken = -1;
4174 }else{
4175 *pnOr += (pExpr->eType==FTSQUERY_OR);
4176 fts3EvalAllocateReaders(pCsr, pExpr->pLeft, pnToken, pnOr, pRc);
4177 fts3EvalAllocateReaders(pCsr, pExpr->pRight, pnToken, pnOr, pRc);
4183 ** Arguments pList/nList contain the doclist for token iToken of phrase p.
4184 ** It is merged into the main doclist stored in p->doclist.aAll/nAll.
4186 ** This function assumes that pList points to a buffer allocated using
4187 ** sqlite3_malloc(). This function takes responsibility for eventually
4188 ** freeing the buffer.
4190 ** SQLITE_OK is returned if successful, or SQLITE_NOMEM if an error occurs.
4192 static int fts3EvalPhraseMergeToken(
4193 Fts3Table *pTab, /* FTS Table pointer */
4194 Fts3Phrase *p, /* Phrase to merge pList/nList into */
4195 int iToken, /* Token pList/nList corresponds to */
4196 char *pList, /* Pointer to doclist */
4197 int nList /* Number of bytes in pList */
4199 int rc = SQLITE_OK;
4200 assert( iToken!=p->iDoclistToken );
4202 if( pList==0 ){
4203 sqlite3_free(p->doclist.aAll);
4204 p->doclist.aAll = 0;
4205 p->doclist.nAll = 0;
4208 else if( p->iDoclistToken<0 ){
4209 p->doclist.aAll = pList;
4210 p->doclist.nAll = nList;
4213 else if( p->doclist.aAll==0 ){
4214 sqlite3_free(pList);
4217 else {
4218 char *pLeft;
4219 char *pRight;
4220 int nLeft;
4221 int nRight;
4222 int nDiff;
4224 if( p->iDoclistToken<iToken ){
4225 pLeft = p->doclist.aAll;
4226 nLeft = p->doclist.nAll;
4227 pRight = pList;
4228 nRight = nList;
4229 nDiff = iToken - p->iDoclistToken;
4230 }else{
4231 pRight = p->doclist.aAll;
4232 nRight = p->doclist.nAll;
4233 pLeft = pList;
4234 nLeft = nList;
4235 nDiff = p->iDoclistToken - iToken;
4238 rc = fts3DoclistPhraseMerge(
4239 pTab->bDescIdx, nDiff, pLeft, nLeft, &pRight, &nRight
4241 sqlite3_free(pLeft);
4242 p->doclist.aAll = pRight;
4243 p->doclist.nAll = nRight;
4246 if( iToken>p->iDoclistToken ) p->iDoclistToken = iToken;
4247 return rc;
4251 ** Load the doclist for phrase p into p->doclist.aAll/nAll. The loaded doclist
4252 ** does not take deferred tokens into account.
4254 ** SQLITE_OK is returned if no error occurs, otherwise an SQLite error code.
4256 static int fts3EvalPhraseLoad(
4257 Fts3Cursor *pCsr, /* FTS Cursor handle */
4258 Fts3Phrase *p /* Phrase object */
4260 Fts3Table *pTab = (Fts3Table *)pCsr->base.pVtab;
4261 int iToken;
4262 int rc = SQLITE_OK;
4264 for(iToken=0; rc==SQLITE_OK && iToken<p->nToken; iToken++){
4265 Fts3PhraseToken *pToken = &p->aToken[iToken];
4266 assert( pToken->pDeferred==0 || pToken->pSegcsr==0 );
4268 if( pToken->pSegcsr ){
4269 int nThis = 0;
4270 char *pThis = 0;
4271 rc = fts3TermSelect(pTab, pToken, p->iColumn, &nThis, &pThis);
4272 if( rc==SQLITE_OK ){
4273 rc = fts3EvalPhraseMergeToken(pTab, p, iToken, pThis, nThis);
4276 assert( pToken->pSegcsr==0 );
4279 return rc;
4282 #ifndef SQLITE_DISABLE_FTS4_DEFERRED
4284 ** This function is called on each phrase after the position lists for
4285 ** any deferred tokens have been loaded into memory. It updates the phrases
4286 ** current position list to include only those positions that are really
4287 ** instances of the phrase (after considering deferred tokens). If this
4288 ** means that the phrase does not appear in the current row, doclist.pList
4289 ** and doclist.nList are both zeroed.
4291 ** SQLITE_OK is returned if no error occurs, otherwise an SQLite error code.
4293 static int fts3EvalDeferredPhrase(Fts3Cursor *pCsr, Fts3Phrase *pPhrase){
4294 int iToken; /* Used to iterate through phrase tokens */
4295 char *aPoslist = 0; /* Position list for deferred tokens */
4296 int nPoslist = 0; /* Number of bytes in aPoslist */
4297 int iPrev = -1; /* Token number of previous deferred token */
4299 assert( pPhrase->doclist.bFreeList==0 );
4301 for(iToken=0; iToken<pPhrase->nToken; iToken++){
4302 Fts3PhraseToken *pToken = &pPhrase->aToken[iToken];
4303 Fts3DeferredToken *pDeferred = pToken->pDeferred;
4305 if( pDeferred ){
4306 char *pList;
4307 int nList;
4308 int rc = sqlite3Fts3DeferredTokenList(pDeferred, &pList, &nList);
4309 if( rc!=SQLITE_OK ) return rc;
4311 if( pList==0 ){
4312 sqlite3_free(aPoslist);
4313 pPhrase->doclist.pList = 0;
4314 pPhrase->doclist.nList = 0;
4315 return SQLITE_OK;
4317 }else if( aPoslist==0 ){
4318 aPoslist = pList;
4319 nPoslist = nList;
4321 }else{
4322 char *aOut = pList;
4323 char *p1 = aPoslist;
4324 char *p2 = aOut;
4326 assert( iPrev>=0 );
4327 fts3PoslistPhraseMerge(&aOut, iToken-iPrev, 0, 1, &p1, &p2);
4328 sqlite3_free(aPoslist);
4329 aPoslist = pList;
4330 nPoslist = (int)(aOut - aPoslist);
4331 if( nPoslist==0 ){
4332 sqlite3_free(aPoslist);
4333 pPhrase->doclist.pList = 0;
4334 pPhrase->doclist.nList = 0;
4335 return SQLITE_OK;
4338 iPrev = iToken;
4342 if( iPrev>=0 ){
4343 int nMaxUndeferred = pPhrase->iDoclistToken;
4344 if( nMaxUndeferred<0 ){
4345 pPhrase->doclist.pList = aPoslist;
4346 pPhrase->doclist.nList = nPoslist;
4347 pPhrase->doclist.iDocid = pCsr->iPrevId;
4348 pPhrase->doclist.bFreeList = 1;
4349 }else{
4350 int nDistance;
4351 char *p1;
4352 char *p2;
4353 char *aOut;
4355 if( nMaxUndeferred>iPrev ){
4356 p1 = aPoslist;
4357 p2 = pPhrase->doclist.pList;
4358 nDistance = nMaxUndeferred - iPrev;
4359 }else{
4360 p1 = pPhrase->doclist.pList;
4361 p2 = aPoslist;
4362 nDistance = iPrev - nMaxUndeferred;
4365 aOut = (char *)sqlite3_malloc(nPoslist+8);
4366 if( !aOut ){
4367 sqlite3_free(aPoslist);
4368 return SQLITE_NOMEM;
4371 pPhrase->doclist.pList = aOut;
4372 if( fts3PoslistPhraseMerge(&aOut, nDistance, 0, 1, &p1, &p2) ){
4373 pPhrase->doclist.bFreeList = 1;
4374 pPhrase->doclist.nList = (int)(aOut - pPhrase->doclist.pList);
4375 }else{
4376 sqlite3_free(aOut);
4377 pPhrase->doclist.pList = 0;
4378 pPhrase->doclist.nList = 0;
4380 sqlite3_free(aPoslist);
4384 return SQLITE_OK;
4386 #endif /* SQLITE_DISABLE_FTS4_DEFERRED */
4389 ** Maximum number of tokens a phrase may have to be considered for the
4390 ** incremental doclists strategy.
4392 #define MAX_INCR_PHRASE_TOKENS 4
4395 ** This function is called for each Fts3Phrase in a full-text query
4396 ** expression to initialize the mechanism for returning rows. Once this
4397 ** function has been called successfully on an Fts3Phrase, it may be
4398 ** used with fts3EvalPhraseNext() to iterate through the matching docids.
4400 ** If parameter bOptOk is true, then the phrase may (or may not) use the
4401 ** incremental loading strategy. Otherwise, the entire doclist is loaded into
4402 ** memory within this call.
4404 ** SQLITE_OK is returned if no error occurs, otherwise an SQLite error code.
4406 static int fts3EvalPhraseStart(Fts3Cursor *pCsr, int bOptOk, Fts3Phrase *p){
4407 Fts3Table *pTab = (Fts3Table *)pCsr->base.pVtab;
4408 int rc = SQLITE_OK; /* Error code */
4409 int i;
4411 /* Determine if doclists may be loaded from disk incrementally. This is
4412 ** possible if the bOptOk argument is true, the FTS doclists will be
4413 ** scanned in forward order, and the phrase consists of
4414 ** MAX_INCR_PHRASE_TOKENS or fewer tokens, none of which are are "^first"
4415 ** tokens or prefix tokens that cannot use a prefix-index. */
4416 int bHaveIncr = 0;
4417 int bIncrOk = (bOptOk
4418 && pCsr->bDesc==pTab->bDescIdx
4419 && p->nToken<=MAX_INCR_PHRASE_TOKENS && p->nToken>0
4420 #if defined(SQLITE_DEBUG) || defined(SQLITE_TEST)
4421 && pTab->bNoIncrDoclist==0
4422 #endif
4424 for(i=0; bIncrOk==1 && i<p->nToken; i++){
4425 Fts3PhraseToken *pToken = &p->aToken[i];
4426 if( pToken->bFirst || (pToken->pSegcsr!=0 && !pToken->pSegcsr->bLookup) ){
4427 bIncrOk = 0;
4429 if( pToken->pSegcsr ) bHaveIncr = 1;
4432 if( bIncrOk && bHaveIncr ){
4433 /* Use the incremental approach. */
4434 int iCol = (p->iColumn >= pTab->nColumn ? -1 : p->iColumn);
4435 for(i=0; rc==SQLITE_OK && i<p->nToken; i++){
4436 Fts3PhraseToken *pToken = &p->aToken[i];
4437 Fts3MultiSegReader *pSegcsr = pToken->pSegcsr;
4438 if( pSegcsr ){
4439 rc = sqlite3Fts3MsrIncrStart(pTab, pSegcsr, iCol, pToken->z, pToken->n);
4442 p->bIncr = 1;
4443 }else{
4444 /* Load the full doclist for the phrase into memory. */
4445 rc = fts3EvalPhraseLoad(pCsr, p);
4446 p->bIncr = 0;
4449 assert( rc!=SQLITE_OK || p->nToken<1 || p->aToken[0].pSegcsr==0 || p->bIncr );
4450 return rc;
4454 ** This function is used to iterate backwards (from the end to start)
4455 ** through doclists. It is used by this module to iterate through phrase
4456 ** doclists in reverse and by the fts3_write.c module to iterate through
4457 ** pending-terms lists when writing to databases with "order=desc".
4459 ** The doclist may be sorted in ascending (parameter bDescIdx==0) or
4460 ** descending (parameter bDescIdx==1) order of docid. Regardless, this
4461 ** function iterates from the end of the doclist to the beginning.
4463 void sqlite3Fts3DoclistPrev(
4464 int bDescIdx, /* True if the doclist is desc */
4465 char *aDoclist, /* Pointer to entire doclist */
4466 int nDoclist, /* Length of aDoclist in bytes */
4467 char **ppIter, /* IN/OUT: Iterator pointer */
4468 sqlite3_int64 *piDocid, /* IN/OUT: Docid pointer */
4469 int *pnList, /* OUT: List length pointer */
4470 u8 *pbEof /* OUT: End-of-file flag */
4472 char *p = *ppIter;
4474 assert( nDoclist>0 );
4475 assert( *pbEof==0 );
4476 assert_fts3_nc( p || *piDocid==0 );
4477 assert( !p || (p>aDoclist && p<&aDoclist[nDoclist]) );
4479 if( p==0 ){
4480 sqlite3_int64 iDocid = 0;
4481 char *pNext = 0;
4482 char *pDocid = aDoclist;
4483 char *pEnd = &aDoclist[nDoclist];
4484 int iMul = 1;
4486 while( pDocid<pEnd ){
4487 sqlite3_int64 iDelta;
4488 pDocid += sqlite3Fts3GetVarint(pDocid, &iDelta);
4489 iDocid += (iMul * iDelta);
4490 pNext = pDocid;
4491 fts3PoslistCopy(0, &pDocid);
4492 while( pDocid<pEnd && *pDocid==0 ) pDocid++;
4493 iMul = (bDescIdx ? -1 : 1);
4496 *pnList = (int)(pEnd - pNext);
4497 *ppIter = pNext;
4498 *piDocid = iDocid;
4499 }else{
4500 int iMul = (bDescIdx ? -1 : 1);
4501 sqlite3_int64 iDelta;
4502 fts3GetReverseVarint(&p, aDoclist, &iDelta);
4503 *piDocid -= (iMul * iDelta);
4505 if( p==aDoclist ){
4506 *pbEof = 1;
4507 }else{
4508 char *pSave = p;
4509 fts3ReversePoslist(aDoclist, &p);
4510 *pnList = (int)(pSave - p);
4512 *ppIter = p;
4517 ** Iterate forwards through a doclist.
4519 void sqlite3Fts3DoclistNext(
4520 int bDescIdx, /* True if the doclist is desc */
4521 char *aDoclist, /* Pointer to entire doclist */
4522 int nDoclist, /* Length of aDoclist in bytes */
4523 char **ppIter, /* IN/OUT: Iterator pointer */
4524 sqlite3_int64 *piDocid, /* IN/OUT: Docid pointer */
4525 u8 *pbEof /* OUT: End-of-file flag */
4527 char *p = *ppIter;
4529 assert( nDoclist>0 );
4530 assert( *pbEof==0 );
4531 assert_fts3_nc( p || *piDocid==0 );
4532 assert( !p || (p>=aDoclist && p<=&aDoclist[nDoclist]) );
4534 if( p==0 ){
4535 p = aDoclist;
4536 p += sqlite3Fts3GetVarint(p, piDocid);
4537 }else{
4538 fts3PoslistCopy(0, &p);
4539 while( p<&aDoclist[nDoclist] && *p==0 ) p++;
4540 if( p>=&aDoclist[nDoclist] ){
4541 *pbEof = 1;
4542 }else{
4543 sqlite3_int64 iVar;
4544 p += sqlite3Fts3GetVarint(p, &iVar);
4545 *piDocid += ((bDescIdx ? -1 : 1) * iVar);
4549 *ppIter = p;
4553 ** Advance the iterator pDL to the next entry in pDL->aAll/nAll. Set *pbEof
4554 ** to true if EOF is reached.
4556 static void fts3EvalDlPhraseNext(
4557 Fts3Table *pTab,
4558 Fts3Doclist *pDL,
4559 u8 *pbEof
4561 char *pIter; /* Used to iterate through aAll */
4562 char *pEnd; /* 1 byte past end of aAll */
4564 if( pDL->pNextDocid ){
4565 pIter = pDL->pNextDocid;
4566 assert( pDL->aAll!=0 || pIter==0 );
4567 }else{
4568 pIter = pDL->aAll;
4571 if( pIter==0 || pIter>=(pEnd = pDL->aAll + pDL->nAll) ){
4572 /* We have already reached the end of this doclist. EOF. */
4573 *pbEof = 1;
4574 }else{
4575 sqlite3_int64 iDelta;
4576 pIter += sqlite3Fts3GetVarint(pIter, &iDelta);
4577 if( pTab->bDescIdx==0 || pDL->pNextDocid==0 ){
4578 pDL->iDocid += iDelta;
4579 }else{
4580 pDL->iDocid -= iDelta;
4582 pDL->pList = pIter;
4583 fts3PoslistCopy(0, &pIter);
4584 pDL->nList = (int)(pIter - pDL->pList);
4586 /* pIter now points just past the 0x00 that terminates the position-
4587 ** list for document pDL->iDocid. However, if this position-list was
4588 ** edited in place by fts3EvalNearTrim(), then pIter may not actually
4589 ** point to the start of the next docid value. The following line deals
4590 ** with this case by advancing pIter past the zero-padding added by
4591 ** fts3EvalNearTrim(). */
4592 while( pIter<pEnd && *pIter==0 ) pIter++;
4594 pDL->pNextDocid = pIter;
4595 assert( pIter>=&pDL->aAll[pDL->nAll] || *pIter );
4596 *pbEof = 0;
4601 ** Helper type used by fts3EvalIncrPhraseNext() and incrPhraseTokenNext().
4603 typedef struct TokenDoclist TokenDoclist;
4604 struct TokenDoclist {
4605 int bIgnore;
4606 sqlite3_int64 iDocid;
4607 char *pList;
4608 int nList;
4612 ** Token pToken is an incrementally loaded token that is part of a
4613 ** multi-token phrase. Advance it to the next matching document in the
4614 ** database and populate output variable *p with the details of the new
4615 ** entry. Or, if the iterator has reached EOF, set *pbEof to true.
4617 ** If an error occurs, return an SQLite error code. Otherwise, return
4618 ** SQLITE_OK.
4620 static int incrPhraseTokenNext(
4621 Fts3Table *pTab, /* Virtual table handle */
4622 Fts3Phrase *pPhrase, /* Phrase to advance token of */
4623 int iToken, /* Specific token to advance */
4624 TokenDoclist *p, /* OUT: Docid and doclist for new entry */
4625 u8 *pbEof /* OUT: True if iterator is at EOF */
4627 int rc = SQLITE_OK;
4629 if( pPhrase->iDoclistToken==iToken ){
4630 assert( p->bIgnore==0 );
4631 assert( pPhrase->aToken[iToken].pSegcsr==0 );
4632 fts3EvalDlPhraseNext(pTab, &pPhrase->doclist, pbEof);
4633 p->pList = pPhrase->doclist.pList;
4634 p->nList = pPhrase->doclist.nList;
4635 p->iDocid = pPhrase->doclist.iDocid;
4636 }else{
4637 Fts3PhraseToken *pToken = &pPhrase->aToken[iToken];
4638 assert( pToken->pDeferred==0 );
4639 assert( pToken->pSegcsr || pPhrase->iDoclistToken>=0 );
4640 if( pToken->pSegcsr ){
4641 assert( p->bIgnore==0 );
4642 rc = sqlite3Fts3MsrIncrNext(
4643 pTab, pToken->pSegcsr, &p->iDocid, &p->pList, &p->nList
4645 if( p->pList==0 ) *pbEof = 1;
4646 }else{
4647 p->bIgnore = 1;
4651 return rc;
4656 ** The phrase iterator passed as the second argument:
4658 ** * features at least one token that uses an incremental doclist, and
4660 ** * does not contain any deferred tokens.
4662 ** Advance it to the next matching documnent in the database and populate
4663 ** the Fts3Doclist.pList and nList fields.
4665 ** If there is no "next" entry and no error occurs, then *pbEof is set to
4666 ** 1 before returning. Otherwise, if no error occurs and the iterator is
4667 ** successfully advanced, *pbEof is set to 0.
4669 ** If an error occurs, return an SQLite error code. Otherwise, return
4670 ** SQLITE_OK.
4672 static int fts3EvalIncrPhraseNext(
4673 Fts3Cursor *pCsr, /* FTS Cursor handle */
4674 Fts3Phrase *p, /* Phrase object to advance to next docid */
4675 u8 *pbEof /* OUT: Set to 1 if EOF */
4677 int rc = SQLITE_OK;
4678 Fts3Doclist *pDL = &p->doclist;
4679 Fts3Table *pTab = (Fts3Table *)pCsr->base.pVtab;
4680 u8 bEof = 0;
4682 /* This is only called if it is guaranteed that the phrase has at least
4683 ** one incremental token. In which case the bIncr flag is set. */
4684 assert( p->bIncr==1 );
4686 if( p->nToken==1 ){
4687 rc = sqlite3Fts3MsrIncrNext(pTab, p->aToken[0].pSegcsr,
4688 &pDL->iDocid, &pDL->pList, &pDL->nList
4690 if( pDL->pList==0 ) bEof = 1;
4691 }else{
4692 int bDescDoclist = pCsr->bDesc;
4693 struct TokenDoclist a[MAX_INCR_PHRASE_TOKENS];
4695 memset(a, 0, sizeof(a));
4696 assert( p->nToken<=MAX_INCR_PHRASE_TOKENS );
4697 assert( p->iDoclistToken<MAX_INCR_PHRASE_TOKENS );
4699 while( bEof==0 ){
4700 int bMaxSet = 0;
4701 sqlite3_int64 iMax = 0; /* Largest docid for all iterators */
4702 int i; /* Used to iterate through tokens */
4704 /* Advance the iterator for each token in the phrase once. */
4705 for(i=0; rc==SQLITE_OK && i<p->nToken && bEof==0; i++){
4706 rc = incrPhraseTokenNext(pTab, p, i, &a[i], &bEof);
4707 if( a[i].bIgnore==0 && (bMaxSet==0 || DOCID_CMP(iMax, a[i].iDocid)<0) ){
4708 iMax = a[i].iDocid;
4709 bMaxSet = 1;
4712 assert( rc!=SQLITE_OK || (p->nToken>=1 && a[p->nToken-1].bIgnore==0) );
4713 assert( rc!=SQLITE_OK || bMaxSet );
4715 /* Keep advancing iterators until they all point to the same document */
4716 for(i=0; i<p->nToken; i++){
4717 while( rc==SQLITE_OK && bEof==0
4718 && a[i].bIgnore==0 && DOCID_CMP(a[i].iDocid, iMax)<0
4720 rc = incrPhraseTokenNext(pTab, p, i, &a[i], &bEof);
4721 if( DOCID_CMP(a[i].iDocid, iMax)>0 ){
4722 iMax = a[i].iDocid;
4723 i = 0;
4728 /* Check if the current entries really are a phrase match */
4729 if( bEof==0 ){
4730 int nList = 0;
4731 int nByte = a[p->nToken-1].nList;
4732 char *aDoclist = sqlite3_malloc(nByte+FTS3_BUFFER_PADDING);
4733 if( !aDoclist ) return SQLITE_NOMEM;
4734 memcpy(aDoclist, a[p->nToken-1].pList, nByte+1);
4735 memset(&aDoclist[nByte], 0, FTS3_BUFFER_PADDING);
4737 for(i=0; i<(p->nToken-1); i++){
4738 if( a[i].bIgnore==0 ){
4739 char *pL = a[i].pList;
4740 char *pR = aDoclist;
4741 char *pOut = aDoclist;
4742 int nDist = p->nToken-1-i;
4743 int res = fts3PoslistPhraseMerge(&pOut, nDist, 0, 1, &pL, &pR);
4744 if( res==0 ) break;
4745 nList = (int)(pOut - aDoclist);
4748 if( i==(p->nToken-1) ){
4749 pDL->iDocid = iMax;
4750 pDL->pList = aDoclist;
4751 pDL->nList = nList;
4752 pDL->bFreeList = 1;
4753 break;
4755 sqlite3_free(aDoclist);
4760 *pbEof = bEof;
4761 return rc;
4765 ** Attempt to move the phrase iterator to point to the next matching docid.
4766 ** If an error occurs, return an SQLite error code. Otherwise, return
4767 ** SQLITE_OK.
4769 ** If there is no "next" entry and no error occurs, then *pbEof is set to
4770 ** 1 before returning. Otherwise, if no error occurs and the iterator is
4771 ** successfully advanced, *pbEof is set to 0.
4773 static int fts3EvalPhraseNext(
4774 Fts3Cursor *pCsr, /* FTS Cursor handle */
4775 Fts3Phrase *p, /* Phrase object to advance to next docid */
4776 u8 *pbEof /* OUT: Set to 1 if EOF */
4778 int rc = SQLITE_OK;
4779 Fts3Doclist *pDL = &p->doclist;
4780 Fts3Table *pTab = (Fts3Table *)pCsr->base.pVtab;
4782 if( p->bIncr ){
4783 rc = fts3EvalIncrPhraseNext(pCsr, p, pbEof);
4784 }else if( pCsr->bDesc!=pTab->bDescIdx && pDL->nAll ){
4785 sqlite3Fts3DoclistPrev(pTab->bDescIdx, pDL->aAll, pDL->nAll,
4786 &pDL->pNextDocid, &pDL->iDocid, &pDL->nList, pbEof
4788 pDL->pList = pDL->pNextDocid;
4789 }else{
4790 fts3EvalDlPhraseNext(pTab, pDL, pbEof);
4793 return rc;
4798 ** If *pRc is not SQLITE_OK when this function is called, it is a no-op.
4799 ** Otherwise, fts3EvalPhraseStart() is called on all phrases within the
4800 ** expression. Also the Fts3Expr.bDeferred variable is set to true for any
4801 ** expressions for which all descendent tokens are deferred.
4803 ** If parameter bOptOk is zero, then it is guaranteed that the
4804 ** Fts3Phrase.doclist.aAll/nAll variables contain the entire doclist for
4805 ** each phrase in the expression (subject to deferred token processing).
4806 ** Or, if bOptOk is non-zero, then one or more tokens within the expression
4807 ** may be loaded incrementally, meaning doclist.aAll/nAll is not available.
4809 ** If an error occurs within this function, *pRc is set to an SQLite error
4810 ** code before returning.
4812 static void fts3EvalStartReaders(
4813 Fts3Cursor *pCsr, /* FTS Cursor handle */
4814 Fts3Expr *pExpr, /* Expression to initialize phrases in */
4815 int *pRc /* IN/OUT: Error code */
4817 if( pExpr && SQLITE_OK==*pRc ){
4818 if( pExpr->eType==FTSQUERY_PHRASE ){
4819 int nToken = pExpr->pPhrase->nToken;
4820 if( nToken ){
4821 int i;
4822 for(i=0; i<nToken; i++){
4823 if( pExpr->pPhrase->aToken[i].pDeferred==0 ) break;
4825 pExpr->bDeferred = (i==nToken);
4827 *pRc = fts3EvalPhraseStart(pCsr, 1, pExpr->pPhrase);
4828 }else{
4829 fts3EvalStartReaders(pCsr, pExpr->pLeft, pRc);
4830 fts3EvalStartReaders(pCsr, pExpr->pRight, pRc);
4831 pExpr->bDeferred = (pExpr->pLeft->bDeferred && pExpr->pRight->bDeferred);
4837 ** An array of the following structures is assembled as part of the process
4838 ** of selecting tokens to defer before the query starts executing (as part
4839 ** of the xFilter() method). There is one element in the array for each
4840 ** token in the FTS expression.
4842 ** Tokens are divided into AND/NEAR clusters. All tokens in a cluster belong
4843 ** to phrases that are connected only by AND and NEAR operators (not OR or
4844 ** NOT). When determining tokens to defer, each AND/NEAR cluster is considered
4845 ** separately. The root of a tokens AND/NEAR cluster is stored in
4846 ** Fts3TokenAndCost.pRoot.
4848 typedef struct Fts3TokenAndCost Fts3TokenAndCost;
4849 struct Fts3TokenAndCost {
4850 Fts3Phrase *pPhrase; /* The phrase the token belongs to */
4851 int iToken; /* Position of token in phrase */
4852 Fts3PhraseToken *pToken; /* The token itself */
4853 Fts3Expr *pRoot; /* Root of NEAR/AND cluster */
4854 int nOvfl; /* Number of overflow pages to load doclist */
4855 int iCol; /* The column the token must match */
4859 ** This function is used to populate an allocated Fts3TokenAndCost array.
4861 ** If *pRc is not SQLITE_OK when this function is called, it is a no-op.
4862 ** Otherwise, if an error occurs during execution, *pRc is set to an
4863 ** SQLite error code.
4865 static void fts3EvalTokenCosts(
4866 Fts3Cursor *pCsr, /* FTS Cursor handle */
4867 Fts3Expr *pRoot, /* Root of current AND/NEAR cluster */
4868 Fts3Expr *pExpr, /* Expression to consider */
4869 Fts3TokenAndCost **ppTC, /* Write new entries to *(*ppTC)++ */
4870 Fts3Expr ***ppOr, /* Write new OR root to *(*ppOr)++ */
4871 int *pRc /* IN/OUT: Error code */
4873 if( *pRc==SQLITE_OK ){
4874 if( pExpr->eType==FTSQUERY_PHRASE ){
4875 Fts3Phrase *pPhrase = pExpr->pPhrase;
4876 int i;
4877 for(i=0; *pRc==SQLITE_OK && i<pPhrase->nToken; i++){
4878 Fts3TokenAndCost *pTC = (*ppTC)++;
4879 pTC->pPhrase = pPhrase;
4880 pTC->iToken = i;
4881 pTC->pRoot = pRoot;
4882 pTC->pToken = &pPhrase->aToken[i];
4883 pTC->iCol = pPhrase->iColumn;
4884 *pRc = sqlite3Fts3MsrOvfl(pCsr, pTC->pToken->pSegcsr, &pTC->nOvfl);
4886 }else if( pExpr->eType!=FTSQUERY_NOT ){
4887 assert( pExpr->eType==FTSQUERY_OR
4888 || pExpr->eType==FTSQUERY_AND
4889 || pExpr->eType==FTSQUERY_NEAR
4891 assert( pExpr->pLeft && pExpr->pRight );
4892 if( pExpr->eType==FTSQUERY_OR ){
4893 pRoot = pExpr->pLeft;
4894 **ppOr = pRoot;
4895 (*ppOr)++;
4897 fts3EvalTokenCosts(pCsr, pRoot, pExpr->pLeft, ppTC, ppOr, pRc);
4898 if( pExpr->eType==FTSQUERY_OR ){
4899 pRoot = pExpr->pRight;
4900 **ppOr = pRoot;
4901 (*ppOr)++;
4903 fts3EvalTokenCosts(pCsr, pRoot, pExpr->pRight, ppTC, ppOr, pRc);
4909 ** Determine the average document (row) size in pages. If successful,
4910 ** write this value to *pnPage and return SQLITE_OK. Otherwise, return
4911 ** an SQLite error code.
4913 ** The average document size in pages is calculated by first calculating
4914 ** determining the average size in bytes, B. If B is less than the amount
4915 ** of data that will fit on a single leaf page of an intkey table in
4916 ** this database, then the average docsize is 1. Otherwise, it is 1 plus
4917 ** the number of overflow pages consumed by a record B bytes in size.
4919 static int fts3EvalAverageDocsize(Fts3Cursor *pCsr, int *pnPage){
4920 int rc = SQLITE_OK;
4921 if( pCsr->nRowAvg==0 ){
4922 /* The average document size, which is required to calculate the cost
4923 ** of each doclist, has not yet been determined. Read the required
4924 ** data from the %_stat table to calculate it.
4926 ** Entry 0 of the %_stat table is a blob containing (nCol+1) FTS3
4927 ** varints, where nCol is the number of columns in the FTS3 table.
4928 ** The first varint is the number of documents currently stored in
4929 ** the table. The following nCol varints contain the total amount of
4930 ** data stored in all rows of each column of the table, from left
4931 ** to right.
4933 Fts3Table *p = (Fts3Table*)pCsr->base.pVtab;
4934 sqlite3_stmt *pStmt;
4935 sqlite3_int64 nDoc = 0;
4936 sqlite3_int64 nByte = 0;
4937 const char *pEnd;
4938 const char *a;
4940 rc = sqlite3Fts3SelectDoctotal(p, &pStmt);
4941 if( rc!=SQLITE_OK ) return rc;
4942 a = sqlite3_column_blob(pStmt, 0);
4943 testcase( a==0 ); /* If %_stat.value set to X'' */
4944 if( a ){
4945 pEnd = &a[sqlite3_column_bytes(pStmt, 0)];
4946 a += sqlite3Fts3GetVarintBounded(a, pEnd, &nDoc);
4947 while( a<pEnd ){
4948 a += sqlite3Fts3GetVarintBounded(a, pEnd, &nByte);
4951 if( nDoc==0 || nByte==0 ){
4952 sqlite3_reset(pStmt);
4953 return FTS_CORRUPT_VTAB;
4956 pCsr->nDoc = nDoc;
4957 pCsr->nRowAvg = (int)(((nByte / nDoc) + p->nPgsz) / p->nPgsz);
4958 assert( pCsr->nRowAvg>0 );
4959 rc = sqlite3_reset(pStmt);
4962 *pnPage = pCsr->nRowAvg;
4963 return rc;
4967 ** This function is called to select the tokens (if any) that will be
4968 ** deferred. The array aTC[] has already been populated when this is
4969 ** called.
4971 ** This function is called once for each AND/NEAR cluster in the
4972 ** expression. Each invocation determines which tokens to defer within
4973 ** the cluster with root node pRoot. See comments above the definition
4974 ** of struct Fts3TokenAndCost for more details.
4976 ** If no error occurs, SQLITE_OK is returned and sqlite3Fts3DeferToken()
4977 ** called on each token to defer. Otherwise, an SQLite error code is
4978 ** returned.
4980 static int fts3EvalSelectDeferred(
4981 Fts3Cursor *pCsr, /* FTS Cursor handle */
4982 Fts3Expr *pRoot, /* Consider tokens with this root node */
4983 Fts3TokenAndCost *aTC, /* Array of expression tokens and costs */
4984 int nTC /* Number of entries in aTC[] */
4986 Fts3Table *pTab = (Fts3Table *)pCsr->base.pVtab;
4987 int nDocSize = 0; /* Number of pages per doc loaded */
4988 int rc = SQLITE_OK; /* Return code */
4989 int ii; /* Iterator variable for various purposes */
4990 int nOvfl = 0; /* Total overflow pages used by doclists */
4991 int nToken = 0; /* Total number of tokens in cluster */
4993 int nMinEst = 0; /* The minimum count for any phrase so far. */
4994 int nLoad4 = 1; /* (Phrases that will be loaded)^4. */
4996 /* Tokens are never deferred for FTS tables created using the content=xxx
4997 ** option. The reason being that it is not guaranteed that the content
4998 ** table actually contains the same data as the index. To prevent this from
4999 ** causing any problems, the deferred token optimization is completely
5000 ** disabled for content=xxx tables. */
5001 if( pTab->zContentTbl ){
5002 return SQLITE_OK;
5005 /* Count the tokens in this AND/NEAR cluster. If none of the doclists
5006 ** associated with the tokens spill onto overflow pages, or if there is
5007 ** only 1 token, exit early. No tokens to defer in this case. */
5008 for(ii=0; ii<nTC; ii++){
5009 if( aTC[ii].pRoot==pRoot ){
5010 nOvfl += aTC[ii].nOvfl;
5011 nToken++;
5014 if( nOvfl==0 || nToken<2 ) return SQLITE_OK;
5016 /* Obtain the average docsize (in pages). */
5017 rc = fts3EvalAverageDocsize(pCsr, &nDocSize);
5018 assert( rc!=SQLITE_OK || nDocSize>0 );
5021 /* Iterate through all tokens in this AND/NEAR cluster, in ascending order
5022 ** of the number of overflow pages that will be loaded by the pager layer
5023 ** to retrieve the entire doclist for the token from the full-text index.
5024 ** Load the doclists for tokens that are either:
5026 ** a. The cheapest token in the entire query (i.e. the one visited by the
5027 ** first iteration of this loop), or
5029 ** b. Part of a multi-token phrase.
5031 ** After each token doclist is loaded, merge it with the others from the
5032 ** same phrase and count the number of documents that the merged doclist
5033 ** contains. Set variable "nMinEst" to the smallest number of documents in
5034 ** any phrase doclist for which 1 or more token doclists have been loaded.
5035 ** Let nOther be the number of other phrases for which it is certain that
5036 ** one or more tokens will not be deferred.
5038 ** Then, for each token, defer it if loading the doclist would result in
5039 ** loading N or more overflow pages into memory, where N is computed as:
5041 ** (nMinEst + 4^nOther - 1) / (4^nOther)
5043 for(ii=0; ii<nToken && rc==SQLITE_OK; ii++){
5044 int iTC; /* Used to iterate through aTC[] array. */
5045 Fts3TokenAndCost *pTC = 0; /* Set to cheapest remaining token. */
5047 /* Set pTC to point to the cheapest remaining token. */
5048 for(iTC=0; iTC<nTC; iTC++){
5049 if( aTC[iTC].pToken && aTC[iTC].pRoot==pRoot
5050 && (!pTC || aTC[iTC].nOvfl<pTC->nOvfl)
5052 pTC = &aTC[iTC];
5055 assert( pTC );
5057 if( ii && pTC->nOvfl>=((nMinEst+(nLoad4/4)-1)/(nLoad4/4))*nDocSize ){
5058 /* The number of overflow pages to load for this (and therefore all
5059 ** subsequent) tokens is greater than the estimated number of pages
5060 ** that will be loaded if all subsequent tokens are deferred.
5062 Fts3PhraseToken *pToken = pTC->pToken;
5063 rc = sqlite3Fts3DeferToken(pCsr, pToken, pTC->iCol);
5064 fts3SegReaderCursorFree(pToken->pSegcsr);
5065 pToken->pSegcsr = 0;
5066 }else{
5067 /* Set nLoad4 to the value of (4^nOther) for the next iteration of the
5068 ** for-loop. Except, limit the value to 2^24 to prevent it from
5069 ** overflowing the 32-bit integer it is stored in. */
5070 if( ii<12 ) nLoad4 = nLoad4*4;
5072 if( ii==0 || (pTC->pPhrase->nToken>1 && ii!=nToken-1) ){
5073 /* Either this is the cheapest token in the entire query, or it is
5074 ** part of a multi-token phrase. Either way, the entire doclist will
5075 ** (eventually) be loaded into memory. It may as well be now. */
5076 Fts3PhraseToken *pToken = pTC->pToken;
5077 int nList = 0;
5078 char *pList = 0;
5079 rc = fts3TermSelect(pTab, pToken, pTC->iCol, &nList, &pList);
5080 assert( rc==SQLITE_OK || pList==0 );
5081 if( rc==SQLITE_OK ){
5082 rc = fts3EvalPhraseMergeToken(
5083 pTab, pTC->pPhrase, pTC->iToken,pList,nList
5086 if( rc==SQLITE_OK ){
5087 int nCount;
5088 nCount = fts3DoclistCountDocids(
5089 pTC->pPhrase->doclist.aAll, pTC->pPhrase->doclist.nAll
5091 if( ii==0 || nCount<nMinEst ) nMinEst = nCount;
5095 pTC->pToken = 0;
5098 return rc;
5102 ** This function is called from within the xFilter method. It initializes
5103 ** the full-text query currently stored in pCsr->pExpr. To iterate through
5104 ** the results of a query, the caller does:
5106 ** fts3EvalStart(pCsr);
5107 ** while( 1 ){
5108 ** fts3EvalNext(pCsr);
5109 ** if( pCsr->bEof ) break;
5110 ** ... return row pCsr->iPrevId to the caller ...
5111 ** }
5113 static int fts3EvalStart(Fts3Cursor *pCsr){
5114 Fts3Table *pTab = (Fts3Table *)pCsr->base.pVtab;
5115 int rc = SQLITE_OK;
5116 int nToken = 0;
5117 int nOr = 0;
5119 /* Allocate a MultiSegReader for each token in the expression. */
5120 fts3EvalAllocateReaders(pCsr, pCsr->pExpr, &nToken, &nOr, &rc);
5122 /* Determine which, if any, tokens in the expression should be deferred. */
5123 #ifndef SQLITE_DISABLE_FTS4_DEFERRED
5124 if( rc==SQLITE_OK && nToken>1 && pTab->bFts4 ){
5125 Fts3TokenAndCost *aTC;
5126 aTC = (Fts3TokenAndCost *)sqlite3_malloc64(
5127 sizeof(Fts3TokenAndCost) * nToken
5128 + sizeof(Fts3Expr *) * nOr * 2
5131 if( !aTC ){
5132 rc = SQLITE_NOMEM;
5133 }else{
5134 Fts3Expr **apOr = (Fts3Expr **)&aTC[nToken];
5135 int ii;
5136 Fts3TokenAndCost *pTC = aTC;
5137 Fts3Expr **ppOr = apOr;
5139 fts3EvalTokenCosts(pCsr, 0, pCsr->pExpr, &pTC, &ppOr, &rc);
5140 nToken = (int)(pTC-aTC);
5141 nOr = (int)(ppOr-apOr);
5143 if( rc==SQLITE_OK ){
5144 rc = fts3EvalSelectDeferred(pCsr, 0, aTC, nToken);
5145 for(ii=0; rc==SQLITE_OK && ii<nOr; ii++){
5146 rc = fts3EvalSelectDeferred(pCsr, apOr[ii], aTC, nToken);
5150 sqlite3_free(aTC);
5153 #endif
5155 fts3EvalStartReaders(pCsr, pCsr->pExpr, &rc);
5156 return rc;
5160 ** Invalidate the current position list for phrase pPhrase.
5162 static void fts3EvalInvalidatePoslist(Fts3Phrase *pPhrase){
5163 if( pPhrase->doclist.bFreeList ){
5164 sqlite3_free(pPhrase->doclist.pList);
5166 pPhrase->doclist.pList = 0;
5167 pPhrase->doclist.nList = 0;
5168 pPhrase->doclist.bFreeList = 0;
5172 ** This function is called to edit the position list associated with
5173 ** the phrase object passed as the fifth argument according to a NEAR
5174 ** condition. For example:
5176 ** abc NEAR/5 "def ghi"
5178 ** Parameter nNear is passed the NEAR distance of the expression (5 in
5179 ** the example above). When this function is called, *paPoslist points to
5180 ** the position list, and *pnToken is the number of phrase tokens in the
5181 ** phrase on the other side of the NEAR operator to pPhrase. For example,
5182 ** if pPhrase refers to the "def ghi" phrase, then *paPoslist points to
5183 ** the position list associated with phrase "abc".
5185 ** All positions in the pPhrase position list that are not sufficiently
5186 ** close to a position in the *paPoslist position list are removed. If this
5187 ** leaves 0 positions, zero is returned. Otherwise, non-zero.
5189 ** Before returning, *paPoslist is set to point to the position lsit
5190 ** associated with pPhrase. And *pnToken is set to the number of tokens in
5191 ** pPhrase.
5193 static int fts3EvalNearTrim(
5194 int nNear, /* NEAR distance. As in "NEAR/nNear". */
5195 char *aTmp, /* Temporary space to use */
5196 char **paPoslist, /* IN/OUT: Position list */
5197 int *pnToken, /* IN/OUT: Tokens in phrase of *paPoslist */
5198 Fts3Phrase *pPhrase /* The phrase object to trim the doclist of */
5200 int nParam1 = nNear + pPhrase->nToken;
5201 int nParam2 = nNear + *pnToken;
5202 int nNew;
5203 char *p2;
5204 char *pOut;
5205 int res;
5207 assert( pPhrase->doclist.pList );
5209 p2 = pOut = pPhrase->doclist.pList;
5210 res = fts3PoslistNearMerge(
5211 &pOut, aTmp, nParam1, nParam2, paPoslist, &p2
5213 if( res ){
5214 nNew = (int)(pOut - pPhrase->doclist.pList) - 1;
5215 assert_fts3_nc( nNew<=pPhrase->doclist.nList && nNew>0 );
5216 if( nNew>=0 && nNew<=pPhrase->doclist.nList ){
5217 assert( pPhrase->doclist.pList[nNew]=='\0' );
5218 memset(&pPhrase->doclist.pList[nNew], 0, pPhrase->doclist.nList - nNew);
5219 pPhrase->doclist.nList = nNew;
5221 *paPoslist = pPhrase->doclist.pList;
5222 *pnToken = pPhrase->nToken;
5225 return res;
5229 ** This function is a no-op if *pRc is other than SQLITE_OK when it is called.
5230 ** Otherwise, it advances the expression passed as the second argument to
5231 ** point to the next matching row in the database. Expressions iterate through
5232 ** matching rows in docid order. Ascending order if Fts3Cursor.bDesc is zero,
5233 ** or descending if it is non-zero.
5235 ** If an error occurs, *pRc is set to an SQLite error code. Otherwise, if
5236 ** successful, the following variables in pExpr are set:
5238 ** Fts3Expr.bEof (non-zero if EOF - there is no next row)
5239 ** Fts3Expr.iDocid (valid if bEof==0. The docid of the next row)
5241 ** If the expression is of type FTSQUERY_PHRASE, and the expression is not
5242 ** at EOF, then the following variables are populated with the position list
5243 ** for the phrase for the visited row:
5245 ** FTs3Expr.pPhrase->doclist.nList (length of pList in bytes)
5246 ** FTs3Expr.pPhrase->doclist.pList (pointer to position list)
5248 ** It says above that this function advances the expression to the next
5249 ** matching row. This is usually true, but there are the following exceptions:
5251 ** 1. Deferred tokens are not taken into account. If a phrase consists
5252 ** entirely of deferred tokens, it is assumed to match every row in
5253 ** the db. In this case the position-list is not populated at all.
5255 ** Or, if a phrase contains one or more deferred tokens and one or
5256 ** more non-deferred tokens, then the expression is advanced to the
5257 ** next possible match, considering only non-deferred tokens. In other
5258 ** words, if the phrase is "A B C", and "B" is deferred, the expression
5259 ** is advanced to the next row that contains an instance of "A * C",
5260 ** where "*" may match any single token. The position list in this case
5261 ** is populated as for "A * C" before returning.
5263 ** 2. NEAR is treated as AND. If the expression is "x NEAR y", it is
5264 ** advanced to point to the next row that matches "x AND y".
5266 ** See sqlite3Fts3EvalTestDeferred() for details on testing if a row is
5267 ** really a match, taking into account deferred tokens and NEAR operators.
5269 static void fts3EvalNextRow(
5270 Fts3Cursor *pCsr, /* FTS Cursor handle */
5271 Fts3Expr *pExpr, /* Expr. to advance to next matching row */
5272 int *pRc /* IN/OUT: Error code */
5274 if( *pRc==SQLITE_OK ){
5275 int bDescDoclist = pCsr->bDesc; /* Used by DOCID_CMP() macro */
5276 assert( pExpr->bEof==0 );
5277 pExpr->bStart = 1;
5279 switch( pExpr->eType ){
5280 case FTSQUERY_NEAR:
5281 case FTSQUERY_AND: {
5282 Fts3Expr *pLeft = pExpr->pLeft;
5283 Fts3Expr *pRight = pExpr->pRight;
5284 assert( !pLeft->bDeferred || !pRight->bDeferred );
5286 if( pLeft->bDeferred ){
5287 /* LHS is entirely deferred. So we assume it matches every row.
5288 ** Advance the RHS iterator to find the next row visited. */
5289 fts3EvalNextRow(pCsr, pRight, pRc);
5290 pExpr->iDocid = pRight->iDocid;
5291 pExpr->bEof = pRight->bEof;
5292 }else if( pRight->bDeferred ){
5293 /* RHS is entirely deferred. So we assume it matches every row.
5294 ** Advance the LHS iterator to find the next row visited. */
5295 fts3EvalNextRow(pCsr, pLeft, pRc);
5296 pExpr->iDocid = pLeft->iDocid;
5297 pExpr->bEof = pLeft->bEof;
5298 }else{
5299 /* Neither the RHS or LHS are deferred. */
5300 fts3EvalNextRow(pCsr, pLeft, pRc);
5301 fts3EvalNextRow(pCsr, pRight, pRc);
5302 while( !pLeft->bEof && !pRight->bEof && *pRc==SQLITE_OK ){
5303 sqlite3_int64 iDiff = DOCID_CMP(pLeft->iDocid, pRight->iDocid);
5304 if( iDiff==0 ) break;
5305 if( iDiff<0 ){
5306 fts3EvalNextRow(pCsr, pLeft, pRc);
5307 }else{
5308 fts3EvalNextRow(pCsr, pRight, pRc);
5311 pExpr->iDocid = pLeft->iDocid;
5312 pExpr->bEof = (pLeft->bEof || pRight->bEof);
5313 if( pExpr->eType==FTSQUERY_NEAR && pExpr->bEof ){
5314 assert( pRight->eType==FTSQUERY_PHRASE );
5315 if( pRight->pPhrase->doclist.aAll ){
5316 Fts3Doclist *pDl = &pRight->pPhrase->doclist;
5317 while( *pRc==SQLITE_OK && pRight->bEof==0 ){
5318 memset(pDl->pList, 0, pDl->nList);
5319 fts3EvalNextRow(pCsr, pRight, pRc);
5322 if( pLeft->pPhrase && pLeft->pPhrase->doclist.aAll ){
5323 Fts3Doclist *pDl = &pLeft->pPhrase->doclist;
5324 while( *pRc==SQLITE_OK && pLeft->bEof==0 ){
5325 memset(pDl->pList, 0, pDl->nList);
5326 fts3EvalNextRow(pCsr, pLeft, pRc);
5329 pRight->bEof = pLeft->bEof = 1;
5332 break;
5335 case FTSQUERY_OR: {
5336 Fts3Expr *pLeft = pExpr->pLeft;
5337 Fts3Expr *pRight = pExpr->pRight;
5338 sqlite3_int64 iCmp = DOCID_CMP(pLeft->iDocid, pRight->iDocid);
5340 assert_fts3_nc( pLeft->bStart || pLeft->iDocid==pRight->iDocid );
5341 assert_fts3_nc( pRight->bStart || pLeft->iDocid==pRight->iDocid );
5343 if( pRight->bEof || (pLeft->bEof==0 && iCmp<0) ){
5344 fts3EvalNextRow(pCsr, pLeft, pRc);
5345 }else if( pLeft->bEof || iCmp>0 ){
5346 fts3EvalNextRow(pCsr, pRight, pRc);
5347 }else{
5348 fts3EvalNextRow(pCsr, pLeft, pRc);
5349 fts3EvalNextRow(pCsr, pRight, pRc);
5352 pExpr->bEof = (pLeft->bEof && pRight->bEof);
5353 iCmp = DOCID_CMP(pLeft->iDocid, pRight->iDocid);
5354 if( pRight->bEof || (pLeft->bEof==0 && iCmp<0) ){
5355 pExpr->iDocid = pLeft->iDocid;
5356 }else{
5357 pExpr->iDocid = pRight->iDocid;
5360 break;
5363 case FTSQUERY_NOT: {
5364 Fts3Expr *pLeft = pExpr->pLeft;
5365 Fts3Expr *pRight = pExpr->pRight;
5367 if( pRight->bStart==0 ){
5368 fts3EvalNextRow(pCsr, pRight, pRc);
5369 assert( *pRc!=SQLITE_OK || pRight->bStart );
5372 fts3EvalNextRow(pCsr, pLeft, pRc);
5373 if( pLeft->bEof==0 ){
5374 while( !*pRc
5375 && !pRight->bEof
5376 && DOCID_CMP(pLeft->iDocid, pRight->iDocid)>0
5378 fts3EvalNextRow(pCsr, pRight, pRc);
5381 pExpr->iDocid = pLeft->iDocid;
5382 pExpr->bEof = pLeft->bEof;
5383 break;
5386 default: {
5387 Fts3Phrase *pPhrase = pExpr->pPhrase;
5388 fts3EvalInvalidatePoslist(pPhrase);
5389 *pRc = fts3EvalPhraseNext(pCsr, pPhrase, &pExpr->bEof);
5390 pExpr->iDocid = pPhrase->doclist.iDocid;
5391 break;
5398 ** If *pRc is not SQLITE_OK, or if pExpr is not the root node of a NEAR
5399 ** cluster, then this function returns 1 immediately.
5401 ** Otherwise, it checks if the current row really does match the NEAR
5402 ** expression, using the data currently stored in the position lists
5403 ** (Fts3Expr->pPhrase.doclist.pList/nList) for each phrase in the expression.
5405 ** If the current row is a match, the position list associated with each
5406 ** phrase in the NEAR expression is edited in place to contain only those
5407 ** phrase instances sufficiently close to their peers to satisfy all NEAR
5408 ** constraints. In this case it returns 1. If the NEAR expression does not
5409 ** match the current row, 0 is returned. The position lists may or may not
5410 ** be edited if 0 is returned.
5412 static int fts3EvalNearTest(Fts3Expr *pExpr, int *pRc){
5413 int res = 1;
5415 /* The following block runs if pExpr is the root of a NEAR query.
5416 ** For example, the query:
5418 ** "w" NEAR "x" NEAR "y" NEAR "z"
5420 ** which is represented in tree form as:
5422 ** |
5423 ** +--NEAR--+ <-- root of NEAR query
5424 ** | |
5425 ** +--NEAR--+ "z"
5426 ** | |
5427 ** +--NEAR--+ "y"
5428 ** | |
5429 ** "w" "x"
5431 ** The right-hand child of a NEAR node is always a phrase. The
5432 ** left-hand child may be either a phrase or a NEAR node. There are
5433 ** no exceptions to this - it's the way the parser in fts3_expr.c works.
5435 if( *pRc==SQLITE_OK
5436 && pExpr->eType==FTSQUERY_NEAR
5437 && (pExpr->pParent==0 || pExpr->pParent->eType!=FTSQUERY_NEAR)
5439 Fts3Expr *p;
5440 sqlite3_int64 nTmp = 0; /* Bytes of temp space */
5441 char *aTmp; /* Temp space for PoslistNearMerge() */
5443 /* Allocate temporary working space. */
5444 for(p=pExpr; p->pLeft; p=p->pLeft){
5445 assert( p->pRight->pPhrase->doclist.nList>0 );
5446 nTmp += p->pRight->pPhrase->doclist.nList;
5448 nTmp += p->pPhrase->doclist.nList;
5449 aTmp = sqlite3_malloc64(nTmp*2);
5450 if( !aTmp ){
5451 *pRc = SQLITE_NOMEM;
5452 res = 0;
5453 }else{
5454 char *aPoslist = p->pPhrase->doclist.pList;
5455 int nToken = p->pPhrase->nToken;
5457 for(p=p->pParent;res && p && p->eType==FTSQUERY_NEAR; p=p->pParent){
5458 Fts3Phrase *pPhrase = p->pRight->pPhrase;
5459 int nNear = p->nNear;
5460 res = fts3EvalNearTrim(nNear, aTmp, &aPoslist, &nToken, pPhrase);
5463 aPoslist = pExpr->pRight->pPhrase->doclist.pList;
5464 nToken = pExpr->pRight->pPhrase->nToken;
5465 for(p=pExpr->pLeft; p && res; p=p->pLeft){
5466 int nNear;
5467 Fts3Phrase *pPhrase;
5468 assert( p->pParent && p->pParent->pLeft==p );
5469 nNear = p->pParent->nNear;
5470 pPhrase = (
5471 p->eType==FTSQUERY_NEAR ? p->pRight->pPhrase : p->pPhrase
5473 res = fts3EvalNearTrim(nNear, aTmp, &aPoslist, &nToken, pPhrase);
5477 sqlite3_free(aTmp);
5480 return res;
5484 ** This function is a helper function for sqlite3Fts3EvalTestDeferred().
5485 ** Assuming no error occurs or has occurred, It returns non-zero if the
5486 ** expression passed as the second argument matches the row that pCsr
5487 ** currently points to, or zero if it does not.
5489 ** If *pRc is not SQLITE_OK when this function is called, it is a no-op.
5490 ** If an error occurs during execution of this function, *pRc is set to
5491 ** the appropriate SQLite error code. In this case the returned value is
5492 ** undefined.
5494 static int fts3EvalTestExpr(
5495 Fts3Cursor *pCsr, /* FTS cursor handle */
5496 Fts3Expr *pExpr, /* Expr to test. May or may not be root. */
5497 int *pRc /* IN/OUT: Error code */
5499 int bHit = 1; /* Return value */
5500 if( *pRc==SQLITE_OK ){
5501 switch( pExpr->eType ){
5502 case FTSQUERY_NEAR:
5503 case FTSQUERY_AND:
5504 bHit = (
5505 fts3EvalTestExpr(pCsr, pExpr->pLeft, pRc)
5506 && fts3EvalTestExpr(pCsr, pExpr->pRight, pRc)
5507 && fts3EvalNearTest(pExpr, pRc)
5510 /* If the NEAR expression does not match any rows, zero the doclist for
5511 ** all phrases involved in the NEAR. This is because the snippet(),
5512 ** offsets() and matchinfo() functions are not supposed to recognize
5513 ** any instances of phrases that are part of unmatched NEAR queries.
5514 ** For example if this expression:
5516 ** ... MATCH 'a OR (b NEAR c)'
5518 ** is matched against a row containing:
5520 ** 'a b d e'
5522 ** then any snippet() should ony highlight the "a" term, not the "b"
5523 ** (as "b" is part of a non-matching NEAR clause).
5525 if( bHit==0
5526 && pExpr->eType==FTSQUERY_NEAR
5527 && (pExpr->pParent==0 || pExpr->pParent->eType!=FTSQUERY_NEAR)
5529 Fts3Expr *p;
5530 for(p=pExpr; p->pPhrase==0; p=p->pLeft){
5531 if( p->pRight->iDocid==pCsr->iPrevId ){
5532 fts3EvalInvalidatePoslist(p->pRight->pPhrase);
5535 if( p->iDocid==pCsr->iPrevId ){
5536 fts3EvalInvalidatePoslist(p->pPhrase);
5540 break;
5542 case FTSQUERY_OR: {
5543 int bHit1 = fts3EvalTestExpr(pCsr, pExpr->pLeft, pRc);
5544 int bHit2 = fts3EvalTestExpr(pCsr, pExpr->pRight, pRc);
5545 bHit = bHit1 || bHit2;
5546 break;
5549 case FTSQUERY_NOT:
5550 bHit = (
5551 fts3EvalTestExpr(pCsr, pExpr->pLeft, pRc)
5552 && !fts3EvalTestExpr(pCsr, pExpr->pRight, pRc)
5554 break;
5556 default: {
5557 #ifndef SQLITE_DISABLE_FTS4_DEFERRED
5558 if( pCsr->pDeferred
5559 && (pExpr->iDocid==pCsr->iPrevId || pExpr->bDeferred)
5561 Fts3Phrase *pPhrase = pExpr->pPhrase;
5562 assert( pExpr->bDeferred || pPhrase->doclist.bFreeList==0 );
5563 if( pExpr->bDeferred ){
5564 fts3EvalInvalidatePoslist(pPhrase);
5566 *pRc = fts3EvalDeferredPhrase(pCsr, pPhrase);
5567 bHit = (pPhrase->doclist.pList!=0);
5568 pExpr->iDocid = pCsr->iPrevId;
5569 }else
5570 #endif
5572 bHit = (
5573 pExpr->bEof==0 && pExpr->iDocid==pCsr->iPrevId
5574 && pExpr->pPhrase->doclist.nList>0
5577 break;
5581 return bHit;
5585 ** This function is called as the second part of each xNext operation when
5586 ** iterating through the results of a full-text query. At this point the
5587 ** cursor points to a row that matches the query expression, with the
5588 ** following caveats:
5590 ** * Up until this point, "NEAR" operators in the expression have been
5591 ** treated as "AND".
5593 ** * Deferred tokens have not yet been considered.
5595 ** If *pRc is not SQLITE_OK when this function is called, it immediately
5596 ** returns 0. Otherwise, it tests whether or not after considering NEAR
5597 ** operators and deferred tokens the current row is still a match for the
5598 ** expression. It returns 1 if both of the following are true:
5600 ** 1. *pRc is SQLITE_OK when this function returns, and
5602 ** 2. After scanning the current FTS table row for the deferred tokens,
5603 ** it is determined that the row does *not* match the query.
5605 ** Or, if no error occurs and it seems the current row does match the FTS
5606 ** query, return 0.
5608 int sqlite3Fts3EvalTestDeferred(Fts3Cursor *pCsr, int *pRc){
5609 int rc = *pRc;
5610 int bMiss = 0;
5611 if( rc==SQLITE_OK ){
5613 /* If there are one or more deferred tokens, load the current row into
5614 ** memory and scan it to determine the position list for each deferred
5615 ** token. Then, see if this row is really a match, considering deferred
5616 ** tokens and NEAR operators (neither of which were taken into account
5617 ** earlier, by fts3EvalNextRow()).
5619 if( pCsr->pDeferred ){
5620 rc = fts3CursorSeek(0, pCsr);
5621 if( rc==SQLITE_OK ){
5622 rc = sqlite3Fts3CacheDeferredDoclists(pCsr);
5625 bMiss = (0==fts3EvalTestExpr(pCsr, pCsr->pExpr, &rc));
5627 /* Free the position-lists accumulated for each deferred token above. */
5628 sqlite3Fts3FreeDeferredDoclists(pCsr);
5629 *pRc = rc;
5631 return (rc==SQLITE_OK && bMiss);
5635 ** Advance to the next document that matches the FTS expression in
5636 ** Fts3Cursor.pExpr.
5638 static int fts3EvalNext(Fts3Cursor *pCsr){
5639 int rc = SQLITE_OK; /* Return Code */
5640 Fts3Expr *pExpr = pCsr->pExpr;
5641 assert( pCsr->isEof==0 );
5642 if( pExpr==0 ){
5643 pCsr->isEof = 1;
5644 }else{
5645 do {
5646 if( pCsr->isRequireSeek==0 ){
5647 sqlite3_reset(pCsr->pStmt);
5649 assert( sqlite3_data_count(pCsr->pStmt)==0 );
5650 fts3EvalNextRow(pCsr, pExpr, &rc);
5651 pCsr->isEof = pExpr->bEof;
5652 pCsr->isRequireSeek = 1;
5653 pCsr->isMatchinfoNeeded = 1;
5654 pCsr->iPrevId = pExpr->iDocid;
5655 }while( pCsr->isEof==0 && sqlite3Fts3EvalTestDeferred(pCsr, &rc) );
5658 /* Check if the cursor is past the end of the docid range specified
5659 ** by Fts3Cursor.iMinDocid/iMaxDocid. If so, set the EOF flag. */
5660 if( rc==SQLITE_OK && (
5661 (pCsr->bDesc==0 && pCsr->iPrevId>pCsr->iMaxDocid)
5662 || (pCsr->bDesc!=0 && pCsr->iPrevId<pCsr->iMinDocid)
5664 pCsr->isEof = 1;
5667 return rc;
5671 ** Restart interation for expression pExpr so that the next call to
5672 ** fts3EvalNext() visits the first row. Do not allow incremental
5673 ** loading or merging of phrase doclists for this iteration.
5675 ** If *pRc is other than SQLITE_OK when this function is called, it is
5676 ** a no-op. If an error occurs within this function, *pRc is set to an
5677 ** SQLite error code before returning.
5679 static void fts3EvalRestart(
5680 Fts3Cursor *pCsr,
5681 Fts3Expr *pExpr,
5682 int *pRc
5684 if( pExpr && *pRc==SQLITE_OK ){
5685 Fts3Phrase *pPhrase = pExpr->pPhrase;
5687 if( pPhrase ){
5688 fts3EvalInvalidatePoslist(pPhrase);
5689 if( pPhrase->bIncr ){
5690 int i;
5691 for(i=0; i<pPhrase->nToken; i++){
5692 Fts3PhraseToken *pToken = &pPhrase->aToken[i];
5693 assert( pToken->pDeferred==0 );
5694 if( pToken->pSegcsr ){
5695 sqlite3Fts3MsrIncrRestart(pToken->pSegcsr);
5698 *pRc = fts3EvalPhraseStart(pCsr, 0, pPhrase);
5700 pPhrase->doclist.pNextDocid = 0;
5701 pPhrase->doclist.iDocid = 0;
5702 pPhrase->pOrPoslist = 0;
5705 pExpr->iDocid = 0;
5706 pExpr->bEof = 0;
5707 pExpr->bStart = 0;
5709 fts3EvalRestart(pCsr, pExpr->pLeft, pRc);
5710 fts3EvalRestart(pCsr, pExpr->pRight, pRc);
5715 ** After allocating the Fts3Expr.aMI[] array for each phrase in the
5716 ** expression rooted at pExpr, the cursor iterates through all rows matched
5717 ** by pExpr, calling this function for each row. This function increments
5718 ** the values in Fts3Expr.aMI[] according to the position-list currently
5719 ** found in Fts3Expr.pPhrase->doclist.pList for each of the phrase
5720 ** expression nodes.
5722 static void fts3EvalUpdateCounts(Fts3Expr *pExpr, int nCol){
5723 if( pExpr ){
5724 Fts3Phrase *pPhrase = pExpr->pPhrase;
5725 if( pPhrase && pPhrase->doclist.pList ){
5726 int iCol = 0;
5727 char *p = pPhrase->doclist.pList;
5730 u8 c = 0;
5731 int iCnt = 0;
5732 while( 0xFE & (*p | c) ){
5733 if( (c&0x80)==0 ) iCnt++;
5734 c = *p++ & 0x80;
5737 /* aMI[iCol*3 + 1] = Number of occurrences
5738 ** aMI[iCol*3 + 2] = Number of rows containing at least one instance
5740 pExpr->aMI[iCol*3 + 1] += iCnt;
5741 pExpr->aMI[iCol*3 + 2] += (iCnt>0);
5742 if( *p==0x00 ) break;
5743 p++;
5744 p += fts3GetVarint32(p, &iCol);
5745 }while( iCol<nCol );
5748 fts3EvalUpdateCounts(pExpr->pLeft, nCol);
5749 fts3EvalUpdateCounts(pExpr->pRight, nCol);
5754 ** Expression pExpr must be of type FTSQUERY_PHRASE.
5756 ** If it is not already allocated and populated, this function allocates and
5757 ** populates the Fts3Expr.aMI[] array for expression pExpr. If pExpr is part
5758 ** of a NEAR expression, then it also allocates and populates the same array
5759 ** for all other phrases that are part of the NEAR expression.
5761 ** SQLITE_OK is returned if the aMI[] array is successfully allocated and
5762 ** populated. Otherwise, if an error occurs, an SQLite error code is returned.
5764 static int fts3EvalGatherStats(
5765 Fts3Cursor *pCsr, /* Cursor object */
5766 Fts3Expr *pExpr /* FTSQUERY_PHRASE expression */
5768 int rc = SQLITE_OK; /* Return code */
5770 assert( pExpr->eType==FTSQUERY_PHRASE );
5771 if( pExpr->aMI==0 ){
5772 Fts3Table *pTab = (Fts3Table *)pCsr->base.pVtab;
5773 Fts3Expr *pRoot; /* Root of NEAR expression */
5774 Fts3Expr *p; /* Iterator used for several purposes */
5776 sqlite3_int64 iPrevId = pCsr->iPrevId;
5777 sqlite3_int64 iDocid;
5778 u8 bEof;
5780 /* Find the root of the NEAR expression */
5781 pRoot = pExpr;
5782 while( pRoot->pParent && pRoot->pParent->eType==FTSQUERY_NEAR ){
5783 pRoot = pRoot->pParent;
5785 iDocid = pRoot->iDocid;
5786 bEof = pRoot->bEof;
5787 assert( pRoot->bStart );
5789 /* Allocate space for the aMSI[] array of each FTSQUERY_PHRASE node */
5790 for(p=pRoot; p; p=p->pLeft){
5791 Fts3Expr *pE = (p->eType==FTSQUERY_PHRASE?p:p->pRight);
5792 assert( pE->aMI==0 );
5793 pE->aMI = (u32 *)sqlite3_malloc64(pTab->nColumn * 3 * sizeof(u32));
5794 if( !pE->aMI ) return SQLITE_NOMEM;
5795 memset(pE->aMI, 0, pTab->nColumn * 3 * sizeof(u32));
5798 fts3EvalRestart(pCsr, pRoot, &rc);
5800 while( pCsr->isEof==0 && rc==SQLITE_OK ){
5802 do {
5803 /* Ensure the %_content statement is reset. */
5804 if( pCsr->isRequireSeek==0 ) sqlite3_reset(pCsr->pStmt);
5805 assert( sqlite3_data_count(pCsr->pStmt)==0 );
5807 /* Advance to the next document */
5808 fts3EvalNextRow(pCsr, pRoot, &rc);
5809 pCsr->isEof = pRoot->bEof;
5810 pCsr->isRequireSeek = 1;
5811 pCsr->isMatchinfoNeeded = 1;
5812 pCsr->iPrevId = pRoot->iDocid;
5813 }while( pCsr->isEof==0
5814 && pRoot->eType==FTSQUERY_NEAR
5815 && sqlite3Fts3EvalTestDeferred(pCsr, &rc)
5818 if( rc==SQLITE_OK && pCsr->isEof==0 ){
5819 fts3EvalUpdateCounts(pRoot, pTab->nColumn);
5823 pCsr->isEof = 0;
5824 pCsr->iPrevId = iPrevId;
5826 if( bEof ){
5827 pRoot->bEof = bEof;
5828 }else{
5829 /* Caution: pRoot may iterate through docids in ascending or descending
5830 ** order. For this reason, even though it seems more defensive, the
5831 ** do loop can not be written:
5833 ** do {...} while( pRoot->iDocid<iDocid && rc==SQLITE_OK );
5835 fts3EvalRestart(pCsr, pRoot, &rc);
5836 do {
5837 fts3EvalNextRow(pCsr, pRoot, &rc);
5838 assert_fts3_nc( pRoot->bEof==0 );
5839 if( pRoot->bEof ) rc = FTS_CORRUPT_VTAB;
5840 }while( pRoot->iDocid!=iDocid && rc==SQLITE_OK );
5843 return rc;
5847 ** This function is used by the matchinfo() module to query a phrase
5848 ** expression node for the following information:
5850 ** 1. The total number of occurrences of the phrase in each column of
5851 ** the FTS table (considering all rows), and
5853 ** 2. For each column, the number of rows in the table for which the
5854 ** column contains at least one instance of the phrase.
5856 ** If no error occurs, SQLITE_OK is returned and the values for each column
5857 ** written into the array aiOut as follows:
5859 ** aiOut[iCol*3 + 1] = Number of occurrences
5860 ** aiOut[iCol*3 + 2] = Number of rows containing at least one instance
5862 ** Caveats:
5864 ** * If a phrase consists entirely of deferred tokens, then all output
5865 ** values are set to the number of documents in the table. In other
5866 ** words we assume that very common tokens occur exactly once in each
5867 ** column of each row of the table.
5869 ** * If a phrase contains some deferred tokens (and some non-deferred
5870 ** tokens), count the potential occurrence identified by considering
5871 ** the non-deferred tokens instead of actual phrase occurrences.
5873 ** * If the phrase is part of a NEAR expression, then only phrase instances
5874 ** that meet the NEAR constraint are included in the counts.
5876 int sqlite3Fts3EvalPhraseStats(
5877 Fts3Cursor *pCsr, /* FTS cursor handle */
5878 Fts3Expr *pExpr, /* Phrase expression */
5879 u32 *aiOut /* Array to write results into (see above) */
5881 Fts3Table *pTab = (Fts3Table *)pCsr->base.pVtab;
5882 int rc = SQLITE_OK;
5883 int iCol;
5885 if( pExpr->bDeferred && pExpr->pParent->eType!=FTSQUERY_NEAR ){
5886 assert( pCsr->nDoc>0 );
5887 for(iCol=0; iCol<pTab->nColumn; iCol++){
5888 aiOut[iCol*3 + 1] = (u32)pCsr->nDoc;
5889 aiOut[iCol*3 + 2] = (u32)pCsr->nDoc;
5891 }else{
5892 rc = fts3EvalGatherStats(pCsr, pExpr);
5893 if( rc==SQLITE_OK ){
5894 assert( pExpr->aMI );
5895 for(iCol=0; iCol<pTab->nColumn; iCol++){
5896 aiOut[iCol*3 + 1] = pExpr->aMI[iCol*3 + 1];
5897 aiOut[iCol*3 + 2] = pExpr->aMI[iCol*3 + 2];
5902 return rc;
5906 ** The expression pExpr passed as the second argument to this function
5907 ** must be of type FTSQUERY_PHRASE.
5909 ** The returned value is either NULL or a pointer to a buffer containing
5910 ** a position-list indicating the occurrences of the phrase in column iCol
5911 ** of the current row.
5913 ** More specifically, the returned buffer contains 1 varint for each
5914 ** occurrence of the phrase in the column, stored using the normal (delta+2)
5915 ** compression and is terminated by either an 0x01 or 0x00 byte. For example,
5916 ** if the requested column contains "a b X c d X X" and the position-list
5917 ** for 'X' is requested, the buffer returned may contain:
5919 ** 0x04 0x05 0x03 0x01 or 0x04 0x05 0x03 0x00
5921 ** This function works regardless of whether or not the phrase is deferred,
5922 ** incremental, or neither.
5924 int sqlite3Fts3EvalPhrasePoslist(
5925 Fts3Cursor *pCsr, /* FTS3 cursor object */
5926 Fts3Expr *pExpr, /* Phrase to return doclist for */
5927 int iCol, /* Column to return position list for */
5928 char **ppOut /* OUT: Pointer to position list */
5930 Fts3Phrase *pPhrase = pExpr->pPhrase;
5931 Fts3Table *pTab = (Fts3Table *)pCsr->base.pVtab;
5932 char *pIter;
5933 int iThis;
5934 sqlite3_int64 iDocid;
5936 /* If this phrase is applies specifically to some column other than
5937 ** column iCol, return a NULL pointer. */
5938 *ppOut = 0;
5939 assert( iCol>=0 && iCol<pTab->nColumn );
5940 if( (pPhrase->iColumn<pTab->nColumn && pPhrase->iColumn!=iCol) ){
5941 return SQLITE_OK;
5944 iDocid = pExpr->iDocid;
5945 pIter = pPhrase->doclist.pList;
5946 if( iDocid!=pCsr->iPrevId || pExpr->bEof ){
5947 int rc = SQLITE_OK;
5948 int bDescDoclist = pTab->bDescIdx; /* For DOCID_CMP macro */
5949 int bOr = 0;
5950 u8 bTreeEof = 0;
5951 Fts3Expr *p; /* Used to iterate from pExpr to root */
5952 Fts3Expr *pNear; /* Most senior NEAR ancestor (or pExpr) */
5953 int bMatch;
5955 /* Check if this phrase descends from an OR expression node. If not,
5956 ** return NULL. Otherwise, the entry that corresponds to docid
5957 ** pCsr->iPrevId may lie earlier in the doclist buffer. Or, if the
5958 ** tree that the node is part of has been marked as EOF, but the node
5959 ** itself is not EOF, then it may point to an earlier entry. */
5960 pNear = pExpr;
5961 for(p=pExpr->pParent; p; p=p->pParent){
5962 if( p->eType==FTSQUERY_OR ) bOr = 1;
5963 if( p->eType==FTSQUERY_NEAR ) pNear = p;
5964 if( p->bEof ) bTreeEof = 1;
5966 if( bOr==0 ) return SQLITE_OK;
5968 /* This is the descendent of an OR node. In this case we cannot use
5969 ** an incremental phrase. Load the entire doclist for the phrase
5970 ** into memory in this case. */
5971 if( pPhrase->bIncr ){
5972 int bEofSave = pNear->bEof;
5973 fts3EvalRestart(pCsr, pNear, &rc);
5974 while( rc==SQLITE_OK && !pNear->bEof ){
5975 fts3EvalNextRow(pCsr, pNear, &rc);
5976 if( bEofSave==0 && pNear->iDocid==iDocid ) break;
5978 assert( rc!=SQLITE_OK || pPhrase->bIncr==0 );
5979 if( rc==SQLITE_OK && pNear->bEof!=bEofSave ){
5980 rc = FTS_CORRUPT_VTAB;
5983 if( bTreeEof ){
5984 while( rc==SQLITE_OK && !pNear->bEof ){
5985 fts3EvalNextRow(pCsr, pNear, &rc);
5988 if( rc!=SQLITE_OK ) return rc;
5990 bMatch = 1;
5991 for(p=pNear; p; p=p->pLeft){
5992 u8 bEof = 0;
5993 Fts3Expr *pTest = p;
5994 Fts3Phrase *pPh;
5995 assert( pTest->eType==FTSQUERY_NEAR || pTest->eType==FTSQUERY_PHRASE );
5996 if( pTest->eType==FTSQUERY_NEAR ) pTest = pTest->pRight;
5997 assert( pTest->eType==FTSQUERY_PHRASE );
5998 pPh = pTest->pPhrase;
6000 pIter = pPh->pOrPoslist;
6001 iDocid = pPh->iOrDocid;
6002 if( pCsr->bDesc==bDescDoclist ){
6003 bEof = !pPh->doclist.nAll ||
6004 (pIter >= (pPh->doclist.aAll + pPh->doclist.nAll));
6005 while( (pIter==0 || DOCID_CMP(iDocid, pCsr->iPrevId)<0 ) && bEof==0 ){
6006 sqlite3Fts3DoclistNext(
6007 bDescDoclist, pPh->doclist.aAll, pPh->doclist.nAll,
6008 &pIter, &iDocid, &bEof
6011 }else{
6012 bEof = !pPh->doclist.nAll || (pIter && pIter<=pPh->doclist.aAll);
6013 while( (pIter==0 || DOCID_CMP(iDocid, pCsr->iPrevId)>0 ) && bEof==0 ){
6014 int dummy;
6015 sqlite3Fts3DoclistPrev(
6016 bDescDoclist, pPh->doclist.aAll, pPh->doclist.nAll,
6017 &pIter, &iDocid, &dummy, &bEof
6021 pPh->pOrPoslist = pIter;
6022 pPh->iOrDocid = iDocid;
6023 if( bEof || iDocid!=pCsr->iPrevId ) bMatch = 0;
6026 if( bMatch ){
6027 pIter = pPhrase->pOrPoslist;
6028 }else{
6029 pIter = 0;
6032 if( pIter==0 ) return SQLITE_OK;
6034 if( *pIter==0x01 ){
6035 pIter++;
6036 pIter += fts3GetVarint32(pIter, &iThis);
6037 }else{
6038 iThis = 0;
6040 while( iThis<iCol ){
6041 fts3ColumnlistCopy(0, &pIter);
6042 if( *pIter==0x00 ) return SQLITE_OK;
6043 pIter++;
6044 pIter += fts3GetVarint32(pIter, &iThis);
6046 if( *pIter==0x00 ){
6047 pIter = 0;
6050 *ppOut = ((iCol==iThis)?pIter:0);
6051 return SQLITE_OK;
6055 ** Free all components of the Fts3Phrase structure that were allocated by
6056 ** the eval module. Specifically, this means to free:
6058 ** * the contents of pPhrase->doclist, and
6059 ** * any Fts3MultiSegReader objects held by phrase tokens.
6061 void sqlite3Fts3EvalPhraseCleanup(Fts3Phrase *pPhrase){
6062 if( pPhrase ){
6063 int i;
6064 sqlite3_free(pPhrase->doclist.aAll);
6065 fts3EvalInvalidatePoslist(pPhrase);
6066 memset(&pPhrase->doclist, 0, sizeof(Fts3Doclist));
6067 for(i=0; i<pPhrase->nToken; i++){
6068 fts3SegReaderCursorFree(pPhrase->aToken[i].pSegcsr);
6069 pPhrase->aToken[i].pSegcsr = 0;
6076 ** Return SQLITE_CORRUPT_VTAB.
6078 #ifdef SQLITE_DEBUG
6079 int sqlite3Fts3Corrupt(){
6080 return SQLITE_CORRUPT_VTAB;
6082 #endif
6084 #if !SQLITE_CORE
6086 ** Initialize API pointer table, if required.
6088 #ifdef _WIN32
6089 __declspec(dllexport)
6090 #endif
6091 int sqlite3_fts3_init(
6092 sqlite3 *db,
6093 char **pzErrMsg,
6094 const sqlite3_api_routines *pApi
6096 SQLITE_EXTENSION_INIT2(pApi)
6097 return sqlite3Fts3Init(db);
6099 #endif
6101 #endif