OLD | NEW |
1 /* | 1 /* |
2 ** 2006 Oct 10 | 2 ** 2006 Oct 10 |
3 ** | 3 ** |
4 ** The author disclaims copyright to this source code. In place of | 4 ** The author disclaims copyright to this source code. In place of |
5 ** a legal notice, here is a blessing: | 5 ** a legal notice, here is a blessing: |
6 ** | 6 ** |
7 ** May you do good and not evil. | 7 ** May you do good and not evil. |
8 ** May you find forgiveness for yourself and forgive others. | 8 ** May you find forgiveness for yourself and forgive others. |
9 ** May you share freely, never taking more than you give. | 9 ** May you share freely, never taking more than you give. |
10 ** | 10 ** |
11 ****************************************************************************** | 11 ****************************************************************************** |
12 ** | 12 ** |
13 ** This is an SQLite module implementing full-text search. | 13 ** This is an SQLite module implementing full-text search. |
14 */ | 14 */ |
15 | 15 |
16 /* | 16 /* |
17 ** The code in this file is only compiled if: | 17 ** The code in this file is only compiled if: |
18 ** | 18 ** |
19 ** * The FTS3 module is being built as an extension | 19 ** * The FTS3 module is being built as an extension |
20 ** (in which case SQLITE_CORE is not defined), or | 20 ** (in which case SQLITE_CORE is not defined), or |
21 ** | 21 ** |
22 ** * The FTS3 module is being built into the core of | 22 ** * The FTS3 module is being built into the core of |
23 ** SQLite (in which case SQLITE_ENABLE_FTS3 is defined). | 23 ** SQLite (in which case SQLITE_ENABLE_FTS3 is defined). |
24 */ | 24 */ |
25 | 25 |
26 /* TODO(shess) Consider exporting this comment to an HTML file or the | |
27 ** wiki. | |
28 */ | |
29 /* The full-text index is stored in a series of b+tree (-like) | 26 /* The full-text index is stored in a series of b+tree (-like) |
30 ** structures called segments which map terms to doclists. The | 27 ** structures called segments which map terms to doclists. The |
31 ** structures are like b+trees in layout, but are constructed from the | 28 ** structures are like b+trees in layout, but are constructed from the |
32 ** bottom up in optimal fashion and are not updatable. Since trees | 29 ** bottom up in optimal fashion and are not updatable. Since trees |
33 ** are built from the bottom up, things will be described from the | 30 ** are built from the bottom up, things will be described from the |
34 ** bottom up. | 31 ** bottom up. |
35 ** | 32 ** |
36 ** | 33 ** |
37 **** Varints **** | 34 **** Varints **** |
38 ** The basic unit of encoding is a variable-length integer called a | 35 ** The basic unit of encoding is a variable-length integer called a |
39 ** varint. We encode variable-length integers in little-endian order | 36 ** varint. We encode variable-length integers in little-endian order |
40 ** using seven bits * per byte as follows: | 37 ** using seven bits * per byte as follows: |
41 ** | 38 ** |
42 ** KEY: | 39 ** KEY: |
43 ** A = 0xxxxxxx 7 bits of data and one flag bit | 40 ** A = 0xxxxxxx 7 bits of data and one flag bit |
44 ** B = 1xxxxxxx 7 bits of data and one flag bit | 41 ** B = 1xxxxxxx 7 bits of data and one flag bit |
45 ** | 42 ** |
46 ** 7 bits - A | 43 ** 7 bits - A |
47 ** 14 bits - BA | 44 ** 14 bits - BA |
48 ** 21 bits - BBA | 45 ** 21 bits - BBA |
49 ** and so on. | 46 ** and so on. |
50 ** | 47 ** |
51 ** This is identical to how sqlite encodes varints (see util.c). | 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). |
| 52 ** |
| 53 ** Example encodings: |
| 54 ** |
| 55 ** 1: 0x01 |
| 56 ** 127: 0x7f |
| 57 ** 128: 0x81 0x00 |
52 ** | 58 ** |
53 ** | 59 ** |
54 **** Document lists **** | 60 **** Document lists **** |
55 ** A doclist (document list) holds a docid-sorted list of hits for a | 61 ** A doclist (document list) holds a docid-sorted list of hits for a |
56 ** given term. Doclists hold docids, and can optionally associate | 62 ** given term. Doclists hold docids and associated token positions. |
57 ** token positions and offsets with docids. | 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. |
58 ** | 66 ** |
59 ** A DL_POSITIONS_OFFSETS doclist is stored like this: | 67 ** FTS3 used to optionally store character offsets using a compile-time |
| 68 ** option. But that functionality is no longer supported. |
| 69 ** |
| 70 ** A doclist is stored like this: |
60 ** | 71 ** |
61 ** array { | 72 ** array { |
62 ** varint docid; | 73 ** varint docid; |
63 ** array { (position list for column 0) | 74 ** array { (position list for column 0) |
64 ** varint position; (delta from previous position plus POS_BASE) | 75 ** varint position; (2 more than the delta from previous position) |
65 ** varint startOffset; (delta from previous startOffset) | |
66 ** varint endOffset; (delta from startOffset) | |
67 ** } | 76 ** } |
68 ** array { | 77 ** array { |
69 ** varint POS_COLUMN; (marks start of position list for new column) | 78 ** varint POS_COLUMN; (marks start of position list for new column) |
70 ** varint column; (index of new column) | 79 ** varint column; (index of new column) |
71 ** array { | 80 ** array { |
72 ** varint position; (delta from previous position plus POS_BASE) | 81 ** varint position; (2 more than the delta from previous position) |
73 ** varint startOffset;(delta from previous startOffset) | |
74 ** varint endOffset; (delta from startOffset) | |
75 ** } | 82 ** } |
76 ** } | 83 ** } |
77 ** varint POS_END; (marks end of positions for this document. | 84 ** varint POS_END; (marks end of positions for this document. |
78 ** } | 85 ** } |
79 ** | 86 ** |
80 ** Here, array { X } means zero or more occurrences of X, adjacent in | 87 ** Here, array { X } means zero or more occurrences of X, adjacent in |
81 ** memory. A "position" is an index of a token in the token stream | 88 ** memory. A "position" is an index of a token in the token stream |
82 ** generated by the tokenizer, while an "offset" is a byte offset, | 89 ** generated by the tokenizer. Note that POS_END and POS_COLUMN occur |
83 ** both based at 0. Note that POS_END and POS_COLUMN occur in the | 90 ** in the same logical place as the position element, and act as sentinals |
84 ** 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. |
85 ** ending a position list array. | 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: |
86 ** | 95 ** |
87 ** A DL_POSITIONS doclist omits the startOffset and endOffset | 96 ** label: A B C D E F G H I J K |
88 ** information. A DL_DOCIDS doclist omits both the position and | 97 ** value: 123 5 9 1 1 14 35 0 234 72 0 |
89 ** offset information, becoming an array of varint-encoded docids. | |
90 ** | 98 ** |
91 ** On-disk data is stored as type DL_DEFAULT, so we don't serialize | 99 ** The 123 value is the first docid. For column zero in this document |
92 ** the type. Due to how deletion is implemented in the segmentation | 100 ** there are two matches at positions 3 and 10 (5-2 and 9-2+3). The 1 |
93 ** system, on-disk doclists MUST store at least positions. | 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 next docid. It has one position 72 (72-2) and then |
| 105 ** terminates with the 0 at K. |
94 ** | 106 ** |
| 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. |
| 112 ** |
| 113 ** A bare doclist omits the position information, becoming an |
| 114 ** array of varint-encoded docids. |
95 ** | 115 ** |
96 **** Segment leaf nodes **** | 116 **** Segment leaf nodes **** |
97 ** Segment leaf nodes store terms and doclists, ordered by term. Leaf | 117 ** Segment leaf nodes store terms and doclists, ordered by term. Leaf |
98 ** nodes are written using LeafWriter, and read using LeafReader (to | 118 ** nodes are written using LeafWriter, and read using LeafReader (to |
99 ** iterate through a single leaf node's data) and LeavesReader (to | 119 ** iterate through a single leaf node's data) and LeavesReader (to |
100 ** iterate through a segment's entire leaf layer). Leaf nodes have | 120 ** iterate through a segment's entire leaf layer). Leaf nodes have |
101 ** the format: | 121 ** the format: |
102 ** | 122 ** |
103 ** varint iHeight; (height from leaf level, always 0) | 123 ** varint iHeight; (height from leaf level, always 0) |
104 ** varint nTerm; (length of first term) | 124 ** varint nTerm; (length of first term) |
(...skipping 167 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
272 ** into a single segment. | 292 ** into a single segment. |
273 */ | 293 */ |
274 #define CHROMIUM_FTS3_CHANGES 1 | 294 #define CHROMIUM_FTS3_CHANGES 1 |
275 | 295 |
276 #if !defined(SQLITE_CORE) || defined(SQLITE_ENABLE_FTS3) | 296 #if !defined(SQLITE_CORE) || defined(SQLITE_ENABLE_FTS3) |
277 | 297 |
278 #if defined(SQLITE_ENABLE_FTS3) && !defined(SQLITE_CORE) | 298 #if defined(SQLITE_ENABLE_FTS3) && !defined(SQLITE_CORE) |
279 # define SQLITE_CORE 1 | 299 # define SQLITE_CORE 1 |
280 #endif | 300 #endif |
281 | 301 |
| 302 #include "fts3Int.h" |
| 303 |
282 #include <assert.h> | 304 #include <assert.h> |
283 #include <stdlib.h> | 305 #include <stdlib.h> |
| 306 #include <stddef.h> |
284 #include <stdio.h> | 307 #include <stdio.h> |
285 #include <string.h> | 308 #include <string.h> |
286 #include <ctype.h> | 309 #include <stdarg.h> |
287 | 310 |
288 #include "fts3.h" | 311 #include "fts3.h" |
289 #include "fts3_expr.h" | |
290 #include "fts3_hash.h" | |
291 #include "fts3_tokenizer.h" | |
292 #ifndef SQLITE_CORE | 312 #ifndef SQLITE_CORE |
293 # include "sqlite3ext.h" | 313 # include "sqlite3ext.h" |
294 SQLITE_EXTENSION_INIT1 | 314 SQLITE_EXTENSION_INIT1 |
295 #endif | 315 #endif |
296 | 316 |
297 | 317 /* |
298 /* TODO(shess) MAN, this thing needs some refactoring. At minimum, it | 318 ** Write a 64-bit variable-length integer to memory starting at p[0]. |
299 ** would be nice to order the file better, perhaps something along the | 319 ** The length of data written will be between 1 and FTS3_VARINT_MAX bytes. |
300 ** lines of: | 320 ** The number of bytes written is returned. |
301 ** | |
302 ** - utility functions | |
303 ** - table setup functions | |
304 ** - table update functions | |
305 ** - table query functions | |
306 ** | |
307 ** Put the query functions last because they're likely to reference | |
308 ** typedefs or functions from the table update section. | |
309 */ | 321 */ |
310 | 322 int sqlite3Fts3PutVarint(char *p, sqlite_int64 v){ |
311 #if 0 | |
312 # define FTSTRACE(A) printf A; fflush(stdout) | |
313 #else | |
314 # define FTSTRACE(A) | |
315 #endif | |
316 | |
317 #if 0 | |
318 /* Useful to set breakpoints. See main.c sqlite3Corrupt(). */ | |
319 static int fts3Corrupt(void){ | |
320 return SQLITE_CORRUPT; | |
321 } | |
322 # define SQLITE_CORRUPT_BKPT fts3Corrupt() | |
323 #else | |
324 # define SQLITE_CORRUPT_BKPT SQLITE_CORRUPT | |
325 #endif | |
326 | |
327 /* It is not safe to call isspace(), tolower(), or isalnum() on | |
328 ** hi-bit-set characters. This is the same solution used in the | |
329 ** tokenizer. | |
330 */ | |
331 /* TODO(shess) The snippet-generation code should be using the | |
332 ** tokenizer-generated tokens rather than doing its own local | |
333 ** tokenization. | |
334 */ | |
335 /* TODO(shess) Is __isascii() a portable version of (c&0x80)==0? */ | |
336 static int safe_isspace(char c){ | |
337 return (c&0x80)==0 ? isspace(c) : 0; | |
338 } | |
339 static int safe_tolower(char c){ | |
340 return (c>='A' && c<='Z') ? (c-'A'+'a') : c; | |
341 } | |
342 static int safe_isalnum(char c){ | |
343 return (c&0x80)==0 ? isalnum(c) : 0; | |
344 } | |
345 | |
346 typedef enum DocListType { | |
347 DL_DOCIDS, /* docids only */ | |
348 DL_POSITIONS, /* docids + positions */ | |
349 DL_POSITIONS_OFFSETS /* docids + positions + offsets */ | |
350 } DocListType; | |
351 | |
352 /* | |
353 ** By default, only positions and not offsets are stored in the doclists. | |
354 ** To change this so that offsets are stored too, compile with | |
355 ** | |
356 ** -DDL_DEFAULT=DL_POSITIONS_OFFSETS | |
357 ** | |
358 ** If DL_DEFAULT is set to DL_DOCIDS, your table can only be inserted | |
359 ** into (no deletes or updates). | |
360 */ | |
361 #ifndef DL_DEFAULT | |
362 # define DL_DEFAULT DL_POSITIONS | |
363 #endif | |
364 | |
365 enum { | |
366 POS_END = 0, /* end of this position list */ | |
367 POS_COLUMN, /* followed by new column number */ | |
368 POS_BASE | |
369 }; | |
370 | |
371 /* MERGE_COUNT controls how often we merge segments (see comment at | |
372 ** top of file). | |
373 */ | |
374 #define MERGE_COUNT 16 | |
375 | |
376 /* utility functions */ | |
377 | |
378 /* CLEAR() and SCRAMBLE() abstract memset() on a pointer to a single | |
379 ** record to prevent errors of the form: | |
380 ** | |
381 ** my_function(SomeType *b){ | |
382 ** memset(b, '\0', sizeof(b)); // sizeof(b)!=sizeof(*b) | |
383 ** } | |
384 */ | |
385 /* TODO(shess) Obvious candidates for a header file. */ | |
386 #define CLEAR(b) memset(b, '\0', sizeof(*(b))) | |
387 | |
388 #ifndef NDEBUG | |
389 # define SCRAMBLE(b) memset(b, 0x55, sizeof(*(b))) | |
390 #else | |
391 # define SCRAMBLE(b) | |
392 #endif | |
393 | |
394 /* We may need up to VARINT_MAX bytes to store an encoded 64-bit integer. */ | |
395 #define VARINT_MAX 10 | |
396 | |
397 /* Write a 64-bit variable-length integer to memory starting at p[0]. | |
398 * The length of data written will be between 1 and VARINT_MAX bytes. | |
399 * The number of bytes written is returned. */ | |
400 static int fts3PutVarint(char *p, sqlite_int64 v){ | |
401 unsigned char *q = (unsigned char *) p; | 323 unsigned char *q = (unsigned char *) p; |
402 sqlite_uint64 vu = v; | 324 sqlite_uint64 vu = v; |
403 do{ | 325 do{ |
404 *q++ = (unsigned char) ((vu & 0x7f) | 0x80); | 326 *q++ = (unsigned char) ((vu & 0x7f) | 0x80); |
405 vu >>= 7; | 327 vu >>= 7; |
406 }while( vu!=0 ); | 328 }while( vu!=0 ); |
407 q[-1] &= 0x7f; /* turn off high bit in final byte */ | 329 q[-1] &= 0x7f; /* turn off high bit in final byte */ |
408 assert( q - (unsigned char *)p <= VARINT_MAX ); | 330 assert( q - (unsigned char *)p <= FTS3_VARINT_MAX ); |
409 return (int) (q - (unsigned char *)p); | 331 return (int) (q - (unsigned char *)p); |
410 } | 332 } |
411 | 333 |
412 /* Read a 64-bit variable-length integer from memory starting at p[0]. | 334 /* |
413 * Return the number of bytes read, or 0 on error. | 335 ** Read a 64-bit variable-length integer from memory starting at p[0]. |
414 * The value is stored in *v. */ | 336 ** Return the number of bytes read, or 0 on error. |
415 static int fts3GetVarintSafe(const char *p, sqlite_int64 *v, int max){ | 337 ** The value is stored in *v. |
| 338 */ |
| 339 int sqlite3Fts3GetVarint(const char *p, sqlite_int64 *v){ |
416 const unsigned char *q = (const unsigned char *) p; | 340 const unsigned char *q = (const unsigned char *) p; |
417 sqlite_uint64 x = 0, y = 1; | 341 sqlite_uint64 x = 0, y = 1; |
418 if( max>VARINT_MAX ) max = VARINT_MAX; | 342 while( (*q&0x80)==0x80 && q-(unsigned char *)p<FTS3_VARINT_MAX ){ |
419 while( max && (*q & 0x80) == 0x80 ){ | |
420 max--; | |
421 x += y * (*q++ & 0x7f); | 343 x += y * (*q++ & 0x7f); |
422 y <<= 7; | 344 y <<= 7; |
423 } | 345 } |
424 if( !max ){ | |
425 assert( 0 ); | |
426 return 0; /* tried to read too much; bad data */ | |
427 } | |
428 x += y * (*q++); | 346 x += y * (*q++); |
429 *v = (sqlite_int64) x; | 347 *v = (sqlite_int64) x; |
430 return (int) (q - (unsigned char *)p); | 348 return (int) (q - (unsigned char *)p); |
431 } | 349 } |
432 | 350 |
433 static int fts3GetVarint(const char *p, sqlite_int64 *v){ | 351 /* |
434 return fts3GetVarintSafe(p, v, VARINT_MAX); | 352 ** Similar to sqlite3Fts3GetVarint(), except that the output is truncated to a |
435 } | 353 ** 32-bit integer before it is returned. |
436 | 354 */ |
437 static int fts3GetVarint32Safe(const char *p, int *pi, int max){ | 355 int sqlite3Fts3GetVarint32(const char *p, int *pi){ |
438 sqlite_int64 i; | 356 sqlite_int64 i; |
439 int ret = fts3GetVarintSafe(p, &i, max); | 357 int ret = sqlite3Fts3GetVarint(p, &i); |
440 if( !ret ) return ret; | |
441 *pi = (int) i; | 358 *pi = (int) i; |
442 assert( *pi==i ); | |
443 return ret; | 359 return ret; |
444 } | 360 } |
445 | 361 |
446 static int fts3GetVarint32(const char* p, int *pi){ | 362 /* |
447 return fts3GetVarint32Safe(p, pi, VARINT_MAX); | 363 ** Return the number of bytes required to encode v as a varint |
448 } | 364 */ |
449 | 365 int sqlite3Fts3VarintLen(sqlite3_uint64 v){ |
450 /*******************************************************************/ | 366 int i = 0; |
451 /* DataBuffer is used to collect data into a buffer in piecemeal | 367 do{ |
452 ** fashion. It implements the usual distinction between amount of | 368 i++; |
453 ** data currently stored (nData) and buffer capacity (nCapacity). | 369 v >>= 7; |
454 ** | 370 }while( v!=0 ); |
455 ** dataBufferInit - create a buffer with given initial capacity. | 371 return i; |
456 ** dataBufferReset - forget buffer's data, retaining capacity. | |
457 ** dataBufferDestroy - free buffer's data. | |
458 ** dataBufferSwap - swap contents of two buffers. | |
459 ** dataBufferExpand - expand capacity without adding data. | |
460 ** dataBufferAppend - append data. | |
461 ** dataBufferAppend2 - append two pieces of data at once. | |
462 ** dataBufferReplace - replace buffer's data. | |
463 */ | |
464 typedef struct DataBuffer { | |
465 char *pData; /* Pointer to malloc'ed buffer. */ | |
466 int nCapacity; /* Size of pData buffer. */ | |
467 int nData; /* End of data loaded into pData. */ | |
468 } DataBuffer; | |
469 | |
470 static void dataBufferInit(DataBuffer *pBuffer, int nCapacity){ | |
471 assert( nCapacity>=0 ); | |
472 pBuffer->nData = 0; | |
473 pBuffer->nCapacity = nCapacity; | |
474 pBuffer->pData = nCapacity==0 ? NULL : sqlite3_malloc(nCapacity); | |
475 } | |
476 static void dataBufferReset(DataBuffer *pBuffer){ | |
477 pBuffer->nData = 0; | |
478 } | |
479 static void dataBufferDestroy(DataBuffer *pBuffer){ | |
480 if( pBuffer->pData!=NULL ) sqlite3_free(pBuffer->pData); | |
481 SCRAMBLE(pBuffer); | |
482 } | |
483 static void dataBufferSwap(DataBuffer *pBuffer1, DataBuffer *pBuffer2){ | |
484 DataBuffer tmp = *pBuffer1; | |
485 *pBuffer1 = *pBuffer2; | |
486 *pBuffer2 = tmp; | |
487 } | |
488 static void dataBufferExpand(DataBuffer *pBuffer, int nAddCapacity){ | |
489 assert( nAddCapacity>0 ); | |
490 /* TODO(shess) Consider expanding more aggressively. Note that the | |
491 ** underlying malloc implementation may take care of such things for | |
492 ** us already. | |
493 */ | |
494 if( pBuffer->nData+nAddCapacity>pBuffer->nCapacity ){ | |
495 pBuffer->nCapacity = pBuffer->nData+nAddCapacity; | |
496 pBuffer->pData = sqlite3_realloc(pBuffer->pData, pBuffer->nCapacity); | |
497 } | |
498 } | |
499 static void dataBufferAppend(DataBuffer *pBuffer, | |
500 const char *pSource, int nSource){ | |
501 assert( nSource>0 && pSource!=NULL ); | |
502 dataBufferExpand(pBuffer, nSource); | |
503 memcpy(pBuffer->pData+pBuffer->nData, pSource, nSource); | |
504 pBuffer->nData += nSource; | |
505 } | |
506 static void dataBufferAppend2(DataBuffer *pBuffer, | |
507 const char *pSource1, int nSource1, | |
508 const char *pSource2, int nSource2){ | |
509 assert( nSource1>0 && pSource1!=NULL ); | |
510 assert( nSource2>0 && pSource2!=NULL ); | |
511 dataBufferExpand(pBuffer, nSource1+nSource2); | |
512 memcpy(pBuffer->pData+pBuffer->nData, pSource1, nSource1); | |
513 memcpy(pBuffer->pData+pBuffer->nData+nSource1, pSource2, nSource2); | |
514 pBuffer->nData += nSource1+nSource2; | |
515 } | |
516 static void dataBufferReplace(DataBuffer *pBuffer, | |
517 const char *pSource, int nSource){ | |
518 dataBufferReset(pBuffer); | |
519 dataBufferAppend(pBuffer, pSource, nSource); | |
520 } | |
521 | |
522 /* StringBuffer is a null-terminated version of DataBuffer. */ | |
523 typedef struct StringBuffer { | |
524 DataBuffer b; /* Includes null terminator. */ | |
525 } StringBuffer; | |
526 | |
527 static void initStringBuffer(StringBuffer *sb){ | |
528 dataBufferInit(&sb->b, 100); | |
529 dataBufferReplace(&sb->b, "", 1); | |
530 } | |
531 static int stringBufferLength(StringBuffer *sb){ | |
532 return sb->b.nData-1; | |
533 } | |
534 static char *stringBufferData(StringBuffer *sb){ | |
535 return sb->b.pData; | |
536 } | |
537 static void stringBufferDestroy(StringBuffer *sb){ | |
538 dataBufferDestroy(&sb->b); | |
539 } | |
540 | |
541 static void nappend(StringBuffer *sb, const char *zFrom, int nFrom){ | |
542 assert( sb->b.nData>0 ); | |
543 if( nFrom>0 ){ | |
544 sb->b.nData--; | |
545 dataBufferAppend2(&sb->b, zFrom, nFrom, "", 1); | |
546 } | |
547 } | |
548 static void append(StringBuffer *sb, const char *zFrom){ | |
549 nappend(sb, zFrom, strlen(zFrom)); | |
550 } | |
551 | |
552 /* Append a list of strings separated by commas. */ | |
553 static void appendList(StringBuffer *sb, int nString, char **azString){ | |
554 int i; | |
555 for(i=0; i<nString; ++i){ | |
556 if( i>0 ) append(sb, ", "); | |
557 append(sb, azString[i]); | |
558 } | |
559 } | |
560 | |
561 static int endsInWhiteSpace(StringBuffer *p){ | |
562 return stringBufferLength(p)>0 && | |
563 safe_isspace(stringBufferData(p)[stringBufferLength(p)-1]); | |
564 } | |
565 | |
566 /* If the StringBuffer ends in something other than white space, add a | |
567 ** single space character to the end. | |
568 */ | |
569 static void appendWhiteSpace(StringBuffer *p){ | |
570 if( stringBufferLength(p)==0 ) return; | |
571 if( !endsInWhiteSpace(p) ) append(p, " "); | |
572 } | |
573 | |
574 /* Remove white space from the end of the StringBuffer */ | |
575 static void trimWhiteSpace(StringBuffer *p){ | |
576 while( endsInWhiteSpace(p) ){ | |
577 p->b.pData[--p->b.nData-1] = '\0'; | |
578 } | |
579 } | |
580 | |
581 /*******************************************************************/ | |
582 /* DLReader is used to read document elements from a doclist. The | |
583 ** current docid is cached, so dlrDocid() is fast. DLReader does not | |
584 ** own the doclist buffer. | |
585 ** | |
586 ** dlrAtEnd - true if there's no more data to read. | |
587 ** dlrDocid - docid of current document. | |
588 ** dlrDocData - doclist data for current document (including docid). | |
589 ** dlrDocDataBytes - length of same. | |
590 ** dlrAllDataBytes - length of all remaining data. | |
591 ** dlrPosData - position data for current document. | |
592 ** dlrPosDataLen - length of pos data for current document (incl POS_END). | |
593 ** dlrStep - step to current document. | |
594 ** dlrInit - initial for doclist of given type against given data. | |
595 ** dlrDestroy - clean up. | |
596 ** | |
597 ** Expected usage is something like: | |
598 ** | |
599 ** DLReader reader; | |
600 ** dlrInit(&reader, pData, nData); | |
601 ** while( !dlrAtEnd(&reader) ){ | |
602 ** // calls to dlrDocid() and kin. | |
603 ** dlrStep(&reader); | |
604 ** } | |
605 ** dlrDestroy(&reader); | |
606 */ | |
607 typedef struct DLReader { | |
608 DocListType iType; | |
609 const char *pData; | |
610 int nData; | |
611 | |
612 sqlite_int64 iDocid; | |
613 int nElement; | |
614 } DLReader; | |
615 | |
616 static int dlrAtEnd(DLReader *pReader){ | |
617 assert( pReader->nData>=0 ); | |
618 return pReader->nData<=0; | |
619 } | |
620 static sqlite_int64 dlrDocid(DLReader *pReader){ | |
621 assert( !dlrAtEnd(pReader) ); | |
622 return pReader->iDocid; | |
623 } | |
624 static const char *dlrDocData(DLReader *pReader){ | |
625 assert( !dlrAtEnd(pReader) ); | |
626 return pReader->pData; | |
627 } | |
628 static int dlrDocDataBytes(DLReader *pReader){ | |
629 assert( !dlrAtEnd(pReader) ); | |
630 return pReader->nElement; | |
631 } | |
632 static int dlrAllDataBytes(DLReader *pReader){ | |
633 assert( !dlrAtEnd(pReader) ); | |
634 return pReader->nData; | |
635 } | |
636 /* TODO(shess) Consider adding a field to track iDocid varint length | |
637 ** to make these two functions faster. This might matter (a tiny bit) | |
638 ** for queries. | |
639 */ | |
640 static const char *dlrPosData(DLReader *pReader){ | |
641 sqlite_int64 iDummy; | |
642 int n = fts3GetVarintSafe(pReader->pData, &iDummy, pReader->nElement); | |
643 if( !n ) return NULL; | |
644 assert( !dlrAtEnd(pReader) ); | |
645 return pReader->pData+n; | |
646 } | |
647 static int dlrPosDataLen(DLReader *pReader){ | |
648 sqlite_int64 iDummy; | |
649 int n = fts3GetVarint(pReader->pData, &iDummy); | |
650 assert( !dlrAtEnd(pReader) ); | |
651 return pReader->nElement-n; | |
652 } | |
653 static int dlrStep(DLReader *pReader){ | |
654 assert( !dlrAtEnd(pReader) ); | |
655 | |
656 /* Skip past current doclist element. */ | |
657 assert( pReader->nElement<=pReader->nData ); | |
658 pReader->pData += pReader->nElement; | |
659 pReader->nData -= pReader->nElement; | |
660 | |
661 /* If there is more data, read the next doclist element. */ | |
662 if( pReader->nData>0 ){ | |
663 sqlite_int64 iDocidDelta; | |
664 int nTotal = 0; | |
665 int iDummy, n = fts3GetVarintSafe(pReader->pData, &iDocidDelta, pReader->nDa
ta); | |
666 if( !n ) return SQLITE_CORRUPT_BKPT; | |
667 nTotal += n; | |
668 pReader->iDocid += iDocidDelta; | |
669 if( pReader->iType>=DL_POSITIONS ){ | |
670 while( 1 ){ | |
671 n = fts3GetVarint32Safe(pReader->pData+nTotal, &iDummy, pReader->nData-n
Total); | |
672 if( !n ) return SQLITE_CORRUPT_BKPT; | |
673 nTotal += n; | |
674 if( iDummy==POS_END ) break; | |
675 if( iDummy==POS_COLUMN ){ | |
676 n = fts3GetVarint32Safe(pReader->pData+nTotal, &iDummy, pReader->nData
-nTotal); | |
677 if( !n ) return SQLITE_CORRUPT_BKPT; | |
678 nTotal += n; | |
679 }else if( pReader->iType==DL_POSITIONS_OFFSETS ){ | |
680 n = fts3GetVarint32Safe(pReader->pData+nTotal, &iDummy, pReader->nData
-nTotal); | |
681 if( !n ) return SQLITE_CORRUPT_BKPT; | |
682 nTotal += n; | |
683 n = fts3GetVarint32Safe(pReader->pData+nTotal, &iDummy, pReader->nData
-nTotal); | |
684 if( !n ) return SQLITE_CORRUPT_BKPT; | |
685 nTotal += n; | |
686 } | |
687 } | |
688 } | |
689 pReader->nElement = nTotal; | |
690 assert( pReader->nElement<=pReader->nData ); | |
691 } | |
692 return SQLITE_OK; | |
693 } | |
694 static void dlrDestroy(DLReader *pReader){ | |
695 SCRAMBLE(pReader); | |
696 } | |
697 static int dlrInit(DLReader *pReader, DocListType iType, | |
698 const char *pData, int nData){ | |
699 int rc; | |
700 assert( pData!=NULL && nData!=0 ); | |
701 pReader->iType = iType; | |
702 pReader->pData = pData; | |
703 pReader->nData = nData; | |
704 pReader->nElement = 0; | |
705 pReader->iDocid = 0; | |
706 | |
707 /* Load the first element's data. There must be a first element. */ | |
708 rc = dlrStep(pReader); | |
709 if( rc!=SQLITE_OK ) dlrDestroy(pReader); | |
710 return rc; | |
711 } | |
712 | |
713 #ifndef NDEBUG | |
714 /* Verify that the doclist can be validly decoded. Also returns the | |
715 ** last docid found because it is convenient in other assertions for | |
716 ** DLWriter. | |
717 */ | |
718 static void docListValidate(DocListType iType, const char *pData, int nData, | |
719 sqlite_int64 *pLastDocid){ | |
720 sqlite_int64 iPrevDocid = 0; | |
721 assert( nData>0 ); | |
722 assert( pData!=0 ); | |
723 assert( pData+nData>pData ); | |
724 while( nData!=0 ){ | |
725 sqlite_int64 iDocidDelta; | |
726 int n = fts3GetVarint(pData, &iDocidDelta); | |
727 iPrevDocid += iDocidDelta; | |
728 if( iType>DL_DOCIDS ){ | |
729 int iDummy; | |
730 while( 1 ){ | |
731 n += fts3GetVarint32(pData+n, &iDummy); | |
732 if( iDummy==POS_END ) break; | |
733 if( iDummy==POS_COLUMN ){ | |
734 n += fts3GetVarint32(pData+n, &iDummy); | |
735 }else if( iType>DL_POSITIONS ){ | |
736 n += fts3GetVarint32(pData+n, &iDummy); | |
737 n += fts3GetVarint32(pData+n, &iDummy); | |
738 } | |
739 assert( n<=nData ); | |
740 } | |
741 } | |
742 assert( n<=nData ); | |
743 pData += n; | |
744 nData -= n; | |
745 } | |
746 if( pLastDocid ) *pLastDocid = iPrevDocid; | |
747 } | |
748 #define ASSERT_VALID_DOCLIST(i, p, n, o) docListValidate(i, p, n, o) | |
749 #else | |
750 #define ASSERT_VALID_DOCLIST(i, p, n, o) assert( 1 ) | |
751 #endif | |
752 | |
753 /*******************************************************************/ | |
754 /* DLWriter is used to write doclist data to a DataBuffer. DLWriter | |
755 ** always appends to the buffer and does not own it. | |
756 ** | |
757 ** dlwInit - initialize to write a given type doclistto a buffer. | |
758 ** dlwDestroy - clear the writer's memory. Does not free buffer. | |
759 ** dlwAppend - append raw doclist data to buffer. | |
760 ** dlwCopy - copy next doclist from reader to writer. | |
761 ** dlwAdd - construct doclist element and append to buffer. | |
762 ** Only apply dlwAdd() to DL_DOCIDS doclists (else use PLWriter). | |
763 */ | |
764 typedef struct DLWriter { | |
765 DocListType iType; | |
766 DataBuffer *b; | |
767 sqlite_int64 iPrevDocid; | |
768 #ifndef NDEBUG | |
769 int has_iPrevDocid; | |
770 #endif | |
771 } DLWriter; | |
772 | |
773 static void dlwInit(DLWriter *pWriter, DocListType iType, DataBuffer *b){ | |
774 pWriter->b = b; | |
775 pWriter->iType = iType; | |
776 pWriter->iPrevDocid = 0; | |
777 #ifndef NDEBUG | |
778 pWriter->has_iPrevDocid = 0; | |
779 #endif | |
780 } | |
781 static void dlwDestroy(DLWriter *pWriter){ | |
782 SCRAMBLE(pWriter); | |
783 } | |
784 /* iFirstDocid is the first docid in the doclist in pData. It is | |
785 ** needed because pData may point within a larger doclist, in which | |
786 ** case the first item would be delta-encoded. | |
787 ** | |
788 ** iLastDocid is the final docid in the doclist in pData. It is | |
789 ** needed to create the new iPrevDocid for future delta-encoding. The | |
790 ** code could decode the passed doclist to recreate iLastDocid, but | |
791 ** the only current user (docListMerge) already has decoded this | |
792 ** information. | |
793 */ | |
794 /* TODO(shess) This has become just a helper for docListMerge. | |
795 ** Consider a refactor to make this cleaner. | |
796 */ | |
797 static int dlwAppend(DLWriter *pWriter, | |
798 const char *pData, int nData, | |
799 sqlite_int64 iFirstDocid, sqlite_int64 iLastDocid){ | |
800 sqlite_int64 iDocid = 0; | |
801 char c[VARINT_MAX]; | |
802 int nFirstOld, nFirstNew; /* Old and new varint len of first docid. */ | |
803 #ifndef NDEBUG | |
804 sqlite_int64 iLastDocidDelta; | |
805 #endif | |
806 | |
807 /* Recode the initial docid as delta from iPrevDocid. */ | |
808 nFirstOld = fts3GetVarintSafe(pData, &iDocid, nData); | |
809 if( !nFirstOld ) return SQLITE_CORRUPT_BKPT; | |
810 assert( nFirstOld<nData || (nFirstOld==nData && pWriter->iType==DL_DOCIDS) ); | |
811 nFirstNew = fts3PutVarint(c, iFirstDocid-pWriter->iPrevDocid); | |
812 | |
813 /* Verify that the incoming doclist is valid AND that it ends with | |
814 ** the expected docid. This is essential because we'll trust this | |
815 ** docid in future delta-encoding. | |
816 */ | |
817 ASSERT_VALID_DOCLIST(pWriter->iType, pData, nData, &iLastDocidDelta); | |
818 assert( iLastDocid==iFirstDocid-iDocid+iLastDocidDelta ); | |
819 | |
820 /* Append recoded initial docid and everything else. Rest of docids | |
821 ** should have been delta-encoded from previous initial docid. | |
822 */ | |
823 if( nFirstOld<nData ){ | |
824 dataBufferAppend2(pWriter->b, c, nFirstNew, | |
825 pData+nFirstOld, nData-nFirstOld); | |
826 }else{ | |
827 dataBufferAppend(pWriter->b, c, nFirstNew); | |
828 } | |
829 pWriter->iPrevDocid = iLastDocid; | |
830 return SQLITE_OK; | |
831 } | |
832 static int dlwCopy(DLWriter *pWriter, DLReader *pReader){ | |
833 return dlwAppend(pWriter, dlrDocData(pReader), dlrDocDataBytes(pReader), | |
834 dlrDocid(pReader), dlrDocid(pReader)); | |
835 } | |
836 static void dlwAdd(DLWriter *pWriter, sqlite_int64 iDocid){ | |
837 char c[VARINT_MAX]; | |
838 int n = fts3PutVarint(c, iDocid-pWriter->iPrevDocid); | |
839 | |
840 /* Docids must ascend. */ | |
841 assert( !pWriter->has_iPrevDocid || iDocid>pWriter->iPrevDocid ); | |
842 assert( pWriter->iType==DL_DOCIDS ); | |
843 | |
844 dataBufferAppend(pWriter->b, c, n); | |
845 pWriter->iPrevDocid = iDocid; | |
846 #ifndef NDEBUG | |
847 pWriter->has_iPrevDocid = 1; | |
848 #endif | |
849 } | |
850 | |
851 /*******************************************************************/ | |
852 /* PLReader is used to read data from a document's position list. As | |
853 ** the caller steps through the list, data is cached so that varints | |
854 ** only need to be decoded once. | |
855 ** | |
856 ** plrInit, plrDestroy - create/destroy a reader. | |
857 ** plrColumn, plrPosition, plrStartOffset, plrEndOffset - accessors | |
858 ** plrAtEnd - at end of stream, only call plrDestroy once true. | |
859 ** plrStep - step to the next element. | |
860 */ | |
861 typedef struct PLReader { | |
862 /* These refer to the next position's data. nData will reach 0 when | |
863 ** reading the last position, so plrStep() signals EOF by setting | |
864 ** pData to NULL. | |
865 */ | |
866 const char *pData; | |
867 int nData; | |
868 | |
869 DocListType iType; | |
870 int iColumn; /* the last column read */ | |
871 int iPosition; /* the last position read */ | |
872 int iStartOffset; /* the last start offset read */ | |
873 int iEndOffset; /* the last end offset read */ | |
874 } PLReader; | |
875 | |
876 static int plrAtEnd(PLReader *pReader){ | |
877 return pReader->pData==NULL; | |
878 } | |
879 static int plrColumn(PLReader *pReader){ | |
880 assert( !plrAtEnd(pReader) ); | |
881 return pReader->iColumn; | |
882 } | |
883 static int plrPosition(PLReader *pReader){ | |
884 assert( !plrAtEnd(pReader) ); | |
885 return pReader->iPosition; | |
886 } | |
887 static int plrStartOffset(PLReader *pReader){ | |
888 assert( !plrAtEnd(pReader) ); | |
889 return pReader->iStartOffset; | |
890 } | |
891 static int plrEndOffset(PLReader *pReader){ | |
892 assert( !plrAtEnd(pReader) ); | |
893 return pReader->iEndOffset; | |
894 } | |
895 static int plrStep(PLReader *pReader){ | |
896 int i, n, nTotal = 0; | |
897 | |
898 assert( !plrAtEnd(pReader) ); | |
899 | |
900 if( pReader->nData<=0 ){ | |
901 pReader->pData = NULL; | |
902 return SQLITE_OK; | |
903 } | |
904 | |
905 n = fts3GetVarint32Safe(pReader->pData, &i, pReader->nData); | |
906 if( !n ) return SQLITE_CORRUPT_BKPT; | |
907 nTotal += n; | |
908 if( i==POS_COLUMN ){ | |
909 n = fts3GetVarint32Safe(pReader->pData+nTotal, &pReader->iColumn, pReader->n
Data-nTotal); | |
910 if( !n ) return SQLITE_CORRUPT_BKPT; | |
911 nTotal += n; | |
912 pReader->iPosition = 0; | |
913 pReader->iStartOffset = 0; | |
914 n = fts3GetVarint32Safe(pReader->pData+nTotal, &i, pReader->nData-nTotal); | |
915 if( !n ) return SQLITE_CORRUPT_BKPT; | |
916 nTotal += n; | |
917 } | |
918 /* Should never see adjacent column changes. */ | |
919 assert( i!=POS_COLUMN ); | |
920 | |
921 if( i==POS_END ){ | |
922 assert( nTotal<=pReader->nData ); | |
923 pReader->nData = 0; | |
924 pReader->pData = NULL; | |
925 return SQLITE_OK; | |
926 } | |
927 | |
928 pReader->iPosition += i-POS_BASE; | |
929 if( pReader->iType==DL_POSITIONS_OFFSETS ){ | |
930 n = fts3GetVarint32Safe(pReader->pData+nTotal, &i, pReader->nData-nTotal); | |
931 if( !n ) return SQLITE_CORRUPT_BKPT; | |
932 nTotal += n; | |
933 pReader->iStartOffset += i; | |
934 n = fts3GetVarint32Safe(pReader->pData+nTotal, &i, pReader->nData-nTotal); | |
935 if( !n ) return SQLITE_CORRUPT_BKPT; | |
936 nTotal += n; | |
937 pReader->iEndOffset = pReader->iStartOffset+i; | |
938 } | |
939 assert( nTotal<=pReader->nData ); | |
940 pReader->pData += nTotal; | |
941 pReader->nData -= nTotal; | |
942 return SQLITE_OK; | |
943 } | |
944 | |
945 static void plrDestroy(PLReader *pReader){ | |
946 SCRAMBLE(pReader); | |
947 } | |
948 static int plrInit(PLReader *pReader, DLReader *pDLReader){ | |
949 int rc; | |
950 pReader->pData = dlrPosData(pDLReader); | |
951 pReader->nData = dlrPosDataLen(pDLReader); | |
952 pReader->iType = pDLReader->iType; | |
953 pReader->iColumn = 0; | |
954 pReader->iPosition = 0; | |
955 pReader->iStartOffset = 0; | |
956 pReader->iEndOffset = 0; | |
957 rc = plrStep(pReader); | |
958 if( rc!=SQLITE_OK ) plrDestroy(pReader); | |
959 return rc; | |
960 } | |
961 | |
962 /*******************************************************************/ | |
963 /* PLWriter is used in constructing a document's position list. As a | |
964 ** convenience, if iType is DL_DOCIDS, PLWriter becomes a no-op. | |
965 ** PLWriter writes to the associated DLWriter's buffer. | |
966 ** | |
967 ** plwInit - init for writing a document's poslist. | |
968 ** plwDestroy - clear a writer. | |
969 ** plwAdd - append position and offset information. | |
970 ** plwCopy - copy next position's data from reader to writer. | |
971 ** plwTerminate - add any necessary doclist terminator. | |
972 ** | |
973 ** Calling plwAdd() after plwTerminate() may result in a corrupt | |
974 ** doclist. | |
975 */ | |
976 /* TODO(shess) Until we've written the second item, we can cache the | |
977 ** first item's information. Then we'd have three states: | |
978 ** | |
979 ** - initialized with docid, no positions. | |
980 ** - docid and one position. | |
981 ** - docid and multiple positions. | |
982 ** | |
983 ** Only the last state needs to actually write to dlw->b, which would | |
984 ** be an improvement in the DLCollector case. | |
985 */ | |
986 typedef struct PLWriter { | |
987 DLWriter *dlw; | |
988 | |
989 int iColumn; /* the last column written */ | |
990 int iPos; /* the last position written */ | |
991 int iOffset; /* the last start offset written */ | |
992 } PLWriter; | |
993 | |
994 /* TODO(shess) In the case where the parent is reading these values | |
995 ** from a PLReader, we could optimize to a copy if that PLReader has | |
996 ** the same type as pWriter. | |
997 */ | |
998 static void plwAdd(PLWriter *pWriter, int iColumn, int iPos, | |
999 int iStartOffset, int iEndOffset){ | |
1000 /* Worst-case space for POS_COLUMN, iColumn, iPosDelta, | |
1001 ** iStartOffsetDelta, and iEndOffsetDelta. | |
1002 */ | |
1003 char c[5*VARINT_MAX]; | |
1004 int n = 0; | |
1005 | |
1006 /* Ban plwAdd() after plwTerminate(). */ | |
1007 assert( pWriter->iPos!=-1 ); | |
1008 | |
1009 if( pWriter->dlw->iType==DL_DOCIDS ) return; | |
1010 | |
1011 if( iColumn!=pWriter->iColumn ){ | |
1012 n += fts3PutVarint(c+n, POS_COLUMN); | |
1013 n += fts3PutVarint(c+n, iColumn); | |
1014 pWriter->iColumn = iColumn; | |
1015 pWriter->iPos = 0; | |
1016 pWriter->iOffset = 0; | |
1017 } | |
1018 assert( iPos>=pWriter->iPos ); | |
1019 n += fts3PutVarint(c+n, POS_BASE+(iPos-pWriter->iPos)); | |
1020 pWriter->iPos = iPos; | |
1021 if( pWriter->dlw->iType==DL_POSITIONS_OFFSETS ){ | |
1022 assert( iStartOffset>=pWriter->iOffset ); | |
1023 n += fts3PutVarint(c+n, iStartOffset-pWriter->iOffset); | |
1024 pWriter->iOffset = iStartOffset; | |
1025 assert( iEndOffset>=iStartOffset ); | |
1026 n += fts3PutVarint(c+n, iEndOffset-iStartOffset); | |
1027 } | |
1028 dataBufferAppend(pWriter->dlw->b, c, n); | |
1029 } | |
1030 static void plwCopy(PLWriter *pWriter, PLReader *pReader){ | |
1031 plwAdd(pWriter, plrColumn(pReader), plrPosition(pReader), | |
1032 plrStartOffset(pReader), plrEndOffset(pReader)); | |
1033 } | |
1034 static void plwInit(PLWriter *pWriter, DLWriter *dlw, sqlite_int64 iDocid){ | |
1035 char c[VARINT_MAX]; | |
1036 int n; | |
1037 | |
1038 pWriter->dlw = dlw; | |
1039 | |
1040 /* Docids must ascend. */ | |
1041 assert( !pWriter->dlw->has_iPrevDocid || iDocid>pWriter->dlw->iPrevDocid ); | |
1042 n = fts3PutVarint(c, iDocid-pWriter->dlw->iPrevDocid); | |
1043 dataBufferAppend(pWriter->dlw->b, c, n); | |
1044 pWriter->dlw->iPrevDocid = iDocid; | |
1045 #ifndef NDEBUG | |
1046 pWriter->dlw->has_iPrevDocid = 1; | |
1047 #endif | |
1048 | |
1049 pWriter->iColumn = 0; | |
1050 pWriter->iPos = 0; | |
1051 pWriter->iOffset = 0; | |
1052 } | |
1053 /* TODO(shess) Should plwDestroy() also terminate the doclist? But | |
1054 ** then plwDestroy() would no longer be just a destructor, it would | |
1055 ** also be doing work, which isn't consistent with the overall idiom. | |
1056 ** Another option would be for plwAdd() to always append any necessary | |
1057 ** terminator, so that the output is always correct. But that would | |
1058 ** add incremental work to the common case with the only benefit being | |
1059 ** API elegance. Punt for now. | |
1060 */ | |
1061 static void plwTerminate(PLWriter *pWriter){ | |
1062 if( pWriter->dlw->iType>DL_DOCIDS ){ | |
1063 char c[VARINT_MAX]; | |
1064 int n = fts3PutVarint(c, POS_END); | |
1065 dataBufferAppend(pWriter->dlw->b, c, n); | |
1066 } | |
1067 #ifndef NDEBUG | |
1068 /* Mark as terminated for assert in plwAdd(). */ | |
1069 pWriter->iPos = -1; | |
1070 #endif | |
1071 } | |
1072 static void plwDestroy(PLWriter *pWriter){ | |
1073 SCRAMBLE(pWriter); | |
1074 } | |
1075 | |
1076 /*******************************************************************/ | |
1077 /* DLCollector wraps PLWriter and DLWriter to provide a | |
1078 ** dynamically-allocated doclist area to use during tokenization. | |
1079 ** | |
1080 ** dlcNew - malloc up and initialize a collector. | |
1081 ** dlcDelete - destroy a collector and all contained items. | |
1082 ** dlcAddPos - append position and offset information. | |
1083 ** dlcAddDoclist - add the collected doclist to the given buffer. | |
1084 ** dlcNext - terminate the current document and open another. | |
1085 */ | |
1086 typedef struct DLCollector { | |
1087 DataBuffer b; | |
1088 DLWriter dlw; | |
1089 PLWriter plw; | |
1090 } DLCollector; | |
1091 | |
1092 /* TODO(shess) This could also be done by calling plwTerminate() and | |
1093 ** dataBufferAppend(). I tried that, expecting nominal performance | |
1094 ** differences, but it seemed to pretty reliably be worth 1% to code | |
1095 ** it this way. I suspect it is the incremental malloc overhead (some | |
1096 ** percentage of the plwTerminate() calls will cause a realloc), so | |
1097 ** this might be worth revisiting if the DataBuffer implementation | |
1098 ** changes. | |
1099 */ | |
1100 static void dlcAddDoclist(DLCollector *pCollector, DataBuffer *b){ | |
1101 if( pCollector->dlw.iType>DL_DOCIDS ){ | |
1102 char c[VARINT_MAX]; | |
1103 int n = fts3PutVarint(c, POS_END); | |
1104 dataBufferAppend2(b, pCollector->b.pData, pCollector->b.nData, c, n); | |
1105 }else{ | |
1106 dataBufferAppend(b, pCollector->b.pData, pCollector->b.nData); | |
1107 } | |
1108 } | |
1109 static void dlcNext(DLCollector *pCollector, sqlite_int64 iDocid){ | |
1110 plwTerminate(&pCollector->plw); | |
1111 plwDestroy(&pCollector->plw); | |
1112 plwInit(&pCollector->plw, &pCollector->dlw, iDocid); | |
1113 } | |
1114 static void dlcAddPos(DLCollector *pCollector, int iColumn, int iPos, | |
1115 int iStartOffset, int iEndOffset){ | |
1116 plwAdd(&pCollector->plw, iColumn, iPos, iStartOffset, iEndOffset); | |
1117 } | |
1118 | |
1119 static DLCollector *dlcNew(sqlite_int64 iDocid, DocListType iType){ | |
1120 DLCollector *pCollector = sqlite3_malloc(sizeof(DLCollector)); | |
1121 dataBufferInit(&pCollector->b, 0); | |
1122 dlwInit(&pCollector->dlw, iType, &pCollector->b); | |
1123 plwInit(&pCollector->plw, &pCollector->dlw, iDocid); | |
1124 return pCollector; | |
1125 } | |
1126 static void dlcDelete(DLCollector *pCollector){ | |
1127 plwDestroy(&pCollector->plw); | |
1128 dlwDestroy(&pCollector->dlw); | |
1129 dataBufferDestroy(&pCollector->b); | |
1130 SCRAMBLE(pCollector); | |
1131 sqlite3_free(pCollector); | |
1132 } | |
1133 | |
1134 | |
1135 /* Copy the doclist data of iType in pData/nData into *out, trimming | |
1136 ** unnecessary data as we go. Only columns matching iColumn are | |
1137 ** copied, all columns copied if iColumn is -1. Elements with no | |
1138 ** matching columns are dropped. The output is an iOutType doclist. | |
1139 */ | |
1140 /* NOTE(shess) This code is only valid after all doclists are merged. | |
1141 ** If this is run before merges, then doclist items which represent | |
1142 ** deletion will be trimmed, and will thus not effect a deletion | |
1143 ** during the merge. | |
1144 */ | |
1145 static int docListTrim(DocListType iType, const char *pData, int nData, | |
1146 int iColumn, DocListType iOutType, DataBuffer *out){ | |
1147 DLReader dlReader; | |
1148 DLWriter dlWriter; | |
1149 int rc; | |
1150 | |
1151 assert( iOutType<=iType ); | |
1152 | |
1153 rc = dlrInit(&dlReader, iType, pData, nData); | |
1154 if( rc!=SQLITE_OK ) return rc; | |
1155 dlwInit(&dlWriter, iOutType, out); | |
1156 | |
1157 while( !dlrAtEnd(&dlReader) ){ | |
1158 PLReader plReader; | |
1159 PLWriter plWriter; | |
1160 int match = 0; | |
1161 | |
1162 rc = plrInit(&plReader, &dlReader); | |
1163 if( rc!=SQLITE_OK ) break; | |
1164 | |
1165 while( !plrAtEnd(&plReader) ){ | |
1166 if( iColumn==-1 || plrColumn(&plReader)==iColumn ){ | |
1167 if( !match ){ | |
1168 plwInit(&plWriter, &dlWriter, dlrDocid(&dlReader)); | |
1169 match = 1; | |
1170 } | |
1171 plwAdd(&plWriter, plrColumn(&plReader), plrPosition(&plReader), | |
1172 plrStartOffset(&plReader), plrEndOffset(&plReader)); | |
1173 } | |
1174 rc = plrStep(&plReader); | |
1175 if( rc!=SQLITE_OK ){ | |
1176 plrDestroy(&plReader); | |
1177 goto err; | |
1178 } | |
1179 } | |
1180 if( match ){ | |
1181 plwTerminate(&plWriter); | |
1182 plwDestroy(&plWriter); | |
1183 } | |
1184 | |
1185 plrDestroy(&plReader); | |
1186 rc = dlrStep(&dlReader); | |
1187 if( rc!=SQLITE_OK ) break; | |
1188 } | |
1189 err: | |
1190 dlwDestroy(&dlWriter); | |
1191 dlrDestroy(&dlReader); | |
1192 return rc; | |
1193 } | |
1194 | |
1195 /* Used by docListMerge() to keep doclists in the ascending order by | |
1196 ** docid, then ascending order by age (so the newest comes first). | |
1197 */ | |
1198 typedef struct OrderedDLReader { | |
1199 DLReader *pReader; | |
1200 | |
1201 /* TODO(shess) If we assume that docListMerge pReaders is ordered by | |
1202 ** age (which we do), then we could use pReader comparisons to break | |
1203 ** ties. | |
1204 */ | |
1205 int idx; | |
1206 } OrderedDLReader; | |
1207 | |
1208 /* Order eof to end, then by docid asc, idx desc. */ | |
1209 static int orderedDLReaderCmp(OrderedDLReader *r1, OrderedDLReader *r2){ | |
1210 if( dlrAtEnd(r1->pReader) ){ | |
1211 if( dlrAtEnd(r2->pReader) ) return 0; /* Both atEnd(). */ | |
1212 return 1; /* Only r1 atEnd(). */ | |
1213 } | |
1214 if( dlrAtEnd(r2->pReader) ) return -1; /* Only r2 atEnd(). */ | |
1215 | |
1216 if( dlrDocid(r1->pReader)<dlrDocid(r2->pReader) ) return -1; | |
1217 if( dlrDocid(r1->pReader)>dlrDocid(r2->pReader) ) return 1; | |
1218 | |
1219 /* Descending on idx. */ | |
1220 return r2->idx-r1->idx; | |
1221 } | |
1222 | |
1223 /* Bubble p[0] to appropriate place in p[1..n-1]. Assumes that | |
1224 ** p[1..n-1] is already sorted. | |
1225 */ | |
1226 /* TODO(shess) Is this frequent enough to warrant a binary search? | |
1227 ** Before implementing that, instrument the code to check. In most | |
1228 ** current usage, I expect that p[0] will be less than p[1] a very | |
1229 ** high proportion of the time. | |
1230 */ | |
1231 static void orderedDLReaderReorder(OrderedDLReader *p, int n){ | |
1232 while( n>1 && orderedDLReaderCmp(p, p+1)>0 ){ | |
1233 OrderedDLReader tmp = p[0]; | |
1234 p[0] = p[1]; | |
1235 p[1] = tmp; | |
1236 n--; | |
1237 p++; | |
1238 } | |
1239 } | |
1240 | |
1241 /* Given an array of doclist readers, merge their doclist elements | |
1242 ** into out in sorted order (by docid), dropping elements from older | |
1243 ** readers when there is a duplicate docid. pReaders is assumed to be | |
1244 ** ordered by age, oldest first. | |
1245 */ | |
1246 /* TODO(shess) nReaders must be <= MERGE_COUNT. This should probably | |
1247 ** be fixed. | |
1248 */ | |
1249 static int docListMerge(DataBuffer *out, | |
1250 DLReader *pReaders, int nReaders){ | |
1251 OrderedDLReader readers[MERGE_COUNT]; | |
1252 DLWriter writer; | |
1253 int i, n; | |
1254 const char *pStart = 0; | |
1255 int nStart = 0; | |
1256 sqlite_int64 iFirstDocid = 0, iLastDocid = 0; | |
1257 int rc = SQLITE_OK; | |
1258 | |
1259 assert( nReaders>0 ); | |
1260 if( nReaders==1 ){ | |
1261 dataBufferAppend(out, dlrDocData(pReaders), dlrAllDataBytes(pReaders)); | |
1262 return SQLITE_OK; | |
1263 } | |
1264 | |
1265 assert( nReaders<=MERGE_COUNT ); | |
1266 n = 0; | |
1267 for(i=0; i<nReaders; i++){ | |
1268 assert( pReaders[i].iType==pReaders[0].iType ); | |
1269 readers[i].pReader = pReaders+i; | |
1270 readers[i].idx = i; | |
1271 n += dlrAllDataBytes(&pReaders[i]); | |
1272 } | |
1273 /* Conservatively size output to sum of inputs. Output should end | |
1274 ** up strictly smaller than input. | |
1275 */ | |
1276 dataBufferExpand(out, n); | |
1277 | |
1278 /* Get the readers into sorted order. */ | |
1279 while( i-->0 ){ | |
1280 orderedDLReaderReorder(readers+i, nReaders-i); | |
1281 } | |
1282 | |
1283 dlwInit(&writer, pReaders[0].iType, out); | |
1284 while( !dlrAtEnd(readers[0].pReader) ){ | |
1285 sqlite_int64 iDocid = dlrDocid(readers[0].pReader); | |
1286 | |
1287 /* If this is a continuation of the current buffer to copy, extend | |
1288 ** that buffer. memcpy() seems to be more efficient if it has a | |
1289 ** lots of data to copy. | |
1290 */ | |
1291 if( dlrDocData(readers[0].pReader)==pStart+nStart ){ | |
1292 nStart += dlrDocDataBytes(readers[0].pReader); | |
1293 }else{ | |
1294 if( pStart!=0 ){ | |
1295 rc = dlwAppend(&writer, pStart, nStart, iFirstDocid, iLastDocid); | |
1296 if( rc!=SQLITE_OK ) goto err; | |
1297 } | |
1298 pStart = dlrDocData(readers[0].pReader); | |
1299 nStart = dlrDocDataBytes(readers[0].pReader); | |
1300 iFirstDocid = iDocid; | |
1301 } | |
1302 iLastDocid = iDocid; | |
1303 rc = dlrStep(readers[0].pReader); | |
1304 if( rc!= SQLITE_OK ) goto err; | |
1305 | |
1306 /* Drop all of the older elements with the same docid. */ | |
1307 for(i=1; i<nReaders && | |
1308 !dlrAtEnd(readers[i].pReader) && | |
1309 dlrDocid(readers[i].pReader)==iDocid; i++){ | |
1310 rc = dlrStep(readers[i].pReader); | |
1311 if( rc!=SQLITE_OK ) goto err; | |
1312 } | |
1313 | |
1314 /* Get the readers back into order. */ | |
1315 while( i-->0 ){ | |
1316 orderedDLReaderReorder(readers+i, nReaders-i); | |
1317 } | |
1318 } | |
1319 | |
1320 /* Copy over any remaining elements. */ | |
1321 if( nStart>0 ) rc = dlwAppend(&writer, pStart, nStart, iFirstDocid, iLastDocid
); | |
1322 err: | |
1323 dlwDestroy(&writer); | |
1324 return rc; | |
1325 } | |
1326 | |
1327 /* Helper function for posListUnion(). Compares the current position | |
1328 ** between left and right, returning as standard C idiom of <0 if | |
1329 ** left<right, >0 if left>right, and 0 if left==right. "End" always | |
1330 ** compares greater. | |
1331 */ | |
1332 static int posListCmp(PLReader *pLeft, PLReader *pRight){ | |
1333 assert( pLeft->iType==pRight->iType ); | |
1334 if( pLeft->iType==DL_DOCIDS ) return 0; | |
1335 | |
1336 if( plrAtEnd(pLeft) ) return plrAtEnd(pRight) ? 0 : 1; | |
1337 if( plrAtEnd(pRight) ) return -1; | |
1338 | |
1339 if( plrColumn(pLeft)<plrColumn(pRight) ) return -1; | |
1340 if( plrColumn(pLeft)>plrColumn(pRight) ) return 1; | |
1341 | |
1342 if( plrPosition(pLeft)<plrPosition(pRight) ) return -1; | |
1343 if( plrPosition(pLeft)>plrPosition(pRight) ) return 1; | |
1344 if( pLeft->iType==DL_POSITIONS ) return 0; | |
1345 | |
1346 if( plrStartOffset(pLeft)<plrStartOffset(pRight) ) return -1; | |
1347 if( plrStartOffset(pLeft)>plrStartOffset(pRight) ) return 1; | |
1348 | |
1349 if( plrEndOffset(pLeft)<plrEndOffset(pRight) ) return -1; | |
1350 if( plrEndOffset(pLeft)>plrEndOffset(pRight) ) return 1; | |
1351 | |
1352 return 0; | |
1353 } | |
1354 | |
1355 /* Write the union of position lists in pLeft and pRight to pOut. | |
1356 ** "Union" in this case meaning "All unique position tuples". Should | |
1357 ** work with any doclist type, though both inputs and the output | |
1358 ** should be the same type. | |
1359 */ | |
1360 static int posListUnion(DLReader *pLeft, DLReader *pRight, DLWriter *pOut){ | |
1361 PLReader left, right; | |
1362 PLWriter writer; | |
1363 int rc; | |
1364 | |
1365 assert( dlrDocid(pLeft)==dlrDocid(pRight) ); | |
1366 assert( pLeft->iType==pRight->iType ); | |
1367 assert( pLeft->iType==pOut->iType ); | |
1368 | |
1369 rc = plrInit(&left, pLeft); | |
1370 if( rc!=SQLITE_OK ) return rc; | |
1371 rc = plrInit(&right, pRight); | |
1372 if( rc!=SQLITE_OK ){ | |
1373 plrDestroy(&left); | |
1374 return rc; | |
1375 } | |
1376 plwInit(&writer, pOut, dlrDocid(pLeft)); | |
1377 | |
1378 while( !plrAtEnd(&left) || !plrAtEnd(&right) ){ | |
1379 int c = posListCmp(&left, &right); | |
1380 if( c<0 ){ | |
1381 plwCopy(&writer, &left); | |
1382 rc = plrStep(&left); | |
1383 if( rc!=SQLITE_OK ) break; | |
1384 }else if( c>0 ){ | |
1385 plwCopy(&writer, &right); | |
1386 rc = plrStep(&right); | |
1387 if( rc!=SQLITE_OK ) break; | |
1388 }else{ | |
1389 plwCopy(&writer, &left); | |
1390 rc = plrStep(&left); | |
1391 if( rc!=SQLITE_OK ) break; | |
1392 rc = plrStep(&right); | |
1393 if( rc!=SQLITE_OK ) break; | |
1394 } | |
1395 } | |
1396 | |
1397 plwTerminate(&writer); | |
1398 plwDestroy(&writer); | |
1399 plrDestroy(&left); | |
1400 plrDestroy(&right); | |
1401 return rc; | |
1402 } | |
1403 | |
1404 /* Write the union of doclists in pLeft and pRight to pOut. For | |
1405 ** docids in common between the inputs, the union of the position | |
1406 ** lists is written. Inputs and outputs are always type DL_DEFAULT. | |
1407 */ | |
1408 static int docListUnion( | |
1409 const char *pLeft, int nLeft, | |
1410 const char *pRight, int nRight, | |
1411 DataBuffer *pOut /* Write the combined doclist here */ | |
1412 ){ | |
1413 DLReader left, right; | |
1414 DLWriter writer; | |
1415 int rc; | |
1416 | |
1417 if( nLeft==0 ){ | |
1418 if( nRight!=0) dataBufferAppend(pOut, pRight, nRight); | |
1419 return SQLITE_OK; | |
1420 } | |
1421 if( nRight==0 ){ | |
1422 dataBufferAppend(pOut, pLeft, nLeft); | |
1423 return SQLITE_OK; | |
1424 } | |
1425 | |
1426 rc = dlrInit(&left, DL_DEFAULT, pLeft, nLeft); | |
1427 if( rc!=SQLITE_OK ) return rc; | |
1428 rc = dlrInit(&right, DL_DEFAULT, pRight, nRight); | |
1429 if( rc!=SQLITE_OK){ | |
1430 dlrDestroy(&left); | |
1431 return rc; | |
1432 } | |
1433 dlwInit(&writer, DL_DEFAULT, pOut); | |
1434 | |
1435 while( !dlrAtEnd(&left) || !dlrAtEnd(&right) ){ | |
1436 if( dlrAtEnd(&right) ){ | |
1437 rc = dlwCopy(&writer, &left); | |
1438 if( rc!=SQLITE_OK) break; | |
1439 rc = dlrStep(&left); | |
1440 if( rc!=SQLITE_OK) break; | |
1441 }else if( dlrAtEnd(&left) ){ | |
1442 rc = dlwCopy(&writer, &right); | |
1443 if( rc!=SQLITE_OK ) break; | |
1444 rc = dlrStep(&right); | |
1445 if( rc!=SQLITE_OK ) break; | |
1446 }else if( dlrDocid(&left)<dlrDocid(&right) ){ | |
1447 rc = dlwCopy(&writer, &left); | |
1448 if( rc!=SQLITE_OK ) break; | |
1449 rc = dlrStep(&left); | |
1450 if( rc!=SQLITE_OK ) break; | |
1451 }else if( dlrDocid(&left)>dlrDocid(&right) ){ | |
1452 rc = dlwCopy(&writer, &right); | |
1453 if( rc!=SQLITE_OK ) break; | |
1454 rc = dlrStep(&right); | |
1455 if( rc!=SQLITE_OK ) break; | |
1456 }else{ | |
1457 rc = posListUnion(&left, &right, &writer); | |
1458 if( rc!=SQLITE_OK ) break; | |
1459 rc = dlrStep(&left); | |
1460 if( rc!=SQLITE_OK ) break; | |
1461 rc = dlrStep(&right); | |
1462 if( rc!=SQLITE_OK ) break; | |
1463 } | |
1464 } | |
1465 | |
1466 dlrDestroy(&left); | |
1467 dlrDestroy(&right); | |
1468 dlwDestroy(&writer); | |
1469 return rc; | |
1470 } | |
1471 | |
1472 /* | |
1473 ** This function is used as part of the implementation of phrase and | |
1474 ** NEAR matching. | |
1475 ** | |
1476 ** pLeft and pRight are DLReaders positioned to the same docid in | |
1477 ** lists of type DL_POSITION. This function writes an entry to the | |
1478 ** DLWriter pOut for each position in pRight that is less than | |
1479 ** (nNear+1) greater (but not equal to or smaller) than a position | |
1480 ** in pLeft. For example, if nNear is 0, and the positions contained | |
1481 ** by pLeft and pRight are: | |
1482 ** | |
1483 ** pLeft: 5 10 15 20 | |
1484 ** pRight: 6 9 17 21 | |
1485 ** | |
1486 ** then the docid is added to pOut. If pOut is of type DL_POSITIONS, | |
1487 ** then a positionids "6" and "21" are also added to pOut. | |
1488 ** | |
1489 ** If boolean argument isSaveLeft is true, then positionids are copied | |
1490 ** from pLeft instead of pRight. In the example above, the positions "5" | |
1491 ** and "20" would be added instead of "6" and "21". | |
1492 */ | |
1493 static int posListPhraseMerge( | |
1494 DLReader *pLeft, | |
1495 DLReader *pRight, | |
1496 int nNear, | |
1497 int isSaveLeft, | |
1498 DLWriter *pOut | |
1499 ){ | |
1500 PLReader left, right; | |
1501 PLWriter writer; | |
1502 int match = 0; | |
1503 int rc; | |
1504 | |
1505 assert( dlrDocid(pLeft)==dlrDocid(pRight) ); | |
1506 assert( pOut->iType!=DL_POSITIONS_OFFSETS ); | |
1507 | |
1508 rc = plrInit(&left, pLeft); | |
1509 if( rc!=SQLITE_OK ) return rc; | |
1510 rc = plrInit(&right, pRight); | |
1511 if( rc!=SQLITE_OK ){ | |
1512 plrDestroy(&left); | |
1513 return rc; | |
1514 } | |
1515 | |
1516 while( !plrAtEnd(&left) && !plrAtEnd(&right) ){ | |
1517 if( plrColumn(&left)<plrColumn(&right) ){ | |
1518 rc = plrStep(&left); | |
1519 if( rc!=SQLITE_OK ) break; | |
1520 }else if( plrColumn(&left)>plrColumn(&right) ){ | |
1521 rc = plrStep(&right); | |
1522 if( rc!=SQLITE_OK ) break; | |
1523 }else if( plrPosition(&left)>=plrPosition(&right) ){ | |
1524 rc = plrStep(&right); | |
1525 if( rc!=SQLITE_OK ) break; | |
1526 }else{ | |
1527 if( (plrPosition(&right)-plrPosition(&left))<=(nNear+1) ){ | |
1528 if( !match ){ | |
1529 plwInit(&writer, pOut, dlrDocid(pLeft)); | |
1530 match = 1; | |
1531 } | |
1532 if( !isSaveLeft ){ | |
1533 plwAdd(&writer, plrColumn(&right), plrPosition(&right), 0, 0); | |
1534 }else{ | |
1535 plwAdd(&writer, plrColumn(&left), plrPosition(&left), 0, 0); | |
1536 } | |
1537 rc = plrStep(&right); | |
1538 if( rc!=SQLITE_OK ) break; | |
1539 }else{ | |
1540 rc = plrStep(&left); | |
1541 if( rc!=SQLITE_OK ) break; | |
1542 } | |
1543 } | |
1544 } | |
1545 | |
1546 if( match ){ | |
1547 plwTerminate(&writer); | |
1548 plwDestroy(&writer); | |
1549 } | |
1550 | |
1551 plrDestroy(&left); | |
1552 plrDestroy(&right); | |
1553 return rc; | |
1554 } | |
1555 | |
1556 /* | |
1557 ** Compare the values pointed to by the PLReaders passed as arguments. | |
1558 ** Return -1 if the value pointed to by pLeft is considered less than | |
1559 ** the value pointed to by pRight, +1 if it is considered greater | |
1560 ** than it, or 0 if it is equal. i.e. | |
1561 ** | |
1562 ** (*pLeft - *pRight) | |
1563 ** | |
1564 ** A PLReader that is in the EOF condition is considered greater than | |
1565 ** any other. If neither argument is in EOF state, the return value of | |
1566 ** plrColumn() is used. If the plrColumn() values are equal, the | |
1567 ** comparison is on the basis of plrPosition(). | |
1568 */ | |
1569 static int plrCompare(PLReader *pLeft, PLReader *pRight){ | |
1570 assert(!plrAtEnd(pLeft) || !plrAtEnd(pRight)); | |
1571 | |
1572 if( plrAtEnd(pRight) || plrAtEnd(pLeft) ){ | |
1573 return (plrAtEnd(pRight) ? -1 : 1); | |
1574 } | |
1575 if( plrColumn(pLeft)!=plrColumn(pRight) ){ | |
1576 return ((plrColumn(pLeft)<plrColumn(pRight)) ? -1 : 1); | |
1577 } | |
1578 if( plrPosition(pLeft)!=plrPosition(pRight) ){ | |
1579 return ((plrPosition(pLeft)<plrPosition(pRight)) ? -1 : 1); | |
1580 } | |
1581 return 0; | |
1582 } | |
1583 | |
1584 /* We have two doclists with positions: pLeft and pRight. Depending | |
1585 ** on the value of the nNear parameter, perform either a phrase | |
1586 ** intersection (if nNear==0) or a NEAR intersection (if nNear>0) | |
1587 ** and write the results into pOut. | |
1588 ** | |
1589 ** A phrase intersection means that two documents only match | |
1590 ** if pLeft.iPos+1==pRight.iPos. | |
1591 ** | |
1592 ** A NEAR intersection means that two documents only match if | |
1593 ** (abs(pLeft.iPos-pRight.iPos)<nNear). | |
1594 ** | |
1595 ** If a NEAR intersection is requested, then the nPhrase argument should | |
1596 ** be passed the number of tokens in the two operands to the NEAR operator | |
1597 ** combined. For example: | |
1598 ** | |
1599 ** Query syntax nPhrase | |
1600 ** ------------------------------------ | |
1601 ** "A B C" NEAR "D E" 5 | |
1602 ** A NEAR B 2 | |
1603 ** | |
1604 ** iType controls the type of data written to pOut. If iType is | |
1605 ** DL_POSITIONS, the positions are those from pRight. | |
1606 */ | |
1607 static int docListPhraseMerge( | |
1608 const char *pLeft, int nLeft, | |
1609 const char *pRight, int nRight, | |
1610 int nNear, /* 0 for a phrase merge, non-zero for a NEAR merge */ | |
1611 int nPhrase, /* Number of tokens in left+right operands to NEAR */ | |
1612 DocListType iType, /* Type of doclist to write to pOut */ | |
1613 DataBuffer *pOut /* Write the combined doclist here */ | |
1614 ){ | |
1615 DLReader left, right; | |
1616 DLWriter writer; | |
1617 int rc; | |
1618 | |
1619 /* These two buffers are used in the 'while', but are declared here | |
1620 ** to simplify error-handling. | |
1621 */ | |
1622 DataBuffer one = {0, 0, 0}; | |
1623 DataBuffer two = {0, 0, 0}; | |
1624 | |
1625 if( nLeft==0 || nRight==0 ) return SQLITE_OK; | |
1626 | |
1627 assert( iType!=DL_POSITIONS_OFFSETS ); | |
1628 | |
1629 rc = dlrInit(&left, DL_POSITIONS, pLeft, nLeft); | |
1630 if( rc!=SQLITE_OK ) return rc; | |
1631 rc = dlrInit(&right, DL_POSITIONS, pRight, nRight); | |
1632 if( rc!=SQLITE_OK ){ | |
1633 dlrDestroy(&left); | |
1634 return rc; | |
1635 } | |
1636 dlwInit(&writer, iType, pOut); | |
1637 | |
1638 while( !dlrAtEnd(&left) && !dlrAtEnd(&right) ){ | |
1639 if( dlrDocid(&left)<dlrDocid(&right) ){ | |
1640 rc = dlrStep(&left); | |
1641 if( rc!=SQLITE_OK ) goto err; | |
1642 }else if( dlrDocid(&right)<dlrDocid(&left) ){ | |
1643 rc = dlrStep(&right); | |
1644 if( rc!=SQLITE_OK ) goto err; | |
1645 }else{ | |
1646 if( nNear==0 ){ | |
1647 rc = posListPhraseMerge(&left, &right, 0, 0, &writer); | |
1648 if( rc!=SQLITE_OK ) goto err; | |
1649 }else{ | |
1650 /* This case occurs when two terms (simple terms or phrases) are | |
1651 * connected by a NEAR operator, span (nNear+1). i.e. | |
1652 * | |
1653 * '"terrible company" NEAR widget' | |
1654 */ | |
1655 DLWriter dlwriter2; | |
1656 DLReader dr1 = {0, 0, 0, 0, 0}; | |
1657 DLReader dr2 = {0, 0, 0, 0, 0}; | |
1658 | |
1659 dlwInit(&dlwriter2, iType, &one); | |
1660 rc = posListPhraseMerge(&right, &left, nNear-3+nPhrase, 1, &dlwriter2); | |
1661 if( rc!=SQLITE_OK ) goto err; | |
1662 dlwInit(&dlwriter2, iType, &two); | |
1663 rc = posListPhraseMerge(&left, &right, nNear-1, 0, &dlwriter2); | |
1664 if( rc!=SQLITE_OK ) goto err; | |
1665 | |
1666 if( one.nData){ | |
1667 rc = dlrInit(&dr1, iType, one.pData, one.nData); | |
1668 if( rc!=SQLITE_OK ) goto err; | |
1669 } | |
1670 if( two.nData){ | |
1671 rc = dlrInit(&dr2, iType, two.pData, two.nData); | |
1672 if( rc!=SQLITE_OK ) goto err; | |
1673 } | |
1674 | |
1675 if( !dlrAtEnd(&dr1) || !dlrAtEnd(&dr2) ){ | |
1676 PLReader pr1 = {0}; | |
1677 PLReader pr2 = {0}; | |
1678 | |
1679 PLWriter plwriter; | |
1680 plwInit(&plwriter, &writer, dlrDocid(dlrAtEnd(&dr1)?&dr2:&dr1)); | |
1681 | |
1682 if( one.nData ){ | |
1683 rc = plrInit(&pr1, &dr1); | |
1684 if( rc!=SQLITE_OK ) goto err; | |
1685 } | |
1686 if( two.nData ){ | |
1687 rc = plrInit(&pr2, &dr2); | |
1688 if( rc!=SQLITE_OK ) goto err; | |
1689 } | |
1690 while( !plrAtEnd(&pr1) || !plrAtEnd(&pr2) ){ | |
1691 int iCompare = plrCompare(&pr1, &pr2); | |
1692 switch( iCompare ){ | |
1693 case -1: | |
1694 plwCopy(&plwriter, &pr1); | |
1695 rc = plrStep(&pr1); | |
1696 if( rc!=SQLITE_OK ) goto err; | |
1697 break; | |
1698 case 1: | |
1699 plwCopy(&plwriter, &pr2); | |
1700 rc = plrStep(&pr2); | |
1701 if( rc!=SQLITE_OK ) goto err; | |
1702 break; | |
1703 case 0: | |
1704 plwCopy(&plwriter, &pr1); | |
1705 rc = plrStep(&pr1); | |
1706 if( rc!=SQLITE_OK ) goto err; | |
1707 rc = plrStep(&pr2); | |
1708 if( rc!=SQLITE_OK ) goto err; | |
1709 break; | |
1710 } | |
1711 } | |
1712 plwTerminate(&plwriter); | |
1713 } | |
1714 dataBufferReset(&one); | |
1715 dataBufferReset(&two); | |
1716 } | |
1717 rc = dlrStep(&left); | |
1718 if( rc!=SQLITE_OK ) goto err; | |
1719 rc = dlrStep(&right); | |
1720 if( rc!=SQLITE_OK ) goto err; | |
1721 } | |
1722 } | |
1723 | |
1724 err: | |
1725 dataBufferDestroy(&one); | |
1726 dataBufferDestroy(&two); | |
1727 dlrDestroy(&left); | |
1728 dlrDestroy(&right); | |
1729 dlwDestroy(&writer); | |
1730 return rc; | |
1731 } | |
1732 | |
1733 /* We have two DL_DOCIDS doclists: pLeft and pRight. | |
1734 ** Write the intersection of these two doclists into pOut as a | |
1735 ** DL_DOCIDS doclist. | |
1736 */ | |
1737 static int docListAndMerge( | |
1738 const char *pLeft, int nLeft, | |
1739 const char *pRight, int nRight, | |
1740 DataBuffer *pOut /* Write the combined doclist here */ | |
1741 ){ | |
1742 DLReader left, right; | |
1743 DLWriter writer; | |
1744 int rc; | |
1745 | |
1746 if( nLeft==0 || nRight==0 ) return SQLITE_OK; | |
1747 | |
1748 rc = dlrInit(&left, DL_DOCIDS, pLeft, nLeft); | |
1749 if( rc!=SQLITE_OK ) return rc; | |
1750 rc = dlrInit(&right, DL_DOCIDS, pRight, nRight); | |
1751 if( rc!=SQLITE_OK ){ | |
1752 dlrDestroy(&left); | |
1753 return rc; | |
1754 } | |
1755 dlwInit(&writer, DL_DOCIDS, pOut); | |
1756 | |
1757 while( !dlrAtEnd(&left) && !dlrAtEnd(&right) ){ | |
1758 if( dlrDocid(&left)<dlrDocid(&right) ){ | |
1759 rc = dlrStep(&left); | |
1760 if( rc!=SQLITE_OK ) break; | |
1761 }else if( dlrDocid(&right)<dlrDocid(&left) ){ | |
1762 rc = dlrStep(&right); | |
1763 if( rc!=SQLITE_OK ) break; | |
1764 }else{ | |
1765 dlwAdd(&writer, dlrDocid(&left)); | |
1766 rc = dlrStep(&left); | |
1767 if( rc!=SQLITE_OK ) break; | |
1768 rc = dlrStep(&right); | |
1769 if( rc!=SQLITE_OK ) break; | |
1770 } | |
1771 } | |
1772 | |
1773 dlrDestroy(&left); | |
1774 dlrDestroy(&right); | |
1775 dlwDestroy(&writer); | |
1776 return rc; | |
1777 } | |
1778 | |
1779 /* We have two DL_DOCIDS doclists: pLeft and pRight. | |
1780 ** Write the union of these two doclists into pOut as a | |
1781 ** DL_DOCIDS doclist. | |
1782 */ | |
1783 static int docListOrMerge( | |
1784 const char *pLeft, int nLeft, | |
1785 const char *pRight, int nRight, | |
1786 DataBuffer *pOut /* Write the combined doclist here */ | |
1787 ){ | |
1788 DLReader left, right; | |
1789 DLWriter writer; | |
1790 int rc; | |
1791 | |
1792 if( nLeft==0 ){ | |
1793 if( nRight!=0 ) dataBufferAppend(pOut, pRight, nRight); | |
1794 return SQLITE_OK; | |
1795 } | |
1796 if( nRight==0 ){ | |
1797 dataBufferAppend(pOut, pLeft, nLeft); | |
1798 return SQLITE_OK; | |
1799 } | |
1800 | |
1801 rc = dlrInit(&left, DL_DOCIDS, pLeft, nLeft); | |
1802 if( rc!=SQLITE_OK ) return rc; | |
1803 rc = dlrInit(&right, DL_DOCIDS, pRight, nRight); | |
1804 if( rc!=SQLITE_OK ){ | |
1805 dlrDestroy(&left); | |
1806 return rc; | |
1807 } | |
1808 dlwInit(&writer, DL_DOCIDS, pOut); | |
1809 | |
1810 while( !dlrAtEnd(&left) || !dlrAtEnd(&right) ){ | |
1811 if( dlrAtEnd(&right) ){ | |
1812 dlwAdd(&writer, dlrDocid(&left)); | |
1813 rc = dlrStep(&left); | |
1814 if( rc!=SQLITE_OK ) break; | |
1815 }else if( dlrAtEnd(&left) ){ | |
1816 dlwAdd(&writer, dlrDocid(&right)); | |
1817 rc = dlrStep(&right); | |
1818 if( rc!=SQLITE_OK ) break; | |
1819 }else if( dlrDocid(&left)<dlrDocid(&right) ){ | |
1820 dlwAdd(&writer, dlrDocid(&left)); | |
1821 rc = dlrStep(&left); | |
1822 if( rc!=SQLITE_OK ) break; | |
1823 }else if( dlrDocid(&right)<dlrDocid(&left) ){ | |
1824 dlwAdd(&writer, dlrDocid(&right)); | |
1825 rc = dlrStep(&right); | |
1826 if( rc!=SQLITE_OK ) break; | |
1827 }else{ | |
1828 dlwAdd(&writer, dlrDocid(&left)); | |
1829 rc = dlrStep(&left); | |
1830 if( rc!=SQLITE_OK ) break; | |
1831 rc = dlrStep(&right); | |
1832 if( rc!=SQLITE_OK ) break; | |
1833 } | |
1834 } | |
1835 | |
1836 dlrDestroy(&left); | |
1837 dlrDestroy(&right); | |
1838 dlwDestroy(&writer); | |
1839 return rc; | |
1840 } | |
1841 | |
1842 /* We have two DL_DOCIDS doclists: pLeft and pRight. | |
1843 ** Write into pOut as DL_DOCIDS doclist containing all documents that | |
1844 ** occur in pLeft but not in pRight. | |
1845 */ | |
1846 static int docListExceptMerge( | |
1847 const char *pLeft, int nLeft, | |
1848 const char *pRight, int nRight, | |
1849 DataBuffer *pOut /* Write the combined doclist here */ | |
1850 ){ | |
1851 DLReader left, right; | |
1852 DLWriter writer; | |
1853 int rc; | |
1854 | |
1855 if( nLeft==0 ) return SQLITE_OK; | |
1856 if( nRight==0 ){ | |
1857 dataBufferAppend(pOut, pLeft, nLeft); | |
1858 return SQLITE_OK; | |
1859 } | |
1860 | |
1861 rc = dlrInit(&left, DL_DOCIDS, pLeft, nLeft); | |
1862 if( rc!=SQLITE_OK ) return rc; | |
1863 rc = dlrInit(&right, DL_DOCIDS, pRight, nRight); | |
1864 if( rc!=SQLITE_OK ){ | |
1865 dlrDestroy(&left); | |
1866 return rc; | |
1867 } | |
1868 dlwInit(&writer, DL_DOCIDS, pOut); | |
1869 | |
1870 while( !dlrAtEnd(&left) ){ | |
1871 while( !dlrAtEnd(&right) && dlrDocid(&right)<dlrDocid(&left) ){ | |
1872 rc = dlrStep(&right); | |
1873 if( rc!=SQLITE_OK ) goto err; | |
1874 } | |
1875 if( dlrAtEnd(&right) || dlrDocid(&left)<dlrDocid(&right) ){ | |
1876 dlwAdd(&writer, dlrDocid(&left)); | |
1877 } | |
1878 rc = dlrStep(&left); | |
1879 if( rc!=SQLITE_OK ) break; | |
1880 } | |
1881 | |
1882 err: | |
1883 dlrDestroy(&left); | |
1884 dlrDestroy(&right); | |
1885 dlwDestroy(&writer); | |
1886 return rc; | |
1887 } | |
1888 | |
1889 static char *string_dup_n(const char *s, int n){ | |
1890 char *str = sqlite3_malloc(n + 1); | |
1891 memcpy(str, s, n); | |
1892 str[n] = '\0'; | |
1893 return str; | |
1894 } | |
1895 | |
1896 /* Duplicate a string; the caller must free() the returned string. | |
1897 * (We don't use strdup() since it is not part of the standard C library and | |
1898 * may not be available everywhere.) */ | |
1899 static char *string_dup(const char *s){ | |
1900 return string_dup_n(s, strlen(s)); | |
1901 } | |
1902 | |
1903 /* Format a string, replacing each occurrence of the % character with | |
1904 * zDb.zName. This may be more convenient than sqlite_mprintf() | |
1905 * when one string is used repeatedly in a format string. | |
1906 * The caller must free() the returned string. */ | |
1907 static char *string_format(const char *zFormat, | |
1908 const char *zDb, const char *zName){ | |
1909 const char *p; | |
1910 size_t len = 0; | |
1911 size_t nDb = strlen(zDb); | |
1912 size_t nName = strlen(zName); | |
1913 size_t nFullTableName = nDb+1+nName; | |
1914 char *result; | |
1915 char *r; | |
1916 | |
1917 /* first compute length needed */ | |
1918 for(p = zFormat ; *p ; ++p){ | |
1919 len += (*p=='%' ? nFullTableName : 1); | |
1920 } | |
1921 len += 1; /* for null terminator */ | |
1922 | |
1923 r = result = sqlite3_malloc(len); | |
1924 for(p = zFormat; *p; ++p){ | |
1925 if( *p=='%' ){ | |
1926 memcpy(r, zDb, nDb); | |
1927 r += nDb; | |
1928 *r++ = '.'; | |
1929 memcpy(r, zName, nName); | |
1930 r += nName; | |
1931 } else { | |
1932 *r++ = *p; | |
1933 } | |
1934 } | |
1935 *r++ = '\0'; | |
1936 assert( r == result + len ); | |
1937 return result; | |
1938 } | |
1939 | |
1940 static int sql_exec(sqlite3 *db, const char *zDb, const char *zName, | |
1941 const char *zFormat){ | |
1942 char *zCommand = string_format(zFormat, zDb, zName); | |
1943 int rc; | |
1944 FTSTRACE(("FTS3 sql: %s\n", zCommand)); | |
1945 rc = sqlite3_exec(db, zCommand, NULL, 0, NULL); | |
1946 sqlite3_free(zCommand); | |
1947 return rc; | |
1948 } | |
1949 | |
1950 static int sql_prepare(sqlite3 *db, const char *zDb, const char *zName, | |
1951 sqlite3_stmt **ppStmt, const char *zFormat){ | |
1952 char *zCommand = string_format(zFormat, zDb, zName); | |
1953 int rc; | |
1954 FTSTRACE(("FTS3 prepare: %s\n", zCommand)); | |
1955 rc = sqlite3_prepare_v2(db, zCommand, -1, ppStmt, NULL); | |
1956 sqlite3_free(zCommand); | |
1957 return rc; | |
1958 } | |
1959 | |
1960 /* end utility functions */ | |
1961 | |
1962 /* Forward reference */ | |
1963 typedef struct fulltext_vtab fulltext_vtab; | |
1964 | |
1965 /* | |
1966 ** An instance of the following structure keeps track of generated | |
1967 ** matching-word offset information and snippets. | |
1968 */ | |
1969 typedef struct Snippet { | |
1970 int nMatch; /* Total number of matches */ | |
1971 int nAlloc; /* Space allocated for aMatch[] */ | |
1972 struct snippetMatch { /* One entry for each matching term */ | |
1973 char snStatus; /* Status flag for use while constructing snippets */ | |
1974 short int iCol; /* The column that contains the match */ | |
1975 short int iTerm; /* The index in Query.pTerms[] of the matching term */ | |
1976 int iToken; /* The index of the matching document token */ | |
1977 short int nByte; /* Number of bytes in the term */ | |
1978 int iStart; /* The offset to the first character of the term */ | |
1979 } *aMatch; /* Points to space obtained from malloc */ | |
1980 char *zOffset; /* Text rendering of aMatch[] */ | |
1981 int nOffset; /* strlen(zOffset) */ | |
1982 char *zSnippet; /* Snippet text */ | |
1983 int nSnippet; /* strlen(zSnippet) */ | |
1984 } Snippet; | |
1985 | |
1986 | |
1987 typedef enum QueryType { | |
1988 QUERY_GENERIC, /* table scan */ | |
1989 QUERY_DOCID, /* lookup by docid */ | |
1990 QUERY_FULLTEXT /* QUERY_FULLTEXT + [i] is a full-text search for column i*/ | |
1991 } QueryType; | |
1992 | |
1993 typedef enum fulltext_statement { | |
1994 CONTENT_INSERT_STMT, | |
1995 CONTENT_SELECT_STMT, | |
1996 CONTENT_UPDATE_STMT, | |
1997 CONTENT_DELETE_STMT, | |
1998 CONTENT_EXISTS_STMT, | |
1999 | |
2000 BLOCK_INSERT_STMT, | |
2001 BLOCK_SELECT_STMT, | |
2002 BLOCK_DELETE_STMT, | |
2003 BLOCK_DELETE_ALL_STMT, | |
2004 | |
2005 SEGDIR_MAX_INDEX_STMT, | |
2006 SEGDIR_SET_STMT, | |
2007 SEGDIR_SELECT_LEVEL_STMT, | |
2008 SEGDIR_SPAN_STMT, | |
2009 SEGDIR_DELETE_STMT, | |
2010 SEGDIR_SELECT_SEGMENT_STMT, | |
2011 SEGDIR_SELECT_ALL_STMT, | |
2012 SEGDIR_DELETE_ALL_STMT, | |
2013 SEGDIR_COUNT_STMT, | |
2014 | |
2015 MAX_STMT /* Always at end! */ | |
2016 } fulltext_statement; | |
2017 | |
2018 /* These must exactly match the enum above. */ | |
2019 /* TODO(shess): Is there some risk that a statement will be used in two | |
2020 ** cursors at once, e.g. if a query joins a virtual table to itself? | |
2021 ** If so perhaps we should move some of these to the cursor object. | |
2022 */ | |
2023 static const char *const fulltext_zStatement[MAX_STMT] = { | |
2024 /* CONTENT_INSERT */ NULL, /* generated in contentInsertStatement() */ | |
2025 /* CONTENT_SELECT */ NULL, /* generated in contentSelectStatement() */ | |
2026 /* CONTENT_UPDATE */ NULL, /* generated in contentUpdateStatement() */ | |
2027 /* CONTENT_DELETE */ "delete from %_content where docid = ?", | |
2028 /* CONTENT_EXISTS */ "select docid from %_content limit 1", | |
2029 | |
2030 /* BLOCK_INSERT */ | |
2031 "insert into %_segments (blockid, block) values (null, ?)", | |
2032 /* BLOCK_SELECT */ "select block from %_segments where blockid = ?", | |
2033 /* BLOCK_DELETE */ "delete from %_segments where blockid between ? and ?", | |
2034 /* BLOCK_DELETE_ALL */ "delete from %_segments", | |
2035 | |
2036 /* SEGDIR_MAX_INDEX */ "select max(idx) from %_segdir where level = ?", | |
2037 /* SEGDIR_SET */ "insert into %_segdir values (?, ?, ?, ?, ?, ?)", | |
2038 /* SEGDIR_SELECT_LEVEL */ | |
2039 "select start_block, leaves_end_block, root, idx from %_segdir " | |
2040 " where level = ? order by idx", | |
2041 /* SEGDIR_SPAN */ | |
2042 "select min(start_block), max(end_block) from %_segdir " | |
2043 " where level = ? and start_block <> 0", | |
2044 /* SEGDIR_DELETE */ "delete from %_segdir where level = ?", | |
2045 | |
2046 /* NOTE(shess): The first three results of the following two | |
2047 ** statements must match. | |
2048 */ | |
2049 /* SEGDIR_SELECT_SEGMENT */ | |
2050 "select start_block, leaves_end_block, root from %_segdir " | |
2051 " where level = ? and idx = ?", | |
2052 /* SEGDIR_SELECT_ALL */ | |
2053 "select start_block, leaves_end_block, root from %_segdir " | |
2054 " order by level desc, idx asc", | |
2055 /* SEGDIR_DELETE_ALL */ "delete from %_segdir", | |
2056 /* SEGDIR_COUNT */ "select count(*), ifnull(max(level),0) from %_segdir", | |
2057 }; | |
2058 | |
2059 /* | |
2060 ** A connection to a fulltext index is an instance of the following | |
2061 ** structure. The xCreate and xConnect methods create an instance | |
2062 ** of this structure and xDestroy and xDisconnect free that instance. | |
2063 ** All other methods receive a pointer to the structure as one of their | |
2064 ** arguments. | |
2065 */ | |
2066 struct fulltext_vtab { | |
2067 sqlite3_vtab base; /* Base class used by SQLite core */ | |
2068 sqlite3 *db; /* The database connection */ | |
2069 const char *zDb; /* logical database name */ | |
2070 const char *zName; /* virtual table name */ | |
2071 int nColumn; /* number of columns in virtual table */ | |
2072 char **azColumn; /* column names. malloced */ | |
2073 char **azContentColumn; /* column names in content table; malloced */ | |
2074 sqlite3_tokenizer *pTokenizer; /* tokenizer for inserts and queries */ | |
2075 | |
2076 /* Precompiled statements which we keep as long as the table is | |
2077 ** open. | |
2078 */ | |
2079 sqlite3_stmt *pFulltextStatements[MAX_STMT]; | |
2080 | |
2081 /* Precompiled statements used for segment merges. We run a | |
2082 ** separate select across the leaf level of each tree being merged. | |
2083 */ | |
2084 sqlite3_stmt *pLeafSelectStmts[MERGE_COUNT]; | |
2085 /* The statement used to prepare pLeafSelectStmts. */ | |
2086 #define LEAF_SELECT \ | |
2087 "select block from %_segments where blockid between ? and ? order by blockid" | |
2088 | |
2089 /* These buffer pending index updates during transactions. | |
2090 ** nPendingData estimates the memory size of the pending data. It | |
2091 ** doesn't include the hash-bucket overhead, nor any malloc | |
2092 ** overhead. When nPendingData exceeds kPendingThreshold, the | |
2093 ** buffer is flushed even before the transaction closes. | |
2094 ** pendingTerms stores the data, and is only valid when nPendingData | |
2095 ** is >=0 (nPendingData<0 means pendingTerms has not been | |
2096 ** initialized). iPrevDocid is the last docid written, used to make | |
2097 ** certain we're inserting in sorted order. | |
2098 */ | |
2099 int nPendingData; | |
2100 #define kPendingThreshold (1*1024*1024) | |
2101 sqlite_int64 iPrevDocid; | |
2102 fts3Hash pendingTerms; | |
2103 }; | |
2104 | |
2105 /* | |
2106 ** When the core wants to do a query, it create a cursor using a | |
2107 ** call to xOpen. This structure is an instance of a cursor. It | |
2108 ** is destroyed by xClose. | |
2109 */ | |
2110 typedef struct fulltext_cursor { | |
2111 sqlite3_vtab_cursor base; /* Base class used by SQLite core */ | |
2112 QueryType iCursorType; /* Copy of sqlite3_index_info.idxNum */ | |
2113 sqlite3_stmt *pStmt; /* Prepared statement in use by the cursor */ | |
2114 int eof; /* True if at End Of Results */ | |
2115 Fts3Expr *pExpr; /* Parsed MATCH query string */ | |
2116 Snippet snippet; /* Cached snippet for the current row */ | |
2117 int iColumn; /* Column being searched */ | |
2118 DataBuffer result; /* Doclist results from fulltextQuery */ | |
2119 DLReader reader; /* Result reader if result not empty */ | |
2120 } fulltext_cursor; | |
2121 | |
2122 static fulltext_vtab *cursor_vtab(fulltext_cursor *c){ | |
2123 return (fulltext_vtab *) c->base.pVtab; | |
2124 } | |
2125 | |
2126 static const sqlite3_module fts3Module; /* forward declaration */ | |
2127 | |
2128 /* Return a dynamically generated statement of the form | |
2129 * insert into %_content (docid, ...) values (?, ...) | |
2130 */ | |
2131 static const char *contentInsertStatement(fulltext_vtab *v){ | |
2132 StringBuffer sb; | |
2133 int i; | |
2134 | |
2135 initStringBuffer(&sb); | |
2136 append(&sb, "insert into %_content (docid, "); | |
2137 appendList(&sb, v->nColumn, v->azContentColumn); | |
2138 append(&sb, ") values (?"); | |
2139 for(i=0; i<v->nColumn; ++i) | |
2140 append(&sb, ", ?"); | |
2141 append(&sb, ")"); | |
2142 return stringBufferData(&sb); | |
2143 } | |
2144 | |
2145 /* Return a dynamically generated statement of the form | |
2146 * select <content columns> from %_content where docid = ? | |
2147 */ | |
2148 static const char *contentSelectStatement(fulltext_vtab *v){ | |
2149 StringBuffer sb; | |
2150 initStringBuffer(&sb); | |
2151 append(&sb, "SELECT "); | |
2152 appendList(&sb, v->nColumn, v->azContentColumn); | |
2153 append(&sb, " FROM %_content WHERE docid = ?"); | |
2154 return stringBufferData(&sb); | |
2155 } | |
2156 | |
2157 /* Return a dynamically generated statement of the form | |
2158 * update %_content set [col_0] = ?, [col_1] = ?, ... | |
2159 * where docid = ? | |
2160 */ | |
2161 static const char *contentUpdateStatement(fulltext_vtab *v){ | |
2162 StringBuffer sb; | |
2163 int i; | |
2164 | |
2165 initStringBuffer(&sb); | |
2166 append(&sb, "update %_content set "); | |
2167 for(i=0; i<v->nColumn; ++i) { | |
2168 if( i>0 ){ | |
2169 append(&sb, ", "); | |
2170 } | |
2171 append(&sb, v->azContentColumn[i]); | |
2172 append(&sb, " = ?"); | |
2173 } | |
2174 append(&sb, " where docid = ?"); | |
2175 return stringBufferData(&sb); | |
2176 } | |
2177 | |
2178 /* Puts a freshly-prepared statement determined by iStmt in *ppStmt. | |
2179 ** If the indicated statement has never been prepared, it is prepared | |
2180 ** and cached, otherwise the cached version is reset. | |
2181 */ | |
2182 static int sql_get_statement(fulltext_vtab *v, fulltext_statement iStmt, | |
2183 sqlite3_stmt **ppStmt){ | |
2184 assert( iStmt<MAX_STMT ); | |
2185 if( v->pFulltextStatements[iStmt]==NULL ){ | |
2186 const char *zStmt; | |
2187 int rc; | |
2188 switch( iStmt ){ | |
2189 case CONTENT_INSERT_STMT: | |
2190 zStmt = contentInsertStatement(v); break; | |
2191 case CONTENT_SELECT_STMT: | |
2192 zStmt = contentSelectStatement(v); break; | |
2193 case CONTENT_UPDATE_STMT: | |
2194 zStmt = contentUpdateStatement(v); break; | |
2195 default: | |
2196 zStmt = fulltext_zStatement[iStmt]; | |
2197 } | |
2198 rc = sql_prepare(v->db, v->zDb, v->zName, &v->pFulltextStatements[iStmt], | |
2199 zStmt); | |
2200 if( zStmt != fulltext_zStatement[iStmt]) sqlite3_free((void *) zStmt); | |
2201 if( rc!=SQLITE_OK ) return rc; | |
2202 } else { | |
2203 int rc = sqlite3_reset(v->pFulltextStatements[iStmt]); | |
2204 if( rc!=SQLITE_OK ) return rc; | |
2205 } | |
2206 | |
2207 *ppStmt = v->pFulltextStatements[iStmt]; | |
2208 return SQLITE_OK; | |
2209 } | |
2210 | |
2211 /* Like sqlite3_step(), but convert SQLITE_DONE to SQLITE_OK and | |
2212 ** SQLITE_ROW to SQLITE_ERROR. Useful for statements like UPDATE, | |
2213 ** where we expect no results. | |
2214 */ | |
2215 static int sql_single_step(sqlite3_stmt *s){ | |
2216 int rc = sqlite3_step(s); | |
2217 return (rc==SQLITE_DONE) ? SQLITE_OK : rc; | |
2218 } | |
2219 | |
2220 /* Like sql_get_statement(), but for special replicated LEAF_SELECT | |
2221 ** statements. idx -1 is a special case for an uncached version of | |
2222 ** the statement (used in the optimize implementation). | |
2223 */ | |
2224 /* TODO(shess) Write version for generic statements and then share | |
2225 ** that between the cached-statement functions. | |
2226 */ | |
2227 static int sql_get_leaf_statement(fulltext_vtab *v, int idx, | |
2228 sqlite3_stmt **ppStmt){ | |
2229 assert( idx>=-1 && idx<MERGE_COUNT ); | |
2230 if( idx==-1 ){ | |
2231 return sql_prepare(v->db, v->zDb, v->zName, ppStmt, LEAF_SELECT); | |
2232 }else if( v->pLeafSelectStmts[idx]==NULL ){ | |
2233 int rc = sql_prepare(v->db, v->zDb, v->zName, &v->pLeafSelectStmts[idx], | |
2234 LEAF_SELECT); | |
2235 if( rc!=SQLITE_OK ) return rc; | |
2236 }else{ | |
2237 int rc = sqlite3_reset(v->pLeafSelectStmts[idx]); | |
2238 if( rc!=SQLITE_OK ) return rc; | |
2239 } | |
2240 | |
2241 *ppStmt = v->pLeafSelectStmts[idx]; | |
2242 return SQLITE_OK; | |
2243 } | |
2244 | |
2245 /* insert into %_content (docid, ...) values ([docid], [pValues]) | |
2246 ** If the docid contains SQL NULL, then a unique docid will be | |
2247 ** generated. | |
2248 */ | |
2249 static int content_insert(fulltext_vtab *v, sqlite3_value *docid, | |
2250 sqlite3_value **pValues){ | |
2251 sqlite3_stmt *s; | |
2252 int i; | |
2253 int rc = sql_get_statement(v, CONTENT_INSERT_STMT, &s); | |
2254 if( rc!=SQLITE_OK ) return rc; | |
2255 | |
2256 rc = sqlite3_bind_value(s, 1, docid); | |
2257 if( rc!=SQLITE_OK ) return rc; | |
2258 | |
2259 for(i=0; i<v->nColumn; ++i){ | |
2260 rc = sqlite3_bind_value(s, 2+i, pValues[i]); | |
2261 if( rc!=SQLITE_OK ) return rc; | |
2262 } | |
2263 | |
2264 return sql_single_step(s); | |
2265 } | |
2266 | |
2267 /* update %_content set col0 = pValues[0], col1 = pValues[1], ... | |
2268 * where docid = [iDocid] */ | |
2269 static int content_update(fulltext_vtab *v, sqlite3_value **pValues, | |
2270 sqlite_int64 iDocid){ | |
2271 sqlite3_stmt *s; | |
2272 int i; | |
2273 int rc = sql_get_statement(v, CONTENT_UPDATE_STMT, &s); | |
2274 if( rc!=SQLITE_OK ) return rc; | |
2275 | |
2276 for(i=0; i<v->nColumn; ++i){ | |
2277 rc = sqlite3_bind_value(s, 1+i, pValues[i]); | |
2278 if( rc!=SQLITE_OK ) return rc; | |
2279 } | |
2280 | |
2281 rc = sqlite3_bind_int64(s, 1+v->nColumn, iDocid); | |
2282 if( rc!=SQLITE_OK ) return rc; | |
2283 | |
2284 return sql_single_step(s); | |
2285 } | |
2286 | |
2287 static void freeStringArray(int nString, const char **pString){ | |
2288 int i; | |
2289 | |
2290 for (i=0 ; i < nString ; ++i) { | |
2291 if( pString[i]!=NULL ) sqlite3_free((void *) pString[i]); | |
2292 } | |
2293 sqlite3_free((void *) pString); | |
2294 } | |
2295 | |
2296 /* select * from %_content where docid = [iDocid] | |
2297 * The caller must delete the returned array and all strings in it. | |
2298 * null fields will be NULL in the returned array. | |
2299 * | |
2300 * TODO: Perhaps we should return pointer/length strings here for consistency | |
2301 * with other code which uses pointer/length. */ | |
2302 static int content_select(fulltext_vtab *v, sqlite_int64 iDocid, | |
2303 const char ***pValues){ | |
2304 sqlite3_stmt *s; | |
2305 const char **values; | |
2306 int i; | |
2307 int rc; | |
2308 | |
2309 *pValues = NULL; | |
2310 | |
2311 rc = sql_get_statement(v, CONTENT_SELECT_STMT, &s); | |
2312 if( rc!=SQLITE_OK ) return rc; | |
2313 | |
2314 rc = sqlite3_bind_int64(s, 1, iDocid); | |
2315 if( rc!=SQLITE_OK ) return rc; | |
2316 | |
2317 rc = sqlite3_step(s); | |
2318 if( rc!=SQLITE_ROW ) return rc; | |
2319 | |
2320 values = (const char **) sqlite3_malloc(v->nColumn * sizeof(const char *)); | |
2321 for(i=0; i<v->nColumn; ++i){ | |
2322 if( sqlite3_column_type(s, i)==SQLITE_NULL ){ | |
2323 values[i] = NULL; | |
2324 }else{ | |
2325 values[i] = string_dup((char*)sqlite3_column_text(s, i)); | |
2326 } | |
2327 } | |
2328 | |
2329 /* We expect only one row. We must execute another sqlite3_step() | |
2330 * to complete the iteration; otherwise the table will remain locked. */ | |
2331 rc = sqlite3_step(s); | |
2332 if( rc==SQLITE_DONE ){ | |
2333 *pValues = values; | |
2334 return SQLITE_OK; | |
2335 } | |
2336 | |
2337 freeStringArray(v->nColumn, values); | |
2338 return rc; | |
2339 } | |
2340 | |
2341 /* delete from %_content where docid = [iDocid ] */ | |
2342 static int content_delete(fulltext_vtab *v, sqlite_int64 iDocid){ | |
2343 sqlite3_stmt *s; | |
2344 int rc = sql_get_statement(v, CONTENT_DELETE_STMT, &s); | |
2345 if( rc!=SQLITE_OK ) return rc; | |
2346 | |
2347 rc = sqlite3_bind_int64(s, 1, iDocid); | |
2348 if( rc!=SQLITE_OK ) return rc; | |
2349 | |
2350 return sql_single_step(s); | |
2351 } | |
2352 | |
2353 /* Returns SQLITE_ROW if any rows exist in %_content, SQLITE_DONE if | |
2354 ** no rows exist, and any error in case of failure. | |
2355 */ | |
2356 static int content_exists(fulltext_vtab *v){ | |
2357 sqlite3_stmt *s; | |
2358 int rc = sql_get_statement(v, CONTENT_EXISTS_STMT, &s); | |
2359 if( rc!=SQLITE_OK ) return rc; | |
2360 | |
2361 rc = sqlite3_step(s); | |
2362 if( rc!=SQLITE_ROW ) return rc; | |
2363 | |
2364 /* We expect only one row. We must execute another sqlite3_step() | |
2365 * to complete the iteration; otherwise the table will remain locked. */ | |
2366 rc = sqlite3_step(s); | |
2367 if( rc==SQLITE_DONE ) return SQLITE_ROW; | |
2368 if( rc==SQLITE_ROW ) return SQLITE_ERROR; | |
2369 return rc; | |
2370 } | |
2371 | |
2372 /* insert into %_segments values ([pData]) | |
2373 ** returns assigned blockid in *piBlockid | |
2374 */ | |
2375 static int block_insert(fulltext_vtab *v, const char *pData, int nData, | |
2376 sqlite_int64 *piBlockid){ | |
2377 sqlite3_stmt *s; | |
2378 int rc = sql_get_statement(v, BLOCK_INSERT_STMT, &s); | |
2379 if( rc!=SQLITE_OK ) return rc; | |
2380 | |
2381 rc = sqlite3_bind_blob(s, 1, pData, nData, SQLITE_STATIC); | |
2382 if( rc!=SQLITE_OK ) return rc; | |
2383 | |
2384 rc = sqlite3_step(s); | |
2385 if( rc==SQLITE_ROW ) return SQLITE_ERROR; | |
2386 if( rc!=SQLITE_DONE ) return rc; | |
2387 | |
2388 /* blockid column is an alias for rowid. */ | |
2389 *piBlockid = sqlite3_last_insert_rowid(v->db); | |
2390 return SQLITE_OK; | |
2391 } | |
2392 | |
2393 /* delete from %_segments | |
2394 ** where blockid between [iStartBlockid] and [iEndBlockid] | |
2395 ** | |
2396 ** Deletes the range of blocks, inclusive, used to delete the blocks | |
2397 ** which form a segment. | |
2398 */ | |
2399 static int block_delete(fulltext_vtab *v, | |
2400 sqlite_int64 iStartBlockid, sqlite_int64 iEndBlockid){ | |
2401 sqlite3_stmt *s; | |
2402 int rc = sql_get_statement(v, BLOCK_DELETE_STMT, &s); | |
2403 if( rc!=SQLITE_OK ) return rc; | |
2404 | |
2405 rc = sqlite3_bind_int64(s, 1, iStartBlockid); | |
2406 if( rc!=SQLITE_OK ) return rc; | |
2407 | |
2408 rc = sqlite3_bind_int64(s, 2, iEndBlockid); | |
2409 if( rc!=SQLITE_OK ) return rc; | |
2410 | |
2411 return sql_single_step(s); | |
2412 } | |
2413 | |
2414 /* Returns SQLITE_ROW with *pidx set to the maximum segment idx found | |
2415 ** at iLevel. Returns SQLITE_DONE if there are no segments at | |
2416 ** iLevel. Otherwise returns an error. | |
2417 */ | |
2418 static int segdir_max_index(fulltext_vtab *v, int iLevel, int *pidx){ | |
2419 sqlite3_stmt *s; | |
2420 int rc = sql_get_statement(v, SEGDIR_MAX_INDEX_STMT, &s); | |
2421 if( rc!=SQLITE_OK ) return rc; | |
2422 | |
2423 rc = sqlite3_bind_int(s, 1, iLevel); | |
2424 if( rc!=SQLITE_OK ) return rc; | |
2425 | |
2426 rc = sqlite3_step(s); | |
2427 /* Should always get at least one row due to how max() works. */ | |
2428 if( rc==SQLITE_DONE ) return SQLITE_DONE; | |
2429 if( rc!=SQLITE_ROW ) return rc; | |
2430 | |
2431 /* NULL means that there were no inputs to max(). */ | |
2432 if( SQLITE_NULL==sqlite3_column_type(s, 0) ){ | |
2433 rc = sqlite3_step(s); | |
2434 if( rc==SQLITE_ROW ) return SQLITE_ERROR; | |
2435 return rc; | |
2436 } | |
2437 | |
2438 *pidx = sqlite3_column_int(s, 0); | |
2439 | |
2440 /* We expect only one row. We must execute another sqlite3_step() | |
2441 * to complete the iteration; otherwise the table will remain locked. */ | |
2442 rc = sqlite3_step(s); | |
2443 if( rc==SQLITE_ROW ) return SQLITE_ERROR; | |
2444 if( rc!=SQLITE_DONE ) return rc; | |
2445 return SQLITE_ROW; | |
2446 } | |
2447 | |
2448 /* insert into %_segdir values ( | |
2449 ** [iLevel], [idx], | |
2450 ** [iStartBlockid], [iLeavesEndBlockid], [iEndBlockid], | |
2451 ** [pRootData] | |
2452 ** ) | |
2453 */ | |
2454 static int segdir_set(fulltext_vtab *v, int iLevel, int idx, | |
2455 sqlite_int64 iStartBlockid, | |
2456 sqlite_int64 iLeavesEndBlockid, | |
2457 sqlite_int64 iEndBlockid, | |
2458 const char *pRootData, int nRootData){ | |
2459 sqlite3_stmt *s; | |
2460 int rc = sql_get_statement(v, SEGDIR_SET_STMT, &s); | |
2461 if( rc!=SQLITE_OK ) return rc; | |
2462 | |
2463 rc = sqlite3_bind_int(s, 1, iLevel); | |
2464 if( rc!=SQLITE_OK ) return rc; | |
2465 | |
2466 rc = sqlite3_bind_int(s, 2, idx); | |
2467 if( rc!=SQLITE_OK ) return rc; | |
2468 | |
2469 rc = sqlite3_bind_int64(s, 3, iStartBlockid); | |
2470 if( rc!=SQLITE_OK ) return rc; | |
2471 | |
2472 rc = sqlite3_bind_int64(s, 4, iLeavesEndBlockid); | |
2473 if( rc!=SQLITE_OK ) return rc; | |
2474 | |
2475 rc = sqlite3_bind_int64(s, 5, iEndBlockid); | |
2476 if( rc!=SQLITE_OK ) return rc; | |
2477 | |
2478 rc = sqlite3_bind_blob(s, 6, pRootData, nRootData, SQLITE_STATIC); | |
2479 if( rc!=SQLITE_OK ) return rc; | |
2480 | |
2481 return sql_single_step(s); | |
2482 } | |
2483 | |
2484 /* Queries %_segdir for the block span of the segments in level | |
2485 ** iLevel. Returns SQLITE_DONE if there are no blocks for iLevel, | |
2486 ** SQLITE_ROW if there are blocks, else an error. | |
2487 */ | |
2488 static int segdir_span(fulltext_vtab *v, int iLevel, | |
2489 sqlite_int64 *piStartBlockid, | |
2490 sqlite_int64 *piEndBlockid){ | |
2491 sqlite3_stmt *s; | |
2492 int rc = sql_get_statement(v, SEGDIR_SPAN_STMT, &s); | |
2493 if( rc!=SQLITE_OK ) return rc; | |
2494 | |
2495 rc = sqlite3_bind_int(s, 1, iLevel); | |
2496 if( rc!=SQLITE_OK ) return rc; | |
2497 | |
2498 rc = sqlite3_step(s); | |
2499 if( rc==SQLITE_DONE ) return SQLITE_DONE; /* Should never happen */ | |
2500 if( rc!=SQLITE_ROW ) return rc; | |
2501 | |
2502 /* This happens if all segments at this level are entirely inline. */ | |
2503 if( SQLITE_NULL==sqlite3_column_type(s, 0) ){ | |
2504 /* We expect only one row. We must execute another sqlite3_step() | |
2505 * to complete the iteration; otherwise the table will remain locked. */ | |
2506 int rc2 = sqlite3_step(s); | |
2507 if( rc2==SQLITE_ROW ) return SQLITE_ERROR; | |
2508 return rc2; | |
2509 } | |
2510 | |
2511 *piStartBlockid = sqlite3_column_int64(s, 0); | |
2512 *piEndBlockid = sqlite3_column_int64(s, 1); | |
2513 | |
2514 /* We expect only one row. We must execute another sqlite3_step() | |
2515 * to complete the iteration; otherwise the table will remain locked. */ | |
2516 rc = sqlite3_step(s); | |
2517 if( rc==SQLITE_ROW ) return SQLITE_ERROR; | |
2518 if( rc!=SQLITE_DONE ) return rc; | |
2519 return SQLITE_ROW; | |
2520 } | |
2521 | |
2522 /* Delete the segment blocks and segment directory records for all | |
2523 ** segments at iLevel. | |
2524 */ | |
2525 static int segdir_delete(fulltext_vtab *v, int iLevel){ | |
2526 sqlite3_stmt *s; | |
2527 sqlite_int64 iStartBlockid, iEndBlockid; | |
2528 int rc = segdir_span(v, iLevel, &iStartBlockid, &iEndBlockid); | |
2529 if( rc!=SQLITE_ROW && rc!=SQLITE_DONE ) return rc; | |
2530 | |
2531 if( rc==SQLITE_ROW ){ | |
2532 rc = block_delete(v, iStartBlockid, iEndBlockid); | |
2533 if( rc!=SQLITE_OK ) return rc; | |
2534 } | |
2535 | |
2536 /* Delete the segment directory itself. */ | |
2537 rc = sql_get_statement(v, SEGDIR_DELETE_STMT, &s); | |
2538 if( rc!=SQLITE_OK ) return rc; | |
2539 | |
2540 rc = sqlite3_bind_int64(s, 1, iLevel); | |
2541 if( rc!=SQLITE_OK ) return rc; | |
2542 | |
2543 return sql_single_step(s); | |
2544 } | |
2545 | |
2546 /* Delete entire fts index, SQLITE_OK on success, relevant error on | |
2547 ** failure. | |
2548 */ | |
2549 static int segdir_delete_all(fulltext_vtab *v){ | |
2550 sqlite3_stmt *s; | |
2551 int rc = sql_get_statement(v, SEGDIR_DELETE_ALL_STMT, &s); | |
2552 if( rc!=SQLITE_OK ) return rc; | |
2553 | |
2554 rc = sql_single_step(s); | |
2555 if( rc!=SQLITE_OK ) return rc; | |
2556 | |
2557 rc = sql_get_statement(v, BLOCK_DELETE_ALL_STMT, &s); | |
2558 if( rc!=SQLITE_OK ) return rc; | |
2559 | |
2560 return sql_single_step(s); | |
2561 } | |
2562 | |
2563 /* Returns SQLITE_OK with *pnSegments set to the number of entries in | |
2564 ** %_segdir and *piMaxLevel set to the highest level which has a | |
2565 ** segment. Otherwise returns the SQLite error which caused failure. | |
2566 */ | |
2567 static int segdir_count(fulltext_vtab *v, int *pnSegments, int *piMaxLevel){ | |
2568 sqlite3_stmt *s; | |
2569 int rc = sql_get_statement(v, SEGDIR_COUNT_STMT, &s); | |
2570 if( rc!=SQLITE_OK ) return rc; | |
2571 | |
2572 rc = sqlite3_step(s); | |
2573 /* TODO(shess): This case should not be possible? Should stronger | |
2574 ** measures be taken if it happens? | |
2575 */ | |
2576 if( rc==SQLITE_DONE ){ | |
2577 *pnSegments = 0; | |
2578 *piMaxLevel = 0; | |
2579 return SQLITE_OK; | |
2580 } | |
2581 if( rc!=SQLITE_ROW ) return rc; | |
2582 | |
2583 *pnSegments = sqlite3_column_int(s, 0); | |
2584 *piMaxLevel = sqlite3_column_int(s, 1); | |
2585 | |
2586 /* We expect only one row. We must execute another sqlite3_step() | |
2587 * to complete the iteration; otherwise the table will remain locked. */ | |
2588 rc = sqlite3_step(s); | |
2589 if( rc==SQLITE_DONE ) return SQLITE_OK; | |
2590 if( rc==SQLITE_ROW ) return SQLITE_ERROR; | |
2591 return rc; | |
2592 } | |
2593 | |
2594 /* TODO(shess) clearPendingTerms() is far down the file because | |
2595 ** writeZeroSegment() is far down the file because LeafWriter is far | |
2596 ** down the file. Consider refactoring the code to move the non-vtab | |
2597 ** code above the vtab code so that we don't need this forward | |
2598 ** reference. | |
2599 */ | |
2600 static int clearPendingTerms(fulltext_vtab *v); | |
2601 | |
2602 /* | |
2603 ** Free the memory used to contain a fulltext_vtab structure. | |
2604 */ | |
2605 static void fulltext_vtab_destroy(fulltext_vtab *v){ | |
2606 int iStmt, i; | |
2607 | |
2608 FTSTRACE(("FTS3 Destroy %p\n", v)); | |
2609 for( iStmt=0; iStmt<MAX_STMT; iStmt++ ){ | |
2610 if( v->pFulltextStatements[iStmt]!=NULL ){ | |
2611 sqlite3_finalize(v->pFulltextStatements[iStmt]); | |
2612 v->pFulltextStatements[iStmt] = NULL; | |
2613 } | |
2614 } | |
2615 | |
2616 for( i=0; i<MERGE_COUNT; i++ ){ | |
2617 if( v->pLeafSelectStmts[i]!=NULL ){ | |
2618 sqlite3_finalize(v->pLeafSelectStmts[i]); | |
2619 v->pLeafSelectStmts[i] = NULL; | |
2620 } | |
2621 } | |
2622 | |
2623 if( v->pTokenizer!=NULL ){ | |
2624 v->pTokenizer->pModule->xDestroy(v->pTokenizer); | |
2625 v->pTokenizer = NULL; | |
2626 } | |
2627 | |
2628 clearPendingTerms(v); | |
2629 | |
2630 sqlite3_free(v->azColumn); | |
2631 for(i = 0; i < v->nColumn; ++i) { | |
2632 sqlite3_free(v->azContentColumn[i]); | |
2633 } | |
2634 sqlite3_free(v->azContentColumn); | |
2635 sqlite3_free(v); | |
2636 } | |
2637 | |
2638 /* | |
2639 ** Token types for parsing the arguments to xConnect or xCreate. | |
2640 */ | |
2641 #define TOKEN_EOF 0 /* End of file */ | |
2642 #define TOKEN_SPACE 1 /* Any kind of whitespace */ | |
2643 #define TOKEN_ID 2 /* An identifier */ | |
2644 #define TOKEN_STRING 3 /* A string literal */ | |
2645 #define TOKEN_PUNCT 4 /* A single punctuation character */ | |
2646 | |
2647 /* | |
2648 ** If X is a character that can be used in an identifier then | |
2649 ** ftsIdChar(X) will be true. Otherwise it is false. | |
2650 ** | |
2651 ** For ASCII, any character with the high-order bit set is | |
2652 ** allowed in an identifier. For 7-bit characters, | |
2653 ** isFtsIdChar[X] must be 1. | |
2654 ** | |
2655 ** Ticket #1066. the SQL standard does not allow '$' in the | |
2656 ** middle of identfiers. But many SQL implementations do. | |
2657 ** SQLite will allow '$' in identifiers for compatibility. | |
2658 ** But the feature is undocumented. | |
2659 */ | |
2660 static const char isFtsIdChar[] = { | |
2661 /* x0 x1 x2 x3 x4 x5 x6 x7 x8 x9 xA xB xC xD xE xF */ | |
2662 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, /* 2x */ | |
2663 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, /* 3x */ | |
2664 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, /* 4x */ | |
2665 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 1, /* 5x */ | |
2666 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, /* 6x */ | |
2667 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, /* 7x */ | |
2668 }; | |
2669 #define ftsIdChar(C) (((c=C)&0x80)!=0 || (c>0x1f && isFtsIdChar[c-0x20])) | |
2670 | |
2671 | |
2672 /* | |
2673 ** Return the length of the token that begins at z[0]. | |
2674 ** Store the token type in *tokenType before returning. | |
2675 */ | |
2676 static int ftsGetToken(const char *z, int *tokenType){ | |
2677 int i, c; | |
2678 switch( *z ){ | |
2679 case 0: { | |
2680 *tokenType = TOKEN_EOF; | |
2681 return 0; | |
2682 } | |
2683 case ' ': case '\t': case '\n': case '\f': case '\r': { | |
2684 for(i=1; safe_isspace(z[i]); i++){} | |
2685 *tokenType = TOKEN_SPACE; | |
2686 return i; | |
2687 } | |
2688 case '`': | |
2689 case '\'': | |
2690 case '"': { | |
2691 int delim = z[0]; | |
2692 for(i=1; (c=z[i])!=0; i++){ | |
2693 if( c==delim ){ | |
2694 if( z[i+1]==delim ){ | |
2695 i++; | |
2696 }else{ | |
2697 break; | |
2698 } | |
2699 } | |
2700 } | |
2701 *tokenType = TOKEN_STRING; | |
2702 return i + (c!=0); | |
2703 } | |
2704 case '[': { | |
2705 for(i=1, c=z[0]; c!=']' && (c=z[i])!=0; i++){} | |
2706 *tokenType = TOKEN_ID; | |
2707 return i; | |
2708 } | |
2709 default: { | |
2710 if( !ftsIdChar(*z) ){ | |
2711 break; | |
2712 } | |
2713 for(i=1; ftsIdChar(z[i]); i++){} | |
2714 *tokenType = TOKEN_ID; | |
2715 return i; | |
2716 } | |
2717 } | |
2718 *tokenType = TOKEN_PUNCT; | |
2719 return 1; | |
2720 } | |
2721 | |
2722 /* | |
2723 ** A token extracted from a string is an instance of the following | |
2724 ** structure. | |
2725 */ | |
2726 typedef struct FtsToken { | |
2727 const char *z; /* Pointer to token text. Not '\000' terminated */ | |
2728 short int n; /* Length of the token text in bytes. */ | |
2729 } FtsToken; | |
2730 | |
2731 /* | |
2732 ** Given a input string (which is really one of the argv[] parameters | |
2733 ** passed into xConnect or xCreate) split the string up into tokens. | |
2734 ** Return an array of pointers to '\000' terminated strings, one string | |
2735 ** for each non-whitespace token. | |
2736 ** | |
2737 ** The returned array is terminated by a single NULL pointer. | |
2738 ** | |
2739 ** Space to hold the returned array is obtained from a single | |
2740 ** malloc and should be freed by passing the return value to free(). | |
2741 ** The individual strings within the token list are all a part of | |
2742 ** the single memory allocation and will all be freed at once. | |
2743 */ | |
2744 static char **tokenizeString(const char *z, int *pnToken){ | |
2745 int nToken = 0; | |
2746 FtsToken *aToken = sqlite3_malloc( strlen(z) * sizeof(aToken[0]) ); | |
2747 int n = 1; | |
2748 int e, i; | |
2749 int totalSize = 0; | |
2750 char **azToken; | |
2751 char *zCopy; | |
2752 while( n>0 ){ | |
2753 n = ftsGetToken(z, &e); | |
2754 if( e!=TOKEN_SPACE ){ | |
2755 aToken[nToken].z = z; | |
2756 aToken[nToken].n = n; | |
2757 nToken++; | |
2758 totalSize += n+1; | |
2759 } | |
2760 z += n; | |
2761 } | |
2762 azToken = (char**)sqlite3_malloc( nToken*sizeof(char*) + totalSize ); | |
2763 zCopy = (char*)&azToken[nToken]; | |
2764 nToken--; | |
2765 for(i=0; i<nToken; i++){ | |
2766 azToken[i] = zCopy; | |
2767 n = aToken[i].n; | |
2768 memcpy(zCopy, aToken[i].z, n); | |
2769 zCopy[n] = 0; | |
2770 zCopy += n+1; | |
2771 } | |
2772 azToken[nToken] = 0; | |
2773 sqlite3_free(aToken); | |
2774 *pnToken = nToken; | |
2775 return azToken; | |
2776 } | 372 } |
2777 | 373 |
2778 /* | 374 /* |
2779 ** Convert an SQL-style quoted string into a normal string by removing | 375 ** Convert an SQL-style quoted string into a normal string by removing |
2780 ** the quote characters. The conversion is done in-place. If the | 376 ** the quote characters. The conversion is done in-place. If the |
2781 ** input does not begin with a quote character, then this routine | 377 ** input does not begin with a quote character, then this routine |
2782 ** is a no-op. | 378 ** is a no-op. |
2783 ** | 379 ** |
2784 ** Examples: | 380 ** Examples: |
2785 ** | 381 ** |
2786 ** "abc" becomes abc | 382 ** "abc" becomes abc |
2787 ** 'xyz' becomes xyz | 383 ** 'xyz' becomes xyz |
2788 ** [pqr] becomes pqr | 384 ** [pqr] becomes pqr |
2789 ** `mno` becomes mno | 385 ** `mno` becomes mno |
2790 */ | 386 ** |
2791 static void dequoteString(char *z){ | 387 */ |
2792 int quote; | 388 void sqlite3Fts3Dequote(char *z){ |
2793 int i, j; | 389 char quote; /* Quote character (if any ) */ |
2794 if( z==0 ) return; | 390 |
2795 quote = z[0]; | 391 quote = z[0]; |
2796 switch( quote ){ | 392 if( quote=='[' || quote=='\'' || quote=='"' || quote=='`' ){ |
2797 case '\'': break; | 393 int iIn = 1; /* Index of next byte to read from input */ |
2798 case '"': break; | 394 int iOut = 0; /* Index of next byte to write to output */ |
2799 case '`': break; /* For MySQL compatibility */ | 395 |
2800 case '[': quote = ']'; break; /* For MS SqlServer compatibility */ | 396 /* If the first byte was a '[', then the close-quote character is a ']' */ |
2801 default: return; | 397 if( quote=='[' ) quote = ']'; |
2802 } | 398 |
2803 for(i=1, j=0; z[i]; i++){ | 399 while( ALWAYS(z[iIn]) ){ |
2804 if( z[i]==quote ){ | 400 if( z[iIn]==quote ){ |
2805 if( z[i+1]==quote ){ | 401 if( z[iIn+1]!=quote ) break; |
2806 z[j++] = quote; | 402 z[iOut++] = quote; |
2807 i++; | 403 iIn += 2; |
2808 }else{ | 404 }else{ |
2809 z[j++] = 0; | 405 z[iOut++] = z[iIn++]; |
| 406 } |
| 407 } |
| 408 z[iOut] = '\0'; |
| 409 } |
| 410 } |
| 411 |
| 412 /* |
| 413 ** Read a single varint from the doclist at *pp and advance *pp to point |
| 414 ** to the first byte past the end of the varint. Add the value of the varint |
| 415 ** to *pVal. |
| 416 */ |
| 417 static void fts3GetDeltaVarint(char **pp, sqlite3_int64 *pVal){ |
| 418 sqlite3_int64 iVal; |
| 419 *pp += sqlite3Fts3GetVarint(*pp, &iVal); |
| 420 *pVal += iVal; |
| 421 } |
| 422 |
| 423 /* |
| 424 ** As long as *pp has not reached its end (pEnd), then do the same |
| 425 ** as fts3GetDeltaVarint(): read a single varint and add it to *pVal. |
| 426 ** But if we have reached the end of the varint, just set *pp=0 and |
| 427 ** leave *pVal unchanged. |
| 428 */ |
| 429 static void fts3GetDeltaVarint2(char **pp, char *pEnd, sqlite3_int64 *pVal){ |
| 430 if( *pp>=pEnd ){ |
| 431 *pp = 0; |
| 432 }else{ |
| 433 fts3GetDeltaVarint(pp, pVal); |
| 434 } |
| 435 } |
| 436 |
| 437 /* |
| 438 ** The xDisconnect() virtual table method. |
| 439 */ |
| 440 static int fts3DisconnectMethod(sqlite3_vtab *pVtab){ |
| 441 Fts3Table *p = (Fts3Table *)pVtab; |
| 442 int i; |
| 443 |
| 444 assert( p->nPendingData==0 ); |
| 445 assert( p->pSegments==0 ); |
| 446 |
| 447 /* Free any prepared statements held */ |
| 448 for(i=0; i<SizeofArray(p->aStmt); i++){ |
| 449 sqlite3_finalize(p->aStmt[i]); |
| 450 } |
| 451 sqlite3_free(p->zSegmentsTbl); |
| 452 sqlite3_free(p->zReadExprlist); |
| 453 sqlite3_free(p->zWriteExprlist); |
| 454 |
| 455 /* Invoke the tokenizer destructor to free the tokenizer. */ |
| 456 p->pTokenizer->pModule->xDestroy(p->pTokenizer); |
| 457 |
| 458 sqlite3_free(p); |
| 459 return SQLITE_OK; |
| 460 } |
| 461 |
| 462 /* |
| 463 ** Construct one or more SQL statements from the format string given |
| 464 ** and then evaluate those statements. The success code is written |
| 465 ** into *pRc. |
| 466 ** |
| 467 ** If *pRc is initially non-zero then this routine is a no-op. |
| 468 */ |
| 469 static void fts3DbExec( |
| 470 int *pRc, /* Success code */ |
| 471 sqlite3 *db, /* Database in which to run SQL */ |
| 472 const char *zFormat, /* Format string for SQL */ |
| 473 ... /* Arguments to the format string */ |
| 474 ){ |
| 475 va_list ap; |
| 476 char *zSql; |
| 477 if( *pRc ) return; |
| 478 va_start(ap, zFormat); |
| 479 zSql = sqlite3_vmprintf(zFormat, ap); |
| 480 va_end(ap); |
| 481 if( zSql==0 ){ |
| 482 *pRc = SQLITE_NOMEM; |
| 483 }else{ |
| 484 *pRc = sqlite3_exec(db, zSql, 0, 0, 0); |
| 485 sqlite3_free(zSql); |
| 486 } |
| 487 } |
| 488 |
| 489 /* |
| 490 ** The xDestroy() virtual table method. |
| 491 */ |
| 492 static int fts3DestroyMethod(sqlite3_vtab *pVtab){ |
| 493 int rc = SQLITE_OK; /* Return code */ |
| 494 Fts3Table *p = (Fts3Table *)pVtab; |
| 495 sqlite3 *db = p->db; |
| 496 |
| 497 /* Drop the shadow tables */ |
| 498 fts3DbExec(&rc, db, "DROP TABLE IF EXISTS %Q.'%q_content'", p->zDb, p->zName); |
| 499 fts3DbExec(&rc, db, "DROP TABLE IF EXISTS %Q.'%q_segments'", p->zDb,p->zName); |
| 500 fts3DbExec(&rc, db, "DROP TABLE IF EXISTS %Q.'%q_segdir'", p->zDb, p->zName); |
| 501 fts3DbExec(&rc, db, "DROP TABLE IF EXISTS %Q.'%q_docsize'", p->zDb, p->zName); |
| 502 fts3DbExec(&rc, db, "DROP TABLE IF EXISTS %Q.'%q_stat'", p->zDb, p->zName); |
| 503 |
| 504 /* If everything has worked, invoke fts3DisconnectMethod() to free the |
| 505 ** memory associated with the Fts3Table structure and return SQLITE_OK. |
| 506 ** Otherwise, return an SQLite error code. |
| 507 */ |
| 508 return (rc==SQLITE_OK ? fts3DisconnectMethod(pVtab) : rc); |
| 509 } |
| 510 |
| 511 |
| 512 /* |
| 513 ** Invoke sqlite3_declare_vtab() to declare the schema for the FTS3 table |
| 514 ** passed as the first argument. This is done as part of the xConnect() |
| 515 ** and xCreate() methods. |
| 516 ** |
| 517 ** If *pRc is non-zero when this function is called, it is a no-op. |
| 518 ** Otherwise, if an error occurs, an SQLite error code is stored in *pRc |
| 519 ** before returning. |
| 520 */ |
| 521 static void fts3DeclareVtab(int *pRc, Fts3Table *p){ |
| 522 if( *pRc==SQLITE_OK ){ |
| 523 int i; /* Iterator variable */ |
| 524 int rc; /* Return code */ |
| 525 char *zSql; /* SQL statement passed to declare_vtab() */ |
| 526 char *zCols; /* List of user defined columns */ |
| 527 |
| 528 /* Create a list of user columns for the virtual table */ |
| 529 zCols = sqlite3_mprintf("%Q, ", p->azColumn[0]); |
| 530 for(i=1; zCols && i<p->nColumn; i++){ |
| 531 zCols = sqlite3_mprintf("%z%Q, ", zCols, p->azColumn[i]); |
| 532 } |
| 533 |
| 534 /* Create the whole "CREATE TABLE" statement to pass to SQLite */ |
| 535 zSql = sqlite3_mprintf( |
| 536 "CREATE TABLE x(%s %Q HIDDEN, docid HIDDEN)", zCols, p->zName |
| 537 ); |
| 538 if( !zCols || !zSql ){ |
| 539 rc = SQLITE_NOMEM; |
| 540 }else{ |
| 541 rc = sqlite3_declare_vtab(p->db, zSql); |
| 542 } |
| 543 |
| 544 sqlite3_free(zSql); |
| 545 sqlite3_free(zCols); |
| 546 *pRc = rc; |
| 547 } |
| 548 } |
| 549 |
| 550 /* |
| 551 ** Create the backing store tables (%_content, %_segments and %_segdir) |
| 552 ** required by the FTS3 table passed as the only argument. This is done |
| 553 ** as part of the vtab xCreate() method. |
| 554 ** |
| 555 ** If the p->bHasDocsize boolean is true (indicating that this is an |
| 556 ** FTS4 table, not an FTS3 table) then also create the %_docsize and |
| 557 ** %_stat tables required by FTS4. |
| 558 */ |
| 559 static int fts3CreateTables(Fts3Table *p){ |
| 560 int rc = SQLITE_OK; /* Return code */ |
| 561 int i; /* Iterator variable */ |
| 562 char *zContentCols; /* Columns of %_content table */ |
| 563 sqlite3 *db = p->db; /* The database connection */ |
| 564 |
| 565 /* Create a list of user columns for the content table */ |
| 566 zContentCols = sqlite3_mprintf("docid INTEGER PRIMARY KEY"); |
| 567 for(i=0; zContentCols && i<p->nColumn; i++){ |
| 568 char *z = p->azColumn[i]; |
| 569 zContentCols = sqlite3_mprintf("%z, 'c%d%q'", zContentCols, i, z); |
| 570 } |
| 571 if( zContentCols==0 ) rc = SQLITE_NOMEM; |
| 572 |
| 573 /* Create the content table */ |
| 574 fts3DbExec(&rc, db, |
| 575 "CREATE TABLE %Q.'%q_content'(%s)", |
| 576 p->zDb, p->zName, zContentCols |
| 577 ); |
| 578 sqlite3_free(zContentCols); |
| 579 /* Create other tables */ |
| 580 fts3DbExec(&rc, db, |
| 581 "CREATE TABLE %Q.'%q_segments'(blockid INTEGER PRIMARY KEY, block BLOB);", |
| 582 p->zDb, p->zName |
| 583 ); |
| 584 fts3DbExec(&rc, db, |
| 585 "CREATE TABLE %Q.'%q_segdir'(" |
| 586 "level INTEGER," |
| 587 "idx INTEGER," |
| 588 "start_block INTEGER," |
| 589 "leaves_end_block INTEGER," |
| 590 "end_block INTEGER," |
| 591 "root BLOB," |
| 592 "PRIMARY KEY(level, idx)" |
| 593 ");", |
| 594 p->zDb, p->zName |
| 595 ); |
| 596 if( p->bHasDocsize ){ |
| 597 fts3DbExec(&rc, db, |
| 598 "CREATE TABLE %Q.'%q_docsize'(docid INTEGER PRIMARY KEY, size BLOB);", |
| 599 p->zDb, p->zName |
| 600 ); |
| 601 } |
| 602 if( p->bHasStat ){ |
| 603 fts3DbExec(&rc, db, |
| 604 "CREATE TABLE %Q.'%q_stat'(id INTEGER PRIMARY KEY, value BLOB);", |
| 605 p->zDb, p->zName |
| 606 ); |
| 607 } |
| 608 return rc; |
| 609 } |
| 610 |
| 611 /* |
| 612 ** Store the current database page-size in bytes in p->nPgsz. |
| 613 ** |
| 614 ** If *pRc is non-zero when this function is called, it is a no-op. |
| 615 ** Otherwise, if an error occurs, an SQLite error code is stored in *pRc |
| 616 ** before returning. |
| 617 */ |
| 618 static void fts3DatabasePageSize(int *pRc, Fts3Table *p){ |
| 619 if( *pRc==SQLITE_OK ){ |
| 620 int rc; /* Return code */ |
| 621 char *zSql; /* SQL text "PRAGMA %Q.page_size" */ |
| 622 sqlite3_stmt *pStmt; /* Compiled "PRAGMA %Q.page_size" statement */ |
| 623 |
| 624 zSql = sqlite3_mprintf("PRAGMA %Q.page_size", p->zDb); |
| 625 if( !zSql ){ |
| 626 rc = SQLITE_NOMEM; |
| 627 }else{ |
| 628 rc = sqlite3_prepare(p->db, zSql, -1, &pStmt, 0); |
| 629 if( rc==SQLITE_OK ){ |
| 630 sqlite3_step(pStmt); |
| 631 p->nPgsz = sqlite3_column_int(pStmt, 0); |
| 632 rc = sqlite3_finalize(pStmt); |
| 633 } |
| 634 } |
| 635 assert( p->nPgsz>0 || rc!=SQLITE_OK ); |
| 636 sqlite3_free(zSql); |
| 637 *pRc = rc; |
| 638 } |
| 639 } |
| 640 |
| 641 /* |
| 642 ** "Special" FTS4 arguments are column specifications of the following form: |
| 643 ** |
| 644 ** <key> = <value> |
| 645 ** |
| 646 ** There may not be whitespace surrounding the "=" character. The <value> |
| 647 ** term may be quoted, but the <key> may not. |
| 648 */ |
| 649 static int fts3IsSpecialColumn( |
| 650 const char *z, |
| 651 int *pnKey, |
| 652 char **pzValue |
| 653 ){ |
| 654 char *zValue; |
| 655 const char *zCsr = z; |
| 656 |
| 657 while( *zCsr!='=' ){ |
| 658 if( *zCsr=='\0' ) return 0; |
| 659 zCsr++; |
| 660 } |
| 661 |
| 662 *pnKey = (int)(zCsr-z); |
| 663 zValue = sqlite3_mprintf("%s", &zCsr[1]); |
| 664 if( zValue ){ |
| 665 sqlite3Fts3Dequote(zValue); |
| 666 } |
| 667 *pzValue = zValue; |
| 668 return 1; |
| 669 } |
| 670 |
| 671 /* |
| 672 ** Append the output of a printf() style formatting to an existing string. |
| 673 */ |
| 674 static void fts3Appendf( |
| 675 int *pRc, /* IN/OUT: Error code */ |
| 676 char **pz, /* IN/OUT: Pointer to string buffer */ |
| 677 const char *zFormat, /* Printf format string to append */ |
| 678 ... /* Arguments for printf format string */ |
| 679 ){ |
| 680 if( *pRc==SQLITE_OK ){ |
| 681 va_list ap; |
| 682 char *z; |
| 683 va_start(ap, zFormat); |
| 684 z = sqlite3_vmprintf(zFormat, ap); |
| 685 if( z && *pz ){ |
| 686 char *z2 = sqlite3_mprintf("%s%s", *pz, z); |
| 687 sqlite3_free(z); |
| 688 z = z2; |
| 689 } |
| 690 if( z==0 ) *pRc = SQLITE_NOMEM; |
| 691 sqlite3_free(*pz); |
| 692 *pz = z; |
| 693 } |
| 694 } |
| 695 |
| 696 /* |
| 697 ** Return a copy of input string zInput enclosed in double-quotes (") and |
| 698 ** with all double quote characters escaped. For example: |
| 699 ** |
| 700 ** fts3QuoteId("un \"zip\"") -> "un \"\"zip\"\"" |
| 701 ** |
| 702 ** The pointer returned points to memory obtained from sqlite3_malloc(). It |
| 703 ** is the callers responsibility to call sqlite3_free() to release this |
| 704 ** memory. |
| 705 */ |
| 706 static char *fts3QuoteId(char const *zInput){ |
| 707 int nRet; |
| 708 char *zRet; |
| 709 nRet = 2 + strlen(zInput)*2 + 1; |
| 710 zRet = sqlite3_malloc(nRet); |
| 711 if( zRet ){ |
| 712 int i; |
| 713 char *z = zRet; |
| 714 *(z++) = '"'; |
| 715 for(i=0; zInput[i]; i++){ |
| 716 if( zInput[i]=='"' ) *(z++) = '"'; |
| 717 *(z++) = zInput[i]; |
| 718 } |
| 719 *(z++) = '"'; |
| 720 *(z++) = '\0'; |
| 721 } |
| 722 return zRet; |
| 723 } |
| 724 |
| 725 /* |
| 726 ** Return a list of comma separated SQL expressions that could be used |
| 727 ** in a SELECT statement such as the following: |
| 728 ** |
| 729 ** SELECT <list of expressions> FROM %_content AS x ... |
| 730 ** |
| 731 ** to return the docid, followed by each column of text data in order |
| 732 ** from left to write. If parameter zFunc is not NULL, then instead of |
| 733 ** being returned directly each column of text data is passed to an SQL |
| 734 ** function named zFunc first. For example, if zFunc is "unzip" and the |
| 735 ** table has the three user-defined columns "a", "b", and "c", the following |
| 736 ** string is returned: |
| 737 ** |
| 738 ** "docid, unzip(x.'a'), unzip(x.'b'), unzip(x.'c')" |
| 739 ** |
| 740 ** The pointer returned points to a buffer allocated by sqlite3_malloc(). It |
| 741 ** is the responsibility of the caller to eventually free it. |
| 742 ** |
| 743 ** If *pRc is not SQLITE_OK when this function is called, it is a no-op (and |
| 744 ** a NULL pointer is returned). Otherwise, if an OOM error is encountered |
| 745 ** by this function, NULL is returned and *pRc is set to SQLITE_NOMEM. If |
| 746 ** no error occurs, *pRc is left unmodified. |
| 747 */ |
| 748 static char *fts3ReadExprList(Fts3Table *p, const char *zFunc, int *pRc){ |
| 749 char *zRet = 0; |
| 750 char *zFree = 0; |
| 751 char *zFunction; |
| 752 int i; |
| 753 |
| 754 if( !zFunc ){ |
| 755 zFunction = ""; |
| 756 }else{ |
| 757 zFree = zFunction = fts3QuoteId(zFunc); |
| 758 } |
| 759 fts3Appendf(pRc, &zRet, "docid"); |
| 760 for(i=0; i<p->nColumn; i++){ |
| 761 fts3Appendf(pRc, &zRet, ",%s(x.'c%d%q')", zFunction, i, p->azColumn[i]); |
| 762 } |
| 763 sqlite3_free(zFree); |
| 764 return zRet; |
| 765 } |
| 766 |
| 767 /* |
| 768 ** Return a list of N comma separated question marks, where N is the number |
| 769 ** of columns in the %_content table (one for the docid plus one for each |
| 770 ** user-defined text column). |
| 771 ** |
| 772 ** If argument zFunc is not NULL, then all but the first question mark |
| 773 ** is preceded by zFunc and an open bracket, and followed by a closed |
| 774 ** bracket. For example, if zFunc is "zip" and the FTS3 table has three |
| 775 ** user-defined text columns, the following string is returned: |
| 776 ** |
| 777 ** "?, zip(?), zip(?), zip(?)" |
| 778 ** |
| 779 ** The pointer returned points to a buffer allocated by sqlite3_malloc(). It |
| 780 ** is the responsibility of the caller to eventually free it. |
| 781 ** |
| 782 ** If *pRc is not SQLITE_OK when this function is called, it is a no-op (and |
| 783 ** a NULL pointer is returned). Otherwise, if an OOM error is encountered |
| 784 ** by this function, NULL is returned and *pRc is set to SQLITE_NOMEM. If |
| 785 ** no error occurs, *pRc is left unmodified. |
| 786 */ |
| 787 static char *fts3WriteExprList(Fts3Table *p, const char *zFunc, int *pRc){ |
| 788 char *zRet = 0; |
| 789 char *zFree = 0; |
| 790 char *zFunction; |
| 791 int i; |
| 792 |
| 793 if( !zFunc ){ |
| 794 zFunction = ""; |
| 795 }else{ |
| 796 zFree = zFunction = fts3QuoteId(zFunc); |
| 797 } |
| 798 fts3Appendf(pRc, &zRet, "?"); |
| 799 for(i=0; i<p->nColumn; i++){ |
| 800 fts3Appendf(pRc, &zRet, ",%s(?)", zFunction); |
| 801 } |
| 802 sqlite3_free(zFree); |
| 803 return zRet; |
| 804 } |
| 805 |
| 806 /* |
| 807 ** This function is the implementation of both the xConnect and xCreate |
| 808 ** methods of the FTS3 virtual table. |
| 809 ** |
| 810 ** The argv[] array contains the following: |
| 811 ** |
| 812 ** argv[0] -> module name ("fts3" or "fts4") |
| 813 ** argv[1] -> database name |
| 814 ** argv[2] -> table name |
| 815 ** argv[...] -> "column name" and other module argument fields. |
| 816 */ |
| 817 static int fts3InitVtab( |
| 818 int isCreate, /* True for xCreate, false for xConnect */ |
| 819 sqlite3 *db, /* The SQLite database connection */ |
| 820 void *pAux, /* Hash table containing tokenizers */ |
| 821 int argc, /* Number of elements in argv array */ |
| 822 const char * const *argv, /* xCreate/xConnect argument array */ |
| 823 sqlite3_vtab **ppVTab, /* Write the resulting vtab structure here */ |
| 824 char **pzErr /* Write any error message here */ |
| 825 ){ |
| 826 Fts3Hash *pHash = (Fts3Hash *)pAux; |
| 827 Fts3Table *p = 0; /* Pointer to allocated vtab */ |
| 828 int rc = SQLITE_OK; /* Return code */ |
| 829 int i; /* Iterator variable */ |
| 830 int nByte; /* Size of allocation used for *p */ |
| 831 int iCol; /* Column index */ |
| 832 int nString = 0; /* Bytes required to hold all column names */ |
| 833 int nCol = 0; /* Number of columns in the FTS table */ |
| 834 char *zCsr; /* Space for holding column names */ |
| 835 int nDb; /* Bytes required to hold database name */ |
| 836 int nName; /* Bytes required to hold table name */ |
| 837 int isFts4 = (argv[0][3]=='4'); /* True for FTS4, false for FTS3 */ |
| 838 int bNoDocsize = 0; /* True to omit %_docsize table */ |
| 839 const char **aCol; /* Array of column names */ |
| 840 sqlite3_tokenizer *pTokenizer = 0; /* Tokenizer for this table */ |
| 841 |
| 842 char *zCompress = 0; |
| 843 char *zUncompress = 0; |
| 844 |
| 845 assert( strlen(argv[0])==4 ); |
| 846 assert( (sqlite3_strnicmp(argv[0], "fts4", 4)==0 && isFts4) |
| 847 || (sqlite3_strnicmp(argv[0], "fts3", 4)==0 && !isFts4) |
| 848 ); |
| 849 |
| 850 nDb = (int)strlen(argv[1]) + 1; |
| 851 nName = (int)strlen(argv[2]) + 1; |
| 852 |
| 853 aCol = (const char **)sqlite3_malloc(sizeof(const char *) * (argc-2) ); |
| 854 if( !aCol ) return SQLITE_NOMEM; |
| 855 memset((void *)aCol, 0, sizeof(const char *) * (argc-2)); |
| 856 |
| 857 /* Loop through all of the arguments passed by the user to the FTS3/4 |
| 858 ** module (i.e. all the column names and special arguments). This loop |
| 859 ** does the following: |
| 860 ** |
| 861 ** + Figures out the number of columns the FTSX table will have, and |
| 862 ** the number of bytes of space that must be allocated to store copies |
| 863 ** of the column names. |
| 864 ** |
| 865 ** + If there is a tokenizer specification included in the arguments, |
| 866 ** initializes the tokenizer pTokenizer. |
| 867 */ |
| 868 for(i=3; rc==SQLITE_OK && i<argc; i++){ |
| 869 char const *z = argv[i]; |
| 870 int nKey; |
| 871 char *zVal; |
| 872 |
| 873 /* Check if this is a tokenizer specification */ |
| 874 if( !pTokenizer |
| 875 && strlen(z)>8 |
| 876 && 0==sqlite3_strnicmp(z, "tokenize", 8) |
| 877 && 0==sqlite3Fts3IsIdChar(z[8]) |
| 878 ){ |
| 879 rc = sqlite3Fts3InitTokenizer(pHash, &z[9], &pTokenizer, pzErr); |
| 880 } |
| 881 |
| 882 /* Check if it is an FTS4 special argument. */ |
| 883 else if( isFts4 && fts3IsSpecialColumn(z, &nKey, &zVal) ){ |
| 884 if( !zVal ){ |
| 885 rc = SQLITE_NOMEM; |
| 886 goto fts3_init_out; |
| 887 } |
| 888 if( nKey==9 && 0==sqlite3_strnicmp(z, "matchinfo", 9) ){ |
| 889 if( strlen(zVal)==4 && 0==sqlite3_strnicmp(zVal, "fts3", 4) ){ |
| 890 bNoDocsize = 1; |
| 891 }else{ |
| 892 *pzErr = sqlite3_mprintf("unrecognized matchinfo: %s", zVal); |
| 893 rc = SQLITE_ERROR; |
| 894 } |
| 895 }else if( nKey==8 && 0==sqlite3_strnicmp(z, "compress", 8) ){ |
| 896 zCompress = zVal; |
| 897 zVal = 0; |
| 898 }else if( nKey==10 && 0==sqlite3_strnicmp(z, "uncompress", 10) ){ |
| 899 zUncompress = zVal; |
| 900 zVal = 0; |
| 901 }else{ |
| 902 *pzErr = sqlite3_mprintf("unrecognized parameter: %s", z); |
| 903 rc = SQLITE_ERROR; |
| 904 } |
| 905 sqlite3_free(zVal); |
| 906 } |
| 907 |
| 908 /* Otherwise, the argument is a column name. */ |
| 909 else { |
| 910 nString += (int)(strlen(z) + 1); |
| 911 aCol[nCol++] = z; |
| 912 } |
| 913 } |
| 914 if( rc!=SQLITE_OK ) goto fts3_init_out; |
| 915 |
| 916 if( nCol==0 ){ |
| 917 assert( nString==0 ); |
| 918 aCol[0] = "content"; |
| 919 nString = 8; |
| 920 nCol = 1; |
| 921 } |
| 922 |
| 923 if( pTokenizer==0 ){ |
| 924 rc = sqlite3Fts3InitTokenizer(pHash, "simple", &pTokenizer, pzErr); |
| 925 if( rc!=SQLITE_OK ) goto fts3_init_out; |
| 926 } |
| 927 assert( pTokenizer ); |
| 928 |
| 929 |
| 930 /* Allocate and populate the Fts3Table structure. */ |
| 931 nByte = sizeof(Fts3Table) + /* Fts3Table */ |
| 932 nCol * sizeof(char *) + /* azColumn */ |
| 933 nName + /* zName */ |
| 934 nDb + /* zDb */ |
| 935 nString; /* Space for azColumn strings */ |
| 936 p = (Fts3Table*)sqlite3_malloc(nByte); |
| 937 if( p==0 ){ |
| 938 rc = SQLITE_NOMEM; |
| 939 goto fts3_init_out; |
| 940 } |
| 941 memset(p, 0, nByte); |
| 942 p->db = db; |
| 943 p->nColumn = nCol; |
| 944 p->nPendingData = 0; |
| 945 p->azColumn = (char **)&p[1]; |
| 946 p->pTokenizer = pTokenizer; |
| 947 p->nNodeSize = 1000; |
| 948 p->nMaxPendingData = FTS3_MAX_PENDING_DATA; |
| 949 p->bHasDocsize = (isFts4 && bNoDocsize==0); |
| 950 p->bHasStat = isFts4; |
| 951 fts3HashInit(&p->pendingTerms, FTS3_HASH_STRING, 1); |
| 952 |
| 953 /* Fill in the zName and zDb fields of the vtab structure. */ |
| 954 zCsr = (char *)&p->azColumn[nCol]; |
| 955 p->zName = zCsr; |
| 956 memcpy(zCsr, argv[2], nName); |
| 957 zCsr += nName; |
| 958 p->zDb = zCsr; |
| 959 memcpy(zCsr, argv[1], nDb); |
| 960 zCsr += nDb; |
| 961 |
| 962 /* Fill in the azColumn array */ |
| 963 for(iCol=0; iCol<nCol; iCol++){ |
| 964 char *z; |
| 965 int n; |
| 966 z = (char *)sqlite3Fts3NextToken(aCol[iCol], &n); |
| 967 memcpy(zCsr, z, n); |
| 968 zCsr[n] = '\0'; |
| 969 sqlite3Fts3Dequote(zCsr); |
| 970 p->azColumn[iCol] = zCsr; |
| 971 zCsr += n+1; |
| 972 assert( zCsr <= &((char *)p)[nByte] ); |
| 973 } |
| 974 |
| 975 if( (zCompress==0)!=(zUncompress==0) ){ |
| 976 char const *zMiss = (zCompress==0 ? "compress" : "uncompress"); |
| 977 rc = SQLITE_ERROR; |
| 978 *pzErr = sqlite3_mprintf("missing %s parameter in fts4 constructor", zMiss); |
| 979 } |
| 980 p->zReadExprlist = fts3ReadExprList(p, zUncompress, &rc); |
| 981 p->zWriteExprlist = fts3WriteExprList(p, zCompress, &rc); |
| 982 if( rc!=SQLITE_OK ) goto fts3_init_out; |
| 983 |
| 984 /* If this is an xCreate call, create the underlying tables in the |
| 985 ** database. TODO: For xConnect(), it could verify that said tables exist. |
| 986 */ |
| 987 if( isCreate ){ |
| 988 rc = fts3CreateTables(p); |
| 989 } |
| 990 |
| 991 /* Figure out the page-size for the database. This is required in order to |
| 992 ** estimate the cost of loading large doclists from the database (see |
| 993 ** function sqlite3Fts3SegReaderCost() for details). |
| 994 */ |
| 995 fts3DatabasePageSize(&rc, p); |
| 996 |
| 997 /* Declare the table schema to SQLite. */ |
| 998 fts3DeclareVtab(&rc, p); |
| 999 |
| 1000 fts3_init_out: |
| 1001 sqlite3_free(zCompress); |
| 1002 sqlite3_free(zUncompress); |
| 1003 sqlite3_free((void *)aCol); |
| 1004 if( rc!=SQLITE_OK ){ |
| 1005 if( p ){ |
| 1006 fts3DisconnectMethod((sqlite3_vtab *)p); |
| 1007 }else if( pTokenizer ){ |
| 1008 pTokenizer->pModule->xDestroy(pTokenizer); |
| 1009 } |
| 1010 }else{ |
| 1011 *ppVTab = &p->base; |
| 1012 } |
| 1013 return rc; |
| 1014 } |
| 1015 |
| 1016 /* |
| 1017 ** The xConnect() and xCreate() methods for the virtual table. All the |
| 1018 ** work is done in function fts3InitVtab(). |
| 1019 */ |
| 1020 static int fts3ConnectMethod( |
| 1021 sqlite3 *db, /* Database connection */ |
| 1022 void *pAux, /* Pointer to tokenizer hash table */ |
| 1023 int argc, /* Number of elements in argv array */ |
| 1024 const char * const *argv, /* xCreate/xConnect argument array */ |
| 1025 sqlite3_vtab **ppVtab, /* OUT: New sqlite3_vtab object */ |
| 1026 char **pzErr /* OUT: sqlite3_malloc'd error message */ |
| 1027 ){ |
| 1028 return fts3InitVtab(0, db, pAux, argc, argv, ppVtab, pzErr); |
| 1029 } |
| 1030 static int fts3CreateMethod( |
| 1031 sqlite3 *db, /* Database connection */ |
| 1032 void *pAux, /* Pointer to tokenizer hash table */ |
| 1033 int argc, /* Number of elements in argv array */ |
| 1034 const char * const *argv, /* xCreate/xConnect argument array */ |
| 1035 sqlite3_vtab **ppVtab, /* OUT: New sqlite3_vtab object */ |
| 1036 char **pzErr /* OUT: sqlite3_malloc'd error message */ |
| 1037 ){ |
| 1038 return fts3InitVtab(1, db, pAux, argc, argv, ppVtab, pzErr); |
| 1039 } |
| 1040 |
| 1041 /* |
| 1042 ** Implementation of the xBestIndex method for FTS3 tables. There |
| 1043 ** are three possible strategies, in order of preference: |
| 1044 ** |
| 1045 ** 1. Direct lookup by rowid or docid. |
| 1046 ** 2. Full-text search using a MATCH operator on a non-docid column. |
| 1047 ** 3. Linear scan of %_content table. |
| 1048 */ |
| 1049 static int fts3BestIndexMethod(sqlite3_vtab *pVTab, sqlite3_index_info *pInfo){ |
| 1050 Fts3Table *p = (Fts3Table *)pVTab; |
| 1051 int i; /* Iterator variable */ |
| 1052 int iCons = -1; /* Index of constraint to use */ |
| 1053 |
| 1054 /* By default use a full table scan. This is an expensive option, |
| 1055 ** so search through the constraints to see if a more efficient |
| 1056 ** strategy is possible. |
| 1057 */ |
| 1058 pInfo->idxNum = FTS3_FULLSCAN_SEARCH; |
| 1059 pInfo->estimatedCost = 500000; |
| 1060 for(i=0; i<pInfo->nConstraint; i++){ |
| 1061 struct sqlite3_index_constraint *pCons = &pInfo->aConstraint[i]; |
| 1062 if( pCons->usable==0 ) continue; |
| 1063 |
| 1064 /* A direct lookup on the rowid or docid column. Assign a cost of 1.0. */ |
| 1065 if( pCons->op==SQLITE_INDEX_CONSTRAINT_EQ |
| 1066 && (pCons->iColumn<0 || pCons->iColumn==p->nColumn+1 ) |
| 1067 ){ |
| 1068 pInfo->idxNum = FTS3_DOCID_SEARCH; |
| 1069 pInfo->estimatedCost = 1.0; |
| 1070 iCons = i; |
| 1071 } |
| 1072 |
| 1073 /* A MATCH constraint. Use a full-text search. |
| 1074 ** |
| 1075 ** If there is more than one MATCH constraint available, use the first |
| 1076 ** one encountered. If there is both a MATCH constraint and a direct |
| 1077 ** rowid/docid lookup, prefer the MATCH strategy. This is done even |
| 1078 ** though the rowid/docid lookup is faster than a MATCH query, selecting |
| 1079 ** it would lead to an "unable to use function MATCH in the requested |
| 1080 ** context" error. |
| 1081 */ |
| 1082 if( pCons->op==SQLITE_INDEX_CONSTRAINT_MATCH |
| 1083 && pCons->iColumn>=0 && pCons->iColumn<=p->nColumn |
| 1084 ){ |
| 1085 pInfo->idxNum = FTS3_FULLTEXT_SEARCH + pCons->iColumn; |
| 1086 pInfo->estimatedCost = 2.0; |
| 1087 iCons = i; |
| 1088 break; |
| 1089 } |
| 1090 } |
| 1091 |
| 1092 if( iCons>=0 ){ |
| 1093 pInfo->aConstraintUsage[iCons].argvIndex = 1; |
| 1094 pInfo->aConstraintUsage[iCons].omit = 1; |
| 1095 } |
| 1096 return SQLITE_OK; |
| 1097 } |
| 1098 |
| 1099 /* |
| 1100 ** Implementation of xOpen method. |
| 1101 */ |
| 1102 static int fts3OpenMethod(sqlite3_vtab *pVTab, sqlite3_vtab_cursor **ppCsr){ |
| 1103 sqlite3_vtab_cursor *pCsr; /* Allocated cursor */ |
| 1104 |
| 1105 UNUSED_PARAMETER(pVTab); |
| 1106 |
| 1107 /* Allocate a buffer large enough for an Fts3Cursor structure. If the |
| 1108 ** allocation succeeds, zero it and return SQLITE_OK. Otherwise, |
| 1109 ** if the allocation fails, return SQLITE_NOMEM. |
| 1110 */ |
| 1111 *ppCsr = pCsr = (sqlite3_vtab_cursor *)sqlite3_malloc(sizeof(Fts3Cursor)); |
| 1112 if( !pCsr ){ |
| 1113 return SQLITE_NOMEM; |
| 1114 } |
| 1115 memset(pCsr, 0, sizeof(Fts3Cursor)); |
| 1116 return SQLITE_OK; |
| 1117 } |
| 1118 |
| 1119 /* |
| 1120 ** Close the cursor. For additional information see the documentation |
| 1121 ** on the xClose method of the virtual table interface. |
| 1122 */ |
| 1123 static int fts3CloseMethod(sqlite3_vtab_cursor *pCursor){ |
| 1124 Fts3Cursor *pCsr = (Fts3Cursor *)pCursor; |
| 1125 assert( ((Fts3Table *)pCsr->base.pVtab)->pSegments==0 ); |
| 1126 sqlite3_finalize(pCsr->pStmt); |
| 1127 sqlite3Fts3ExprFree(pCsr->pExpr); |
| 1128 sqlite3Fts3FreeDeferredTokens(pCsr); |
| 1129 sqlite3_free(pCsr->aDoclist); |
| 1130 sqlite3_free(pCsr->aMatchinfo); |
| 1131 sqlite3_free(pCsr); |
| 1132 return SQLITE_OK; |
| 1133 } |
| 1134 |
| 1135 /* |
| 1136 ** Position the pCsr->pStmt statement so that it is on the row |
| 1137 ** of the %_content table that contains the last match. Return |
| 1138 ** SQLITE_OK on success. |
| 1139 */ |
| 1140 static int fts3CursorSeek(sqlite3_context *pContext, Fts3Cursor *pCsr){ |
| 1141 if( pCsr->isRequireSeek ){ |
| 1142 pCsr->isRequireSeek = 0; |
| 1143 sqlite3_bind_int64(pCsr->pStmt, 1, pCsr->iPrevId); |
| 1144 if( SQLITE_ROW==sqlite3_step(pCsr->pStmt) ){ |
| 1145 return SQLITE_OK; |
| 1146 }else{ |
| 1147 int rc = sqlite3_reset(pCsr->pStmt); |
| 1148 if( rc==SQLITE_OK ){ |
| 1149 /* If no row was found and no error has occured, then the %_content |
| 1150 ** table is missing a row that is present in the full-text index. |
| 1151 ** The data structures are corrupt. |
| 1152 */ |
| 1153 rc = SQLITE_CORRUPT; |
| 1154 } |
| 1155 pCsr->isEof = 1; |
| 1156 if( pContext ){ |
| 1157 sqlite3_result_error_code(pContext, rc); |
| 1158 } |
| 1159 return rc; |
| 1160 } |
| 1161 }else{ |
| 1162 return SQLITE_OK; |
| 1163 } |
| 1164 } |
| 1165 |
| 1166 /* |
| 1167 ** This function is used to process a single interior node when searching |
| 1168 ** a b-tree for a term or term prefix. The node data is passed to this |
| 1169 ** function via the zNode/nNode parameters. The term to search for is |
| 1170 ** passed in zTerm/nTerm. |
| 1171 ** |
| 1172 ** If piFirst is not NULL, then this function sets *piFirst to the blockid |
| 1173 ** of the child node that heads the sub-tree that may contain the term. |
| 1174 ** |
| 1175 ** If piLast is not NULL, then *piLast is set to the right-most child node |
| 1176 ** that heads a sub-tree that may contain a term for which zTerm/nTerm is |
| 1177 ** a prefix. |
| 1178 ** |
| 1179 ** If an OOM error occurs, SQLITE_NOMEM is returned. Otherwise, SQLITE_OK. |
| 1180 */ |
| 1181 static int fts3ScanInteriorNode( |
| 1182 const char *zTerm, /* Term to select leaves for */ |
| 1183 int nTerm, /* Size of term zTerm in bytes */ |
| 1184 const char *zNode, /* Buffer containing segment interior node */ |
| 1185 int nNode, /* Size of buffer at zNode */ |
| 1186 sqlite3_int64 *piFirst, /* OUT: Selected child node */ |
| 1187 sqlite3_int64 *piLast /* OUT: Selected child node */ |
| 1188 ){ |
| 1189 int rc = SQLITE_OK; /* Return code */ |
| 1190 const char *zCsr = zNode; /* Cursor to iterate through node */ |
| 1191 const char *zEnd = &zCsr[nNode];/* End of interior node buffer */ |
| 1192 char *zBuffer = 0; /* Buffer to load terms into */ |
| 1193 int nAlloc = 0; /* Size of allocated buffer */ |
| 1194 int isFirstTerm = 1; /* True when processing first term on page */ |
| 1195 sqlite3_int64 iChild; /* Block id of child node to descend to */ |
| 1196 |
| 1197 /* Skip over the 'height' varint that occurs at the start of every |
| 1198 ** interior node. Then load the blockid of the left-child of the b-tree |
| 1199 ** node into variable iChild. |
| 1200 ** |
| 1201 ** Even if the data structure on disk is corrupted, this (reading two |
| 1202 ** varints from the buffer) does not risk an overread. If zNode is a |
| 1203 ** root node, then the buffer comes from a SELECT statement. SQLite does |
| 1204 ** not make this guarantee explicitly, but in practice there are always |
| 1205 ** either more than 20 bytes of allocated space following the nNode bytes of |
| 1206 ** contents, or two zero bytes. Or, if the node is read from the %_segments |
| 1207 ** table, then there are always 20 bytes of zeroed padding following the |
| 1208 ** nNode bytes of content (see sqlite3Fts3ReadBlock() for details). |
| 1209 */ |
| 1210 zCsr += sqlite3Fts3GetVarint(zCsr, &iChild); |
| 1211 zCsr += sqlite3Fts3GetVarint(zCsr, &iChild); |
| 1212 if( zCsr>zEnd ){ |
| 1213 return SQLITE_CORRUPT; |
| 1214 } |
| 1215 |
| 1216 while( zCsr<zEnd && (piFirst || piLast) ){ |
| 1217 int cmp; /* memcmp() result */ |
| 1218 int nSuffix; /* Size of term suffix */ |
| 1219 int nPrefix = 0; /* Size of term prefix */ |
| 1220 int nBuffer; /* Total term size */ |
| 1221 |
| 1222 /* Load the next term on the node into zBuffer. Use realloc() to expand |
| 1223 ** the size of zBuffer if required. */ |
| 1224 if( !isFirstTerm ){ |
| 1225 zCsr += sqlite3Fts3GetVarint32(zCsr, &nPrefix); |
| 1226 } |
| 1227 isFirstTerm = 0; |
| 1228 zCsr += sqlite3Fts3GetVarint32(zCsr, &nSuffix); |
| 1229 |
| 1230 /* NOTE(shess): Previous code checked for negative nPrefix and |
| 1231 ** nSuffix and suffix overrunning zEnd. Additionally corrupt if |
| 1232 ** the prefix is longer than the previous term, or if the suffix |
| 1233 ** causes overflow. |
| 1234 */ |
| 1235 if( nPrefix<0 || nSuffix<0 || nPrefix>nBuffer |
| 1236 || &zCsr[nSuffix]<zCsr || &zCsr[nSuffix]>zEnd ){ |
| 1237 rc = SQLITE_CORRUPT; |
| 1238 goto finish_scan; |
| 1239 } |
| 1240 if( nPrefix+nSuffix>nAlloc ){ |
| 1241 char *zNew; |
| 1242 nAlloc = (nPrefix+nSuffix) * 2; |
| 1243 zNew = (char *)sqlite3_realloc(zBuffer, nAlloc); |
| 1244 if( !zNew ){ |
| 1245 rc = SQLITE_NOMEM; |
| 1246 goto finish_scan; |
| 1247 } |
| 1248 zBuffer = zNew; |
| 1249 } |
| 1250 memcpy(&zBuffer[nPrefix], zCsr, nSuffix); |
| 1251 nBuffer = nPrefix + nSuffix; |
| 1252 zCsr += nSuffix; |
| 1253 |
| 1254 /* Compare the term we are searching for with the term just loaded from |
| 1255 ** the interior node. If the specified term is greater than or equal |
| 1256 ** to the term from the interior node, then all terms on the sub-tree |
| 1257 ** headed by node iChild are smaller than zTerm. No need to search |
| 1258 ** iChild. |
| 1259 ** |
| 1260 ** If the interior node term is larger than the specified term, then |
| 1261 ** the tree headed by iChild may contain the specified term. |
| 1262 */ |
| 1263 cmp = memcmp(zTerm, zBuffer, (nBuffer>nTerm ? nTerm : nBuffer)); |
| 1264 if( piFirst && (cmp<0 || (cmp==0 && nBuffer>nTerm)) ){ |
| 1265 *piFirst = iChild; |
| 1266 piFirst = 0; |
| 1267 } |
| 1268 |
| 1269 if( piLast && cmp<0 ){ |
| 1270 *piLast = iChild; |
| 1271 piLast = 0; |
| 1272 } |
| 1273 |
| 1274 iChild++; |
| 1275 }; |
| 1276 |
| 1277 if( piFirst ) *piFirst = iChild; |
| 1278 if( piLast ) *piLast = iChild; |
| 1279 |
| 1280 finish_scan: |
| 1281 sqlite3_free(zBuffer); |
| 1282 return rc; |
| 1283 } |
| 1284 |
| 1285 |
| 1286 /* |
| 1287 ** The buffer pointed to by argument zNode (size nNode bytes) contains an |
| 1288 ** interior node of a b-tree segment. The zTerm buffer (size nTerm bytes) |
| 1289 ** contains a term. This function searches the sub-tree headed by the zNode |
| 1290 ** node for the range of leaf nodes that may contain the specified term |
| 1291 ** or terms for which the specified term is a prefix. |
| 1292 ** |
| 1293 ** If piLeaf is not NULL, then *piLeaf is set to the blockid of the |
| 1294 ** left-most leaf node in the tree that may contain the specified term. |
| 1295 ** If piLeaf2 is not NULL, then *piLeaf2 is set to the blockid of the |
| 1296 ** right-most leaf node that may contain a term for which the specified |
| 1297 ** term is a prefix. |
| 1298 ** |
| 1299 ** It is possible that the range of returned leaf nodes does not contain |
| 1300 ** the specified term or any terms for which it is a prefix. However, if the |
| 1301 ** segment does contain any such terms, they are stored within the identified |
| 1302 ** range. Because this function only inspects interior segment nodes (and |
| 1303 ** never loads leaf nodes into memory), it is not possible to be sure. |
| 1304 ** |
| 1305 ** If an error occurs, an error code other than SQLITE_OK is returned. |
| 1306 */ |
| 1307 static int fts3SelectLeaf( |
| 1308 Fts3Table *p, /* Virtual table handle */ |
| 1309 const char *zTerm, /* Term to select leaves for */ |
| 1310 int nTerm, /* Size of term zTerm in bytes */ |
| 1311 const char *zNode, /* Buffer containing segment interior node */ |
| 1312 int nNode, /* Size of buffer at zNode */ |
| 1313 sqlite3_int64 *piLeaf, /* Selected leaf node */ |
| 1314 sqlite3_int64 *piLeaf2 /* Selected leaf node */ |
| 1315 ){ |
| 1316 int rc; /* Return code */ |
| 1317 int iHeight; /* Height of this node in tree */ |
| 1318 |
| 1319 assert( piLeaf || piLeaf2 ); |
| 1320 |
| 1321 sqlite3Fts3GetVarint32(zNode, &iHeight); |
| 1322 rc = fts3ScanInteriorNode(zTerm, nTerm, zNode, nNode, piLeaf, piLeaf2); |
| 1323 assert( !piLeaf2 || !piLeaf || rc!=SQLITE_OK || (*piLeaf<=*piLeaf2) ); |
| 1324 |
| 1325 if( rc==SQLITE_OK && iHeight>1 ){ |
| 1326 char *zBlob = 0; /* Blob read from %_segments table */ |
| 1327 int nBlob; /* Size of zBlob in bytes */ |
| 1328 |
| 1329 if( piLeaf && piLeaf2 && (*piLeaf!=*piLeaf2) ){ |
| 1330 rc = sqlite3Fts3ReadBlock(p, *piLeaf, &zBlob, &nBlob); |
| 1331 if( rc==SQLITE_OK ){ |
| 1332 rc = fts3SelectLeaf(p, zTerm, nTerm, zBlob, nBlob, piLeaf, 0); |
| 1333 } |
| 1334 sqlite3_free(zBlob); |
| 1335 piLeaf = 0; |
| 1336 zBlob = 0; |
| 1337 } |
| 1338 |
| 1339 if( rc==SQLITE_OK ){ |
| 1340 rc = sqlite3Fts3ReadBlock(p, piLeaf ? *piLeaf : *piLeaf2, &zBlob, &nBlob); |
| 1341 } |
| 1342 if( rc==SQLITE_OK ){ |
| 1343 rc = fts3SelectLeaf(p, zTerm, nTerm, zBlob, nBlob, piLeaf, piLeaf2); |
| 1344 } |
| 1345 sqlite3_free(zBlob); |
| 1346 } |
| 1347 |
| 1348 return rc; |
| 1349 } |
| 1350 |
| 1351 /* |
| 1352 ** This function is used to create delta-encoded serialized lists of FTS3 |
| 1353 ** varints. Each call to this function appends a single varint to a list. |
| 1354 */ |
| 1355 static void fts3PutDeltaVarint( |
| 1356 char **pp, /* IN/OUT: Output pointer */ |
| 1357 sqlite3_int64 *piPrev, /* IN/OUT: Previous value written to list */ |
| 1358 sqlite3_int64 iVal /* Write this value to the list */ |
| 1359 ){ |
| 1360 assert( iVal-*piPrev > 0 || (*piPrev==0 && iVal==0) ); |
| 1361 *pp += sqlite3Fts3PutVarint(*pp, iVal-*piPrev); |
| 1362 *piPrev = iVal; |
| 1363 } |
| 1364 |
| 1365 /* |
| 1366 ** When this function is called, *ppPoslist is assumed to point to the |
| 1367 ** start of a position-list. After it returns, *ppPoslist points to the |
| 1368 ** first byte after the position-list. |
| 1369 ** |
| 1370 ** A position list is list of positions (delta encoded) and columns for |
| 1371 ** a single document record of a doclist. So, in other words, this |
| 1372 ** routine advances *ppPoslist so that it points to the next docid in |
| 1373 ** the doclist, or to the first byte past the end of the doclist. |
| 1374 ** |
| 1375 ** If pp is not NULL, then the contents of the position list are copied |
| 1376 ** to *pp. *pp is set to point to the first byte past the last byte copied |
| 1377 ** before this function returns. |
| 1378 */ |
| 1379 static void fts3PoslistCopy(char **pp, char **ppPoslist){ |
| 1380 char *pEnd = *ppPoslist; |
| 1381 char c = 0; |
| 1382 |
| 1383 /* The end of a position list is marked by a zero encoded as an FTS3 |
| 1384 ** varint. A single POS_END (0) byte. Except, if the 0 byte is preceded by |
| 1385 ** a byte with the 0x80 bit set, then it is not a varint 0, but the tail |
| 1386 ** of some other, multi-byte, value. |
| 1387 ** |
| 1388 ** The following while-loop moves pEnd to point to the first byte that is not |
| 1389 ** immediately preceded by a byte with the 0x80 bit set. Then increments |
| 1390 ** pEnd once more so that it points to the byte immediately following the |
| 1391 ** last byte in the position-list. |
| 1392 */ |
| 1393 while( *pEnd | c ){ |
| 1394 c = *pEnd++ & 0x80; |
| 1395 testcase( c!=0 && (*pEnd)==0 ); |
| 1396 } |
| 1397 pEnd++; /* Advance past the POS_END terminator byte */ |
| 1398 |
| 1399 if( pp ){ |
| 1400 int n = (int)(pEnd - *ppPoslist); |
| 1401 char *p = *pp; |
| 1402 memcpy(p, *ppPoslist, n); |
| 1403 p += n; |
| 1404 *pp = p; |
| 1405 } |
| 1406 *ppPoslist = pEnd; |
| 1407 } |
| 1408 |
| 1409 /* |
| 1410 ** When this function is called, *ppPoslist is assumed to point to the |
| 1411 ** start of a column-list. After it returns, *ppPoslist points to the |
| 1412 ** to the terminator (POS_COLUMN or POS_END) byte of the column-list. |
| 1413 ** |
| 1414 ** A column-list is list of delta-encoded positions for a single column |
| 1415 ** within a single document within a doclist. |
| 1416 ** |
| 1417 ** The column-list is terminated either by a POS_COLUMN varint (1) or |
| 1418 ** a POS_END varint (0). This routine leaves *ppPoslist pointing to |
| 1419 ** the POS_COLUMN or POS_END that terminates the column-list. |
| 1420 ** |
| 1421 ** If pp is not NULL, then the contents of the column-list are copied |
| 1422 ** to *pp. *pp is set to point to the first byte past the last byte copied |
| 1423 ** before this function returns. The POS_COLUMN or POS_END terminator |
| 1424 ** is not copied into *pp. |
| 1425 */ |
| 1426 static void fts3ColumnlistCopy(char **pp, char **ppPoslist){ |
| 1427 char *pEnd = *ppPoslist; |
| 1428 char c = 0; |
| 1429 |
| 1430 /* A column-list is terminated by either a 0x01 or 0x00 byte that is |
| 1431 ** not part of a multi-byte varint. |
| 1432 */ |
| 1433 while( 0xFE & (*pEnd | c) ){ |
| 1434 c = *pEnd++ & 0x80; |
| 1435 testcase( c!=0 && ((*pEnd)&0xfe)==0 ); |
| 1436 } |
| 1437 if( pp ){ |
| 1438 int n = (int)(pEnd - *ppPoslist); |
| 1439 char *p = *pp; |
| 1440 memcpy(p, *ppPoslist, n); |
| 1441 p += n; |
| 1442 *pp = p; |
| 1443 } |
| 1444 *ppPoslist = pEnd; |
| 1445 } |
| 1446 |
| 1447 /* |
| 1448 ** Value used to signify the end of an position-list. This is safe because |
| 1449 ** it is not possible to have a document with 2^31 terms. |
| 1450 */ |
| 1451 #define POSITION_LIST_END 0x7fffffff |
| 1452 |
| 1453 /* |
| 1454 ** This function is used to help parse position-lists. When this function is |
| 1455 ** called, *pp may point to the start of the next varint in the position-list |
| 1456 ** being parsed, or it may point to 1 byte past the end of the position-list |
| 1457 ** (in which case **pp will be a terminator bytes POS_END (0) or |
| 1458 ** (1)). |
| 1459 ** |
| 1460 ** If *pp points past the end of the current position-list, set *pi to |
| 1461 ** POSITION_LIST_END and return. Otherwise, read the next varint from *pp, |
| 1462 ** increment the current value of *pi by the value read, and set *pp to |
| 1463 ** point to the next value before returning. |
| 1464 ** |
| 1465 ** Before calling this routine *pi must be initialized to the value of |
| 1466 ** the previous position, or zero if we are reading the first position |
| 1467 ** in the position-list. Because positions are delta-encoded, the value |
| 1468 ** of the previous position is needed in order to compute the value of |
| 1469 ** the next position. |
| 1470 */ |
| 1471 static void fts3ReadNextPos( |
| 1472 char **pp, /* IN/OUT: Pointer into position-list buffer */ |
| 1473 sqlite3_int64 *pi /* IN/OUT: Value read from position-list */ |
| 1474 ){ |
| 1475 if( (**pp)&0xFE ){ |
| 1476 fts3GetDeltaVarint(pp, pi); |
| 1477 *pi -= 2; |
| 1478 }else{ |
| 1479 *pi = POSITION_LIST_END; |
| 1480 } |
| 1481 } |
| 1482 |
| 1483 /* |
| 1484 ** If parameter iCol is not 0, write an POS_COLUMN (1) byte followed by |
| 1485 ** the value of iCol encoded as a varint to *pp. This will start a new |
| 1486 ** column list. |
| 1487 ** |
| 1488 ** Set *pp to point to the byte just after the last byte written before |
| 1489 ** returning (do not modify it if iCol==0). Return the total number of bytes |
| 1490 ** written (0 if iCol==0). |
| 1491 */ |
| 1492 static int fts3PutColNumber(char **pp, int iCol){ |
| 1493 int n = 0; /* Number of bytes written */ |
| 1494 if( iCol ){ |
| 1495 char *p = *pp; /* Output pointer */ |
| 1496 n = 1 + sqlite3Fts3PutVarint(&p[1], iCol); |
| 1497 *p = 0x01; |
| 1498 *pp = &p[n]; |
| 1499 } |
| 1500 return n; |
| 1501 } |
| 1502 |
| 1503 /* |
| 1504 ** Compute the union of two position lists. The output written |
| 1505 ** into *pp contains all positions of both *pp1 and *pp2 in sorted |
| 1506 ** order and with any duplicates removed. All pointers are |
| 1507 ** updated appropriately. The caller is responsible for insuring |
| 1508 ** that there is enough space in *pp to hold the complete output. |
| 1509 */ |
| 1510 static void fts3PoslistMerge( |
| 1511 char **pp, /* Output buffer */ |
| 1512 char **pp1, /* Left input list */ |
| 1513 char **pp2 /* Right input list */ |
| 1514 ){ |
| 1515 char *p = *pp; |
| 1516 char *p1 = *pp1; |
| 1517 char *p2 = *pp2; |
| 1518 |
| 1519 while( *p1 || *p2 ){ |
| 1520 int iCol1; /* The current column index in pp1 */ |
| 1521 int iCol2; /* The current column index in pp2 */ |
| 1522 |
| 1523 if( *p1==POS_COLUMN ) sqlite3Fts3GetVarint32(&p1[1], &iCol1); |
| 1524 else if( *p1==POS_END ) iCol1 = POSITION_LIST_END; |
| 1525 else iCol1 = 0; |
| 1526 |
| 1527 if( *p2==POS_COLUMN ) sqlite3Fts3GetVarint32(&p2[1], &iCol2); |
| 1528 else if( *p2==POS_END ) iCol2 = POSITION_LIST_END; |
| 1529 else iCol2 = 0; |
| 1530 |
| 1531 if( iCol1==iCol2 ){ |
| 1532 sqlite3_int64 i1 = 0; /* Last position from pp1 */ |
| 1533 sqlite3_int64 i2 = 0; /* Last position from pp2 */ |
| 1534 sqlite3_int64 iPrev = 0; |
| 1535 int n = fts3PutColNumber(&p, iCol1); |
| 1536 p1 += n; |
| 1537 p2 += n; |
| 1538 |
| 1539 /* At this point, both p1 and p2 point to the start of column-lists |
| 1540 ** for the same column (the column with index iCol1 and iCol2). |
| 1541 ** A column-list is a list of non-negative delta-encoded varints, each |
| 1542 ** incremented by 2 before being stored. Each list is terminated by a |
| 1543 ** POS_END (0) or POS_COLUMN (1). The following block merges the two lists |
| 1544 ** and writes the results to buffer p. p is left pointing to the byte |
| 1545 ** after the list written. No terminator (POS_END or POS_COLUMN) is |
| 1546 ** written to the output. |
| 1547 */ |
| 1548 fts3GetDeltaVarint(&p1, &i1); |
| 1549 fts3GetDeltaVarint(&p2, &i2); |
| 1550 do { |
| 1551 fts3PutDeltaVarint(&p, &iPrev, (i1<i2) ? i1 : i2); |
| 1552 iPrev -= 2; |
| 1553 if( i1==i2 ){ |
| 1554 fts3ReadNextPos(&p1, &i1); |
| 1555 fts3ReadNextPos(&p2, &i2); |
| 1556 }else if( i1<i2 ){ |
| 1557 fts3ReadNextPos(&p1, &i1); |
| 1558 }else{ |
| 1559 fts3ReadNextPos(&p2, &i2); |
| 1560 } |
| 1561 }while( i1!=POSITION_LIST_END || i2!=POSITION_LIST_END ); |
| 1562 }else if( iCol1<iCol2 ){ |
| 1563 p1 += fts3PutColNumber(&p, iCol1); |
| 1564 fts3ColumnlistCopy(&p, &p1); |
| 1565 }else{ |
| 1566 p2 += fts3PutColNumber(&p, iCol2); |
| 1567 fts3ColumnlistCopy(&p, &p2); |
| 1568 } |
| 1569 } |
| 1570 |
| 1571 *p++ = POS_END; |
| 1572 *pp = p; |
| 1573 *pp1 = p1 + 1; |
| 1574 *pp2 = p2 + 1; |
| 1575 } |
| 1576 |
| 1577 /* |
| 1578 ** nToken==1 searches for adjacent positions. |
| 1579 ** |
| 1580 ** This function is used to merge two position lists into one. When it is |
| 1581 ** called, *pp1 and *pp2 must both point to position lists. A position-list is |
| 1582 ** the part of a doclist that follows each document id. For example, if a row |
| 1583 ** contains: |
| 1584 ** |
| 1585 ** 'a b c'|'x y z'|'a b b a' |
| 1586 ** |
| 1587 ** Then the position list for this row for token 'b' would consist of: |
| 1588 ** |
| 1589 ** 0x02 0x01 0x02 0x03 0x03 0x00 |
| 1590 ** |
| 1591 ** When this function returns, both *pp1 and *pp2 are left pointing to the |
| 1592 ** byte following the 0x00 terminator of their respective position lists. |
| 1593 ** |
| 1594 ** If isSaveLeft is 0, an entry is added to the output position list for |
| 1595 ** each position in *pp2 for which there exists one or more positions in |
| 1596 ** *pp1 so that (pos(*pp2)>pos(*pp1) && pos(*pp2)-pos(*pp1)<=nToken). i.e. |
| 1597 ** when the *pp1 token appears before the *pp2 token, but not more than nToken |
| 1598 ** slots before it. |
| 1599 */ |
| 1600 static int fts3PoslistPhraseMerge( |
| 1601 char **pp, /* IN/OUT: Preallocated output buffer */ |
| 1602 int nToken, /* Maximum difference in token positions */ |
| 1603 int isSaveLeft, /* Save the left position */ |
| 1604 int isExact, /* If *pp1 is exactly nTokens before *pp2 */ |
| 1605 char **pp1, /* IN/OUT: Left input list */ |
| 1606 char **pp2 /* IN/OUT: Right input list */ |
| 1607 ){ |
| 1608 char *p = (pp ? *pp : 0); |
| 1609 char *p1 = *pp1; |
| 1610 char *p2 = *pp2; |
| 1611 int iCol1 = 0; |
| 1612 int iCol2 = 0; |
| 1613 |
| 1614 /* Never set both isSaveLeft and isExact for the same invocation. */ |
| 1615 assert( isSaveLeft==0 || isExact==0 ); |
| 1616 |
| 1617 assert( *p1!=0 && *p2!=0 ); |
| 1618 if( *p1==POS_COLUMN ){ |
| 1619 p1++; |
| 1620 p1 += sqlite3Fts3GetVarint32(p1, &iCol1); |
| 1621 } |
| 1622 if( *p2==POS_COLUMN ){ |
| 1623 p2++; |
| 1624 p2 += sqlite3Fts3GetVarint32(p2, &iCol2); |
| 1625 } |
| 1626 |
| 1627 while( 1 ){ |
| 1628 if( iCol1==iCol2 ){ |
| 1629 char *pSave = p; |
| 1630 sqlite3_int64 iPrev = 0; |
| 1631 sqlite3_int64 iPos1 = 0; |
| 1632 sqlite3_int64 iPos2 = 0; |
| 1633 |
| 1634 if( pp && iCol1 ){ |
| 1635 *p++ = POS_COLUMN; |
| 1636 p += sqlite3Fts3PutVarint(p, iCol1); |
| 1637 } |
| 1638 |
| 1639 assert( *p1!=POS_END && *p1!=POS_COLUMN ); |
| 1640 assert( *p2!=POS_END && *p2!=POS_COLUMN ); |
| 1641 fts3GetDeltaVarint(&p1, &iPos1); iPos1 -= 2; |
| 1642 fts3GetDeltaVarint(&p2, &iPos2); iPos2 -= 2; |
| 1643 |
| 1644 while( 1 ){ |
| 1645 if( iPos2==iPos1+nToken |
| 1646 || (isExact==0 && iPos2>iPos1 && iPos2<=iPos1+nToken) |
| 1647 ){ |
| 1648 sqlite3_int64 iSave; |
| 1649 if( !pp ){ |
| 1650 fts3PoslistCopy(0, &p2); |
| 1651 fts3PoslistCopy(0, &p1); |
| 1652 *pp1 = p1; |
| 1653 *pp2 = p2; |
| 1654 return 1; |
| 1655 } |
| 1656 iSave = isSaveLeft ? iPos1 : iPos2; |
| 1657 fts3PutDeltaVarint(&p, &iPrev, iSave+2); iPrev -= 2; |
| 1658 pSave = 0; |
| 1659 } |
| 1660 if( (!isSaveLeft && iPos2<=(iPos1+nToken)) || iPos2<=iPos1 ){ |
| 1661 if( (*p2&0xFE)==0 ) break; |
| 1662 fts3GetDeltaVarint(&p2, &iPos2); iPos2 -= 2; |
| 1663 }else{ |
| 1664 if( (*p1&0xFE)==0 ) break; |
| 1665 fts3GetDeltaVarint(&p1, &iPos1); iPos1 -= 2; |
| 1666 } |
| 1667 } |
| 1668 |
| 1669 if( pSave ){ |
| 1670 assert( pp && p ); |
| 1671 p = pSave; |
| 1672 } |
| 1673 |
| 1674 fts3ColumnlistCopy(0, &p1); |
| 1675 fts3ColumnlistCopy(0, &p2); |
| 1676 assert( (*p1&0xFE)==0 && (*p2&0xFE)==0 ); |
| 1677 if( 0==*p1 || 0==*p2 ) break; |
| 1678 |
| 1679 p1++; |
| 1680 p1 += sqlite3Fts3GetVarint32(p1, &iCol1); |
| 1681 p2++; |
| 1682 p2 += sqlite3Fts3GetVarint32(p2, &iCol2); |
| 1683 } |
| 1684 |
| 1685 /* Advance pointer p1 or p2 (whichever corresponds to the smaller of |
| 1686 ** iCol1 and iCol2) so that it points to either the 0x00 that marks the |
| 1687 ** end of the position list, or the 0x01 that precedes the next |
| 1688 ** column-number in the position list. |
| 1689 */ |
| 1690 else if( iCol1<iCol2 ){ |
| 1691 fts3ColumnlistCopy(0, &p1); |
| 1692 if( 0==*p1 ) break; |
| 1693 p1++; |
| 1694 p1 += sqlite3Fts3GetVarint32(p1, &iCol1); |
| 1695 }else{ |
| 1696 fts3ColumnlistCopy(0, &p2); |
| 1697 if( 0==*p2 ) break; |
| 1698 p2++; |
| 1699 p2 += sqlite3Fts3GetVarint32(p2, &iCol2); |
| 1700 } |
| 1701 } |
| 1702 |
| 1703 fts3PoslistCopy(0, &p2); |
| 1704 fts3PoslistCopy(0, &p1); |
| 1705 *pp1 = p1; |
| 1706 *pp2 = p2; |
| 1707 if( !pp || *pp==p ){ |
| 1708 return 0; |
| 1709 } |
| 1710 *p++ = 0x00; |
| 1711 *pp = p; |
| 1712 return 1; |
| 1713 } |
| 1714 |
| 1715 /* |
| 1716 ** Merge two position-lists as required by the NEAR operator. |
| 1717 */ |
| 1718 static int fts3PoslistNearMerge( |
| 1719 char **pp, /* Output buffer */ |
| 1720 char *aTmp, /* Temporary buffer space */ |
| 1721 int nRight, /* Maximum difference in token positions */ |
| 1722 int nLeft, /* Maximum difference in token positions */ |
| 1723 char **pp1, /* IN/OUT: Left input list */ |
| 1724 char **pp2 /* IN/OUT: Right input list */ |
| 1725 ){ |
| 1726 char *p1 = *pp1; |
| 1727 char *p2 = *pp2; |
| 1728 |
| 1729 if( !pp ){ |
| 1730 if( fts3PoslistPhraseMerge(0, nRight, 0, 0, pp1, pp2) ) return 1; |
| 1731 *pp1 = p1; |
| 1732 *pp2 = p2; |
| 1733 return fts3PoslistPhraseMerge(0, nLeft, 0, 0, pp2, pp1); |
| 1734 }else{ |
| 1735 char *pTmp1 = aTmp; |
| 1736 char *pTmp2; |
| 1737 char *aTmp2; |
| 1738 int res = 1; |
| 1739 |
| 1740 fts3PoslistPhraseMerge(&pTmp1, nRight, 0, 0, pp1, pp2); |
| 1741 aTmp2 = pTmp2 = pTmp1; |
| 1742 *pp1 = p1; |
| 1743 *pp2 = p2; |
| 1744 fts3PoslistPhraseMerge(&pTmp2, nLeft, 1, 0, pp2, pp1); |
| 1745 if( pTmp1!=aTmp && pTmp2!=aTmp2 ){ |
| 1746 fts3PoslistMerge(pp, &aTmp, &aTmp2); |
| 1747 }else if( pTmp1!=aTmp ){ |
| 1748 fts3PoslistCopy(pp, &aTmp); |
| 1749 }else if( pTmp2!=aTmp2 ){ |
| 1750 fts3PoslistCopy(pp, &aTmp2); |
| 1751 }else{ |
| 1752 res = 0; |
| 1753 } |
| 1754 |
| 1755 return res; |
| 1756 } |
| 1757 } |
| 1758 |
| 1759 /* |
| 1760 ** Values that may be used as the first parameter to fts3DoclistMerge(). |
| 1761 */ |
| 1762 #define MERGE_NOT 2 /* D + D -> D */ |
| 1763 #define MERGE_AND 3 /* D + D -> D */ |
| 1764 #define MERGE_OR 4 /* D + D -> D */ |
| 1765 #define MERGE_POS_OR 5 /* P + P -> P */ |
| 1766 #define MERGE_PHRASE 6 /* P + P -> D */ |
| 1767 #define MERGE_POS_PHRASE 7 /* P + P -> P */ |
| 1768 #define MERGE_NEAR 8 /* P + P -> D */ |
| 1769 #define MERGE_POS_NEAR 9 /* P + P -> P */ |
| 1770 |
| 1771 /* |
| 1772 ** Merge the two doclists passed in buffer a1 (size n1 bytes) and a2 |
| 1773 ** (size n2 bytes). The output is written to pre-allocated buffer aBuffer, |
| 1774 ** which is guaranteed to be large enough to hold the results. The number |
| 1775 ** of bytes written to aBuffer is stored in *pnBuffer before returning. |
| 1776 ** |
| 1777 ** If successful, SQLITE_OK is returned. Otherwise, if a malloc error |
| 1778 ** occurs while allocating a temporary buffer as part of the merge operation, |
| 1779 ** SQLITE_NOMEM is returned. |
| 1780 */ |
| 1781 static int fts3DoclistMerge( |
| 1782 int mergetype, /* One of the MERGE_XXX constants */ |
| 1783 int nParam1, /* Used by MERGE_NEAR and MERGE_POS_NEAR */ |
| 1784 int nParam2, /* Used by MERGE_NEAR and MERGE_POS_NEAR */ |
| 1785 char *aBuffer, /* Pre-allocated output buffer */ |
| 1786 int *pnBuffer, /* OUT: Bytes written to aBuffer */ |
| 1787 char *a1, /* Buffer containing first doclist */ |
| 1788 int n1, /* Size of buffer a1 */ |
| 1789 char *a2, /* Buffer containing second doclist */ |
| 1790 int n2, /* Size of buffer a2 */ |
| 1791 int *pnDoc /* OUT: Number of docids in output */ |
| 1792 ){ |
| 1793 sqlite3_int64 i1 = 0; |
| 1794 sqlite3_int64 i2 = 0; |
| 1795 sqlite3_int64 iPrev = 0; |
| 1796 |
| 1797 char *p = aBuffer; |
| 1798 char *p1 = a1; |
| 1799 char *p2 = a2; |
| 1800 char *pEnd1 = &a1[n1]; |
| 1801 char *pEnd2 = &a2[n2]; |
| 1802 int nDoc = 0; |
| 1803 |
| 1804 assert( mergetype==MERGE_OR || mergetype==MERGE_POS_OR |
| 1805 || mergetype==MERGE_AND || mergetype==MERGE_NOT |
| 1806 || mergetype==MERGE_PHRASE || mergetype==MERGE_POS_PHRASE |
| 1807 || mergetype==MERGE_NEAR || mergetype==MERGE_POS_NEAR |
| 1808 ); |
| 1809 |
| 1810 if( !aBuffer ){ |
| 1811 *pnBuffer = 0; |
| 1812 return SQLITE_NOMEM; |
| 1813 } |
| 1814 |
| 1815 /* Read the first docid from each doclist */ |
| 1816 fts3GetDeltaVarint2(&p1, pEnd1, &i1); |
| 1817 fts3GetDeltaVarint2(&p2, pEnd2, &i2); |
| 1818 |
| 1819 switch( mergetype ){ |
| 1820 case MERGE_OR: |
| 1821 case MERGE_POS_OR: |
| 1822 while( p1 || p2 ){ |
| 1823 if( p2 && p1 && i1==i2 ){ |
| 1824 fts3PutDeltaVarint(&p, &iPrev, i1); |
| 1825 if( mergetype==MERGE_POS_OR ) fts3PoslistMerge(&p, &p1, &p2); |
| 1826 fts3GetDeltaVarint2(&p1, pEnd1, &i1); |
| 1827 fts3GetDeltaVarint2(&p2, pEnd2, &i2); |
| 1828 }else if( !p2 || (p1 && i1<i2) ){ |
| 1829 fts3PutDeltaVarint(&p, &iPrev, i1); |
| 1830 if( mergetype==MERGE_POS_OR ) fts3PoslistCopy(&p, &p1); |
| 1831 fts3GetDeltaVarint2(&p1, pEnd1, &i1); |
| 1832 }else{ |
| 1833 fts3PutDeltaVarint(&p, &iPrev, i2); |
| 1834 if( mergetype==MERGE_POS_OR ) fts3PoslistCopy(&p, &p2); |
| 1835 fts3GetDeltaVarint2(&p2, pEnd2, &i2); |
| 1836 } |
| 1837 } |
| 1838 break; |
| 1839 |
| 1840 case MERGE_AND: |
| 1841 while( p1 && p2 ){ |
| 1842 if( i1==i2 ){ |
| 1843 fts3PutDeltaVarint(&p, &iPrev, i1); |
| 1844 fts3GetDeltaVarint2(&p1, pEnd1, &i1); |
| 1845 fts3GetDeltaVarint2(&p2, pEnd2, &i2); |
| 1846 nDoc++; |
| 1847 }else if( i1<i2 ){ |
| 1848 fts3GetDeltaVarint2(&p1, pEnd1, &i1); |
| 1849 }else{ |
| 1850 fts3GetDeltaVarint2(&p2, pEnd2, &i2); |
| 1851 } |
| 1852 } |
| 1853 break; |
| 1854 |
| 1855 case MERGE_NOT: |
| 1856 while( p1 ){ |
| 1857 if( p2 && i1==i2 ){ |
| 1858 fts3GetDeltaVarint2(&p1, pEnd1, &i1); |
| 1859 fts3GetDeltaVarint2(&p2, pEnd2, &i2); |
| 1860 }else if( !p2 || i1<i2 ){ |
| 1861 fts3PutDeltaVarint(&p, &iPrev, i1); |
| 1862 fts3GetDeltaVarint2(&p1, pEnd1, &i1); |
| 1863 }else{ |
| 1864 fts3GetDeltaVarint2(&p2, pEnd2, &i2); |
| 1865 } |
| 1866 } |
| 1867 break; |
| 1868 |
| 1869 case MERGE_POS_PHRASE: |
| 1870 case MERGE_PHRASE: { |
| 1871 char **ppPos = (mergetype==MERGE_PHRASE ? 0 : &p); |
| 1872 while( p1 && p2 ){ |
| 1873 if( i1==i2 ){ |
| 1874 char *pSave = p; |
| 1875 sqlite3_int64 iPrevSave = iPrev; |
| 1876 fts3PutDeltaVarint(&p, &iPrev, i1); |
| 1877 if( 0==fts3PoslistPhraseMerge(ppPos, nParam1, 0, 1, &p1, &p2) ){ |
| 1878 p = pSave; |
| 1879 iPrev = iPrevSave; |
| 1880 }else{ |
| 1881 nDoc++; |
| 1882 } |
| 1883 fts3GetDeltaVarint2(&p1, pEnd1, &i1); |
| 1884 fts3GetDeltaVarint2(&p2, pEnd2, &i2); |
| 1885 }else if( i1<i2 ){ |
| 1886 fts3PoslistCopy(0, &p1); |
| 1887 fts3GetDeltaVarint2(&p1, pEnd1, &i1); |
| 1888 }else{ |
| 1889 fts3PoslistCopy(0, &p2); |
| 1890 fts3GetDeltaVarint2(&p2, pEnd2, &i2); |
| 1891 } |
| 1892 } |
| 1893 break; |
| 1894 } |
| 1895 |
| 1896 default: assert( mergetype==MERGE_POS_NEAR || mergetype==MERGE_NEAR ); { |
| 1897 char *aTmp = 0; |
| 1898 char **ppPos = 0; |
| 1899 |
| 1900 if( mergetype==MERGE_POS_NEAR ){ |
| 1901 ppPos = &p; |
| 1902 aTmp = sqlite3_malloc(2*(n1+n2+1)); |
| 1903 if( !aTmp ){ |
| 1904 return SQLITE_NOMEM; |
| 1905 } |
| 1906 } |
| 1907 |
| 1908 while( p1 && p2 ){ |
| 1909 if( i1==i2 ){ |
| 1910 char *pSave = p; |
| 1911 sqlite3_int64 iPrevSave = iPrev; |
| 1912 fts3PutDeltaVarint(&p, &iPrev, i1); |
| 1913 |
| 1914 if( !fts3PoslistNearMerge(ppPos, aTmp, nParam1, nParam2, &p1, &p2) ){ |
| 1915 iPrev = iPrevSave; |
| 1916 p = pSave; |
| 1917 } |
| 1918 |
| 1919 fts3GetDeltaVarint2(&p1, pEnd1, &i1); |
| 1920 fts3GetDeltaVarint2(&p2, pEnd2, &i2); |
| 1921 }else if( i1<i2 ){ |
| 1922 fts3PoslistCopy(0, &p1); |
| 1923 fts3GetDeltaVarint2(&p1, pEnd1, &i1); |
| 1924 }else{ |
| 1925 fts3PoslistCopy(0, &p2); |
| 1926 fts3GetDeltaVarint2(&p2, pEnd2, &i2); |
| 1927 } |
| 1928 } |
| 1929 sqlite3_free(aTmp); |
| 1930 break; |
| 1931 } |
| 1932 } |
| 1933 |
| 1934 if( pnDoc ) *pnDoc = nDoc; |
| 1935 *pnBuffer = (int)(p-aBuffer); |
| 1936 return SQLITE_OK; |
| 1937 } |
| 1938 |
| 1939 /* |
| 1940 ** A pointer to an instance of this structure is used as the context |
| 1941 ** argument to sqlite3Fts3SegReaderIterate() |
| 1942 */ |
| 1943 typedef struct TermSelect TermSelect; |
| 1944 struct TermSelect { |
| 1945 int isReqPos; |
| 1946 char *aaOutput[16]; /* Malloc'd output buffer */ |
| 1947 int anOutput[16]; /* Size of output in bytes */ |
| 1948 }; |
| 1949 |
| 1950 /* |
| 1951 ** Merge all doclists in the TermSelect.aaOutput[] array into a single |
| 1952 ** doclist stored in TermSelect.aaOutput[0]. If successful, delete all |
| 1953 ** other doclists (except the aaOutput[0] one) and return SQLITE_OK. |
| 1954 ** |
| 1955 ** If an OOM error occurs, return SQLITE_NOMEM. In this case it is |
| 1956 ** the responsibility of the caller to free any doclists left in the |
| 1957 ** TermSelect.aaOutput[] array. |
| 1958 */ |
| 1959 static int fts3TermSelectMerge(TermSelect *pTS){ |
| 1960 int mergetype = (pTS->isReqPos ? MERGE_POS_OR : MERGE_OR); |
| 1961 char *aOut = 0; |
| 1962 int nOut = 0; |
| 1963 int i; |
| 1964 |
| 1965 /* Loop through the doclists in the aaOutput[] array. Merge them all |
| 1966 ** into a single doclist. |
| 1967 */ |
| 1968 for(i=0; i<SizeofArray(pTS->aaOutput); i++){ |
| 1969 if( pTS->aaOutput[i] ){ |
| 1970 if( !aOut ){ |
| 1971 aOut = pTS->aaOutput[i]; |
| 1972 nOut = pTS->anOutput[i]; |
| 1973 pTS->aaOutput[i] = 0; |
| 1974 }else{ |
| 1975 int nNew = nOut + pTS->anOutput[i]; |
| 1976 char *aNew = sqlite3_malloc(nNew); |
| 1977 if( !aNew ){ |
| 1978 sqlite3_free(aOut); |
| 1979 return SQLITE_NOMEM; |
| 1980 } |
| 1981 fts3DoclistMerge(mergetype, 0, 0, |
| 1982 aNew, &nNew, pTS->aaOutput[i], pTS->anOutput[i], aOut, nOut, 0 |
| 1983 ); |
| 1984 sqlite3_free(pTS->aaOutput[i]); |
| 1985 sqlite3_free(aOut); |
| 1986 pTS->aaOutput[i] = 0; |
| 1987 aOut = aNew; |
| 1988 nOut = nNew; |
| 1989 } |
| 1990 } |
| 1991 } |
| 1992 |
| 1993 pTS->aaOutput[0] = aOut; |
| 1994 pTS->anOutput[0] = nOut; |
| 1995 return SQLITE_OK; |
| 1996 } |
| 1997 |
| 1998 /* |
| 1999 ** This function is used as the sqlite3Fts3SegReaderIterate() callback when |
| 2000 ** querying the full-text index for a doclist associated with a term or |
| 2001 ** term-prefix. |
| 2002 */ |
| 2003 static int fts3TermSelectCb( |
| 2004 Fts3Table *p, /* Virtual table object */ |
| 2005 void *pContext, /* Pointer to TermSelect structure */ |
| 2006 char *zTerm, |
| 2007 int nTerm, |
| 2008 char *aDoclist, |
| 2009 int nDoclist |
| 2010 ){ |
| 2011 TermSelect *pTS = (TermSelect *)pContext; |
| 2012 |
| 2013 UNUSED_PARAMETER(p); |
| 2014 UNUSED_PARAMETER(zTerm); |
| 2015 UNUSED_PARAMETER(nTerm); |
| 2016 |
| 2017 if( pTS->aaOutput[0]==0 ){ |
| 2018 /* If this is the first term selected, copy the doclist to the output |
| 2019 ** buffer using memcpy(). TODO: Add a way to transfer control of the |
| 2020 ** aDoclist buffer from the caller so as to avoid the memcpy(). |
| 2021 */ |
| 2022 pTS->aaOutput[0] = sqlite3_malloc(nDoclist); |
| 2023 pTS->anOutput[0] = nDoclist; |
| 2024 if( pTS->aaOutput[0] ){ |
| 2025 memcpy(pTS->aaOutput[0], aDoclist, nDoclist); |
| 2026 }else{ |
| 2027 return SQLITE_NOMEM; |
| 2028 } |
| 2029 }else{ |
| 2030 int mergetype = (pTS->isReqPos ? MERGE_POS_OR : MERGE_OR); |
| 2031 char *aMerge = aDoclist; |
| 2032 int nMerge = nDoclist; |
| 2033 int iOut; |
| 2034 |
| 2035 for(iOut=0; iOut<SizeofArray(pTS->aaOutput); iOut++){ |
| 2036 char *aNew; |
| 2037 int nNew; |
| 2038 if( pTS->aaOutput[iOut]==0 ){ |
| 2039 assert( iOut>0 ); |
| 2040 pTS->aaOutput[iOut] = aMerge; |
| 2041 pTS->anOutput[iOut] = nMerge; |
2810 break; | 2042 break; |
2811 } | 2043 } |
| 2044 |
| 2045 nNew = nMerge + pTS->anOutput[iOut]; |
| 2046 aNew = sqlite3_malloc(nNew); |
| 2047 if( !aNew ){ |
| 2048 if( aMerge!=aDoclist ){ |
| 2049 sqlite3_free(aMerge); |
| 2050 } |
| 2051 return SQLITE_NOMEM; |
| 2052 } |
| 2053 fts3DoclistMerge(mergetype, 0, 0, aNew, &nNew, |
| 2054 pTS->aaOutput[iOut], pTS->anOutput[iOut], aMerge, nMerge, 0 |
| 2055 ); |
| 2056 |
| 2057 if( iOut>0 ) sqlite3_free(aMerge); |
| 2058 sqlite3_free(pTS->aaOutput[iOut]); |
| 2059 pTS->aaOutput[iOut] = 0; |
| 2060 |
| 2061 aMerge = aNew; |
| 2062 nMerge = nNew; |
| 2063 if( (iOut+1)==SizeofArray(pTS->aaOutput) ){ |
| 2064 pTS->aaOutput[iOut] = aMerge; |
| 2065 pTS->anOutput[iOut] = nMerge; |
| 2066 } |
| 2067 } |
| 2068 } |
| 2069 return SQLITE_OK; |
| 2070 } |
| 2071 |
| 2072 static int fts3DeferredTermSelect( |
| 2073 Fts3DeferredToken *pToken, /* Phrase token */ |
| 2074 int isTermPos, /* True to include positions */ |
| 2075 int *pnOut, /* OUT: Size of list */ |
| 2076 char **ppOut /* OUT: Body of list */ |
| 2077 ){ |
| 2078 char *aSource; |
| 2079 int nSource; |
| 2080 |
| 2081 aSource = sqlite3Fts3DeferredDoclist(pToken, &nSource); |
| 2082 if( !aSource ){ |
| 2083 *pnOut = 0; |
| 2084 *ppOut = 0; |
| 2085 }else if( isTermPos ){ |
| 2086 *ppOut = sqlite3_malloc(nSource); |
| 2087 if( !*ppOut ) return SQLITE_NOMEM; |
| 2088 memcpy(*ppOut, aSource, nSource); |
| 2089 *pnOut = nSource; |
| 2090 }else{ |
| 2091 sqlite3_int64 docid; |
| 2092 *pnOut = sqlite3Fts3GetVarint(aSource, &docid); |
| 2093 *ppOut = sqlite3_malloc(*pnOut); |
| 2094 if( !*ppOut ) return SQLITE_NOMEM; |
| 2095 sqlite3Fts3PutVarint(*ppOut, docid); |
| 2096 } |
| 2097 |
| 2098 return SQLITE_OK; |
| 2099 } |
| 2100 |
| 2101 int sqlite3Fts3SegReaderCursor( |
| 2102 Fts3Table *p, /* FTS3 table handle */ |
| 2103 int iLevel, /* Level of segments to scan */ |
| 2104 const char *zTerm, /* Term to query for */ |
| 2105 int nTerm, /* Size of zTerm in bytes */ |
| 2106 int isPrefix, /* True for a prefix search */ |
| 2107 int isScan, /* True to scan from zTerm to EOF */ |
| 2108 Fts3SegReaderCursor *pCsr /* Cursor object to populate */ |
| 2109 ){ |
| 2110 int rc = SQLITE_OK; |
| 2111 int rc2; |
| 2112 int iAge = 0; |
| 2113 sqlite3_stmt *pStmt = 0; |
| 2114 Fts3SegReader *pPending = 0; |
| 2115 |
| 2116 assert( iLevel==FTS3_SEGCURSOR_ALL |
| 2117 || iLevel==FTS3_SEGCURSOR_PENDING |
| 2118 || iLevel>=0 |
| 2119 ); |
| 2120 assert( FTS3_SEGCURSOR_PENDING<0 ); |
| 2121 assert( FTS3_SEGCURSOR_ALL<0 ); |
| 2122 assert( iLevel==FTS3_SEGCURSOR_ALL || (zTerm==0 && isPrefix==1) ); |
| 2123 assert( isPrefix==0 || isScan==0 ); |
| 2124 |
| 2125 |
| 2126 memset(pCsr, 0, sizeof(Fts3SegReaderCursor)); |
| 2127 |
| 2128 /* If iLevel is less than 0, include a seg-reader for the pending-terms. */ |
| 2129 assert( isScan==0 || fts3HashCount(&p->pendingTerms)==0 ); |
| 2130 if( iLevel<0 && isScan==0 ){ |
| 2131 rc = sqlite3Fts3SegReaderPending(p, zTerm, nTerm, isPrefix, &pPending); |
| 2132 if( rc==SQLITE_OK && pPending ){ |
| 2133 int nByte = (sizeof(Fts3SegReader *) * 16); |
| 2134 pCsr->apSegment = (Fts3SegReader **)sqlite3_malloc(nByte); |
| 2135 if( pCsr->apSegment==0 ){ |
| 2136 rc = SQLITE_NOMEM; |
| 2137 }else{ |
| 2138 pCsr->apSegment[0] = pPending; |
| 2139 pCsr->nSegment = 1; |
| 2140 pPending = 0; |
| 2141 } |
| 2142 } |
| 2143 } |
| 2144 |
| 2145 if( iLevel!=FTS3_SEGCURSOR_PENDING ){ |
| 2146 if( rc==SQLITE_OK ){ |
| 2147 rc = sqlite3Fts3AllSegdirs(p, iLevel, &pStmt); |
| 2148 } |
| 2149 while( rc==SQLITE_OK && SQLITE_ROW==(rc = sqlite3_step(pStmt)) ){ |
| 2150 |
| 2151 /* Read the values returned by the SELECT into local variables. */ |
| 2152 sqlite3_int64 iStartBlock = sqlite3_column_int64(pStmt, 1); |
| 2153 sqlite3_int64 iLeavesEndBlock = sqlite3_column_int64(pStmt, 2); |
| 2154 sqlite3_int64 iEndBlock = sqlite3_column_int64(pStmt, 3); |
| 2155 int nRoot = sqlite3_column_bytes(pStmt, 4); |
| 2156 char const *zRoot = sqlite3_column_blob(pStmt, 4); |
| 2157 |
| 2158 /* If nSegment is a multiple of 16 the array needs to be extended. */ |
| 2159 if( (pCsr->nSegment%16)==0 ){ |
| 2160 Fts3SegReader **apNew; |
| 2161 int nByte = (pCsr->nSegment + 16)*sizeof(Fts3SegReader*); |
| 2162 apNew = (Fts3SegReader **)sqlite3_realloc(pCsr->apSegment, nByte); |
| 2163 if( !apNew ){ |
| 2164 rc = SQLITE_NOMEM; |
| 2165 goto finished; |
| 2166 } |
| 2167 pCsr->apSegment = apNew; |
| 2168 } |
| 2169 |
| 2170 /* If zTerm is not NULL, and this segment is not stored entirely on its |
| 2171 ** root node, the range of leaves scanned can be reduced. Do this. */ |
| 2172 if( iStartBlock && zTerm ){ |
| 2173 sqlite3_int64 *pi = (isPrefix ? &iLeavesEndBlock : 0); |
| 2174 rc = fts3SelectLeaf(p, zTerm, nTerm, zRoot, nRoot, &iStartBlock, pi); |
| 2175 if( rc!=SQLITE_OK ) goto finished; |
| 2176 if( isPrefix==0 && isScan==0 ) iLeavesEndBlock = iStartBlock; |
| 2177 } |
| 2178 |
| 2179 rc = sqlite3Fts3SegReaderNew(iAge, iStartBlock, iLeavesEndBlock, |
| 2180 iEndBlock, zRoot, nRoot, &pCsr->apSegment[pCsr->nSegment] |
| 2181 ); |
| 2182 if( rc!=SQLITE_OK ) goto finished; |
| 2183 pCsr->nSegment++; |
| 2184 iAge++; |
| 2185 } |
| 2186 } |
| 2187 |
| 2188 finished: |
| 2189 rc2 = sqlite3_reset(pStmt); |
| 2190 if( rc==SQLITE_DONE ) rc = rc2; |
| 2191 sqlite3Fts3SegReaderFree(pPending); |
| 2192 |
| 2193 return rc; |
| 2194 } |
| 2195 |
| 2196 |
| 2197 static int fts3TermSegReaderCursor( |
| 2198 Fts3Cursor *pCsr, /* Virtual table cursor handle */ |
| 2199 const char *zTerm, /* Term to query for */ |
| 2200 int nTerm, /* Size of zTerm in bytes */ |
| 2201 int isPrefix, /* True for a prefix search */ |
| 2202 Fts3SegReaderCursor **ppSegcsr /* OUT: Allocated seg-reader cursor */ |
| 2203 ){ |
| 2204 Fts3SegReaderCursor *pSegcsr; /* Object to allocate and return */ |
| 2205 int rc = SQLITE_NOMEM; /* Return code */ |
| 2206 |
| 2207 pSegcsr = sqlite3_malloc(sizeof(Fts3SegReaderCursor)); |
| 2208 if( pSegcsr ){ |
| 2209 Fts3Table *p = (Fts3Table *)pCsr->base.pVtab; |
| 2210 int i; |
| 2211 int nCost = 0; |
| 2212 rc = sqlite3Fts3SegReaderCursor( |
| 2213 p, FTS3_SEGCURSOR_ALL, zTerm, nTerm, isPrefix, 0, pSegcsr); |
| 2214 |
| 2215 for(i=0; rc==SQLITE_OK && i<pSegcsr->nSegment; i++){ |
| 2216 rc = sqlite3Fts3SegReaderCost(pCsr, pSegcsr->apSegment[i], &nCost); |
| 2217 } |
| 2218 pSegcsr->nCost = nCost; |
| 2219 } |
| 2220 |
| 2221 *ppSegcsr = pSegcsr; |
| 2222 return rc; |
| 2223 } |
| 2224 |
| 2225 static void fts3SegReaderCursorFree(Fts3SegReaderCursor *pSegcsr){ |
| 2226 sqlite3Fts3SegReaderFinish(pSegcsr); |
| 2227 sqlite3_free(pSegcsr); |
| 2228 } |
| 2229 |
| 2230 /* |
| 2231 ** This function retreives the doclist for the specified term (or term |
| 2232 ** prefix) from the database. |
| 2233 ** |
| 2234 ** The returned doclist may be in one of two formats, depending on the |
| 2235 ** value of parameter isReqPos. If isReqPos is zero, then the doclist is |
| 2236 ** a sorted list of delta-compressed docids (a bare doclist). If isReqPos |
| 2237 ** is non-zero, then the returned list is in the same format as is stored |
| 2238 ** in the database without the found length specifier at the start of on-disk |
| 2239 ** doclists. |
| 2240 */ |
| 2241 static int fts3TermSelect( |
| 2242 Fts3Table *p, /* Virtual table handle */ |
| 2243 Fts3PhraseToken *pTok, /* Token to query for */ |
| 2244 int iColumn, /* Column to query (or -ve for all columns) */ |
| 2245 int isReqPos, /* True to include position lists in output */ |
| 2246 int *pnOut, /* OUT: Size of buffer at *ppOut */ |
| 2247 char **ppOut /* OUT: Malloced result buffer */ |
| 2248 ){ |
| 2249 int rc; /* Return code */ |
| 2250 Fts3SegReaderCursor *pSegcsr; /* Seg-reader cursor for this term */ |
| 2251 TermSelect tsc; /* Context object for fts3TermSelectCb() */ |
| 2252 Fts3SegFilter filter; /* Segment term filter configuration */ |
| 2253 |
| 2254 pSegcsr = pTok->pSegcsr; |
| 2255 memset(&tsc, 0, sizeof(TermSelect)); |
| 2256 tsc.isReqPos = isReqPos; |
| 2257 |
| 2258 filter.flags = FTS3_SEGMENT_IGNORE_EMPTY |
| 2259 | (pTok->isPrefix ? FTS3_SEGMENT_PREFIX : 0) |
| 2260 | (isReqPos ? FTS3_SEGMENT_REQUIRE_POS : 0) |
| 2261 | (iColumn<p->nColumn ? FTS3_SEGMENT_COLUMN_FILTER : 0); |
| 2262 filter.iCol = iColumn; |
| 2263 filter.zTerm = pTok->z; |
| 2264 filter.nTerm = pTok->n; |
| 2265 |
| 2266 rc = sqlite3Fts3SegReaderStart(p, pSegcsr, &filter); |
| 2267 while( SQLITE_OK==rc |
| 2268 && SQLITE_ROW==(rc = sqlite3Fts3SegReaderStep(p, pSegcsr)) |
| 2269 ){ |
| 2270 rc = fts3TermSelectCb(p, (void *)&tsc, |
| 2271 pSegcsr->zTerm, pSegcsr->nTerm, pSegcsr->aDoclist, pSegcsr->nDoclist |
| 2272 ); |
| 2273 } |
| 2274 |
| 2275 if( rc==SQLITE_OK ){ |
| 2276 rc = fts3TermSelectMerge(&tsc); |
| 2277 } |
| 2278 if( rc==SQLITE_OK ){ |
| 2279 *ppOut = tsc.aaOutput[0]; |
| 2280 *pnOut = tsc.anOutput[0]; |
| 2281 }else{ |
| 2282 int i; |
| 2283 for(i=0; i<SizeofArray(tsc.aaOutput); i++){ |
| 2284 sqlite3_free(tsc.aaOutput[i]); |
| 2285 } |
| 2286 } |
| 2287 |
| 2288 fts3SegReaderCursorFree(pSegcsr); |
| 2289 pTok->pSegcsr = 0; |
| 2290 return rc; |
| 2291 } |
| 2292 |
| 2293 /* |
| 2294 ** This function counts the total number of docids in the doclist stored |
| 2295 ** in buffer aList[], size nList bytes. |
| 2296 ** |
| 2297 ** If the isPoslist argument is true, then it is assumed that the doclist |
| 2298 ** contains a position-list following each docid. Otherwise, it is assumed |
| 2299 ** that the doclist is simply a list of docids stored as delta encoded |
| 2300 ** varints. |
| 2301 */ |
| 2302 static int fts3DoclistCountDocids(int isPoslist, char *aList, int nList){ |
| 2303 int nDoc = 0; /* Return value */ |
| 2304 if( aList ){ |
| 2305 char *aEnd = &aList[nList]; /* Pointer to one byte after EOF */ |
| 2306 char *p = aList; /* Cursor */ |
| 2307 if( !isPoslist ){ |
| 2308 /* The number of docids in the list is the same as the number of |
| 2309 ** varints. In FTS3 a varint consists of a single byte with the 0x80 |
| 2310 ** bit cleared and zero or more bytes with the 0x80 bit set. So to |
| 2311 ** count the varints in the buffer, just count the number of bytes |
| 2312 ** with the 0x80 bit clear. */ |
| 2313 while( p<aEnd ) nDoc += (((*p++)&0x80)==0); |
2812 }else{ | 2314 }else{ |
2813 z[j++] = z[i]; | 2315 while( p<aEnd ){ |
2814 } | 2316 nDoc++; |
2815 } | 2317 while( (*p++)&0x80 ); /* Skip docid varint */ |
2816 } | 2318 fts3PoslistCopy(0, &p); /* Skip over position list */ |
2817 | 2319 } |
2818 /* | 2320 } |
2819 ** The input azIn is a NULL-terminated list of tokens. Remove the first | 2321 } |
2820 ** token and all punctuation tokens. Remove the quotes from | 2322 |
2821 ** around string literal tokens. | 2323 return nDoc; |
2822 ** | 2324 } |
2823 ** Example: | 2325 |
2824 ** | 2326 /* |
2825 ** input: tokenize chinese ( 'simplifed' , 'mixed' ) | 2327 ** Call sqlite3Fts3DeferToken() for each token in the expression pExpr. |
2826 ** output: chinese simplifed mixed | 2328 */ |
2827 ** | 2329 static int fts3DeferExpression(Fts3Cursor *pCsr, Fts3Expr *pExpr){ |
2828 ** Another example: | 2330 int rc = SQLITE_OK; |
2829 ** | 2331 if( pExpr ){ |
2830 ** input: delimiters ( '[' , ']' , '...' ) | 2332 rc = fts3DeferExpression(pCsr, pExpr->pLeft); |
2831 ** output: [ ] ... | 2333 if( rc==SQLITE_OK ){ |
2832 */ | 2334 rc = fts3DeferExpression(pCsr, pExpr->pRight); |
2833 static void tokenListToIdList(char **azIn){ | 2335 } |
2834 int i, j; | 2336 if( pExpr->eType==FTSQUERY_PHRASE ){ |
2835 if( azIn ){ | 2337 int iCol = pExpr->pPhrase->iColumn; |
2836 for(i=0, j=-1; azIn[i]; i++){ | 2338 int i; |
2837 if( safe_isalnum(azIn[i][0]) || azIn[i][1] ){ | 2339 for(i=0; rc==SQLITE_OK && i<pExpr->pPhrase->nToken; i++){ |
2838 dequoteString(azIn[i]); | 2340 Fts3PhraseToken *pToken = &pExpr->pPhrase->aToken[i]; |
2839 if( j>=0 ){ | 2341 if( pToken->pDeferred==0 ){ |
2840 azIn[j] = azIn[i]; | 2342 rc = sqlite3Fts3DeferToken(pCsr, pToken, iCol); |
2841 } | 2343 } |
2842 j++; | 2344 } |
2843 } | 2345 } |
2844 } | 2346 } |
2845 azIn[j] = 0; | 2347 return rc; |
2846 } | 2348 } |
2847 } | 2349 |
2848 | 2350 /* |
2849 | 2351 ** This function removes the position information from a doclist. When |
2850 /* | 2352 ** called, buffer aList (size *pnList bytes) contains a doclist that includes |
2851 ** Find the first alphanumeric token in the string zIn. Null-terminate | 2353 ** position information. This function removes the position information so |
2852 ** this token. Remove any quotation marks. And return a pointer to | 2354 ** that aList contains only docids, and adjusts *pnList to reflect the new |
2853 ** the result. | 2355 ** (possibly reduced) size of the doclist. |
2854 */ | 2356 */ |
2855 static char *firstToken(char *zIn, char **pzTail){ | 2357 static void fts3DoclistStripPositions( |
2856 int n, ttype; | 2358 char *aList, /* IN/OUT: Buffer containing doclist */ |
2857 while(1){ | 2359 int *pnList /* IN/OUT: Size of doclist in bytes */ |
2858 n = ftsGetToken(zIn, &ttype); | 2360 ){ |
2859 if( ttype==TOKEN_SPACE ){ | 2361 if( aList ){ |
2860 zIn += n; | 2362 char *aEnd = &aList[*pnList]; /* Pointer to one byte after EOF */ |
2861 }else if( ttype==TOKEN_EOF ){ | 2363 char *p = aList; /* Input cursor */ |
2862 *pzTail = zIn; | 2364 char *pOut = aList; /* Output cursor */ |
2863 return 0; | 2365 |
| 2366 while( p<aEnd ){ |
| 2367 sqlite3_int64 delta; |
| 2368 p += sqlite3Fts3GetVarint(p, &delta); |
| 2369 fts3PoslistCopy(0, &p); |
| 2370 pOut += sqlite3Fts3PutVarint(pOut, delta); |
| 2371 } |
| 2372 |
| 2373 *pnList = (int)(pOut - aList); |
| 2374 } |
| 2375 } |
| 2376 |
| 2377 /* |
| 2378 ** Return a DocList corresponding to the phrase *pPhrase. |
| 2379 ** |
| 2380 ** If this function returns SQLITE_OK, but *pnOut is set to a negative value, |
| 2381 ** then no tokens in the phrase were looked up in the full-text index. This |
| 2382 ** is only possible when this function is called from within xFilter(). The |
| 2383 ** caller should assume that all documents match the phrase. The actual |
| 2384 ** filtering will take place in xNext(). |
| 2385 */ |
| 2386 static int fts3PhraseSelect( |
| 2387 Fts3Cursor *pCsr, /* Virtual table cursor handle */ |
| 2388 Fts3Phrase *pPhrase, /* Phrase to return a doclist for */ |
| 2389 int isReqPos, /* True if output should contain positions */ |
| 2390 char **paOut, /* OUT: Pointer to malloc'd result buffer */ |
| 2391 int *pnOut /* OUT: Size of buffer at *paOut */ |
| 2392 ){ |
| 2393 char *pOut = 0; |
| 2394 int nOut = 0; |
| 2395 int rc = SQLITE_OK; |
| 2396 int ii; |
| 2397 int iCol = pPhrase->iColumn; |
| 2398 int isTermPos = (pPhrase->nToken>1 || isReqPos); |
| 2399 Fts3Table *p = (Fts3Table *)pCsr->base.pVtab; |
| 2400 int isFirst = 1; |
| 2401 |
| 2402 int iPrevTok = 0; |
| 2403 int nDoc = 0; |
| 2404 |
| 2405 /* If this is an xFilter() evaluation, create a segment-reader for each |
| 2406 ** phrase token. Or, if this is an xNext() or snippet/offsets/matchinfo |
| 2407 ** evaluation, only create segment-readers if there are no Fts3DeferredToken |
| 2408 ** objects attached to the phrase-tokens. |
| 2409 */ |
| 2410 for(ii=0; ii<pPhrase->nToken; ii++){ |
| 2411 Fts3PhraseToken *pTok = &pPhrase->aToken[ii]; |
| 2412 if( pTok->pSegcsr==0 ){ |
| 2413 if( (pCsr->eEvalmode==FTS3_EVAL_FILTER) |
| 2414 || (pCsr->eEvalmode==FTS3_EVAL_NEXT && pCsr->pDeferred==0) |
| 2415 || (pCsr->eEvalmode==FTS3_EVAL_MATCHINFO && pTok->bFulltext) |
| 2416 ){ |
| 2417 rc = fts3TermSegReaderCursor( |
| 2418 pCsr, pTok->z, pTok->n, pTok->isPrefix, &pTok->pSegcsr |
| 2419 ); |
| 2420 if( rc!=SQLITE_OK ) return rc; |
| 2421 } |
| 2422 } |
| 2423 } |
| 2424 |
| 2425 for(ii=0; ii<pPhrase->nToken; ii++){ |
| 2426 Fts3PhraseToken *pTok; /* Token to find doclist for */ |
| 2427 int iTok = 0; /* The token being queried this iteration */ |
| 2428 char *pList = 0; /* Pointer to token doclist */ |
| 2429 int nList = 0; /* Size of buffer at pList */ |
| 2430 |
| 2431 /* Select a token to process. If this is an xFilter() call, then tokens |
| 2432 ** are processed in order from least to most costly. Otherwise, tokens |
| 2433 ** are processed in the order in which they occur in the phrase. |
| 2434 */ |
| 2435 if( pCsr->eEvalmode==FTS3_EVAL_MATCHINFO ){ |
| 2436 assert( isReqPos ); |
| 2437 iTok = ii; |
| 2438 pTok = &pPhrase->aToken[iTok]; |
| 2439 if( pTok->bFulltext==0 ) continue; |
| 2440 }else if( pCsr->eEvalmode==FTS3_EVAL_NEXT || isReqPos ){ |
| 2441 iTok = ii; |
| 2442 pTok = &pPhrase->aToken[iTok]; |
2864 }else{ | 2443 }else{ |
2865 zIn[n] = 0; | 2444 int nMinCost = 0x7FFFFFFF; |
2866 *pzTail = &zIn[1]; | 2445 int jj; |
2867 dequoteString(zIn); | 2446 |
2868 return zIn; | 2447 /* Find the remaining token with the lowest cost. */ |
2869 } | 2448 for(jj=0; jj<pPhrase->nToken; jj++){ |
2870 } | 2449 Fts3SegReaderCursor *pSegcsr = pPhrase->aToken[jj].pSegcsr; |
2871 /*NOTREACHED*/ | 2450 if( pSegcsr && pSegcsr->nCost<nMinCost ){ |
2872 } | 2451 iTok = jj; |
2873 | 2452 nMinCost = pSegcsr->nCost; |
2874 /* Return true if... | 2453 } |
2875 ** | 2454 } |
2876 ** * s begins with the string t, ignoring case | 2455 pTok = &pPhrase->aToken[iTok]; |
2877 ** * s is longer than t | 2456 |
2878 ** * The first character of s beyond t is not a alphanumeric | 2457 /* This branch is taken if it is determined that loading the doclist |
2879 ** | 2458 ** for the next token would require more IO than loading all documents |
2880 ** Ignore leading space in *s. | 2459 ** currently identified by doclist pOut/nOut. No further doclists will |
2881 ** | 2460 ** be loaded from the full-text index for this phrase. |
2882 ** To put it another way, return true if the first token of | 2461 */ |
2883 ** s[] is t[]. | 2462 if( nMinCost>nDoc && ii>0 ){ |
2884 */ | 2463 rc = fts3DeferExpression(pCsr, pCsr->pExpr); |
2885 static int startsWith(const char *s, const char *t){ | 2464 break; |
2886 while( safe_isspace(*s) ){ s++; } | 2465 } |
2887 while( *t ){ | 2466 } |
2888 if( safe_tolower(*s++)!=safe_tolower(*t++) ) return 0; | 2467 |
2889 } | 2468 if( pCsr->eEvalmode==FTS3_EVAL_NEXT && pTok->pDeferred ){ |
2890 return *s!='_' && !safe_isalnum(*s); | 2469 rc = fts3DeferredTermSelect(pTok->pDeferred, isTermPos, &nList, &pList); |
2891 } | |
2892 | |
2893 /* | |
2894 ** An instance of this structure defines the "spec" of a | |
2895 ** full text index. This structure is populated by parseSpec | |
2896 ** and use by fulltextConnect and fulltextCreate. | |
2897 */ | |
2898 typedef struct TableSpec { | |
2899 const char *zDb; /* Logical database name */ | |
2900 const char *zName; /* Name of the full-text index */ | |
2901 int nColumn; /* Number of columns to be indexed */ | |
2902 char **azColumn; /* Original names of columns to be indexed */ | |
2903 char **azContentColumn; /* Column names for %_content */ | |
2904 char **azTokenizer; /* Name of tokenizer and its arguments */ | |
2905 } TableSpec; | |
2906 | |
2907 /* | |
2908 ** Reclaim all of the memory used by a TableSpec | |
2909 */ | |
2910 static void clearTableSpec(TableSpec *p) { | |
2911 sqlite3_free(p->azColumn); | |
2912 sqlite3_free(p->azContentColumn); | |
2913 sqlite3_free(p->azTokenizer); | |
2914 } | |
2915 | |
2916 /* Parse a CREATE VIRTUAL TABLE statement, which looks like this: | |
2917 * | |
2918 * CREATE VIRTUAL TABLE email | |
2919 * USING fts3(subject, body, tokenize mytokenizer(myarg)) | |
2920 * | |
2921 * We return parsed information in a TableSpec structure. | |
2922 * | |
2923 */ | |
2924 static int parseSpec(TableSpec *pSpec, int argc, const char *const*argv, | |
2925 char**pzErr){ | |
2926 int i, n; | |
2927 char *z, *zDummy; | |
2928 char **azArg; | |
2929 const char *zTokenizer = 0; /* argv[] entry describing the tokenizer */ | |
2930 | |
2931 assert( argc>=3 ); | |
2932 /* Current interface: | |
2933 ** argv[0] - module name | |
2934 ** argv[1] - database name | |
2935 ** argv[2] - table name | |
2936 ** argv[3..] - columns, optionally followed by tokenizer specification | |
2937 ** and snippet delimiters specification. | |
2938 */ | |
2939 | |
2940 /* Make a copy of the complete argv[][] array in a single allocation. | |
2941 ** The argv[][] array is read-only and transient. We can write to the | |
2942 ** copy in order to modify things and the copy is persistent. | |
2943 */ | |
2944 CLEAR(pSpec); | |
2945 for(i=n=0; i<argc; i++){ | |
2946 n += strlen(argv[i]) + 1; | |
2947 } | |
2948 azArg = sqlite3_malloc( sizeof(char*)*argc + n ); | |
2949 if( azArg==0 ){ | |
2950 return SQLITE_NOMEM; | |
2951 } | |
2952 z = (char*)&azArg[argc]; | |
2953 for(i=0; i<argc; i++){ | |
2954 azArg[i] = z; | |
2955 strcpy(z, argv[i]); | |
2956 z += strlen(z)+1; | |
2957 } | |
2958 | |
2959 /* Identify the column names and the tokenizer and delimiter arguments | |
2960 ** in the argv[][] array. | |
2961 */ | |
2962 pSpec->zDb = azArg[1]; | |
2963 pSpec->zName = azArg[2]; | |
2964 pSpec->nColumn = 0; | |
2965 pSpec->azColumn = azArg; | |
2966 zTokenizer = "tokenize simple"; | |
2967 for(i=3; i<argc; ++i){ | |
2968 if( startsWith(azArg[i],"tokenize") ){ | |
2969 zTokenizer = azArg[i]; | |
2970 }else{ | 2470 }else{ |
2971 z = azArg[pSpec->nColumn] = firstToken(azArg[i], &zDummy); | 2471 if( pTok->pSegcsr ){ |
2972 pSpec->nColumn++; | 2472 rc = fts3TermSelect(p, pTok, iCol, isTermPos, &nList, &pList); |
2973 } | 2473 } |
2974 } | 2474 pTok->bFulltext = 1; |
2975 if( pSpec->nColumn==0 ){ | 2475 } |
2976 azArg[0] = "content"; | 2476 assert( rc!=SQLITE_OK || pCsr->eEvalmode || pTok->pSegcsr==0 ); |
2977 pSpec->nColumn = 1; | 2477 if( rc!=SQLITE_OK ) break; |
2978 } | 2478 |
2979 | 2479 if( isFirst ){ |
2980 /* | 2480 pOut = pList; |
2981 ** Construct the list of content column names. | 2481 nOut = nList; |
2982 ** | 2482 if( pCsr->eEvalmode==FTS3_EVAL_FILTER && pPhrase->nToken>1 ){ |
2983 ** Each content column name will be of the form cNNAAAA | 2483 nDoc = fts3DoclistCountDocids(1, pOut, nOut); |
2984 ** where NN is the column number and AAAA is the sanitized | 2484 } |
2985 ** column name. "sanitized" means that special characters are | 2485 isFirst = 0; |
2986 ** converted to "_". The cNN prefix guarantees that all column | 2486 iPrevTok = iTok; |
2987 ** names are unique. | 2487 }else{ |
2988 ** | 2488 /* Merge the new term list and the current output. */ |
2989 ** The AAAA suffix is not strictly necessary. It is included | 2489 char *aLeft, *aRight; |
2990 ** for the convenience of people who might examine the generated | 2490 int nLeft, nRight; |
2991 ** %_content table and wonder what the columns are used for. | 2491 int nDist; |
2992 */ | 2492 int mt; |
2993 pSpec->azContentColumn = sqlite3_malloc( pSpec->nColumn * sizeof(char *) ); | 2493 |
2994 if( pSpec->azContentColumn==0 ){ | 2494 /* If this is the final token of the phrase, and positions were not |
2995 clearTableSpec(pSpec); | 2495 ** requested by the caller, use MERGE_PHRASE instead of POS_PHRASE. |
2996 return SQLITE_NOMEM; | 2496 ** This drops the position information from the output list. |
2997 } | 2497 */ |
2998 for(i=0; i<pSpec->nColumn; i++){ | 2498 mt = MERGE_POS_PHRASE; |
2999 char *p; | 2499 if( ii==pPhrase->nToken-1 && !isReqPos ) mt = MERGE_PHRASE; |
3000 pSpec->azContentColumn[i] = sqlite3_mprintf("c%d%s", i, azArg[i]); | 2500 |
3001 for (p = pSpec->azContentColumn[i]; *p ; ++p) { | 2501 assert( iPrevTok!=iTok ); |
3002 if( !safe_isalnum(*p) ) *p = '_'; | 2502 if( iPrevTok<iTok ){ |
3003 } | 2503 aLeft = pOut; |
3004 } | 2504 nLeft = nOut; |
3005 | 2505 aRight = pList; |
3006 /* | 2506 nRight = nList; |
3007 ** Parse the tokenizer specification string. | 2507 nDist = iTok-iPrevTok; |
3008 */ | 2508 iPrevTok = iTok; |
3009 pSpec->azTokenizer = tokenizeString(zTokenizer, &n); | 2509 }else{ |
3010 tokenListToIdList(pSpec->azTokenizer); | 2510 aRight = pOut; |
3011 | 2511 nRight = nOut; |
3012 return SQLITE_OK; | 2512 aLeft = pList; |
3013 } | 2513 nLeft = nList; |
3014 | 2514 nDist = iPrevTok-iTok; |
3015 /* | 2515 } |
3016 ** Generate a CREATE TABLE statement that describes the schema of | 2516 pOut = aRight; |
3017 ** the virtual table. Return a pointer to this schema string. | 2517 fts3DoclistMerge( |
3018 ** | 2518 mt, nDist, 0, pOut, &nOut, aLeft, nLeft, aRight, nRight, &nDoc |
3019 ** Space is obtained from sqlite3_mprintf() and should be freed | 2519 ); |
3020 ** using sqlite3_free(). | 2520 sqlite3_free(aLeft); |
3021 */ | 2521 } |
3022 static char *fulltextSchema( | 2522 assert( nOut==0 || pOut!=0 ); |
3023 int nColumn, /* Number of columns */ | 2523 } |
3024 const char *const* azColumn, /* List of columns */ | 2524 |
3025 const char *zTableName /* Name of the table */ | 2525 if( rc==SQLITE_OK ){ |
3026 ){ | 2526 if( ii!=pPhrase->nToken ){ |
3027 int i; | 2527 assert( pCsr->eEvalmode==FTS3_EVAL_FILTER && isReqPos==0 ); |
3028 char *zSchema, *zNext; | 2528 fts3DoclistStripPositions(pOut, &nOut); |
3029 const char *zSep = "("; | 2529 } |
3030 zSchema = sqlite3_mprintf("CREATE TABLE x"); | 2530 *paOut = pOut; |
3031 for(i=0; i<nColumn; i++){ | 2531 *pnOut = nOut; |
3032 zNext = sqlite3_mprintf("%s%s%Q", zSchema, zSep, azColumn[i]); | |
3033 sqlite3_free(zSchema); | |
3034 zSchema = zNext; | |
3035 zSep = ","; | |
3036 } | |
3037 zNext = sqlite3_mprintf("%s,%Q HIDDEN", zSchema, zTableName); | |
3038 sqlite3_free(zSchema); | |
3039 zSchema = zNext; | |
3040 zNext = sqlite3_mprintf("%s,docid HIDDEN)", zSchema); | |
3041 sqlite3_free(zSchema); | |
3042 return zNext; | |
3043 } | |
3044 | |
3045 /* | |
3046 ** Build a new sqlite3_vtab structure that will describe the | |
3047 ** fulltext index defined by spec. | |
3048 */ | |
3049 static int constructVtab( | |
3050 sqlite3 *db, /* The SQLite database connection */ | |
3051 fts3Hash *pHash, /* Hash table containing tokenizers */ | |
3052 TableSpec *spec, /* Parsed spec information from parseSpec() */ | |
3053 sqlite3_vtab **ppVTab, /* Write the resulting vtab structure here */ | |
3054 char **pzErr /* Write any error message here */ | |
3055 ){ | |
3056 int rc; | |
3057 int n; | |
3058 fulltext_vtab *v = 0; | |
3059 const sqlite3_tokenizer_module *m = NULL; | |
3060 char *schema; | |
3061 | |
3062 char const *zTok; /* Name of tokenizer to use for this fts table */ | |
3063 int nTok; /* Length of zTok, including nul terminator */ | |
3064 | |
3065 v = (fulltext_vtab *) sqlite3_malloc(sizeof(fulltext_vtab)); | |
3066 if( v==0 ) return SQLITE_NOMEM; | |
3067 CLEAR(v); | |
3068 /* sqlite will initialize v->base */ | |
3069 v->db = db; | |
3070 v->zDb = spec->zDb; /* Freed when azColumn is freed */ | |
3071 v->zName = spec->zName; /* Freed when azColumn is freed */ | |
3072 v->nColumn = spec->nColumn; | |
3073 v->azContentColumn = spec->azContentColumn; | |
3074 spec->azContentColumn = 0; | |
3075 v->azColumn = spec->azColumn; | |
3076 spec->azColumn = 0; | |
3077 | |
3078 if( spec->azTokenizer==0 ){ | |
3079 return SQLITE_NOMEM; | |
3080 } | |
3081 | |
3082 zTok = spec->azTokenizer[0]; | |
3083 if( !zTok ){ | |
3084 zTok = "simple"; | |
3085 } | |
3086 nTok = strlen(zTok)+1; | |
3087 | |
3088 m = (sqlite3_tokenizer_module *)sqlite3Fts3HashFind(pHash, zTok, nTok); | |
3089 if( !m ){ | |
3090 *pzErr = sqlite3_mprintf("unknown tokenizer: %s", spec->azTokenizer[0]); | |
3091 rc = SQLITE_ERROR; | |
3092 goto err; | |
3093 } | |
3094 | |
3095 for(n=0; spec->azTokenizer[n]; n++){} | |
3096 if( n ){ | |
3097 rc = m->xCreate(n-1, (const char*const*)&spec->azTokenizer[1], | |
3098 &v->pTokenizer); | |
3099 }else{ | 2532 }else{ |
3100 rc = m->xCreate(0, 0, &v->pTokenizer); | 2533 sqlite3_free(pOut); |
3101 } | 2534 } |
3102 if( rc!=SQLITE_OK ) goto err; | |
3103 v->pTokenizer->pModule = m; | |
3104 | |
3105 /* TODO: verify the existence of backing tables foo_content, foo_term */ | |
3106 | |
3107 schema = fulltextSchema(v->nColumn, (const char*const*)v->azColumn, | |
3108 spec->zName); | |
3109 rc = sqlite3_declare_vtab(db, schema); | |
3110 sqlite3_free(schema); | |
3111 if( rc!=SQLITE_OK ) goto err; | |
3112 | |
3113 memset(v->pFulltextStatements, 0, sizeof(v->pFulltextStatements)); | |
3114 | |
3115 /* Indicate that the buffer is not live. */ | |
3116 v->nPendingData = -1; | |
3117 | |
3118 *ppVTab = &v->base; | |
3119 FTSTRACE(("FTS3 Connect %p\n", v)); | |
3120 | |
3121 return rc; | 2535 return rc; |
3122 | 2536 } |
3123 err: | 2537 |
3124 fulltext_vtab_destroy(v); | 2538 /* |
| 2539 ** This function merges two doclists according to the requirements of a |
| 2540 ** NEAR operator. |
| 2541 ** |
| 2542 ** Both input doclists must include position information. The output doclist |
| 2543 ** includes position information if the first argument to this function |
| 2544 ** is MERGE_POS_NEAR, or does not if it is MERGE_NEAR. |
| 2545 */ |
| 2546 static int fts3NearMerge( |
| 2547 int mergetype, /* MERGE_POS_NEAR or MERGE_NEAR */ |
| 2548 int nNear, /* Parameter to NEAR operator */ |
| 2549 int nTokenLeft, /* Number of tokens in LHS phrase arg */ |
| 2550 char *aLeft, /* Doclist for LHS (incl. positions) */ |
| 2551 int nLeft, /* Size of LHS doclist in bytes */ |
| 2552 int nTokenRight, /* As nTokenLeft */ |
| 2553 char *aRight, /* As aLeft */ |
| 2554 int nRight, /* As nRight */ |
| 2555 char **paOut, /* OUT: Results of merge (malloced) */ |
| 2556 int *pnOut /* OUT: Sized of output buffer */ |
| 2557 ){ |
| 2558 char *aOut; /* Buffer to write output doclist to */ |
| 2559 int rc; /* Return code */ |
| 2560 |
| 2561 assert( mergetype==MERGE_POS_NEAR || MERGE_NEAR ); |
| 2562 |
| 2563 aOut = sqlite3_malloc(nLeft+nRight+1); |
| 2564 if( aOut==0 ){ |
| 2565 rc = SQLITE_NOMEM; |
| 2566 }else{ |
| 2567 rc = fts3DoclistMerge(mergetype, nNear+nTokenRight, nNear+nTokenLeft, |
| 2568 aOut, pnOut, aLeft, nLeft, aRight, nRight, 0 |
| 2569 ); |
| 2570 if( rc!=SQLITE_OK ){ |
| 2571 sqlite3_free(aOut); |
| 2572 aOut = 0; |
| 2573 } |
| 2574 } |
| 2575 |
| 2576 *paOut = aOut; |
3125 return rc; | 2577 return rc; |
3126 } | 2578 } |
3127 | 2579 |
3128 static int fulltextConnect( | 2580 /* |
3129 sqlite3 *db, | 2581 ** This function is used as part of the processing for the snippet() and |
3130 void *pAux, | 2582 ** offsets() functions. |
3131 int argc, const char *const*argv, | 2583 ** |
3132 sqlite3_vtab **ppVTab, | 2584 ** Both pLeft and pRight are expression nodes of type FTSQUERY_PHRASE. Both |
3133 char **pzErr | 2585 ** have their respective doclists (including position information) loaded |
3134 ){ | 2586 ** in Fts3Expr.aDoclist/nDoclist. This function removes all entries from |
3135 TableSpec spec; | 2587 ** each doclist that are not within nNear tokens of a corresponding entry |
3136 int rc = parseSpec(&spec, argc, argv, pzErr); | 2588 ** in the other doclist. |
3137 if( rc!=SQLITE_OK ) return rc; | 2589 */ |
3138 | 2590 int sqlite3Fts3ExprNearTrim(Fts3Expr *pLeft, Fts3Expr *pRight, int nNear){ |
3139 rc = constructVtab(db, (fts3Hash *)pAux, &spec, ppVTab, pzErr); | 2591 int rc; /* Return code */ |
3140 clearTableSpec(&spec); | 2592 |
| 2593 assert( pLeft->eType==FTSQUERY_PHRASE ); |
| 2594 assert( pRight->eType==FTSQUERY_PHRASE ); |
| 2595 assert( pLeft->isLoaded && pRight->isLoaded ); |
| 2596 |
| 2597 if( pLeft->aDoclist==0 || pRight->aDoclist==0 ){ |
| 2598 sqlite3_free(pLeft->aDoclist); |
| 2599 sqlite3_free(pRight->aDoclist); |
| 2600 pRight->aDoclist = 0; |
| 2601 pLeft->aDoclist = 0; |
| 2602 rc = SQLITE_OK; |
| 2603 }else{ |
| 2604 char *aOut; /* Buffer in which to assemble new doclist */ |
| 2605 int nOut; /* Size of buffer aOut in bytes */ |
| 2606 |
| 2607 rc = fts3NearMerge(MERGE_POS_NEAR, nNear, |
| 2608 pLeft->pPhrase->nToken, pLeft->aDoclist, pLeft->nDoclist, |
| 2609 pRight->pPhrase->nToken, pRight->aDoclist, pRight->nDoclist, |
| 2610 &aOut, &nOut |
| 2611 ); |
| 2612 if( rc!=SQLITE_OK ) return rc; |
| 2613 sqlite3_free(pRight->aDoclist); |
| 2614 pRight->aDoclist = aOut; |
| 2615 pRight->nDoclist = nOut; |
| 2616 |
| 2617 rc = fts3NearMerge(MERGE_POS_NEAR, nNear, |
| 2618 pRight->pPhrase->nToken, pRight->aDoclist, pRight->nDoclist, |
| 2619 pLeft->pPhrase->nToken, pLeft->aDoclist, pLeft->nDoclist, |
| 2620 &aOut, &nOut |
| 2621 ); |
| 2622 sqlite3_free(pLeft->aDoclist); |
| 2623 pLeft->aDoclist = aOut; |
| 2624 pLeft->nDoclist = nOut; |
| 2625 } |
3141 return rc; | 2626 return rc; |
3142 } | 2627 } |
3143 | 2628 |
3144 /* The %_content table holds the text of each document, with | 2629 |
3145 ** the docid column exposed as the SQLite rowid for the table. | 2630 /* |
3146 */ | 2631 ** Allocate an Fts3SegReaderArray for each token in the expression pExpr. |
3147 /* TODO(shess) This comment needs elaboration to match the updated | 2632 ** The allocated objects are stored in the Fts3PhraseToken.pArray member |
3148 ** code. Work it into the top-of-file comment at that time. | 2633 ** variables of each token structure. |
3149 */ | 2634 */ |
3150 static int fulltextCreate(sqlite3 *db, void *pAux, | 2635 static int fts3ExprAllocateSegReaders( |
3151 int argc, const char * const *argv, | 2636 Fts3Cursor *pCsr, /* FTS3 table */ |
3152 sqlite3_vtab **ppVTab, char **pzErr){ | 2637 Fts3Expr *pExpr, /* Expression to create seg-readers for */ |
3153 int rc; | 2638 int *pnExpr /* OUT: Number of AND'd expressions */ |
3154 TableSpec spec; | 2639 ){ |
3155 StringBuffer schema; | 2640 int rc = SQLITE_OK; /* Return code */ |
3156 FTSTRACE(("FTS3 Create\n")); | 2641 |
3157 | 2642 assert( pCsr->eEvalmode==FTS3_EVAL_FILTER ); |
3158 rc = parseSpec(&spec, argc, argv, pzErr); | 2643 if( pnExpr && pExpr->eType!=FTSQUERY_AND ){ |
3159 if( rc!=SQLITE_OK ) return rc; | 2644 (*pnExpr)++; |
3160 | 2645 pnExpr = 0; |
3161 initStringBuffer(&schema); | 2646 } |
3162 append(&schema, "CREATE TABLE %_content("); | 2647 |
3163 append(&schema, " docid INTEGER PRIMARY KEY,"); | 2648 if( pExpr->eType==FTSQUERY_PHRASE ){ |
3164 appendList(&schema, spec.nColumn, spec.azContentColumn); | 2649 Fts3Phrase *pPhrase = pExpr->pPhrase; |
3165 append(&schema, ")"); | 2650 int ii; |
3166 rc = sql_exec(db, spec.zDb, spec.zName, stringBufferData(&schema)); | 2651 |
3167 stringBufferDestroy(&schema); | 2652 for(ii=0; rc==SQLITE_OK && ii<pPhrase->nToken; ii++){ |
3168 if( rc!=SQLITE_OK ) goto out; | 2653 Fts3PhraseToken *pTok = &pPhrase->aToken[ii]; |
3169 | 2654 if( pTok->pSegcsr==0 ){ |
3170 rc = sql_exec(db, spec.zDb, spec.zName, | 2655 rc = fts3TermSegReaderCursor( |
3171 "create table %_segments(" | 2656 pCsr, pTok->z, pTok->n, pTok->isPrefix, &pTok->pSegcsr |
3172 " blockid INTEGER PRIMARY KEY," | 2657 ); |
3173 " block blob" | 2658 } |
3174 ");" | 2659 } |
3175 ); | 2660 }else{ |
3176 if( rc!=SQLITE_OK ) goto out; | 2661 rc = fts3ExprAllocateSegReaders(pCsr, pExpr->pLeft, pnExpr); |
3177 | 2662 if( rc==SQLITE_OK ){ |
3178 rc = sql_exec(db, spec.zDb, spec.zName, | 2663 rc = fts3ExprAllocateSegReaders(pCsr, pExpr->pRight, pnExpr); |
3179 "create table %_segdir(" | 2664 } |
3180 " level integer," | 2665 } |
3181 " idx integer," | |
3182 " start_block integer," | |
3183 " leaves_end_block integer," | |
3184 " end_block integer," | |
3185 " root blob," | |
3186 " primary key(level, idx)" | |
3187 ");"); | |
3188 if( rc!=SQLITE_OK ) goto out; | |
3189 | |
3190 rc = constructVtab(db, (fts3Hash *)pAux, &spec, ppVTab, pzErr); | |
3191 | |
3192 out: | |
3193 clearTableSpec(&spec); | |
3194 return rc; | 2666 return rc; |
3195 } | 2667 } |
3196 | 2668 |
3197 /* Decide how to handle an SQL query. */ | 2669 /* |
3198 static int fulltextBestIndex(sqlite3_vtab *pVTab, sqlite3_index_info *pInfo){ | 2670 ** Free the Fts3SegReaderArray objects associated with each token in the |
3199 fulltext_vtab *v = (fulltext_vtab *)pVTab; | 2671 ** expression pExpr. In other words, this function frees the resources |
3200 int i; | 2672 ** allocated by fts3ExprAllocateSegReaders(). |
3201 FTSTRACE(("FTS3 BestIndex\n")); | 2673 */ |
3202 | 2674 static void fts3ExprFreeSegReaders(Fts3Expr *pExpr){ |
3203 for(i=0; i<pInfo->nConstraint; ++i){ | 2675 if( pExpr ){ |
3204 const struct sqlite3_index_constraint *pConstraint; | 2676 Fts3Phrase *pPhrase = pExpr->pPhrase; |
3205 pConstraint = &pInfo->aConstraint[i]; | 2677 if( pPhrase ){ |
3206 if( pConstraint->usable ) { | 2678 int kk; |
3207 if( (pConstraint->iColumn==-1 || pConstraint->iColumn==v->nColumn+1) && | 2679 for(kk=0; kk<pPhrase->nToken; kk++){ |
3208 pConstraint->op==SQLITE_INDEX_CONSTRAINT_EQ ){ | 2680 fts3SegReaderCursorFree(pPhrase->aToken[kk].pSegcsr); |
3209 pInfo->idxNum = QUERY_DOCID; /* lookup by docid */ | 2681 pPhrase->aToken[kk].pSegcsr = 0; |
3210 FTSTRACE(("FTS3 QUERY_DOCID\n")); | 2682 } |
3211 } else if( pConstraint->iColumn>=0 && pConstraint->iColumn<=v->nColumn && | 2683 } |
3212 pConstraint->op==SQLITE_INDEX_CONSTRAINT_MATCH ){ | 2684 fts3ExprFreeSegReaders(pExpr->pLeft); |
3213 /* full-text search */ | 2685 fts3ExprFreeSegReaders(pExpr->pRight); |
3214 pInfo->idxNum = QUERY_FULLTEXT + pConstraint->iColumn; | 2686 } |
3215 FTSTRACE(("FTS3 QUERY_FULLTEXT %d\n", pConstraint->iColumn)); | 2687 } |
3216 } else continue; | 2688 |
3217 | 2689 /* |
3218 pInfo->aConstraintUsage[i].argvIndex = 1; | 2690 ** Return the sum of the costs of all tokens in the expression pExpr. This |
3219 pInfo->aConstraintUsage[i].omit = 1; | 2691 ** function must be called after Fts3SegReaderArrays have been allocated |
3220 | 2692 ** for all tokens using fts3ExprAllocateSegReaders(). |
3221 /* An arbitrary value for now. | 2693 */ |
3222 * TODO: Perhaps docid matches should be considered cheaper than | 2694 static int fts3ExprCost(Fts3Expr *pExpr){ |
3223 * full-text searches. */ | 2695 int nCost; /* Return value */ |
3224 pInfo->estimatedCost = 1.0; | 2696 if( pExpr->eType==FTSQUERY_PHRASE ){ |
3225 | 2697 Fts3Phrase *pPhrase = pExpr->pPhrase; |
3226 return SQLITE_OK; | 2698 int ii; |
3227 } | 2699 nCost = 0; |
3228 } | 2700 for(ii=0; ii<pPhrase->nToken; ii++){ |
3229 pInfo->idxNum = QUERY_GENERIC; | 2701 Fts3SegReaderCursor *pSegcsr = pPhrase->aToken[ii].pSegcsr; |
3230 return SQLITE_OK; | 2702 if( pSegcsr ) nCost += pSegcsr->nCost; |
3231 } | 2703 } |
3232 | |
3233 static int fulltextDisconnect(sqlite3_vtab *pVTab){ | |
3234 FTSTRACE(("FTS3 Disconnect %p\n", pVTab)); | |
3235 fulltext_vtab_destroy((fulltext_vtab *)pVTab); | |
3236 return SQLITE_OK; | |
3237 } | |
3238 | |
3239 static int fulltextDestroy(sqlite3_vtab *pVTab){ | |
3240 fulltext_vtab *v = (fulltext_vtab *)pVTab; | |
3241 int rc; | |
3242 | |
3243 FTSTRACE(("FTS3 Destroy %p\n", pVTab)); | |
3244 rc = sql_exec(v->db, v->zDb, v->zName, | |
3245 "drop table if exists %_content;" | |
3246 "drop table if exists %_segments;" | |
3247 "drop table if exists %_segdir;" | |
3248 ); | |
3249 if( rc!=SQLITE_OK ) return rc; | |
3250 | |
3251 fulltext_vtab_destroy((fulltext_vtab *)pVTab); | |
3252 return SQLITE_OK; | |
3253 } | |
3254 | |
3255 static int fulltextOpen(sqlite3_vtab *pVTab, sqlite3_vtab_cursor **ppCursor){ | |
3256 fulltext_cursor *c; | |
3257 | |
3258 c = (fulltext_cursor *) sqlite3_malloc(sizeof(fulltext_cursor)); | |
3259 if( c ){ | |
3260 memset(c, 0, sizeof(fulltext_cursor)); | |
3261 /* sqlite will initialize c->base */ | |
3262 *ppCursor = &c->base; | |
3263 FTSTRACE(("FTS3 Open %p: %p\n", pVTab, c)); | |
3264 return SQLITE_OK; | |
3265 }else{ | 2704 }else{ |
3266 return SQLITE_NOMEM; | 2705 nCost = fts3ExprCost(pExpr->pLeft) + fts3ExprCost(pExpr->pRight); |
3267 } | 2706 } |
3268 } | 2707 return nCost; |
3269 | 2708 } |
3270 /* Free all of the dynamically allocated memory held by the | 2709 |
3271 ** Snippet | 2710 /* |
3272 */ | 2711 ** The following is a helper function (and type) for fts3EvalExpr(). It |
3273 static void snippetClear(Snippet *p){ | 2712 ** must be called after Fts3SegReaders have been allocated for every token |
3274 sqlite3_free(p->aMatch); | 2713 ** in the expression. See the context it is called from in fts3EvalExpr() |
3275 sqlite3_free(p->zOffset); | 2714 ** for further explanation. |
3276 sqlite3_free(p->zSnippet); | 2715 */ |
3277 CLEAR(p); | 2716 typedef struct ExprAndCost ExprAndCost; |
3278 } | 2717 struct ExprAndCost { |
3279 | 2718 Fts3Expr *pExpr; |
3280 /* | 2719 int nCost; |
3281 ** Append a single entry to the p->aMatch[] log. | 2720 }; |
3282 */ | 2721 static void fts3ExprAssignCosts( |
3283 static void snippetAppendMatch( | 2722 Fts3Expr *pExpr, /* Expression to create seg-readers for */ |
3284 Snippet *p, /* Append the entry to this snippet */ | 2723 ExprAndCost **ppExprCost /* OUT: Write to *ppExprCost */ |
3285 int iCol, int iTerm, /* The column and query term */ | 2724 ){ |
3286 int iToken, /* Matching token in document */ | 2725 if( pExpr->eType==FTSQUERY_AND ){ |
3287 int iStart, int nByte /* Offset and size of the match */ | 2726 fts3ExprAssignCosts(pExpr->pLeft, ppExprCost); |
3288 ){ | 2727 fts3ExprAssignCosts(pExpr->pRight, ppExprCost); |
3289 int i; | |
3290 struct snippetMatch *pMatch; | |
3291 if( p->nMatch+1>=p->nAlloc ){ | |
3292 p->nAlloc = p->nAlloc*2 + 10; | |
3293 p->aMatch = sqlite3_realloc(p->aMatch, p->nAlloc*sizeof(p->aMatch[0]) ); | |
3294 if( p->aMatch==0 ){ | |
3295 p->nMatch = 0; | |
3296 p->nAlloc = 0; | |
3297 return; | |
3298 } | |
3299 } | |
3300 i = p->nMatch++; | |
3301 pMatch = &p->aMatch[i]; | |
3302 pMatch->iCol = iCol; | |
3303 pMatch->iTerm = iTerm; | |
3304 pMatch->iToken = iToken; | |
3305 pMatch->iStart = iStart; | |
3306 pMatch->nByte = nByte; | |
3307 } | |
3308 | |
3309 /* | |
3310 ** Sizing information for the circular buffer used in snippetOffsetsOfColumn() | |
3311 */ | |
3312 #define FTS3_ROTOR_SZ (32) | |
3313 #define FTS3_ROTOR_MASK (FTS3_ROTOR_SZ-1) | |
3314 | |
3315 /* | |
3316 ** Function to iterate through the tokens of a compiled expression. | |
3317 ** | |
3318 ** Except, skip all tokens on the right-hand side of a NOT operator. | |
3319 ** This function is used to find tokens as part of snippet and offset | |
3320 ** generation and we do nt want snippets and offsets to report matches | |
3321 ** for tokens on the RHS of a NOT. | |
3322 */ | |
3323 static int fts3NextExprToken(Fts3Expr **ppExpr, int *piToken){ | |
3324 Fts3Expr *p = *ppExpr; | |
3325 int iToken = *piToken; | |
3326 if( iToken<0 ){ | |
3327 /* In this case the expression p is the root of an expression tree. | |
3328 ** Move to the first token in the expression tree. | |
3329 */ | |
3330 while( p->pLeft ){ | |
3331 p = p->pLeft; | |
3332 } | |
3333 iToken = 0; | |
3334 }else{ | 2728 }else{ |
3335 assert(p && p->eType==FTSQUERY_PHRASE ); | 2729 (*ppExprCost)->pExpr = pExpr; |
3336 if( iToken<(p->pPhrase->nToken-1) ){ | 2730 (*ppExprCost)->nCost = fts3ExprCost(pExpr); |
3337 iToken++; | 2731 (*ppExprCost)++; |
3338 }else{ | 2732 } |
3339 iToken = 0; | 2733 } |
3340 while( p->pParent && p->pParent->pLeft!=p ){ | 2734 |
3341 assert( p->pParent->pRight==p ); | 2735 /* |
3342 p = p->pParent; | 2736 ** Evaluate the full-text expression pExpr against FTS3 table pTab. Store |
3343 } | 2737 ** the resulting doclist in *paOut and *pnOut. This routine mallocs for |
3344 p = p->pParent; | 2738 ** the space needed to store the output. The caller is responsible for |
3345 if( p ){ | 2739 ** freeing the space when it has finished. |
3346 assert( p->pRight!=0 ); | 2740 ** |
3347 p = p->pRight; | 2741 ** This function is called in two distinct contexts: |
3348 while( p->pLeft ){ | 2742 ** |
3349 p = p->pLeft; | 2743 ** * From within the virtual table xFilter() method. In this case, the |
3350 } | 2744 ** output doclist contains entries for all rows in the table, based on |
3351 } | 2745 ** data read from the full-text index. |
3352 } | 2746 ** |
3353 } | 2747 ** In this case, if the query expression contains one or more tokens that |
3354 | 2748 ** are very common, then the returned doclist may contain a superset of |
3355 *ppExpr = p; | 2749 ** the documents that actually match the expression. |
3356 *piToken = iToken; | 2750 ** |
3357 return p?1:0; | 2751 ** * From within the virtual table xNext() method. This call is only made |
3358 } | 2752 ** if the call from within xFilter() found that there were very common |
3359 | 2753 ** tokens in the query expression and did return a superset of the |
3360 /* | 2754 ** matching documents. In this case the returned doclist contains only |
3361 ** Return TRUE if the expression node pExpr is located beneath the | 2755 ** entries that correspond to the current row of the table. Instead of |
3362 ** RHS of a NOT operator. | 2756 ** reading the data for each token from the full-text index, the data is |
3363 */ | 2757 ** already available in-memory in the Fts3PhraseToken.pDeferred structures. |
3364 static int fts3ExprBeneathNot(Fts3Expr *p){ | 2758 ** See fts3EvalDeferred() for how it gets there. |
3365 Fts3Expr *pParent; | 2759 ** |
3366 while( p ){ | 2760 ** In the first case above, Fts3Cursor.doDeferred==0. In the second (if it is |
3367 pParent = p->pParent; | 2761 ** required) Fts3Cursor.doDeferred==1. |
3368 if( pParent && pParent->eType==FTSQUERY_NOT && pParent->pRight==p ){ | 2762 ** |
3369 return 1; | 2763 ** If the SQLite invokes the snippet(), offsets() or matchinfo() function |
3370 } | 2764 ** as part of a SELECT on an FTS3 table, this function is called on each |
3371 p = pParent; | 2765 ** individual phrase expression in the query. If there were very common tokens |
3372 } | 2766 ** found in the xFilter() call, then this function is called once for phrase |
3373 return 0; | 2767 ** for each row visited, and the returned doclist contains entries for the |
3374 } | 2768 ** current row only. Otherwise, if there were no very common tokens, then this |
3375 | 2769 ** function is called once only for each phrase in the query and the returned |
3376 /* | 2770 ** doclist contains entries for all rows of the table. |
3377 ** Add entries to pSnippet->aMatch[] for every match that occurs against | 2771 ** |
3378 ** document zDoc[0..nDoc-1] which is stored in column iColumn. | 2772 ** Fts3Cursor.doDeferred==1 when this function is called on phrases as a |
3379 */ | 2773 ** result of a snippet(), offsets() or matchinfo() invocation. |
3380 static void snippetOffsetsOfColumn( | 2774 */ |
3381 fulltext_cursor *pCur, /* The fulltest search cursor */ | 2775 static int fts3EvalExpr( |
3382 Snippet *pSnippet, /* The Snippet object to be filled in */ | 2776 Fts3Cursor *p, /* Virtual table cursor handle */ |
3383 int iColumn, /* Index of fulltext table column */ | 2777 Fts3Expr *pExpr, /* Parsed fts3 expression */ |
3384 const char *zDoc, /* Text of the fulltext table column */ | 2778 char **paOut, /* OUT: Pointer to malloc'd result buffer */ |
3385 int nDoc /* Length of zDoc in bytes */ | 2779 int *pnOut, /* OUT: Size of buffer at *paOut */ |
3386 ){ | 2780 int isReqPos /* Require positions in output buffer */ |
3387 const sqlite3_tokenizer_module *pTModule; /* The tokenizer module */ | 2781 ){ |
3388 sqlite3_tokenizer *pTokenizer; /* The specific tokenizer */ | 2782 int rc = SQLITE_OK; /* Return code */ |
3389 sqlite3_tokenizer_cursor *pTCursor; /* Tokenizer cursor */ | 2783 |
3390 fulltext_vtab *pVtab; /* The full text index */ | 2784 /* Zero the output parameters. */ |
3391 int nColumn; /* Number of columns in the index */ | 2785 *paOut = 0; |
3392 int i, j; /* Loop counters */ | 2786 *pnOut = 0; |
3393 int rc; /* Return code */ | 2787 |
3394 unsigned int match, prevMatch; /* Phrase search bitmasks */ | |
3395 const char *zToken; /* Next token from the tokenizer */ | |
3396 int nToken; /* Size of zToken */ | |
3397 int iBegin, iEnd, iPos; /* Offsets of beginning and end */ | |
3398 | |
3399 /* The following variables keep a circular buffer of the last | |
3400 ** few tokens */ | |
3401 unsigned int iRotor = 0; /* Index of current token */ | |
3402 int iRotorBegin[FTS3_ROTOR_SZ]; /* Beginning offset of token */ | |
3403 int iRotorLen[FTS3_ROTOR_SZ]; /* Length of token */ | |
3404 | |
3405 pVtab = cursor_vtab(pCur); | |
3406 nColumn = pVtab->nColumn; | |
3407 pTokenizer = pVtab->pTokenizer; | |
3408 pTModule = pTokenizer->pModule; | |
3409 rc = pTModule->xOpen(pTokenizer, zDoc, nDoc, &pTCursor); | |
3410 if( rc ) return; | |
3411 pTCursor->pTokenizer = pTokenizer; | |
3412 | |
3413 prevMatch = 0; | |
3414 while( !pTModule->xNext(pTCursor, &zToken, &nToken, &iBegin, &iEnd, &iPos) ){ | |
3415 Fts3Expr *pIter = pCur->pExpr; | |
3416 int iIter = -1; | |
3417 iRotorBegin[iRotor&FTS3_ROTOR_MASK] = iBegin; | |
3418 iRotorLen[iRotor&FTS3_ROTOR_MASK] = iEnd-iBegin; | |
3419 match = 0; | |
3420 for(i=0; i<(FTS3_ROTOR_SZ-1) && fts3NextExprToken(&pIter, &iIter); i++){ | |
3421 int nPhrase; /* Number of tokens in current phrase */ | |
3422 struct PhraseToken *pToken; /* Current token */ | |
3423 int iCol; /* Column index */ | |
3424 | |
3425 if( fts3ExprBeneathNot(pIter) ) continue; | |
3426 nPhrase = pIter->pPhrase->nToken; | |
3427 pToken = &pIter->pPhrase->aToken[iIter]; | |
3428 iCol = pIter->pPhrase->iColumn; | |
3429 if( iCol>=0 && iCol<nColumn && iCol!=iColumn ) continue; | |
3430 if( pToken->n>nToken ) continue; | |
3431 if( !pToken->isPrefix && pToken->n<nToken ) continue; | |
3432 assert( pToken->n<=nToken ); | |
3433 if( memcmp(pToken->z, zToken, pToken->n) ) continue; | |
3434 if( iIter>0 && (prevMatch & (1<<i))==0 ) continue; | |
3435 match |= 1<<i; | |
3436 if( i==(FTS3_ROTOR_SZ-2) || nPhrase==iIter+1 ){ | |
3437 for(j=nPhrase-1; j>=0; j--){ | |
3438 int k = (iRotor-j) & FTS3_ROTOR_MASK; | |
3439 snippetAppendMatch(pSnippet, iColumn, i-j, iPos-j, | |
3440 iRotorBegin[k], iRotorLen[k]); | |
3441 } | |
3442 } | |
3443 } | |
3444 prevMatch = match<<1; | |
3445 iRotor++; | |
3446 } | |
3447 pTModule->xClose(pTCursor); | |
3448 } | |
3449 | |
3450 /* | |
3451 ** Remove entries from the pSnippet structure to account for the NEAR | |
3452 ** operator. When this is called, pSnippet contains the list of token | |
3453 ** offsets produced by treating all NEAR operators as AND operators. | |
3454 ** This function removes any entries that should not be present after | |
3455 ** accounting for the NEAR restriction. For example, if the queried | |
3456 ** document is: | |
3457 ** | |
3458 ** "A B C D E A" | |
3459 ** | |
3460 ** and the query is: | |
3461 ** | |
3462 ** A NEAR/0 E | |
3463 ** | |
3464 ** then when this function is called the Snippet contains token offsets | |
3465 ** 0, 4 and 5. This function removes the "0" entry (because the first A | |
3466 ** is not near enough to an E). | |
3467 ** | |
3468 ** When this function is called, the value pointed to by parameter piLeft is | |
3469 ** the integer id of the left-most token in the expression tree headed by | |
3470 ** pExpr. This function increments *piLeft by the total number of tokens | |
3471 ** in the expression tree headed by pExpr. | |
3472 ** | |
3473 ** Return 1 if any trimming occurs. Return 0 if no trimming is required. | |
3474 */ | |
3475 static int trimSnippetOffsets( | |
3476 Fts3Expr *pExpr, /* The search expression */ | |
3477 Snippet *pSnippet, /* The set of snippet offsets to be trimmed */ | |
3478 int *piLeft /* Index of left-most token in pExpr */ | |
3479 ){ | |
3480 if( pExpr ){ | 2788 if( pExpr ){ |
3481 if( trimSnippetOffsets(pExpr->pLeft, pSnippet, piLeft) ){ | 2789 assert( pExpr->eType==FTSQUERY_NEAR || pExpr->eType==FTSQUERY_OR |
3482 return 1; | 2790 || pExpr->eType==FTSQUERY_AND || pExpr->eType==FTSQUERY_NOT |
3483 } | 2791 || pExpr->eType==FTSQUERY_PHRASE |
3484 | 2792 ); |
3485 switch( pExpr->eType ){ | 2793 assert( pExpr->eType==FTSQUERY_PHRASE || isReqPos==0 ); |
3486 case FTSQUERY_PHRASE: | 2794 |
3487 *piLeft += pExpr->pPhrase->nToken; | 2795 if( pExpr->eType==FTSQUERY_PHRASE ){ |
3488 break; | 2796 rc = fts3PhraseSelect(p, pExpr->pPhrase, |
3489 case FTSQUERY_NEAR: { | 2797 isReqPos || (pExpr->pParent && pExpr->pParent->eType==FTSQUERY_NEAR), |
3490 /* The right-hand-side of a NEAR operator is always a phrase. The | 2798 paOut, pnOut |
3491 ** left-hand-side is either a phrase or an expression tree that is | 2799 ); |
3492 ** itself headed by a NEAR operator. The following initializations | 2800 fts3ExprFreeSegReaders(pExpr); |
3493 ** set local variable iLeft to the token number of the left-most | 2801 }else if( p->eEvalmode==FTS3_EVAL_FILTER && pExpr->eType==FTSQUERY_AND ){ |
3494 ** token in the right-hand phrase, and iRight to the right most | 2802 ExprAndCost *aExpr = 0; /* Array of AND'd expressions and costs */ |
3495 ** token in the same phrase. For example, if we had: | 2803 int nExpr = 0; /* Size of aExpr[] */ |
3496 ** | 2804 char *aRet = 0; /* Doclist to return to caller */ |
3497 ** <col> MATCH '"abc def" NEAR/2 "ghi jkl"' | 2805 int nRet = 0; /* Length of aRet[] in bytes */ |
3498 ** | 2806 int nDoc = 0x7FFFFFFF; |
3499 ** then iLeft will be set to 2 (token number of ghi) and nToken will | 2807 |
3500 ** be set to 4. | 2808 assert( !isReqPos ); |
3501 */ | 2809 |
3502 Fts3Expr *pLeft = pExpr->pLeft; | 2810 rc = fts3ExprAllocateSegReaders(p, pExpr, &nExpr); |
3503 Fts3Expr *pRight = pExpr->pRight; | 2811 if( rc==SQLITE_OK ){ |
3504 int iLeft = *piLeft; | 2812 assert( nExpr>1 ); |
3505 int nNear = pExpr->nNear; | 2813 aExpr = sqlite3_malloc(sizeof(ExprAndCost) * nExpr); |
3506 int nToken = pRight->pPhrase->nToken; | 2814 if( !aExpr ) rc = SQLITE_NOMEM; |
3507 int jj, ii; | 2815 } |
3508 if( pLeft->eType==FTSQUERY_NEAR ){ | 2816 if( rc==SQLITE_OK ){ |
3509 pLeft = pLeft->pRight; | 2817 int ii; /* Used to iterate through expressions */ |
3510 } | 2818 |
3511 assert( pRight->eType==FTSQUERY_PHRASE ); | 2819 fts3ExprAssignCosts(pExpr, &aExpr); |
3512 assert( pLeft->eType==FTSQUERY_PHRASE ); | 2820 aExpr -= nExpr; |
3513 nToken += pLeft->pPhrase->nToken; | 2821 for(ii=0; ii<nExpr; ii++){ |
3514 | 2822 char *aNew; |
3515 for(ii=0; ii<pSnippet->nMatch; ii++){ | 2823 int nNew; |
3516 struct snippetMatch *p = &pSnippet->aMatch[ii]; | 2824 int jj; |
3517 if( p->iTerm==iLeft ){ | 2825 ExprAndCost *pBest = 0; |
3518 int isOk = 0; | 2826 |
3519 /* Snippet ii is an occurence of query term iLeft in the document. | 2827 for(jj=0; jj<nExpr; jj++){ |
3520 ** It occurs at position (p->iToken) of the document. We now | 2828 ExprAndCost *pCand = &aExpr[jj]; |
3521 ** search for an instance of token (iLeft-1) somewhere in the | 2829 if( pCand->pExpr && (pBest==0 || pCand->nCost<pBest->nCost) ){ |
3522 ** range (p->iToken - nNear)...(p->iToken + nNear + nToken) within | 2830 pBest = pCand; |
3523 ** the set of snippetMatch structures. If one is found, proceed. | |
3524 ** If one cannot be found, then remove snippets ii..(ii+N-1) | |
3525 ** from the matching snippets, where N is the number of tokens | |
3526 ** in phrase pRight->pPhrase. | |
3527 */ | |
3528 for(jj=0; isOk==0 && jj<pSnippet->nMatch; jj++){ | |
3529 struct snippetMatch *p2 = &pSnippet->aMatch[jj]; | |
3530 if( p2->iTerm==(iLeft-1) ){ | |
3531 if( p2->iToken>=(p->iToken-nNear-1) | |
3532 && p2->iToken<(p->iToken+nNear+nToken) | |
3533 ){ | |
3534 isOk = 1; | |
3535 } | |
3536 } | |
3537 } | |
3538 if( !isOk ){ | |
3539 int kk; | |
3540 for(kk=0; kk<pRight->pPhrase->nToken; kk++){ | |
3541 pSnippet->aMatch[kk+ii].iTerm = -2; | |
3542 } | |
3543 return 1; | |
3544 } | 2831 } |
3545 } | 2832 } |
3546 if( p->iTerm==(iLeft-1) ){ | 2833 |
3547 int isOk = 0; | 2834 if( pBest->nCost>nDoc ){ |
3548 for(jj=0; isOk==0 && jj<pSnippet->nMatch; jj++){ | 2835 rc = fts3DeferExpression(p, p->pExpr); |
3549 struct snippetMatch *p2 = &pSnippet->aMatch[jj]; | 2836 break; |
3550 if( p2->iTerm==iLeft ){ | 2837 }else{ |
3551 if( p2->iToken<=(p->iToken+nNear+1) | 2838 rc = fts3EvalExpr(p, pBest->pExpr, &aNew, &nNew, 0); |
3552 && p2->iToken>(p->iToken-nNear-nToken) | 2839 if( rc!=SQLITE_OK ) break; |
3553 ){ | 2840 pBest->pExpr = 0; |
3554 isOk = 1; | 2841 if( ii==0 ){ |
3555 } | 2842 aRet = aNew; |
3556 } | 2843 nRet = nNew; |
3557 } | 2844 nDoc = fts3DoclistCountDocids(0, aRet, nRet); |
3558 if( !isOk ){ | 2845 }else{ |
3559 int kk; | 2846 fts3DoclistMerge( |
3560 for(kk=0; kk<pLeft->pPhrase->nToken; kk++){ | 2847 MERGE_AND, 0, 0, aRet, &nRet, aRet, nRet, aNew, nNew, &nDoc |
3561 pSnippet->aMatch[ii-kk].iTerm = -2; | 2848 ); |
3562 } | 2849 sqlite3_free(aNew); |
3563 return 1; | |
3564 } | 2850 } |
3565 } | 2851 } |
3566 } | 2852 } |
3567 break; | 2853 } |
3568 } | 2854 |
3569 } | 2855 if( rc==SQLITE_OK ){ |
3570 | 2856 *paOut = aRet; |
3571 if( trimSnippetOffsets(pExpr->pRight, pSnippet, piLeft) ){ | 2857 *pnOut = nRet; |
3572 return 1; | 2858 }else{ |
3573 } | 2859 assert( *paOut==0 ); |
3574 } | 2860 sqlite3_free(aRet); |
3575 return 0; | 2861 } |
3576 } | 2862 sqlite3_free(aExpr); |
3577 | 2863 fts3ExprFreeSegReaders(pExpr); |
3578 /* | 2864 |
3579 ** Compute all offsets for the current row of the query. | |
3580 ** If the offsets have already been computed, this routine is a no-op. | |
3581 */ | |
3582 static void snippetAllOffsets(fulltext_cursor *p){ | |
3583 int nColumn; | |
3584 int iColumn, i; | |
3585 int iFirst, iLast; | |
3586 int iTerm = 0; | |
3587 fulltext_vtab *pFts = cursor_vtab(p); | |
3588 | |
3589 if( p->snippet.nMatch || p->pExpr==0 ){ | |
3590 return; | |
3591 } | |
3592 nColumn = pFts->nColumn; | |
3593 iColumn = (p->iCursorType - QUERY_FULLTEXT); | |
3594 if( iColumn<0 || iColumn>=nColumn ){ | |
3595 /* Look for matches over all columns of the full-text index */ | |
3596 iFirst = 0; | |
3597 iLast = nColumn-1; | |
3598 }else{ | |
3599 /* Look for matches in the iColumn-th column of the index only */ | |
3600 iFirst = iColumn; | |
3601 iLast = iColumn; | |
3602 } | |
3603 for(i=iFirst; i<=iLast; i++){ | |
3604 const char *zDoc; | |
3605 int nDoc; | |
3606 zDoc = (const char*)sqlite3_column_text(p->pStmt, i+1); | |
3607 nDoc = sqlite3_column_bytes(p->pStmt, i+1); | |
3608 snippetOffsetsOfColumn(p, &p->snippet, i, zDoc, nDoc); | |
3609 } | |
3610 | |
3611 while( trimSnippetOffsets(p->pExpr, &p->snippet, &iTerm) ){ | |
3612 iTerm = 0; | |
3613 } | |
3614 } | |
3615 | |
3616 /* | |
3617 ** Convert the information in the aMatch[] array of the snippet | |
3618 ** into the string zOffset[0..nOffset-1]. This string is used as | |
3619 ** the return of the SQL offsets() function. | |
3620 */ | |
3621 static void snippetOffsetText(Snippet *p){ | |
3622 int i; | |
3623 int cnt = 0; | |
3624 StringBuffer sb; | |
3625 char zBuf[200]; | |
3626 if( p->zOffset ) return; | |
3627 initStringBuffer(&sb); | |
3628 for(i=0; i<p->nMatch; i++){ | |
3629 struct snippetMatch *pMatch = &p->aMatch[i]; | |
3630 if( pMatch->iTerm>=0 ){ | |
3631 /* If snippetMatch.iTerm is less than 0, then the match was | |
3632 ** discarded as part of processing the NEAR operator (see the | |
3633 ** trimSnippetOffsetsForNear() function for details). Ignore | |
3634 ** it in this case | |
3635 */ | |
3636 zBuf[0] = ' '; | |
3637 sqlite3_snprintf(sizeof(zBuf)-1, &zBuf[cnt>0], "%d %d %d %d", | |
3638 pMatch->iCol, pMatch->iTerm, pMatch->iStart, pMatch->nByte); | |
3639 append(&sb, zBuf); | |
3640 cnt++; | |
3641 } | |
3642 } | |
3643 p->zOffset = stringBufferData(&sb); | |
3644 p->nOffset = stringBufferLength(&sb); | |
3645 } | |
3646 | |
3647 /* | |
3648 ** zDoc[0..nDoc-1] is phrase of text. aMatch[0..nMatch-1] are a set | |
3649 ** of matching words some of which might be in zDoc. zDoc is column | |
3650 ** number iCol. | |
3651 ** | |
3652 ** iBreak is suggested spot in zDoc where we could begin or end an | |
3653 ** excerpt. Return a value similar to iBreak but possibly adjusted | |
3654 ** to be a little left or right so that the break point is better. | |
3655 */ | |
3656 static int wordBoundary( | |
3657 int iBreak, /* The suggested break point */ | |
3658 const char *zDoc, /* Document text */ | |
3659 int nDoc, /* Number of bytes in zDoc[] */ | |
3660 struct snippetMatch *aMatch, /* Matching words */ | |
3661 int nMatch, /* Number of entries in aMatch[] */ | |
3662 int iCol /* The column number for zDoc[] */ | |
3663 ){ | |
3664 int i; | |
3665 if( iBreak<=10 ){ | |
3666 return 0; | |
3667 } | |
3668 if( iBreak>=nDoc-10 ){ | |
3669 return nDoc; | |
3670 } | |
3671 for(i=0; i<nMatch && aMatch[i].iCol<iCol; i++){} | |
3672 while( i<nMatch && aMatch[i].iStart+aMatch[i].nByte<iBreak ){ i++; } | |
3673 if( i<nMatch ){ | |
3674 if( aMatch[i].iStart<iBreak+10 ){ | |
3675 return aMatch[i].iStart; | |
3676 } | |
3677 if( i>0 && aMatch[i-1].iStart+aMatch[i-1].nByte>=iBreak ){ | |
3678 return aMatch[i-1].iStart; | |
3679 } | |
3680 } | |
3681 for(i=1; i<=10; i++){ | |
3682 if( safe_isspace(zDoc[iBreak-i]) ){ | |
3683 return iBreak - i + 1; | |
3684 } | |
3685 if( safe_isspace(zDoc[iBreak+i]) ){ | |
3686 return iBreak + i + 1; | |
3687 } | |
3688 } | |
3689 return iBreak; | |
3690 } | |
3691 | |
3692 | |
3693 | |
3694 /* | |
3695 ** Allowed values for Snippet.aMatch[].snStatus | |
3696 */ | |
3697 #define SNIPPET_IGNORE 0 /* It is ok to omit this match from the snippet */ | |
3698 #define SNIPPET_DESIRED 1 /* We want to include this match in the snippet */ | |
3699 | |
3700 /* | |
3701 ** Generate the text of a snippet. | |
3702 */ | |
3703 static void snippetText( | |
3704 fulltext_cursor *pCursor, /* The cursor we need the snippet for */ | |
3705 const char *zStartMark, /* Markup to appear before each match */ | |
3706 const char *zEndMark, /* Markup to appear after each match */ | |
3707 const char *zEllipsis /* Ellipsis mark */ | |
3708 ){ | |
3709 int i, j; | |
3710 struct snippetMatch *aMatch; | |
3711 int nMatch; | |
3712 int nDesired; | |
3713 StringBuffer sb; | |
3714 int tailCol; | |
3715 int tailOffset; | |
3716 int iCol; | |
3717 int nDoc; | |
3718 const char *zDoc; | |
3719 int iStart, iEnd; | |
3720 int tailEllipsis = 0; | |
3721 int iMatch; | |
3722 | |
3723 | |
3724 sqlite3_free(pCursor->snippet.zSnippet); | |
3725 pCursor->snippet.zSnippet = 0; | |
3726 aMatch = pCursor->snippet.aMatch; | |
3727 nMatch = pCursor->snippet.nMatch; | |
3728 initStringBuffer(&sb); | |
3729 | |
3730 for(i=0; i<nMatch; i++){ | |
3731 aMatch[i].snStatus = SNIPPET_IGNORE; | |
3732 } | |
3733 nDesired = 0; | |
3734 for(i=0; i<FTS3_ROTOR_SZ; i++){ | |
3735 for(j=0; j<nMatch; j++){ | |
3736 if( aMatch[j].iTerm==i ){ | |
3737 aMatch[j].snStatus = SNIPPET_DESIRED; | |
3738 nDesired++; | |
3739 break; | |
3740 } | |
3741 } | |
3742 } | |
3743 | |
3744 iMatch = 0; | |
3745 tailCol = -1; | |
3746 tailOffset = 0; | |
3747 for(i=0; i<nMatch && nDesired>0; i++){ | |
3748 if( aMatch[i].snStatus!=SNIPPET_DESIRED ) continue; | |
3749 nDesired--; | |
3750 iCol = aMatch[i].iCol; | |
3751 zDoc = (const char*)sqlite3_column_text(pCursor->pStmt, iCol+1); | |
3752 nDoc = sqlite3_column_bytes(pCursor->pStmt, iCol+1); | |
3753 iStart = aMatch[i].iStart - 40; | |
3754 iStart = wordBoundary(iStart, zDoc, nDoc, aMatch, nMatch, iCol); | |
3755 if( iStart<=10 ){ | |
3756 iStart = 0; | |
3757 } | |
3758 if( iCol==tailCol && iStart<=tailOffset+20 ){ | |
3759 iStart = tailOffset; | |
3760 } | |
3761 if( (iCol!=tailCol && tailCol>=0) || iStart!=tailOffset ){ | |
3762 trimWhiteSpace(&sb); | |
3763 appendWhiteSpace(&sb); | |
3764 append(&sb, zEllipsis); | |
3765 appendWhiteSpace(&sb); | |
3766 } | |
3767 iEnd = aMatch[i].iStart + aMatch[i].nByte + 40; | |
3768 iEnd = wordBoundary(iEnd, zDoc, nDoc, aMatch, nMatch, iCol); | |
3769 if( iEnd>=nDoc-10 ){ | |
3770 iEnd = nDoc; | |
3771 tailEllipsis = 0; | |
3772 }else{ | 2865 }else{ |
3773 tailEllipsis = 1; | 2866 char *aLeft; |
3774 } | 2867 char *aRight; |
3775 while( iMatch<nMatch && aMatch[iMatch].iCol<iCol ){ iMatch++; } | 2868 int nLeft; |
3776 while( iStart<iEnd ){ | 2869 int nRight; |
3777 while( iMatch<nMatch && aMatch[iMatch].iStart<iStart | 2870 |
3778 && aMatch[iMatch].iCol<=iCol ){ | 2871 assert( pExpr->eType==FTSQUERY_NEAR |
3779 iMatch++; | 2872 || pExpr->eType==FTSQUERY_OR |
3780 } | 2873 || pExpr->eType==FTSQUERY_NOT |
3781 if( iMatch<nMatch && aMatch[iMatch].iStart<iEnd | 2874 || (pExpr->eType==FTSQUERY_AND && p->eEvalmode==FTS3_EVAL_NEXT) |
3782 && aMatch[iMatch].iCol==iCol ){ | 2875 ); |
3783 nappend(&sb, &zDoc[iStart], aMatch[iMatch].iStart - iStart); | 2876 |
3784 iStart = aMatch[iMatch].iStart; | 2877 if( 0==(rc = fts3EvalExpr(p, pExpr->pRight, &aRight, &nRight, isReqPos)) |
3785 append(&sb, zStartMark); | 2878 && 0==(rc = fts3EvalExpr(p, pExpr->pLeft, &aLeft, &nLeft, isReqPos)) |
3786 nappend(&sb, &zDoc[iStart], aMatch[iMatch].nByte); | |
3787 append(&sb, zEndMark); | |
3788 iStart += aMatch[iMatch].nByte; | |
3789 for(j=iMatch+1; j<nMatch; j++){ | |
3790 if( aMatch[j].iTerm==aMatch[iMatch].iTerm | |
3791 && aMatch[j].snStatus==SNIPPET_DESIRED ){ | |
3792 nDesired--; | |
3793 aMatch[j].snStatus = SNIPPET_IGNORE; | |
3794 } | |
3795 } | |
3796 }else{ | |
3797 nappend(&sb, &zDoc[iStart], iEnd - iStart); | |
3798 iStart = iEnd; | |
3799 } | |
3800 } | |
3801 tailCol = iCol; | |
3802 tailOffset = iEnd; | |
3803 } | |
3804 trimWhiteSpace(&sb); | |
3805 if( tailEllipsis ){ | |
3806 appendWhiteSpace(&sb); | |
3807 append(&sb, zEllipsis); | |
3808 } | |
3809 pCursor->snippet.zSnippet = stringBufferData(&sb); | |
3810 pCursor->snippet.nSnippet = stringBufferLength(&sb); | |
3811 } | |
3812 | |
3813 | |
3814 /* | |
3815 ** Close the cursor. For additional information see the documentation | |
3816 ** on the xClose method of the virtual table interface. | |
3817 */ | |
3818 static int fulltextClose(sqlite3_vtab_cursor *pCursor){ | |
3819 fulltext_cursor *c = (fulltext_cursor *) pCursor; | |
3820 FTSTRACE(("FTS3 Close %p\n", c)); | |
3821 sqlite3_finalize(c->pStmt); | |
3822 sqlite3Fts3ExprFree(c->pExpr); | |
3823 snippetClear(&c->snippet); | |
3824 if( c->result.nData!=0 ){ | |
3825 dlrDestroy(&c->reader); | |
3826 } | |
3827 dataBufferDestroy(&c->result); | |
3828 sqlite3_free(c); | |
3829 return SQLITE_OK; | |
3830 } | |
3831 | |
3832 static int fulltextNext(sqlite3_vtab_cursor *pCursor){ | |
3833 fulltext_cursor *c = (fulltext_cursor *) pCursor; | |
3834 int rc; | |
3835 | |
3836 FTSTRACE(("FTS3 Next %p\n", pCursor)); | |
3837 snippetClear(&c->snippet); | |
3838 if( c->iCursorType < QUERY_FULLTEXT ){ | |
3839 /* TODO(shess) Handle SQLITE_SCHEMA AND SQLITE_BUSY. */ | |
3840 rc = sqlite3_step(c->pStmt); | |
3841 switch( rc ){ | |
3842 case SQLITE_ROW: | |
3843 c->eof = 0; | |
3844 return SQLITE_OK; | |
3845 case SQLITE_DONE: | |
3846 c->eof = 1; | |
3847 return SQLITE_OK; | |
3848 default: | |
3849 c->eof = 1; | |
3850 return rc; | |
3851 } | |
3852 } else { /* full-text query */ | |
3853 rc = sqlite3_reset(c->pStmt); | |
3854 if( rc!=SQLITE_OK ) return rc; | |
3855 | |
3856 if( c->result.nData==0 || dlrAtEnd(&c->reader) ){ | |
3857 c->eof = 1; | |
3858 return SQLITE_OK; | |
3859 } | |
3860 rc = sqlite3_bind_int64(c->pStmt, 1, dlrDocid(&c->reader)); | |
3861 if( rc!=SQLITE_OK ) return rc; | |
3862 rc = dlrStep(&c->reader); | |
3863 if( rc!=SQLITE_OK ) return rc; | |
3864 /* TODO(shess) Handle SQLITE_SCHEMA AND SQLITE_BUSY. */ | |
3865 rc = sqlite3_step(c->pStmt); | |
3866 if( rc==SQLITE_ROW ){ /* the case we expect */ | |
3867 c->eof = 0; | |
3868 return SQLITE_OK; | |
3869 } | |
3870 /* Corrupt if the index refers to missing document. */ | |
3871 if( rc==SQLITE_DONE ) return SQLITE_CORRUPT_BKPT; | |
3872 | |
3873 return rc; | |
3874 } | |
3875 } | |
3876 | |
3877 | |
3878 /* TODO(shess) If we pushed LeafReader to the top of the file, or to | |
3879 ** another file, term_select() could be pushed above | |
3880 ** docListOfTerm(). | |
3881 */ | |
3882 static int termSelect(fulltext_vtab *v, int iColumn, | |
3883 const char *pTerm, int nTerm, int isPrefix, | |
3884 DocListType iType, DataBuffer *out); | |
3885 | |
3886 /* | |
3887 ** Return a DocList corresponding to the phrase *pPhrase. | |
3888 ** | |
3889 ** The resulting DL_DOCIDS doclist is stored in pResult, which is | |
3890 ** overwritten. | |
3891 */ | |
3892 static int docListOfPhrase( | |
3893 fulltext_vtab *pTab, /* The full text index */ | |
3894 Fts3Phrase *pPhrase, /* Phrase to return a doclist corresponding to */ | |
3895 DocListType eListType, /* Either DL_DOCIDS or DL_POSITIONS */ | |
3896 DataBuffer *pResult /* Write the result here */ | |
3897 ){ | |
3898 int ii; | |
3899 int rc = SQLITE_OK; | |
3900 int iCol = pPhrase->iColumn; | |
3901 DocListType eType = eListType; | |
3902 assert( eType==DL_POSITIONS || eType==DL_DOCIDS ); | |
3903 if( pPhrase->nToken>1 ){ | |
3904 eType = DL_POSITIONS; | |
3905 } | |
3906 | |
3907 /* This code should never be called with buffered updates. */ | |
3908 assert( pTab->nPendingData<0 ); | |
3909 | |
3910 for(ii=0; rc==SQLITE_OK && ii<pPhrase->nToken; ii++){ | |
3911 DataBuffer tmp; | |
3912 struct PhraseToken *p = &pPhrase->aToken[ii]; | |
3913 rc = termSelect(pTab, iCol, p->z, p->n, p->isPrefix, eType, &tmp); | |
3914 if( rc==SQLITE_OK ){ | |
3915 if( ii==0 ){ | |
3916 *pResult = tmp; | |
3917 }else{ | |
3918 DataBuffer res = *pResult; | |
3919 dataBufferInit(pResult, 0); | |
3920 if( ii==(pPhrase->nToken-1) ){ | |
3921 eType = eListType; | |
3922 } | |
3923 rc = docListPhraseMerge( | |
3924 res.pData, res.nData, tmp.pData, tmp.nData, 0, 0, eType, pResult | |
3925 ); | |
3926 dataBufferDestroy(&res); | |
3927 dataBufferDestroy(&tmp); | |
3928 if( rc!= SQLITE_OK ) return rc; | |
3929 } | |
3930 } | |
3931 } | |
3932 | |
3933 return rc; | |
3934 } | |
3935 | |
3936 /* | |
3937 ** Evaluate the full-text expression pExpr against fts3 table pTab. Write | |
3938 ** the results into pRes. | |
3939 */ | |
3940 static int evalFts3Expr( | |
3941 fulltext_vtab *pTab, /* Fts3 Virtual table object */ | |
3942 Fts3Expr *pExpr, /* Parsed fts3 expression */ | |
3943 DataBuffer *pRes /* OUT: Write results of the expression here */ | |
3944 ){ | |
3945 int rc = SQLITE_OK; | |
3946 | |
3947 /* Initialize the output buffer. If this is an empty query (pExpr==0), | |
3948 ** this is all that needs to be done. Empty queries produce empty | |
3949 ** result sets. | |
3950 */ | |
3951 dataBufferInit(pRes, 0); | |
3952 | |
3953 if( pExpr ){ | |
3954 if( pExpr->eType==FTSQUERY_PHRASE ){ | |
3955 DocListType eType = DL_DOCIDS; | |
3956 if( pExpr->pParent && pExpr->pParent->eType==FTSQUERY_NEAR ){ | |
3957 eType = DL_POSITIONS; | |
3958 } | |
3959 rc = docListOfPhrase(pTab, pExpr->pPhrase, eType, pRes); | |
3960 }else{ | |
3961 DataBuffer lhs; | |
3962 DataBuffer rhs; | |
3963 | |
3964 dataBufferInit(&rhs, 0); | |
3965 if( SQLITE_OK==(rc = evalFts3Expr(pTab, pExpr->pLeft, &lhs)) | |
3966 && SQLITE_OK==(rc = evalFts3Expr(pTab, pExpr->pRight, &rhs)) | |
3967 ){ | 2879 ){ |
3968 switch( pExpr->eType ){ | 2880 switch( pExpr->eType ){ |
3969 case FTSQUERY_NEAR: { | 2881 case FTSQUERY_NEAR: { |
3970 int nToken; | |
3971 Fts3Expr *pLeft; | 2882 Fts3Expr *pLeft; |
3972 DocListType eType = DL_DOCIDS; | 2883 Fts3Expr *pRight; |
| 2884 int mergetype = MERGE_NEAR; |
3973 if( pExpr->pParent && pExpr->pParent->eType==FTSQUERY_NEAR ){ | 2885 if( pExpr->pParent && pExpr->pParent->eType==FTSQUERY_NEAR ){ |
3974 eType = DL_POSITIONS; | 2886 mergetype = MERGE_POS_NEAR; |
3975 } | 2887 } |
3976 pLeft = pExpr->pLeft; | 2888 pLeft = pExpr->pLeft; |
3977 while( pLeft->eType==FTSQUERY_NEAR ){ | 2889 while( pLeft->eType==FTSQUERY_NEAR ){ |
3978 pLeft=pLeft->pRight; | 2890 pLeft=pLeft->pRight; |
3979 } | 2891 } |
3980 assert( pExpr->pRight->eType==FTSQUERY_PHRASE ); | 2892 pRight = pExpr->pRight; |
| 2893 assert( pRight->eType==FTSQUERY_PHRASE ); |
3981 assert( pLeft->eType==FTSQUERY_PHRASE ); | 2894 assert( pLeft->eType==FTSQUERY_PHRASE ); |
3982 nToken = pLeft->pPhrase->nToken + pExpr->pRight->pPhrase->nToken; | 2895 |
3983 rc = docListPhraseMerge(lhs.pData, lhs.nData, rhs.pData, rhs.nData, | 2896 rc = fts3NearMerge(mergetype, pExpr->nNear, |
3984 pExpr->nNear+1, nToken, eType, pRes | 2897 pLeft->pPhrase->nToken, aLeft, nLeft, |
| 2898 pRight->pPhrase->nToken, aRight, nRight, |
| 2899 paOut, pnOut |
3985 ); | 2900 ); |
| 2901 sqlite3_free(aLeft); |
3986 break; | 2902 break; |
3987 } | 2903 } |
3988 case FTSQUERY_NOT: { | 2904 |
3989 rc = docListExceptMerge(lhs.pData, lhs.nData, rhs.pData, rhs.nData,p
Res); | 2905 case FTSQUERY_OR: { |
| 2906 /* Allocate a buffer for the output. The maximum size is the |
| 2907 ** sum of the sizes of the two input buffers. The +1 term is |
| 2908 ** so that a buffer of zero bytes is never allocated - this can |
| 2909 ** cause fts3DoclistMerge() to incorrectly return SQLITE_NOMEM. |
| 2910 */ |
| 2911 char *aBuffer = sqlite3_malloc(nRight+nLeft+1); |
| 2912 rc = fts3DoclistMerge(MERGE_OR, 0, 0, aBuffer, pnOut, |
| 2913 aLeft, nLeft, aRight, nRight, 0 |
| 2914 ); |
| 2915 *paOut = aBuffer; |
| 2916 sqlite3_free(aLeft); |
3990 break; | 2917 break; |
3991 } | 2918 } |
3992 case FTSQUERY_AND: { | 2919 |
3993 rc = docListAndMerge(lhs.pData, lhs.nData, rhs.pData, rhs.nData, pRe
s); | 2920 default: { |
3994 break; | 2921 assert( FTSQUERY_NOT==MERGE_NOT && FTSQUERY_AND==MERGE_AND ); |
3995 } | 2922 fts3DoclistMerge(pExpr->eType, 0, 0, aLeft, pnOut, |
3996 case FTSQUERY_OR: { | 2923 aLeft, nLeft, aRight, nRight, 0 |
3997 rc = docListOrMerge(lhs.pData, lhs.nData, rhs.pData, rhs.nData, pRes
); | 2924 ); |
| 2925 *paOut = aLeft; |
3998 break; | 2926 break; |
3999 } | 2927 } |
4000 } | 2928 } |
4001 } | 2929 } |
4002 dataBufferDestroy(&lhs); | 2930 sqlite3_free(aRight); |
4003 dataBufferDestroy(&rhs); | 2931 } |
4004 } | 2932 } |
4005 } | 2933 |
4006 | 2934 assert( rc==SQLITE_OK || *paOut==0 ); |
4007 return rc; | 2935 return rc; |
4008 } | 2936 } |
4009 | 2937 |
4010 /* TODO(shess) Refactor the code to remove this forward decl. */ | 2938 /* |
4011 static int flushPendingTerms(fulltext_vtab *v); | 2939 ** This function is called from within xNext() for each row visited by |
4012 | 2940 ** an FTS3 query. If evaluating the FTS3 query expression within xFilter() |
4013 /* Perform a full-text query using the search expression in | 2941 ** was able to determine the exact set of matching rows, this function sets |
4014 ** zInput[0..nInput-1]. Return a list of matching documents | 2942 ** *pbRes to true and returns SQLITE_IO immediately. |
4015 ** in pResult. | 2943 ** |
4016 ** | 2944 ** Otherwise, if evaluating the query expression within xFilter() returned a |
4017 ** Queries must match column iColumn. Or if iColumn>=nColumn | 2945 ** superset of the matching documents instead of an exact set (this happens |
4018 ** they are allowed to match against any column. | 2946 ** when the query includes very common tokens and it is deemed too expensive to |
4019 */ | 2947 ** load their doclists from disk), this function tests if the current row |
4020 static int fulltextQuery( | 2948 ** really does match the FTS3 query. |
4021 fulltext_vtab *v, /* The full text index */ | 2949 ** |
4022 int iColumn, /* Match against this column by default */ | 2950 ** If an error occurs, an SQLite error code is returned. Otherwise, SQLITE_OK |
4023 const char *zInput, /* The query string */ | 2951 ** is returned and *pbRes is set to true if the current row matches the |
4024 int nInput, /* Number of bytes in zInput[] */ | 2952 ** FTS3 query (and should be included in the results returned to SQLite), or |
4025 DataBuffer *pResult, /* Write the result doclist here */ | 2953 ** false otherwise. |
4026 Fts3Expr **ppExpr /* Put parsed query string here */ | 2954 */ |
4027 ){ | 2955 static int fts3EvalDeferred( |
4028 int rc; | 2956 Fts3Cursor *pCsr, /* FTS3 cursor pointing at row to test */ |
4029 | 2957 int *pbRes /* OUT: Set to true if row is a match */ |
4030 /* TODO(shess) Instead of flushing pendingTerms, we could query for | 2958 ){ |
4031 ** the relevant term and merge the doclist into what we receive from | 2959 int rc = SQLITE_OK; |
4032 ** the database. Wait and see if this is a common issue, first. | 2960 if( pCsr->pDeferred==0 ){ |
4033 ** | 2961 *pbRes = 1; |
4034 ** A good reason not to flush is to not generate update-related | 2962 }else{ |
4035 ** error codes from here. | 2963 rc = fts3CursorSeek(0, pCsr); |
4036 */ | 2964 if( rc==SQLITE_OK ){ |
4037 | 2965 sqlite3Fts3FreeDeferredDoclists(pCsr); |
4038 /* Flush any buffered updates before executing the query. */ | 2966 rc = sqlite3Fts3CacheDeferredDoclists(pCsr); |
4039 rc = flushPendingTerms(v); | 2967 } |
4040 if( rc!=SQLITE_OK ){ | 2968 if( rc==SQLITE_OK ){ |
4041 return rc; | 2969 char *a = 0; |
4042 } | 2970 int n = 0; |
4043 | 2971 rc = fts3EvalExpr(pCsr, pCsr->pExpr, &a, &n, 0); |
4044 /* Parse the query passed to the MATCH operator. */ | 2972 assert( n>=0 ); |
4045 rc = sqlite3Fts3ExprParse(v->pTokenizer, | 2973 *pbRes = (n>0); |
4046 v->azColumn, v->nColumn, iColumn, zInput, nInput, ppExpr | 2974 sqlite3_free(a); |
4047 ); | 2975 } |
4048 if( rc!=SQLITE_OK ){ | 2976 } |
4049 assert( 0==(*ppExpr) ); | 2977 return rc; |
4050 return rc; | 2978 } |
4051 } | 2979 |
4052 | 2980 /* |
4053 return evalFts3Expr(v, *ppExpr, pResult); | 2981 ** Advance the cursor to the next row in the %_content table that |
| 2982 ** matches the search criteria. For a MATCH search, this will be |
| 2983 ** the next row that matches. For a full-table scan, this will be |
| 2984 ** simply the next row in the %_content table. For a docid lookup, |
| 2985 ** this routine simply sets the EOF flag. |
| 2986 ** |
| 2987 ** Return SQLITE_OK if nothing goes wrong. SQLITE_OK is returned |
| 2988 ** even if we reach end-of-file. The fts3EofMethod() will be called |
| 2989 ** subsequently to determine whether or not an EOF was hit. |
| 2990 */ |
| 2991 static int fts3NextMethod(sqlite3_vtab_cursor *pCursor){ |
| 2992 int res; |
| 2993 int rc = SQLITE_OK; /* Return code */ |
| 2994 Fts3Cursor *pCsr = (Fts3Cursor *)pCursor; |
| 2995 |
| 2996 pCsr->eEvalmode = FTS3_EVAL_NEXT; |
| 2997 do { |
| 2998 if( pCsr->aDoclist==0 ){ |
| 2999 if( SQLITE_ROW!=sqlite3_step(pCsr->pStmt) ){ |
| 3000 pCsr->isEof = 1; |
| 3001 rc = sqlite3_reset(pCsr->pStmt); |
| 3002 break; |
| 3003 } |
| 3004 pCsr->iPrevId = sqlite3_column_int64(pCsr->pStmt, 0); |
| 3005 }else{ |
| 3006 if( pCsr->pNextId>=&pCsr->aDoclist[pCsr->nDoclist] ){ |
| 3007 pCsr->isEof = 1; |
| 3008 break; |
| 3009 } |
| 3010 sqlite3_reset(pCsr->pStmt); |
| 3011 fts3GetDeltaVarint(&pCsr->pNextId, &pCsr->iPrevId); |
| 3012 pCsr->isRequireSeek = 1; |
| 3013 pCsr->isMatchinfoNeeded = 1; |
| 3014 } |
| 3015 }while( SQLITE_OK==(rc = fts3EvalDeferred(pCsr, &res)) && res==0 ); |
| 3016 |
| 3017 return rc; |
4054 } | 3018 } |
4055 | 3019 |
4056 /* | 3020 /* |
4057 ** This is the xFilter interface for the virtual table. See | 3021 ** This is the xFilter interface for the virtual table. See |
4058 ** the virtual table xFilter method documentation for additional | 3022 ** the virtual table xFilter method documentation for additional |
4059 ** information. | 3023 ** information. |
4060 ** | 3024 ** |
4061 ** If idxNum==QUERY_GENERIC then do a full table scan against | 3025 ** If idxNum==FTS3_FULLSCAN_SEARCH then do a full table scan against |
4062 ** the %_content table. | 3026 ** the %_content table. |
4063 ** | 3027 ** |
4064 ** If idxNum==QUERY_DOCID then do a docid lookup for a single entry | 3028 ** If idxNum==FTS3_DOCID_SEARCH then do a docid lookup for a single entry |
4065 ** in the %_content table. | 3029 ** in the %_content table. |
4066 ** | 3030 ** |
4067 ** If idxNum>=QUERY_FULLTEXT then use the full text index. The | 3031 ** If idxNum>=FTS3_FULLTEXT_SEARCH then use the full text index. The |
4068 ** column on the left-hand side of the MATCH operator is column | 3032 ** column on the left-hand side of the MATCH operator is column |
4069 ** number idxNum-QUERY_FULLTEXT, 0 indexed. argv[0] is the right-hand | 3033 ** number idxNum-FTS3_FULLTEXT_SEARCH, 0 indexed. argv[0] is the right-hand |
4070 ** side of the MATCH operator. | 3034 ** side of the MATCH operator. |
4071 */ | 3035 */ |
4072 /* TODO(shess) Upgrade the cursor initialization and destruction to | 3036 static int fts3FilterMethod( |
4073 ** account for fulltextFilter() being called multiple times on the | 3037 sqlite3_vtab_cursor *pCursor, /* The cursor used for this query */ |
4074 ** same cursor. The current solution is very fragile. Apply fix to | 3038 int idxNum, /* Strategy index */ |
4075 ** fts3 as appropriate. | 3039 const char *idxStr, /* Unused */ |
4076 */ | 3040 int nVal, /* Number of elements in apVal */ |
4077 static int fulltextFilter( | 3041 sqlite3_value **apVal /* Arguments for the indexing scheme */ |
4078 sqlite3_vtab_cursor *pCursor, /* The cursor used for this query */ | 3042 ){ |
4079 int idxNum, const char *idxStr, /* Which indexing scheme to use */ | 3043 const char *azSql[] = { |
4080 int argc, sqlite3_value **argv /* Arguments for the indexing scheme */ | 3044 "SELECT %s FROM %Q.'%q_content' AS x WHERE docid = ?", /* non-full-scan */ |
4081 ){ | 3045 "SELECT %s FROM %Q.'%q_content' AS x ", /* full-scan */ |
4082 fulltext_cursor *c = (fulltext_cursor *) pCursor; | 3046 }; |
4083 fulltext_vtab *v = cursor_vtab(c); | 3047 int rc; /* Return code */ |
| 3048 char *zSql; /* SQL statement used to access %_content */ |
| 3049 Fts3Table *p = (Fts3Table *)pCursor->pVtab; |
| 3050 Fts3Cursor *pCsr = (Fts3Cursor *)pCursor; |
| 3051 |
| 3052 UNUSED_PARAMETER(idxStr); |
| 3053 UNUSED_PARAMETER(nVal); |
| 3054 |
| 3055 assert( idxNum>=0 && idxNum<=(FTS3_FULLTEXT_SEARCH+p->nColumn) ); |
| 3056 assert( nVal==0 || nVal==1 ); |
| 3057 assert( (nVal==0)==(idxNum==FTS3_FULLSCAN_SEARCH) ); |
| 3058 assert( p->pSegments==0 ); |
| 3059 |
| 3060 /* In case the cursor has been used before, clear it now. */ |
| 3061 sqlite3_finalize(pCsr->pStmt); |
| 3062 sqlite3_free(pCsr->aDoclist); |
| 3063 sqlite3Fts3ExprFree(pCsr->pExpr); |
| 3064 memset(&pCursor[1], 0, sizeof(Fts3Cursor)-sizeof(sqlite3_vtab_cursor)); |
| 3065 |
| 3066 if( idxNum!=FTS3_DOCID_SEARCH && idxNum!=FTS3_FULLSCAN_SEARCH ){ |
| 3067 int iCol = idxNum-FTS3_FULLTEXT_SEARCH; |
| 3068 const char *zQuery = (const char *)sqlite3_value_text(apVal[0]); |
| 3069 |
| 3070 if( zQuery==0 && sqlite3_value_type(apVal[0])!=SQLITE_NULL ){ |
| 3071 return SQLITE_NOMEM; |
| 3072 } |
| 3073 |
| 3074 rc = sqlite3Fts3ExprParse(p->pTokenizer, p->azColumn, p->nColumn, |
| 3075 iCol, zQuery, -1, &pCsr->pExpr |
| 3076 ); |
| 3077 if( rc!=SQLITE_OK ){ |
| 3078 if( rc==SQLITE_ERROR ){ |
| 3079 p->base.zErrMsg = sqlite3_mprintf("malformed MATCH expression: [%s]", |
| 3080 zQuery); |
| 3081 } |
| 3082 return rc; |
| 3083 } |
| 3084 |
| 3085 rc = sqlite3Fts3ReadLock(p); |
| 3086 if( rc!=SQLITE_OK ) return rc; |
| 3087 |
| 3088 rc = fts3EvalExpr(pCsr, pCsr->pExpr, &pCsr->aDoclist, &pCsr->nDoclist, 0); |
| 3089 sqlite3Fts3SegmentsClose(p); |
| 3090 if( rc!=SQLITE_OK ) return rc; |
| 3091 pCsr->pNextId = pCsr->aDoclist; |
| 3092 pCsr->iPrevId = 0; |
| 3093 } |
| 3094 |
| 3095 /* Compile a SELECT statement for this cursor. For a full-table-scan, the |
| 3096 ** statement loops through all rows of the %_content table. For a |
| 3097 ** full-text query or docid lookup, the statement retrieves a single |
| 3098 ** row by docid. |
| 3099 */ |
| 3100 zSql = (char *)azSql[idxNum==FTS3_FULLSCAN_SEARCH]; |
| 3101 zSql = sqlite3_mprintf(zSql, p->zReadExprlist, p->zDb, p->zName); |
| 3102 if( !zSql ){ |
| 3103 rc = SQLITE_NOMEM; |
| 3104 }else{ |
| 3105 rc = sqlite3_prepare_v2(p->db, zSql, -1, &pCsr->pStmt, 0); |
| 3106 sqlite3_free(zSql); |
| 3107 } |
| 3108 if( rc==SQLITE_OK && idxNum==FTS3_DOCID_SEARCH ){ |
| 3109 rc = sqlite3_bind_value(pCsr->pStmt, 1, apVal[0]); |
| 3110 } |
| 3111 pCsr->eSearch = (i16)idxNum; |
| 3112 |
| 3113 if( rc!=SQLITE_OK ) return rc; |
| 3114 return fts3NextMethod(pCursor); |
| 3115 } |
| 3116 |
| 3117 /* |
| 3118 ** This is the xEof method of the virtual table. SQLite calls this |
| 3119 ** routine to find out if it has reached the end of a result set. |
| 3120 */ |
| 3121 static int fts3EofMethod(sqlite3_vtab_cursor *pCursor){ |
| 3122 return ((Fts3Cursor *)pCursor)->isEof; |
| 3123 } |
| 3124 |
| 3125 /* |
| 3126 ** This is the xRowid method. The SQLite core calls this routine to |
| 3127 ** retrieve the rowid for the current row of the result set. fts3 |
| 3128 ** exposes %_content.docid as the rowid for the virtual table. The |
| 3129 ** rowid should be written to *pRowid. |
| 3130 */ |
| 3131 static int fts3RowidMethod(sqlite3_vtab_cursor *pCursor, sqlite_int64 *pRowid){ |
| 3132 Fts3Cursor *pCsr = (Fts3Cursor *) pCursor; |
| 3133 if( pCsr->aDoclist ){ |
| 3134 *pRowid = pCsr->iPrevId; |
| 3135 }else{ |
| 3136 /* This branch runs if the query is implemented using a full-table scan |
| 3137 ** (not using the full-text index). In this case grab the rowid from the |
| 3138 ** SELECT statement. |
| 3139 */ |
| 3140 assert( pCsr->isRequireSeek==0 ); |
| 3141 *pRowid = sqlite3_column_int64(pCsr->pStmt, 0); |
| 3142 } |
| 3143 return SQLITE_OK; |
| 3144 } |
| 3145 |
| 3146 /* |
| 3147 ** This is the xColumn method, called by SQLite to request a value from |
| 3148 ** the row that the supplied cursor currently points to. |
| 3149 */ |
| 3150 static int fts3ColumnMethod( |
| 3151 sqlite3_vtab_cursor *pCursor, /* Cursor to retrieve value from */ |
| 3152 sqlite3_context *pContext, /* Context for sqlite3_result_xxx() calls */ |
| 3153 int iCol /* Index of column to read value from */ |
| 3154 ){ |
| 3155 int rc; /* Return Code */ |
| 3156 Fts3Cursor *pCsr = (Fts3Cursor *) pCursor; |
| 3157 Fts3Table *p = (Fts3Table *)pCursor->pVtab; |
| 3158 |
| 3159 /* The column value supplied by SQLite must be in range. */ |
| 3160 assert( iCol>=0 && iCol<=p->nColumn+1 ); |
| 3161 |
| 3162 if( iCol==p->nColumn+1 ){ |
| 3163 /* This call is a request for the "docid" column. Since "docid" is an |
| 3164 ** alias for "rowid", use the xRowid() method to obtain the value. |
| 3165 */ |
| 3166 sqlite3_int64 iRowid; |
| 3167 rc = fts3RowidMethod(pCursor, &iRowid); |
| 3168 sqlite3_result_int64(pContext, iRowid); |
| 3169 }else if( iCol==p->nColumn ){ |
| 3170 /* The extra column whose name is the same as the table. |
| 3171 ** Return a blob which is a pointer to the cursor. |
| 3172 */ |
| 3173 sqlite3_result_blob(pContext, &pCsr, sizeof(pCsr), SQLITE_TRANSIENT); |
| 3174 rc = SQLITE_OK; |
| 3175 }else{ |
| 3176 rc = fts3CursorSeek(0, pCsr); |
| 3177 if( rc==SQLITE_OK ){ |
| 3178 sqlite3_result_value(pContext, sqlite3_column_value(pCsr->pStmt, iCol+1)); |
| 3179 } |
| 3180 } |
| 3181 return rc; |
| 3182 } |
| 3183 |
| 3184 /* |
| 3185 ** This function is the implementation of the xUpdate callback used by |
| 3186 ** FTS3 virtual tables. It is invoked by SQLite each time a row is to be |
| 3187 ** inserted, updated or deleted. |
| 3188 */ |
| 3189 static int fts3UpdateMethod( |
| 3190 sqlite3_vtab *pVtab, /* Virtual table handle */ |
| 3191 int nArg, /* Size of argument array */ |
| 3192 sqlite3_value **apVal, /* Array of arguments */ |
| 3193 sqlite_int64 *pRowid /* OUT: The affected (or effected) rowid */ |
| 3194 ){ |
| 3195 return sqlite3Fts3UpdateMethod(pVtab, nArg, apVal, pRowid); |
| 3196 } |
| 3197 |
| 3198 /* |
| 3199 ** Implementation of xSync() method. Flush the contents of the pending-terms |
| 3200 ** hash-table to the database. |
| 3201 */ |
| 3202 static int fts3SyncMethod(sqlite3_vtab *pVtab){ |
| 3203 int rc = sqlite3Fts3PendingTermsFlush((Fts3Table *)pVtab); |
| 3204 sqlite3Fts3SegmentsClose((Fts3Table *)pVtab); |
| 3205 return rc; |
| 3206 } |
| 3207 |
| 3208 /* |
| 3209 ** Implementation of xBegin() method. This is a no-op. |
| 3210 */ |
| 3211 static int fts3BeginMethod(sqlite3_vtab *pVtab){ |
| 3212 UNUSED_PARAMETER(pVtab); |
| 3213 assert( ((Fts3Table *)pVtab)->nPendingData==0 ); |
| 3214 return SQLITE_OK; |
| 3215 } |
| 3216 |
| 3217 /* |
| 3218 ** Implementation of xCommit() method. This is a no-op. The contents of |
| 3219 ** the pending-terms hash-table have already been flushed into the database |
| 3220 ** by fts3SyncMethod(). |
| 3221 */ |
| 3222 static int fts3CommitMethod(sqlite3_vtab *pVtab){ |
| 3223 UNUSED_PARAMETER(pVtab); |
| 3224 assert( ((Fts3Table *)pVtab)->nPendingData==0 ); |
| 3225 return SQLITE_OK; |
| 3226 } |
| 3227 |
| 3228 /* |
| 3229 ** Implementation of xRollback(). Discard the contents of the pending-terms |
| 3230 ** hash-table. Any changes made to the database are reverted by SQLite. |
| 3231 */ |
| 3232 static int fts3RollbackMethod(sqlite3_vtab *pVtab){ |
| 3233 sqlite3Fts3PendingTermsClear((Fts3Table *)pVtab); |
| 3234 return SQLITE_OK; |
| 3235 } |
| 3236 |
| 3237 /* |
| 3238 ** Load the doclist associated with expression pExpr to pExpr->aDoclist. |
| 3239 ** The loaded doclist contains positions as well as the document ids. |
| 3240 ** This is used by the matchinfo(), snippet() and offsets() auxillary |
| 3241 ** functions. |
| 3242 */ |
| 3243 int sqlite3Fts3ExprLoadDoclist(Fts3Cursor *pCsr, Fts3Expr *pExpr){ |
4084 int rc; | 3244 int rc; |
4085 | 3245 assert( pExpr->eType==FTSQUERY_PHRASE && pExpr->pPhrase ); |
4086 FTSTRACE(("FTS3 Filter %p\n",pCursor)); | 3246 assert( pCsr->eEvalmode==FTS3_EVAL_NEXT ); |
4087 | 3247 rc = fts3EvalExpr(pCsr, pExpr, &pExpr->aDoclist, &pExpr->nDoclist, 1); |
4088 /* If the cursor has a statement that was not prepared according to | 3248 return rc; |
4089 ** idxNum, clear it. I believe all calls to fulltextFilter with a | 3249 } |
4090 ** given cursor will have the same idxNum , but in this case it's | 3250 |
4091 ** easy to be safe. | 3251 int sqlite3Fts3ExprLoadFtDoclist( |
| 3252 Fts3Cursor *pCsr, |
| 3253 Fts3Expr *pExpr, |
| 3254 char **paDoclist, |
| 3255 int *pnDoclist |
| 3256 ){ |
| 3257 int rc; |
| 3258 assert( pCsr->eEvalmode==FTS3_EVAL_NEXT ); |
| 3259 assert( pExpr->eType==FTSQUERY_PHRASE && pExpr->pPhrase ); |
| 3260 pCsr->eEvalmode = FTS3_EVAL_MATCHINFO; |
| 3261 rc = fts3EvalExpr(pCsr, pExpr, paDoclist, pnDoclist, 1); |
| 3262 pCsr->eEvalmode = FTS3_EVAL_NEXT; |
| 3263 return rc; |
| 3264 } |
| 3265 |
| 3266 /* |
| 3267 ** After ExprLoadDoclist() (see above) has been called, this function is |
| 3268 ** used to iterate/search through the position lists that make up the doclist |
| 3269 ** stored in pExpr->aDoclist. |
| 3270 */ |
| 3271 char *sqlite3Fts3FindPositions( |
| 3272 Fts3Expr *pExpr, /* Access this expressions doclist */ |
| 3273 sqlite3_int64 iDocid, /* Docid associated with requested pos-list */ |
| 3274 int iCol /* Column of requested pos-list */ |
| 3275 ){ |
| 3276 assert( pExpr->isLoaded ); |
| 3277 if( pExpr->aDoclist ){ |
| 3278 char *pEnd = &pExpr->aDoclist[pExpr->nDoclist]; |
| 3279 char *pCsr; |
| 3280 |
| 3281 if( pExpr->pCurrent==0 ){ |
| 3282 pExpr->pCurrent = pExpr->aDoclist; |
| 3283 pExpr->iCurrent = 0; |
| 3284 pExpr->pCurrent += sqlite3Fts3GetVarint(pExpr->pCurrent,&pExpr->iCurrent); |
| 3285 } |
| 3286 pCsr = pExpr->pCurrent; |
| 3287 assert( pCsr ); |
| 3288 |
| 3289 while( pCsr<pEnd ){ |
| 3290 if( pExpr->iCurrent<iDocid ){ |
| 3291 fts3PoslistCopy(0, &pCsr); |
| 3292 if( pCsr<pEnd ){ |
| 3293 fts3GetDeltaVarint(&pCsr, &pExpr->iCurrent); |
| 3294 } |
| 3295 pExpr->pCurrent = pCsr; |
| 3296 }else{ |
| 3297 if( pExpr->iCurrent==iDocid ){ |
| 3298 int iThis = 0; |
| 3299 if( iCol<0 ){ |
| 3300 /* If iCol is negative, return a pointer to the start of the |
| 3301 ** position-list (instead of a pointer to the start of a list |
| 3302 ** of offsets associated with a specific column). |
| 3303 */ |
| 3304 return pCsr; |
| 3305 } |
| 3306 while( iThis<iCol ){ |
| 3307 fts3ColumnlistCopy(0, &pCsr); |
| 3308 if( *pCsr==0x00 ) return 0; |
| 3309 pCsr++; |
| 3310 pCsr += sqlite3Fts3GetVarint32(pCsr, &iThis); |
| 3311 } |
| 3312 if( iCol==iThis && (*pCsr&0xFE) ) return pCsr; |
| 3313 } |
| 3314 return 0; |
| 3315 } |
| 3316 } |
| 3317 } |
| 3318 |
| 3319 return 0; |
| 3320 } |
| 3321 |
| 3322 /* |
| 3323 ** Helper function used by the implementation of the overloaded snippet(), |
| 3324 ** offsets() and optimize() SQL functions. |
| 3325 ** |
| 3326 ** If the value passed as the third argument is a blob of size |
| 3327 ** sizeof(Fts3Cursor*), then the blob contents are copied to the |
| 3328 ** output variable *ppCsr and SQLITE_OK is returned. Otherwise, an error |
| 3329 ** message is written to context pContext and SQLITE_ERROR returned. The |
| 3330 ** string passed via zFunc is used as part of the error message. |
| 3331 */ |
| 3332 static int fts3FunctionArg( |
| 3333 sqlite3_context *pContext, /* SQL function call context */ |
| 3334 const char *zFunc, /* Function name */ |
| 3335 sqlite3_value *pVal, /* argv[0] passed to function */ |
| 3336 Fts3Cursor **ppCsr /* OUT: Store cursor handle here */ |
| 3337 ){ |
| 3338 Fts3Cursor *pRet; |
| 3339 if( sqlite3_value_type(pVal)!=SQLITE_BLOB |
| 3340 || sqlite3_value_bytes(pVal)!=sizeof(Fts3Cursor *) |
| 3341 ){ |
| 3342 char *zErr = sqlite3_mprintf("illegal first argument to %s", zFunc); |
| 3343 sqlite3_result_error(pContext, zErr, -1); |
| 3344 sqlite3_free(zErr); |
| 3345 return SQLITE_ERROR; |
| 3346 } |
| 3347 memcpy(&pRet, sqlite3_value_blob(pVal), sizeof(Fts3Cursor *)); |
| 3348 *ppCsr = pRet; |
| 3349 return SQLITE_OK; |
| 3350 } |
| 3351 |
| 3352 /* |
| 3353 ** Implementation of the snippet() function for FTS3 |
| 3354 */ |
| 3355 static void fts3SnippetFunc( |
| 3356 sqlite3_context *pContext, /* SQLite function call context */ |
| 3357 int nVal, /* Size of apVal[] array */ |
| 3358 sqlite3_value **apVal /* Array of arguments */ |
| 3359 ){ |
| 3360 Fts3Cursor *pCsr; /* Cursor handle passed through apVal[0] */ |
| 3361 const char *zStart = "<b>"; |
| 3362 const char *zEnd = "</b>"; |
| 3363 const char *zEllipsis = "<b>...</b>"; |
| 3364 int iCol = -1; |
| 3365 int nToken = 15; /* Default number of tokens in snippet */ |
| 3366 |
| 3367 /* There must be at least one argument passed to this function (otherwise |
| 3368 ** the non-overloaded version would have been called instead of this one). |
4092 */ | 3369 */ |
4093 if( c->pStmt && c->iCursorType!=idxNum ){ | 3370 assert( nVal>=1 ); |
4094 sqlite3_finalize(c->pStmt); | 3371 |
4095 c->pStmt = NULL; | 3372 if( nVal>6 ){ |
4096 } | 3373 sqlite3_result_error(pContext, |
4097 | 3374 "wrong number of arguments to function snippet()", -1); |
4098 /* Get a fresh statement appropriate to idxNum. */ | 3375 return; |
4099 /* TODO(shess): Add a prepared-statement cache in the vt structure. | 3376 } |
4100 ** The cache must handle multiple open cursors. Easier to cache the | 3377 if( fts3FunctionArg(pContext, "snippet", apVal[0], &pCsr) ) return; |
4101 ** statement variants at the vt to reduce malloc/realloc/free here. | 3378 |
4102 ** Or we could have a StringBuffer variant which allowed stack | 3379 switch( nVal ){ |
4103 ** construction for small values. | 3380 case 6: nToken = sqlite3_value_int(apVal[5]); |
4104 */ | 3381 case 5: iCol = sqlite3_value_int(apVal[4]); |
4105 if( !c->pStmt ){ | 3382 case 4: zEllipsis = (const char*)sqlite3_value_text(apVal[3]); |
4106 StringBuffer sb; | 3383 case 3: zEnd = (const char*)sqlite3_value_text(apVal[2]); |
4107 initStringBuffer(&sb); | 3384 case 2: zStart = (const char*)sqlite3_value_text(apVal[1]); |
4108 append(&sb, "SELECT docid, "); | 3385 } |
4109 appendList(&sb, v->nColumn, v->azContentColumn); | 3386 if( !zEllipsis || !zEnd || !zStart ){ |
4110 append(&sb, " FROM %_content"); | 3387 sqlite3_result_error_nomem(pContext); |
4111 if( idxNum!=QUERY_GENERIC ) append(&sb, " WHERE docid = ?"); | 3388 }else if( SQLITE_OK==fts3CursorSeek(pContext, pCsr) ){ |
4112 rc = sql_prepare(v->db, v->zDb, v->zName, &c->pStmt, | 3389 sqlite3Fts3Snippet(pContext, pCsr, zStart, zEnd, zEllipsis, iCol, nToken); |
4113 stringBufferData(&sb)); | 3390 } |
4114 stringBufferDestroy(&sb); | 3391 } |
4115 if( rc!=SQLITE_OK ) return rc; | 3392 |
4116 c->iCursorType = idxNum; | 3393 /* |
4117 }else{ | 3394 ** Implementation of the offsets() function for FTS3 |
4118 sqlite3_reset(c->pStmt); | 3395 */ |
4119 assert( c->iCursorType==idxNum ); | 3396 static void fts3OffsetsFunc( |
4120 } | 3397 sqlite3_context *pContext, /* SQLite function call context */ |
4121 | 3398 int nVal, /* Size of argument array */ |
4122 switch( idxNum ){ | 3399 sqlite3_value **apVal /* Array of arguments */ |
4123 case QUERY_GENERIC: | 3400 ){ |
| 3401 Fts3Cursor *pCsr; /* Cursor handle passed through apVal[0] */ |
| 3402 |
| 3403 UNUSED_PARAMETER(nVal); |
| 3404 |
| 3405 assert( nVal==1 ); |
| 3406 if( fts3FunctionArg(pContext, "offsets", apVal[0], &pCsr) ) return; |
| 3407 assert( pCsr ); |
| 3408 if( SQLITE_OK==fts3CursorSeek(pContext, pCsr) ){ |
| 3409 sqlite3Fts3Offsets(pContext, pCsr); |
| 3410 } |
| 3411 } |
| 3412 |
| 3413 /* |
| 3414 ** Implementation of the special optimize() function for FTS3. This |
| 3415 ** function merges all segments in the database to a single segment. |
| 3416 ** Example usage is: |
| 3417 ** |
| 3418 ** SELECT optimize(t) FROM t LIMIT 1; |
| 3419 ** |
| 3420 ** where 't' is the name of an FTS3 table. |
| 3421 */ |
| 3422 static void fts3OptimizeFunc( |
| 3423 sqlite3_context *pContext, /* SQLite function call context */ |
| 3424 int nVal, /* Size of argument array */ |
| 3425 sqlite3_value **apVal /* Array of arguments */ |
| 3426 ){ |
| 3427 int rc; /* Return code */ |
| 3428 Fts3Table *p; /* Virtual table handle */ |
| 3429 Fts3Cursor *pCursor; /* Cursor handle passed through apVal[0] */ |
| 3430 |
| 3431 UNUSED_PARAMETER(nVal); |
| 3432 |
| 3433 assert( nVal==1 ); |
| 3434 if( fts3FunctionArg(pContext, "optimize", apVal[0], &pCursor) ) return; |
| 3435 p = (Fts3Table *)pCursor->base.pVtab; |
| 3436 assert( p ); |
| 3437 |
| 3438 rc = sqlite3Fts3Optimize(p); |
| 3439 |
| 3440 switch( rc ){ |
| 3441 case SQLITE_OK: |
| 3442 sqlite3_result_text(pContext, "Index optimized", -1, SQLITE_STATIC); |
4124 break; | 3443 break; |
4125 | 3444 case SQLITE_DONE: |
4126 case QUERY_DOCID: | 3445 sqlite3_result_text(pContext, "Index already optimal", -1, SQLITE_STATIC); |
4127 rc = sqlite3_bind_int64(c->pStmt, 1, sqlite3_value_int64(argv[0])); | |
4128 if( rc!=SQLITE_OK ) return rc; | |
4129 break; | 3446 break; |
4130 | 3447 default: |
4131 default: /* full-text search */ | 3448 sqlite3_result_error_code(pContext, rc); |
4132 { | |
4133 int iCol = idxNum-QUERY_FULLTEXT; | |
4134 const char *zQuery = (const char *)sqlite3_value_text(argv[0]); | |
4135 assert( idxNum<=QUERY_FULLTEXT+v->nColumn); | |
4136 assert( argc==1 ); | |
4137 if( c->result.nData!=0 ){ | |
4138 /* This case happens if the same cursor is used repeatedly. */ | |
4139 dlrDestroy(&c->reader); | |
4140 dataBufferReset(&c->result); | |
4141 }else{ | |
4142 dataBufferInit(&c->result, 0); | |
4143 } | |
4144 rc = fulltextQuery(v, iCol, zQuery, -1, &c->result, &c->pExpr); | |
4145 if( rc!=SQLITE_OK ) return rc; | |
4146 if( c->result.nData!=0 ){ | |
4147 dlrInit(&c->reader, DL_DOCIDS, c->result.pData, c->result.nData); | |
4148 } | |
4149 break; | 3449 break; |
4150 } | 3450 } |
4151 } | 3451 } |
4152 | 3452 |
4153 return fulltextNext(pCursor); | 3453 /* |
4154 } | 3454 ** Implementation of the matchinfo() function for FTS3 |
4155 | 3455 */ |
4156 /* This is the xEof method of the virtual table. The SQLite core | 3456 static void fts3MatchinfoFunc( |
4157 ** calls this routine to find out if it has reached the end of | 3457 sqlite3_context *pContext, /* SQLite function call context */ |
4158 ** a query's results set. | 3458 int nVal, /* Size of argument array */ |
4159 */ | 3459 sqlite3_value **apVal /* Array of arguments */ |
4160 static int fulltextEof(sqlite3_vtab_cursor *pCursor){ | 3460 ){ |
4161 fulltext_cursor *c = (fulltext_cursor *) pCursor; | 3461 Fts3Cursor *pCsr; /* Cursor handle passed through apVal[0] */ |
4162 return c->eof; | 3462 assert( nVal==1 || nVal==2 ); |
4163 } | 3463 if( SQLITE_OK==fts3FunctionArg(pContext, "matchinfo", apVal[0], &pCsr) ){ |
4164 | 3464 const char *zArg = 0; |
4165 /* This is the xColumn method of the virtual table. The SQLite | 3465 if( nVal>1 ){ |
4166 ** core calls this method during a query when it needs the value | 3466 zArg = (const char *)sqlite3_value_text(apVal[1]); |
4167 ** of a column from the virtual table. This method needs to use | 3467 } |
4168 ** one of the sqlite3_result_*() routines to store the requested | 3468 sqlite3Fts3Matchinfo(pContext, pCsr, zArg); |
4169 ** value back in the pContext. | 3469 } |
4170 */ | 3470 } |
4171 static int fulltextColumn(sqlite3_vtab_cursor *pCursor, | |
4172 sqlite3_context *pContext, int idxCol){ | |
4173 fulltext_cursor *c = (fulltext_cursor *) pCursor; | |
4174 fulltext_vtab *v = cursor_vtab(c); | |
4175 | |
4176 if( idxCol<v->nColumn ){ | |
4177 sqlite3_value *pVal = sqlite3_column_value(c->pStmt, idxCol+1); | |
4178 sqlite3_result_value(pContext, pVal); | |
4179 }else if( idxCol==v->nColumn ){ | |
4180 /* The extra column whose name is the same as the table. | |
4181 ** Return a blob which is a pointer to the cursor | |
4182 */ | |
4183 sqlite3_result_blob(pContext, &c, sizeof(c), SQLITE_TRANSIENT); | |
4184 }else if( idxCol==v->nColumn+1 ){ | |
4185 /* The docid column, which is an alias for rowid. */ | |
4186 sqlite3_value *pVal = sqlite3_column_value(c->pStmt, 0); | |
4187 sqlite3_result_value(pContext, pVal); | |
4188 } | |
4189 return SQLITE_OK; | |
4190 } | |
4191 | |
4192 /* This is the xRowid method. The SQLite core calls this routine to | |
4193 ** retrieve the rowid for the current row of the result set. fts3 | |
4194 ** exposes %_content.docid as the rowid for the virtual table. The | |
4195 ** rowid should be written to *pRowid. | |
4196 */ | |
4197 static int fulltextRowid(sqlite3_vtab_cursor *pCursor, sqlite_int64 *pRowid){ | |
4198 fulltext_cursor *c = (fulltext_cursor *) pCursor; | |
4199 | |
4200 *pRowid = sqlite3_column_int64(c->pStmt, 0); | |
4201 return SQLITE_OK; | |
4202 } | |
4203 | |
4204 /* Add all terms in [zText] to pendingTerms table. If [iColumn] > 0, | |
4205 ** we also store positions and offsets in the hash table using that | |
4206 ** column number. | |
4207 */ | |
4208 static int buildTerms(fulltext_vtab *v, sqlite_int64 iDocid, | |
4209 const char *zText, int iColumn){ | |
4210 sqlite3_tokenizer *pTokenizer = v->pTokenizer; | |
4211 sqlite3_tokenizer_cursor *pCursor; | |
4212 const char *pToken; | |
4213 int nTokenBytes; | |
4214 int iStartOffset, iEndOffset, iPosition; | |
4215 int rc; | |
4216 | |
4217 rc = pTokenizer->pModule->xOpen(pTokenizer, zText, -1, &pCursor); | |
4218 if( rc!=SQLITE_OK ) return rc; | |
4219 | |
4220 pCursor->pTokenizer = pTokenizer; | |
4221 while( SQLITE_OK==(rc=pTokenizer->pModule->xNext(pCursor, | |
4222 &pToken, &nTokenBytes, | |
4223 &iStartOffset, &iEndOffset, | |
4224 &iPosition)) ){ | |
4225 DLCollector *p; | |
4226 int nData; /* Size of doclist before our update. */ | |
4227 | |
4228 /* Positions can't be negative; we use -1 as a terminator | |
4229 * internally. Token can't be NULL or empty. */ | |
4230 if( iPosition<0 || pToken == NULL || nTokenBytes == 0 ){ | |
4231 rc = SQLITE_ERROR; | |
4232 break; | |
4233 } | |
4234 | |
4235 p = fts3HashFind(&v->pendingTerms, pToken, nTokenBytes); | |
4236 if( p==NULL ){ | |
4237 nData = 0; | |
4238 p = dlcNew(iDocid, DL_DEFAULT); | |
4239 fts3HashInsert(&v->pendingTerms, pToken, nTokenBytes, p); | |
4240 | |
4241 /* Overhead for our hash table entry, the key, and the value. */ | |
4242 v->nPendingData += sizeof(struct fts3HashElem)+sizeof(*p)+nTokenBytes; | |
4243 }else{ | |
4244 nData = p->b.nData; | |
4245 if( p->dlw.iPrevDocid!=iDocid ) dlcNext(p, iDocid); | |
4246 } | |
4247 if( iColumn>=0 ){ | |
4248 dlcAddPos(p, iColumn, iPosition, iStartOffset, iEndOffset); | |
4249 } | |
4250 | |
4251 /* Accumulate data added by dlcNew or dlcNext, and dlcAddPos. */ | |
4252 v->nPendingData += p->b.nData-nData; | |
4253 } | |
4254 | |
4255 /* TODO(shess) Check return? Should this be able to cause errors at | |
4256 ** this point? Actually, same question about sqlite3_finalize(), | |
4257 ** though one could argue that failure there means that the data is | |
4258 ** not durable. *ponder* | |
4259 */ | |
4260 pTokenizer->pModule->xClose(pCursor); | |
4261 if( SQLITE_DONE == rc ) return SQLITE_OK; | |
4262 return rc; | |
4263 } | |
4264 | |
4265 /* Add doclists for all terms in [pValues] to pendingTerms table. */ | |
4266 static int insertTerms(fulltext_vtab *v, sqlite_int64 iDocid, | |
4267 sqlite3_value **pValues){ | |
4268 int i; | |
4269 for(i = 0; i < v->nColumn ; ++i){ | |
4270 char *zText = (char*)sqlite3_value_text(pValues[i]); | |
4271 int rc = buildTerms(v, iDocid, zText, i); | |
4272 if( rc!=SQLITE_OK ) return rc; | |
4273 } | |
4274 return SQLITE_OK; | |
4275 } | |
4276 | |
4277 /* Add empty doclists for all terms in the given row's content to | |
4278 ** pendingTerms. | |
4279 */ | |
4280 static int deleteTerms(fulltext_vtab *v, sqlite_int64 iDocid){ | |
4281 const char **pValues; | |
4282 int i, rc; | |
4283 | |
4284 /* TODO(shess) Should we allow such tables at all? */ | |
4285 if( DL_DEFAULT==DL_DOCIDS ) return SQLITE_ERROR; | |
4286 | |
4287 rc = content_select(v, iDocid, &pValues); | |
4288 if( rc!=SQLITE_OK ) return rc; | |
4289 | |
4290 for(i = 0 ; i < v->nColumn; ++i) { | |
4291 rc = buildTerms(v, iDocid, pValues[i], -1); | |
4292 if( rc!=SQLITE_OK ) break; | |
4293 } | |
4294 | |
4295 freeStringArray(v->nColumn, pValues); | |
4296 return SQLITE_OK; | |
4297 } | |
4298 | |
4299 /* TODO(shess) Refactor the code to remove this forward decl. */ | |
4300 static int initPendingTerms(fulltext_vtab *v, sqlite_int64 iDocid); | |
4301 | |
4302 /* Insert a row into the %_content table; set *piDocid to be the ID of the | |
4303 ** new row. Add doclists for terms to pendingTerms. | |
4304 */ | |
4305 static int index_insert(fulltext_vtab *v, sqlite3_value *pRequestDocid, | |
4306 sqlite3_value **pValues, sqlite_int64 *piDocid){ | |
4307 int rc; | |
4308 | |
4309 rc = content_insert(v, pRequestDocid, pValues); /* execute an SQL INSERT */ | |
4310 if( rc!=SQLITE_OK ) return rc; | |
4311 | |
4312 /* docid column is an alias for rowid. */ | |
4313 *piDocid = sqlite3_last_insert_rowid(v->db); | |
4314 rc = initPendingTerms(v, *piDocid); | |
4315 if( rc!=SQLITE_OK ) return rc; | |
4316 | |
4317 return insertTerms(v, *piDocid, pValues); | |
4318 } | |
4319 | |
4320 /* Delete a row from the %_content table; add empty doclists for terms | |
4321 ** to pendingTerms. | |
4322 */ | |
4323 static int index_delete(fulltext_vtab *v, sqlite_int64 iRow){ | |
4324 int rc = initPendingTerms(v, iRow); | |
4325 if( rc!=SQLITE_OK ) return rc; | |
4326 | |
4327 rc = deleteTerms(v, iRow); | |
4328 if( rc!=SQLITE_OK ) return rc; | |
4329 | |
4330 return content_delete(v, iRow); /* execute an SQL DELETE */ | |
4331 } | |
4332 | |
4333 /* Update a row in the %_content table; add delete doclists to | |
4334 ** pendingTerms for old terms not in the new data, add insert doclists | |
4335 ** to pendingTerms for terms in the new data. | |
4336 */ | |
4337 static int index_update(fulltext_vtab *v, sqlite_int64 iRow, | |
4338 sqlite3_value **pValues){ | |
4339 int rc = initPendingTerms(v, iRow); | |
4340 if( rc!=SQLITE_OK ) return rc; | |
4341 | |
4342 /* Generate an empty doclist for each term that previously appeared in this | |
4343 * row. */ | |
4344 rc = deleteTerms(v, iRow); | |
4345 if( rc!=SQLITE_OK ) return rc; | |
4346 | |
4347 rc = content_update(v, pValues, iRow); /* execute an SQL UPDATE */ | |
4348 if( rc!=SQLITE_OK ) return rc; | |
4349 | |
4350 /* Now add positions for terms which appear in the updated row. */ | |
4351 return insertTerms(v, iRow, pValues); | |
4352 } | |
4353 | |
4354 /*******************************************************************/ | |
4355 /* InteriorWriter is used to collect terms and block references into | |
4356 ** interior nodes in %_segments. See commentary at top of file for | |
4357 ** format. | |
4358 */ | |
4359 | |
4360 /* How large interior nodes can grow. */ | |
4361 #define INTERIOR_MAX 2048 | |
4362 | |
4363 /* Minimum number of terms per interior node (except the root). This | |
4364 ** prevents large terms from making the tree too skinny - must be >0 | |
4365 ** so that the tree always makes progress. Note that the min tree | |
4366 ** fanout will be INTERIOR_MIN_TERMS+1. | |
4367 */ | |
4368 #define INTERIOR_MIN_TERMS 7 | |
4369 #if INTERIOR_MIN_TERMS<1 | |
4370 # error INTERIOR_MIN_TERMS must be greater than 0. | |
4371 #endif | |
4372 | |
4373 /* ROOT_MAX controls how much data is stored inline in the segment | |
4374 ** directory. | |
4375 */ | |
4376 /* TODO(shess) Push ROOT_MAX down to whoever is writing things. It's | |
4377 ** only here so that interiorWriterRootInfo() and leafWriterRootInfo() | |
4378 ** can both see it, but if the caller passed it in, we wouldn't even | |
4379 ** need a define. | |
4380 */ | |
4381 #define ROOT_MAX 1024 | |
4382 #if ROOT_MAX<VARINT_MAX*2 | |
4383 # error ROOT_MAX must have enough space for a header. | |
4384 #endif | |
4385 | |
4386 /* InteriorBlock stores a linked-list of interior blocks while a lower | |
4387 ** layer is being constructed. | |
4388 */ | |
4389 typedef struct InteriorBlock { | |
4390 DataBuffer term; /* Leftmost term in block's subtree. */ | |
4391 DataBuffer data; /* Accumulated data for the block. */ | |
4392 struct InteriorBlock *next; | |
4393 } InteriorBlock; | |
4394 | |
4395 static InteriorBlock *interiorBlockNew(int iHeight, sqlite_int64 iChildBlock, | |
4396 const char *pTerm, int nTerm){ | |
4397 InteriorBlock *block = sqlite3_malloc(sizeof(InteriorBlock)); | |
4398 char c[VARINT_MAX+VARINT_MAX]; | |
4399 int n; | |
4400 | |
4401 if( block ){ | |
4402 memset(block, 0, sizeof(*block)); | |
4403 dataBufferInit(&block->term, 0); | |
4404 dataBufferReplace(&block->term, pTerm, nTerm); | |
4405 | |
4406 n = fts3PutVarint(c, iHeight); | |
4407 n += fts3PutVarint(c+n, iChildBlock); | |
4408 dataBufferInit(&block->data, INTERIOR_MAX); | |
4409 dataBufferReplace(&block->data, c, n); | |
4410 } | |
4411 return block; | |
4412 } | |
4413 | |
4414 #ifndef NDEBUG | |
4415 /* Verify that the data is readable as an interior node. */ | |
4416 static void interiorBlockValidate(InteriorBlock *pBlock){ | |
4417 const char *pData = pBlock->data.pData; | |
4418 int nData = pBlock->data.nData; | |
4419 int n, iDummy; | |
4420 sqlite_int64 iBlockid; | |
4421 | |
4422 assert( nData>0 ); | |
4423 assert( pData!=0 ); | |
4424 assert( pData+nData>pData ); | |
4425 | |
4426 /* Must lead with height of node as a varint(n), n>0 */ | |
4427 n = fts3GetVarint32(pData, &iDummy); | |
4428 assert( n>0 ); | |
4429 assert( iDummy>0 ); | |
4430 assert( n<nData ); | |
4431 pData += n; | |
4432 nData -= n; | |
4433 | |
4434 /* Must contain iBlockid. */ | |
4435 n = fts3GetVarint(pData, &iBlockid); | |
4436 assert( n>0 ); | |
4437 assert( n<=nData ); | |
4438 pData += n; | |
4439 nData -= n; | |
4440 | |
4441 /* Zero or more terms of positive length */ | |
4442 if( nData!=0 ){ | |
4443 /* First term is not delta-encoded. */ | |
4444 n = fts3GetVarint32(pData, &iDummy); | |
4445 assert( n>0 ); | |
4446 assert( iDummy>0 ); | |
4447 assert( n+iDummy>0); | |
4448 assert( n+iDummy<=nData ); | |
4449 pData += n+iDummy; | |
4450 nData -= n+iDummy; | |
4451 | |
4452 /* Following terms delta-encoded. */ | |
4453 while( nData!=0 ){ | |
4454 /* Length of shared prefix. */ | |
4455 n = fts3GetVarint32(pData, &iDummy); | |
4456 assert( n>0 ); | |
4457 assert( iDummy>=0 ); | |
4458 assert( n<nData ); | |
4459 pData += n; | |
4460 nData -= n; | |
4461 | |
4462 /* Length and data of distinct suffix. */ | |
4463 n = fts3GetVarint32(pData, &iDummy); | |
4464 assert( n>0 ); | |
4465 assert( iDummy>0 ); | |
4466 assert( n+iDummy>0); | |
4467 assert( n+iDummy<=nData ); | |
4468 pData += n+iDummy; | |
4469 nData -= n+iDummy; | |
4470 } | |
4471 } | |
4472 } | |
4473 #define ASSERT_VALID_INTERIOR_BLOCK(x) interiorBlockValidate(x) | |
4474 #else | |
4475 #define ASSERT_VALID_INTERIOR_BLOCK(x) assert( 1 ) | |
4476 #endif | |
4477 | |
4478 typedef struct InteriorWriter { | |
4479 int iHeight; /* from 0 at leaves. */ | |
4480 InteriorBlock *first, *last; | |
4481 struct InteriorWriter *parentWriter; | |
4482 | |
4483 DataBuffer term; /* Last term written to block "last". */ | |
4484 sqlite_int64 iOpeningChildBlock; /* First child block in block "last". */ | |
4485 #ifndef NDEBUG | |
4486 sqlite_int64 iLastChildBlock; /* for consistency checks. */ | |
4487 #endif | |
4488 } InteriorWriter; | |
4489 | |
4490 /* Initialize an interior node where pTerm[nTerm] marks the leftmost | |
4491 ** term in the tree. iChildBlock is the leftmost child block at the | |
4492 ** next level down the tree. | |
4493 */ | |
4494 static void interiorWriterInit(int iHeight, const char *pTerm, int nTerm, | |
4495 sqlite_int64 iChildBlock, | |
4496 InteriorWriter *pWriter){ | |
4497 InteriorBlock *block; | |
4498 assert( iHeight>0 ); | |
4499 CLEAR(pWriter); | |
4500 | |
4501 pWriter->iHeight = iHeight; | |
4502 pWriter->iOpeningChildBlock = iChildBlock; | |
4503 #ifndef NDEBUG | |
4504 pWriter->iLastChildBlock = iChildBlock; | |
4505 #endif | |
4506 block = interiorBlockNew(iHeight, iChildBlock, pTerm, nTerm); | |
4507 pWriter->last = pWriter->first = block; | |
4508 ASSERT_VALID_INTERIOR_BLOCK(pWriter->last); | |
4509 dataBufferInit(&pWriter->term, 0); | |
4510 } | |
4511 | |
4512 /* Append the child node rooted at iChildBlock to the interior node, | |
4513 ** with pTerm[nTerm] as the leftmost term in iChildBlock's subtree. | |
4514 */ | |
4515 static void interiorWriterAppend(InteriorWriter *pWriter, | |
4516 const char *pTerm, int nTerm, | |
4517 sqlite_int64 iChildBlock){ | |
4518 char c[VARINT_MAX+VARINT_MAX]; | |
4519 int n, nPrefix = 0; | |
4520 | |
4521 ASSERT_VALID_INTERIOR_BLOCK(pWriter->last); | |
4522 | |
4523 /* The first term written into an interior node is actually | |
4524 ** associated with the second child added (the first child was added | |
4525 ** in interiorWriterInit, or in the if clause at the bottom of this | |
4526 ** function). That term gets encoded straight up, with nPrefix left | |
4527 ** at 0. | |
4528 */ | |
4529 if( pWriter->term.nData==0 ){ | |
4530 n = fts3PutVarint(c, nTerm); | |
4531 }else{ | |
4532 while( nPrefix<pWriter->term.nData && | |
4533 pTerm[nPrefix]==pWriter->term.pData[nPrefix] ){ | |
4534 nPrefix++; | |
4535 } | |
4536 | |
4537 n = fts3PutVarint(c, nPrefix); | |
4538 n += fts3PutVarint(c+n, nTerm-nPrefix); | |
4539 } | |
4540 | |
4541 #ifndef NDEBUG | |
4542 pWriter->iLastChildBlock++; | |
4543 #endif | |
4544 assert( pWriter->iLastChildBlock==iChildBlock ); | |
4545 | |
4546 /* Overflow to a new block if the new term makes the current block | |
4547 ** too big, and the current block already has enough terms. | |
4548 */ | |
4549 if( pWriter->last->data.nData+n+nTerm-nPrefix>INTERIOR_MAX && | |
4550 iChildBlock-pWriter->iOpeningChildBlock>INTERIOR_MIN_TERMS ){ | |
4551 pWriter->last->next = interiorBlockNew(pWriter->iHeight, iChildBlock, | |
4552 pTerm, nTerm); | |
4553 pWriter->last = pWriter->last->next; | |
4554 pWriter->iOpeningChildBlock = iChildBlock; | |
4555 dataBufferReset(&pWriter->term); | |
4556 }else{ | |
4557 dataBufferAppend2(&pWriter->last->data, c, n, | |
4558 pTerm+nPrefix, nTerm-nPrefix); | |
4559 dataBufferReplace(&pWriter->term, pTerm, nTerm); | |
4560 } | |
4561 ASSERT_VALID_INTERIOR_BLOCK(pWriter->last); | |
4562 } | |
4563 | |
4564 /* Free the space used by pWriter, including the linked-list of | |
4565 ** InteriorBlocks, and parentWriter, if present. | |
4566 */ | |
4567 static int interiorWriterDestroy(InteriorWriter *pWriter){ | |
4568 InteriorBlock *block = pWriter->first; | |
4569 | |
4570 while( block!=NULL ){ | |
4571 InteriorBlock *b = block; | |
4572 block = block->next; | |
4573 dataBufferDestroy(&b->term); | |
4574 dataBufferDestroy(&b->data); | |
4575 sqlite3_free(b); | |
4576 } | |
4577 if( pWriter->parentWriter!=NULL ){ | |
4578 interiorWriterDestroy(pWriter->parentWriter); | |
4579 sqlite3_free(pWriter->parentWriter); | |
4580 } | |
4581 dataBufferDestroy(&pWriter->term); | |
4582 SCRAMBLE(pWriter); | |
4583 return SQLITE_OK; | |
4584 } | |
4585 | |
4586 /* If pWriter can fit entirely in ROOT_MAX, return it as the root info | |
4587 ** directly, leaving *piEndBlockid unchanged. Otherwise, flush | |
4588 ** pWriter to %_segments, building a new layer of interior nodes, and | |
4589 ** recursively ask for their root into. | |
4590 */ | |
4591 static int interiorWriterRootInfo(fulltext_vtab *v, InteriorWriter *pWriter, | |
4592 char **ppRootInfo, int *pnRootInfo, | |
4593 sqlite_int64 *piEndBlockid){ | |
4594 InteriorBlock *block = pWriter->first; | |
4595 sqlite_int64 iBlockid = 0; | |
4596 int rc; | |
4597 | |
4598 /* If we can fit the segment inline */ | |
4599 if( block==pWriter->last && block->data.nData<ROOT_MAX ){ | |
4600 *ppRootInfo = block->data.pData; | |
4601 *pnRootInfo = block->data.nData; | |
4602 return SQLITE_OK; | |
4603 } | |
4604 | |
4605 /* Flush the first block to %_segments, and create a new level of | |
4606 ** interior node. | |
4607 */ | |
4608 ASSERT_VALID_INTERIOR_BLOCK(block); | |
4609 rc = block_insert(v, block->data.pData, block->data.nData, &iBlockid); | |
4610 if( rc!=SQLITE_OK ) return rc; | |
4611 *piEndBlockid = iBlockid; | |
4612 | |
4613 pWriter->parentWriter = sqlite3_malloc(sizeof(*pWriter->parentWriter)); | |
4614 interiorWriterInit(pWriter->iHeight+1, | |
4615 block->term.pData, block->term.nData, | |
4616 iBlockid, pWriter->parentWriter); | |
4617 | |
4618 /* Flush additional blocks and append to the higher interior | |
4619 ** node. | |
4620 */ | |
4621 for(block=block->next; block!=NULL; block=block->next){ | |
4622 ASSERT_VALID_INTERIOR_BLOCK(block); | |
4623 rc = block_insert(v, block->data.pData, block->data.nData, &iBlockid); | |
4624 if( rc!=SQLITE_OK ) return rc; | |
4625 *piEndBlockid = iBlockid; | |
4626 | |
4627 interiorWriterAppend(pWriter->parentWriter, | |
4628 block->term.pData, block->term.nData, iBlockid); | |
4629 } | |
4630 | |
4631 /* Parent node gets the chance to be the root. */ | |
4632 return interiorWriterRootInfo(v, pWriter->parentWriter, | |
4633 ppRootInfo, pnRootInfo, piEndBlockid); | |
4634 } | |
4635 | |
4636 /****************************************************************/ | |
4637 /* InteriorReader is used to read off the data from an interior node | |
4638 ** (see comment at top of file for the format). | |
4639 */ | |
4640 typedef struct InteriorReader { | |
4641 const char *pData; | |
4642 int nData; | |
4643 | |
4644 DataBuffer term; /* previous term, for decoding term delta. */ | |
4645 | |
4646 sqlite_int64 iBlockid; | |
4647 } InteriorReader; | |
4648 | |
4649 static void interiorReaderDestroy(InteriorReader *pReader){ | |
4650 dataBufferDestroy(&pReader->term); | |
4651 SCRAMBLE(pReader); | |
4652 } | |
4653 | |
4654 static int interiorReaderInit(const char *pData, int nData, | |
4655 InteriorReader *pReader){ | |
4656 int n, nTerm; | |
4657 | |
4658 /* These conditions are checked and met by the callers. */ | |
4659 assert( nData>0 ); | |
4660 assert( pData[0]!='\0' ); | |
4661 | |
4662 CLEAR(pReader); | |
4663 | |
4664 /* Decode the base blockid, and set the cursor to the first term. */ | |
4665 n = fts3GetVarintSafe(pData+1, &pReader->iBlockid, nData-1); | |
4666 if( !n ) return SQLITE_CORRUPT_BKPT; | |
4667 pReader->pData = pData+1+n; | |
4668 pReader->nData = nData-(1+n); | |
4669 | |
4670 /* A single-child interior node (such as when a leaf node was too | |
4671 ** large for the segment directory) won't have any terms. | |
4672 ** Otherwise, decode the first term. | |
4673 */ | |
4674 if( pReader->nData==0 ){ | |
4675 dataBufferInit(&pReader->term, 0); | |
4676 }else{ | |
4677 n = fts3GetVarint32Safe(pReader->pData, &nTerm, pReader->nData); | |
4678 if( !n || nTerm<0 || nTerm>pReader->nData-n) return SQLITE_CORRUPT_BKPT; | |
4679 dataBufferInit(&pReader->term, nTerm); | |
4680 dataBufferReplace(&pReader->term, pReader->pData+n, nTerm); | |
4681 pReader->pData += n+nTerm; | |
4682 pReader->nData -= n+nTerm; | |
4683 } | |
4684 return SQLITE_OK; | |
4685 } | |
4686 | |
4687 static int interiorReaderAtEnd(InteriorReader *pReader){ | |
4688 return pReader->term.nData<=0; | |
4689 } | |
4690 | |
4691 static sqlite_int64 interiorReaderCurrentBlockid(InteriorReader *pReader){ | |
4692 return pReader->iBlockid; | |
4693 } | |
4694 | |
4695 static int interiorReaderTermBytes(InteriorReader *pReader){ | |
4696 assert( !interiorReaderAtEnd(pReader) ); | |
4697 return pReader->term.nData; | |
4698 } | |
4699 static const char *interiorReaderTerm(InteriorReader *pReader){ | |
4700 assert( !interiorReaderAtEnd(pReader) ); | |
4701 return pReader->term.pData; | |
4702 } | |
4703 | |
4704 /* Step forward to the next term in the node. */ | |
4705 static int interiorReaderStep(InteriorReader *pReader){ | |
4706 assert( !interiorReaderAtEnd(pReader) ); | |
4707 | |
4708 /* If the last term has been read, signal eof, else construct the | |
4709 ** next term. | |
4710 */ | |
4711 if( pReader->nData==0 ){ | |
4712 dataBufferReset(&pReader->term); | |
4713 }else{ | |
4714 int n, nPrefix, nSuffix; | |
4715 | |
4716 n = fts3GetVarint32Safe(pReader->pData, &nPrefix, pReader->nData); | |
4717 if( !n ) return SQLITE_CORRUPT_BKPT; | |
4718 pReader->nData -= n; | |
4719 pReader->pData += n; | |
4720 n = fts3GetVarint32Safe(pReader->pData, &nSuffix, pReader->nData); | |
4721 if( !n ) return SQLITE_CORRUPT_BKPT; | |
4722 pReader->nData -= n; | |
4723 pReader->pData += n; | |
4724 if( nSuffix<0 || nSuffix>pReader->nData ) return SQLITE_CORRUPT_BKPT; | |
4725 if( nPrefix<0 || nPrefix>pReader->term.nData ) return SQLITE_CORRUPT_BKPT; | |
4726 | |
4727 /* Truncate the current term and append suffix data. */ | |
4728 pReader->term.nData = nPrefix; | |
4729 dataBufferAppend(&pReader->term, pReader->pData, nSuffix); | |
4730 | |
4731 pReader->pData += nSuffix; | |
4732 pReader->nData -= nSuffix; | |
4733 } | |
4734 pReader->iBlockid++; | |
4735 return SQLITE_OK; | |
4736 } | |
4737 | |
4738 /* Compare the current term to pTerm[nTerm], returning strcmp-style | |
4739 ** results. If isPrefix, equality means equal through nTerm bytes. | |
4740 */ | |
4741 static int interiorReaderTermCmp(InteriorReader *pReader, | |
4742 const char *pTerm, int nTerm, int isPrefix){ | |
4743 const char *pReaderTerm = interiorReaderTerm(pReader); | |
4744 int nReaderTerm = interiorReaderTermBytes(pReader); | |
4745 int c, n = nReaderTerm<nTerm ? nReaderTerm : nTerm; | |
4746 | |
4747 if( n==0 ){ | |
4748 if( nReaderTerm>0 ) return -1; | |
4749 if( nTerm>0 ) return 1; | |
4750 return 0; | |
4751 } | |
4752 | |
4753 c = memcmp(pReaderTerm, pTerm, n); | |
4754 if( c!=0 ) return c; | |
4755 if( isPrefix && n==nTerm ) return 0; | |
4756 return nReaderTerm - nTerm; | |
4757 } | |
4758 | |
4759 /****************************************************************/ | |
4760 /* LeafWriter is used to collect terms and associated doclist data | |
4761 ** into leaf blocks in %_segments (see top of file for format info). | |
4762 ** Expected usage is: | |
4763 ** | |
4764 ** LeafWriter writer; | |
4765 ** leafWriterInit(0, 0, &writer); | |
4766 ** while( sorted_terms_left_to_process ){ | |
4767 ** // data is doclist data for that term. | |
4768 ** rc = leafWriterStep(v, &writer, pTerm, nTerm, pData, nData); | |
4769 ** if( rc!=SQLITE_OK ) goto err; | |
4770 ** } | |
4771 ** rc = leafWriterFinalize(v, &writer); | |
4772 **err: | |
4773 ** leafWriterDestroy(&writer); | |
4774 ** return rc; | |
4775 ** | |
4776 ** leafWriterStep() may write a collected leaf out to %_segments. | |
4777 ** leafWriterFinalize() finishes writing any buffered data and stores | |
4778 ** a root node in %_segdir. leafWriterDestroy() frees all buffers and | |
4779 ** InteriorWriters allocated as part of writing this segment. | |
4780 ** | |
4781 ** TODO(shess) Document leafWriterStepMerge(). | |
4782 */ | |
4783 | |
4784 /* Put terms with data this big in their own block. */ | |
4785 #define STANDALONE_MIN 1024 | |
4786 | |
4787 /* Keep leaf blocks below this size. */ | |
4788 #define LEAF_MAX 2048 | |
4789 | |
4790 typedef struct LeafWriter { | |
4791 int iLevel; | |
4792 int idx; | |
4793 sqlite_int64 iStartBlockid; /* needed to create the root info */ | |
4794 sqlite_int64 iEndBlockid; /* when we're done writing. */ | |
4795 | |
4796 DataBuffer term; /* previous encoded term */ | |
4797 DataBuffer data; /* encoding buffer */ | |
4798 | |
4799 /* bytes of first term in the current node which distinguishes that | |
4800 ** term from the last term of the previous node. | |
4801 */ | |
4802 int nTermDistinct; | |
4803 | |
4804 InteriorWriter parentWriter; /* if we overflow */ | |
4805 int has_parent; | |
4806 } LeafWriter; | |
4807 | |
4808 static void leafWriterInit(int iLevel, int idx, LeafWriter *pWriter){ | |
4809 CLEAR(pWriter); | |
4810 pWriter->iLevel = iLevel; | |
4811 pWriter->idx = idx; | |
4812 | |
4813 dataBufferInit(&pWriter->term, 32); | |
4814 | |
4815 /* Start out with a reasonably sized block, though it can grow. */ | |
4816 dataBufferInit(&pWriter->data, LEAF_MAX); | |
4817 } | |
4818 | |
4819 #ifndef NDEBUG | |
4820 /* Verify that the data is readable as a leaf node. */ | |
4821 static void leafNodeValidate(const char *pData, int nData){ | |
4822 int n, iDummy; | |
4823 | |
4824 if( nData==0 ) return; | |
4825 assert( nData>0 ); | |
4826 assert( pData!=0 ); | |
4827 assert( pData+nData>pData ); | |
4828 | |
4829 /* Must lead with a varint(0) */ | |
4830 n = fts3GetVarint32(pData, &iDummy); | |
4831 assert( iDummy==0 ); | |
4832 assert( n>0 ); | |
4833 assert( n<nData ); | |
4834 pData += n; | |
4835 nData -= n; | |
4836 | |
4837 /* Leading term length and data must fit in buffer. */ | |
4838 n = fts3GetVarint32(pData, &iDummy); | |
4839 assert( n>0 ); | |
4840 assert( iDummy>0 ); | |
4841 assert( n+iDummy>0 ); | |
4842 assert( n+iDummy<nData ); | |
4843 pData += n+iDummy; | |
4844 nData -= n+iDummy; | |
4845 | |
4846 /* Leading term's doclist length and data must fit. */ | |
4847 n = fts3GetVarint32(pData, &iDummy); | |
4848 assert( n>0 ); | |
4849 assert( iDummy>0 ); | |
4850 assert( n+iDummy>0 ); | |
4851 assert( n+iDummy<=nData ); | |
4852 ASSERT_VALID_DOCLIST(DL_DEFAULT, pData+n, iDummy, NULL); | |
4853 pData += n+iDummy; | |
4854 nData -= n+iDummy; | |
4855 | |
4856 /* Verify that trailing terms and doclists also are readable. */ | |
4857 while( nData!=0 ){ | |
4858 n = fts3GetVarint32(pData, &iDummy); | |
4859 assert( n>0 ); | |
4860 assert( iDummy>=0 ); | |
4861 assert( n<nData ); | |
4862 pData += n; | |
4863 nData -= n; | |
4864 n = fts3GetVarint32(pData, &iDummy); | |
4865 assert( n>0 ); | |
4866 assert( iDummy>0 ); | |
4867 assert( n+iDummy>0 ); | |
4868 assert( n+iDummy<nData ); | |
4869 pData += n+iDummy; | |
4870 nData -= n+iDummy; | |
4871 | |
4872 n = fts3GetVarint32(pData, &iDummy); | |
4873 assert( n>0 ); | |
4874 assert( iDummy>0 ); | |
4875 assert( n+iDummy>0 ); | |
4876 assert( n+iDummy<=nData ); | |
4877 ASSERT_VALID_DOCLIST(DL_DEFAULT, pData+n, iDummy, NULL); | |
4878 pData += n+iDummy; | |
4879 nData -= n+iDummy; | |
4880 } | |
4881 } | |
4882 #define ASSERT_VALID_LEAF_NODE(p, n) leafNodeValidate(p, n) | |
4883 #else | |
4884 #define ASSERT_VALID_LEAF_NODE(p, n) assert( 1 ) | |
4885 #endif | |
4886 | |
4887 /* Flush the current leaf node to %_segments, and adding the resulting | |
4888 ** blockid and the starting term to the interior node which will | |
4889 ** contain it. | |
4890 */ | |
4891 static int leafWriterInternalFlush(fulltext_vtab *v, LeafWriter *pWriter, | |
4892 int iData, int nData){ | |
4893 sqlite_int64 iBlockid = 0; | |
4894 const char *pStartingTerm; | |
4895 int nStartingTerm, rc, n; | |
4896 | |
4897 /* Must have the leading varint(0) flag, plus at least some | |
4898 ** valid-looking data. | |
4899 */ | |
4900 assert( nData>2 ); | |
4901 assert( iData>=0 ); | |
4902 assert( iData+nData<=pWriter->data.nData ); | |
4903 ASSERT_VALID_LEAF_NODE(pWriter->data.pData+iData, nData); | |
4904 | |
4905 rc = block_insert(v, pWriter->data.pData+iData, nData, &iBlockid); | |
4906 if( rc!=SQLITE_OK ) return rc; | |
4907 assert( iBlockid!=0 ); | |
4908 | |
4909 /* Reconstruct the first term in the leaf for purposes of building | |
4910 ** the interior node. | |
4911 */ | |
4912 n = fts3GetVarint32(pWriter->data.pData+iData+1, &nStartingTerm); | |
4913 pStartingTerm = pWriter->data.pData+iData+1+n; | |
4914 assert( pWriter->data.nData>iData+1+n+nStartingTerm ); | |
4915 assert( pWriter->nTermDistinct>0 ); | |
4916 assert( pWriter->nTermDistinct<=nStartingTerm ); | |
4917 nStartingTerm = pWriter->nTermDistinct; | |
4918 | |
4919 if( pWriter->has_parent ){ | |
4920 interiorWriterAppend(&pWriter->parentWriter, | |
4921 pStartingTerm, nStartingTerm, iBlockid); | |
4922 }else{ | |
4923 interiorWriterInit(1, pStartingTerm, nStartingTerm, iBlockid, | |
4924 &pWriter->parentWriter); | |
4925 pWriter->has_parent = 1; | |
4926 } | |
4927 | |
4928 /* Track the span of this segment's leaf nodes. */ | |
4929 if( pWriter->iEndBlockid==0 ){ | |
4930 pWriter->iEndBlockid = pWriter->iStartBlockid = iBlockid; | |
4931 }else{ | |
4932 pWriter->iEndBlockid++; | |
4933 assert( iBlockid==pWriter->iEndBlockid ); | |
4934 } | |
4935 | |
4936 return SQLITE_OK; | |
4937 } | |
4938 static int leafWriterFlush(fulltext_vtab *v, LeafWriter *pWriter){ | |
4939 int rc = leafWriterInternalFlush(v, pWriter, 0, pWriter->data.nData); | |
4940 if( rc!=SQLITE_OK ) return rc; | |
4941 | |
4942 /* Re-initialize the output buffer. */ | |
4943 dataBufferReset(&pWriter->data); | |
4944 | |
4945 return SQLITE_OK; | |
4946 } | |
4947 | |
4948 /* Fetch the root info for the segment. If the entire leaf fits | |
4949 ** within ROOT_MAX, then it will be returned directly, otherwise it | |
4950 ** will be flushed and the root info will be returned from the | |
4951 ** interior node. *piEndBlockid is set to the blockid of the last | |
4952 ** interior or leaf node written to disk (0 if none are written at | |
4953 ** all). | |
4954 */ | |
4955 static int leafWriterRootInfo(fulltext_vtab *v, LeafWriter *pWriter, | |
4956 char **ppRootInfo, int *pnRootInfo, | |
4957 sqlite_int64 *piEndBlockid){ | |
4958 /* we can fit the segment entirely inline */ | |
4959 if( !pWriter->has_parent && pWriter->data.nData<ROOT_MAX ){ | |
4960 *ppRootInfo = pWriter->data.pData; | |
4961 *pnRootInfo = pWriter->data.nData; | |
4962 *piEndBlockid = 0; | |
4963 return SQLITE_OK; | |
4964 } | |
4965 | |
4966 /* Flush remaining leaf data. */ | |
4967 if( pWriter->data.nData>0 ){ | |
4968 int rc = leafWriterFlush(v, pWriter); | |
4969 if( rc!=SQLITE_OK ) return rc; | |
4970 } | |
4971 | |
4972 /* We must have flushed a leaf at some point. */ | |
4973 assert( pWriter->has_parent ); | |
4974 | |
4975 /* Tenatively set the end leaf blockid as the end blockid. If the | |
4976 ** interior node can be returned inline, this will be the final | |
4977 ** blockid, otherwise it will be overwritten by | |
4978 ** interiorWriterRootInfo(). | |
4979 */ | |
4980 *piEndBlockid = pWriter->iEndBlockid; | |
4981 | |
4982 return interiorWriterRootInfo(v, &pWriter->parentWriter, | |
4983 ppRootInfo, pnRootInfo, piEndBlockid); | |
4984 } | |
4985 | |
4986 /* Collect the rootInfo data and store it into the segment directory. | |
4987 ** This has the effect of flushing the segment's leaf data to | |
4988 ** %_segments, and also flushing any interior nodes to %_segments. | |
4989 */ | |
4990 static int leafWriterFinalize(fulltext_vtab *v, LeafWriter *pWriter){ | |
4991 sqlite_int64 iEndBlockid; | |
4992 char *pRootInfo; | |
4993 int rc, nRootInfo; | |
4994 | |
4995 rc = leafWriterRootInfo(v, pWriter, &pRootInfo, &nRootInfo, &iEndBlockid); | |
4996 if( rc!=SQLITE_OK ) return rc; | |
4997 | |
4998 /* Don't bother storing an entirely empty segment. */ | |
4999 if( iEndBlockid==0 && nRootInfo==0 ) return SQLITE_OK; | |
5000 | |
5001 return segdir_set(v, pWriter->iLevel, pWriter->idx, | |
5002 pWriter->iStartBlockid, pWriter->iEndBlockid, | |
5003 iEndBlockid, pRootInfo, nRootInfo); | |
5004 } | |
5005 | |
5006 static void leafWriterDestroy(LeafWriter *pWriter){ | |
5007 if( pWriter->has_parent ) interiorWriterDestroy(&pWriter->parentWriter); | |
5008 dataBufferDestroy(&pWriter->term); | |
5009 dataBufferDestroy(&pWriter->data); | |
5010 } | |
5011 | |
5012 /* Encode a term into the leafWriter, delta-encoding as appropriate. | |
5013 ** Returns the length of the new term which distinguishes it from the | |
5014 ** previous term, which can be used to set nTermDistinct when a node | |
5015 ** boundary is crossed. | |
5016 */ | |
5017 static int leafWriterEncodeTerm(LeafWriter *pWriter, | |
5018 const char *pTerm, int nTerm){ | |
5019 char c[VARINT_MAX+VARINT_MAX]; | |
5020 int n, nPrefix = 0; | |
5021 | |
5022 assert( nTerm>0 ); | |
5023 while( nPrefix<pWriter->term.nData && | |
5024 pTerm[nPrefix]==pWriter->term.pData[nPrefix] ){ | |
5025 nPrefix++; | |
5026 /* Failing this implies that the terms weren't in order. */ | |
5027 assert( nPrefix<nTerm ); | |
5028 } | |
5029 | |
5030 if( pWriter->data.nData==0 ){ | |
5031 /* Encode the node header and leading term as: | |
5032 ** varint(0) | |
5033 ** varint(nTerm) | |
5034 ** char pTerm[nTerm] | |
5035 */ | |
5036 n = fts3PutVarint(c, '\0'); | |
5037 n += fts3PutVarint(c+n, nTerm); | |
5038 dataBufferAppend2(&pWriter->data, c, n, pTerm, nTerm); | |
5039 }else{ | |
5040 /* Delta-encode the term as: | |
5041 ** varint(nPrefix) | |
5042 ** varint(nSuffix) | |
5043 ** char pTermSuffix[nSuffix] | |
5044 */ | |
5045 n = fts3PutVarint(c, nPrefix); | |
5046 n += fts3PutVarint(c+n, nTerm-nPrefix); | |
5047 dataBufferAppend2(&pWriter->data, c, n, pTerm+nPrefix, nTerm-nPrefix); | |
5048 } | |
5049 dataBufferReplace(&pWriter->term, pTerm, nTerm); | |
5050 | |
5051 return nPrefix+1; | |
5052 } | |
5053 | |
5054 /* Used to avoid a memmove when a large amount of doclist data is in | |
5055 ** the buffer. This constructs a node and term header before | |
5056 ** iDoclistData and flushes the resulting complete node using | |
5057 ** leafWriterInternalFlush(). | |
5058 */ | |
5059 static int leafWriterInlineFlush(fulltext_vtab *v, LeafWriter *pWriter, | |
5060 const char *pTerm, int nTerm, | |
5061 int iDoclistData){ | |
5062 char c[VARINT_MAX+VARINT_MAX]; | |
5063 int iData, n = fts3PutVarint(c, 0); | |
5064 n += fts3PutVarint(c+n, nTerm); | |
5065 | |
5066 /* There should always be room for the header. Even if pTerm shared | |
5067 ** a substantial prefix with the previous term, the entire prefix | |
5068 ** could be constructed from earlier data in the doclist, so there | |
5069 ** should be room. | |
5070 */ | |
5071 assert( iDoclistData>=n+nTerm ); | |
5072 | |
5073 iData = iDoclistData-(n+nTerm); | |
5074 memcpy(pWriter->data.pData+iData, c, n); | |
5075 memcpy(pWriter->data.pData+iData+n, pTerm, nTerm); | |
5076 | |
5077 return leafWriterInternalFlush(v, pWriter, iData, pWriter->data.nData-iData); | |
5078 } | |
5079 | |
5080 /* Push pTerm[nTerm] along with the doclist data to the leaf layer of | |
5081 ** %_segments. | |
5082 */ | |
5083 static int leafWriterStepMerge(fulltext_vtab *v, LeafWriter *pWriter, | |
5084 const char *pTerm, int nTerm, | |
5085 DLReader *pReaders, int nReaders){ | |
5086 char c[VARINT_MAX+VARINT_MAX]; | |
5087 int iTermData = pWriter->data.nData, iDoclistData; | |
5088 int i, nData, n, nActualData, nActual, rc, nTermDistinct; | |
5089 | |
5090 ASSERT_VALID_LEAF_NODE(pWriter->data.pData, pWriter->data.nData); | |
5091 nTermDistinct = leafWriterEncodeTerm(pWriter, pTerm, nTerm); | |
5092 | |
5093 /* Remember nTermDistinct if opening a new node. */ | |
5094 if( iTermData==0 ) pWriter->nTermDistinct = nTermDistinct; | |
5095 | |
5096 iDoclistData = pWriter->data.nData; | |
5097 | |
5098 /* Estimate the length of the merged doclist so we can leave space | |
5099 ** to encode it. | |
5100 */ | |
5101 for(i=0, nData=0; i<nReaders; i++){ | |
5102 nData += dlrAllDataBytes(&pReaders[i]); | |
5103 } | |
5104 n = fts3PutVarint(c, nData); | |
5105 dataBufferAppend(&pWriter->data, c, n); | |
5106 | |
5107 rc = docListMerge(&pWriter->data, pReaders, nReaders); | |
5108 if( rc!=SQLITE_OK ) return rc; | |
5109 ASSERT_VALID_DOCLIST(DL_DEFAULT, | |
5110 pWriter->data.pData+iDoclistData+n, | |
5111 pWriter->data.nData-iDoclistData-n, NULL); | |
5112 | |
5113 /* The actual amount of doclist data at this point could be smaller | |
5114 ** than the length we encoded. Additionally, the space required to | |
5115 ** encode this length could be smaller. For small doclists, this is | |
5116 ** not a big deal, we can just use memmove() to adjust things. | |
5117 */ | |
5118 nActualData = pWriter->data.nData-(iDoclistData+n); | |
5119 nActual = fts3PutVarint(c, nActualData); | |
5120 assert( nActualData<=nData ); | |
5121 assert( nActual<=n ); | |
5122 | |
5123 /* If the new doclist is big enough for force a standalone leaf | |
5124 ** node, we can immediately flush it inline without doing the | |
5125 ** memmove(). | |
5126 */ | |
5127 /* TODO(shess) This test matches leafWriterStep(), which does this | |
5128 ** test before it knows the cost to varint-encode the term and | |
5129 ** doclist lengths. At some point, change to | |
5130 ** pWriter->data.nData-iTermData>STANDALONE_MIN. | |
5131 */ | |
5132 if( nTerm+nActualData>STANDALONE_MIN ){ | |
5133 /* Push leaf node from before this term. */ | |
5134 if( iTermData>0 ){ | |
5135 rc = leafWriterInternalFlush(v, pWriter, 0, iTermData); | |
5136 if( rc!=SQLITE_OK ) return rc; | |
5137 | |
5138 pWriter->nTermDistinct = nTermDistinct; | |
5139 } | |
5140 | |
5141 /* Fix the encoded doclist length. */ | |
5142 iDoclistData += n - nActual; | |
5143 memcpy(pWriter->data.pData+iDoclistData, c, nActual); | |
5144 | |
5145 /* Push the standalone leaf node. */ | |
5146 rc = leafWriterInlineFlush(v, pWriter, pTerm, nTerm, iDoclistData); | |
5147 if( rc!=SQLITE_OK ) return rc; | |
5148 | |
5149 /* Leave the node empty. */ | |
5150 dataBufferReset(&pWriter->data); | |
5151 | |
5152 return rc; | |
5153 } | |
5154 | |
5155 /* At this point, we know that the doclist was small, so do the | |
5156 ** memmove if indicated. | |
5157 */ | |
5158 if( nActual<n ){ | |
5159 memmove(pWriter->data.pData+iDoclistData+nActual, | |
5160 pWriter->data.pData+iDoclistData+n, | |
5161 pWriter->data.nData-(iDoclistData+n)); | |
5162 pWriter->data.nData -= n-nActual; | |
5163 } | |
5164 | |
5165 /* Replace written length with actual length. */ | |
5166 memcpy(pWriter->data.pData+iDoclistData, c, nActual); | |
5167 | |
5168 /* If the node is too large, break things up. */ | |
5169 /* TODO(shess) This test matches leafWriterStep(), which does this | |
5170 ** test before it knows the cost to varint-encode the term and | |
5171 ** doclist lengths. At some point, change to | |
5172 ** pWriter->data.nData>LEAF_MAX. | |
5173 */ | |
5174 if( iTermData+nTerm+nActualData>LEAF_MAX ){ | |
5175 /* Flush out the leading data as a node */ | |
5176 rc = leafWriterInternalFlush(v, pWriter, 0, iTermData); | |
5177 if( rc!=SQLITE_OK ) return rc; | |
5178 | |
5179 pWriter->nTermDistinct = nTermDistinct; | |
5180 | |
5181 /* Rebuild header using the current term */ | |
5182 n = fts3PutVarint(pWriter->data.pData, 0); | |
5183 n += fts3PutVarint(pWriter->data.pData+n, nTerm); | |
5184 memcpy(pWriter->data.pData+n, pTerm, nTerm); | |
5185 n += nTerm; | |
5186 | |
5187 /* There should always be room, because the previous encoding | |
5188 ** included all data necessary to construct the term. | |
5189 */ | |
5190 assert( n<iDoclistData ); | |
5191 /* So long as STANDALONE_MIN is half or less of LEAF_MAX, the | |
5192 ** following memcpy() is safe (as opposed to needing a memmove). | |
5193 */ | |
5194 assert( 2*STANDALONE_MIN<=LEAF_MAX ); | |
5195 assert( n+pWriter->data.nData-iDoclistData<iDoclistData ); | |
5196 memcpy(pWriter->data.pData+n, | |
5197 pWriter->data.pData+iDoclistData, | |
5198 pWriter->data.nData-iDoclistData); | |
5199 pWriter->data.nData -= iDoclistData-n; | |
5200 } | |
5201 ASSERT_VALID_LEAF_NODE(pWriter->data.pData, pWriter->data.nData); | |
5202 | |
5203 return SQLITE_OK; | |
5204 } | |
5205 | |
5206 /* Push pTerm[nTerm] along with the doclist data to the leaf layer of | |
5207 ** %_segments. | |
5208 */ | |
5209 /* TODO(shess) Revise writeZeroSegment() so that doclists are | |
5210 ** constructed directly in pWriter->data. | |
5211 */ | |
5212 static int leafWriterStep(fulltext_vtab *v, LeafWriter *pWriter, | |
5213 const char *pTerm, int nTerm, | |
5214 const char *pData, int nData){ | |
5215 int rc; | |
5216 DLReader reader; | |
5217 | |
5218 rc = dlrInit(&reader, DL_DEFAULT, pData, nData); | |
5219 if( rc!=SQLITE_OK ) return rc; | |
5220 rc = leafWriterStepMerge(v, pWriter, pTerm, nTerm, &reader, 1); | |
5221 dlrDestroy(&reader); | |
5222 | |
5223 return rc; | |
5224 } | |
5225 | |
5226 | |
5227 /****************************************************************/ | |
5228 /* LeafReader is used to iterate over an individual leaf node. */ | |
5229 typedef struct LeafReader { | |
5230 DataBuffer term; /* copy of current term. */ | |
5231 | |
5232 const char *pData; /* data for current term. */ | |
5233 int nData; | |
5234 } LeafReader; | |
5235 | |
5236 static void leafReaderDestroy(LeafReader *pReader){ | |
5237 dataBufferDestroy(&pReader->term); | |
5238 SCRAMBLE(pReader); | |
5239 } | |
5240 | |
5241 static int leafReaderAtEnd(LeafReader *pReader){ | |
5242 return pReader->nData<=0; | |
5243 } | |
5244 | |
5245 /* Access the current term. */ | |
5246 static int leafReaderTermBytes(LeafReader *pReader){ | |
5247 return pReader->term.nData; | |
5248 } | |
5249 static const char *leafReaderTerm(LeafReader *pReader){ | |
5250 assert( pReader->term.nData>0 ); | |
5251 return pReader->term.pData; | |
5252 } | |
5253 | |
5254 /* Access the doclist data for the current term. */ | |
5255 static int leafReaderDataBytes(LeafReader *pReader){ | |
5256 int nData; | |
5257 assert( pReader->term.nData>0 ); | |
5258 fts3GetVarint32(pReader->pData, &nData); | |
5259 return nData; | |
5260 } | |
5261 static const char *leafReaderData(LeafReader *pReader){ | |
5262 int n, nData; | |
5263 assert( pReader->term.nData>0 ); | |
5264 n = fts3GetVarint32Safe(pReader->pData, &nData, pReader->nData); | |
5265 if( !n || nData>pReader->nData-n ) return NULL; | |
5266 return pReader->pData+n; | |
5267 } | |
5268 | |
5269 static int leafReaderInit(const char *pData, int nData, | |
5270 LeafReader *pReader){ | |
5271 int nTerm, n; | |
5272 | |
5273 /* All callers check this precondition. */ | |
5274 assert( nData>0 ); | |
5275 assert( pData[0]=='\0' ); | |
5276 | |
5277 CLEAR(pReader); | |
5278 | |
5279 /* Read the first term, skipping the header byte. */ | |
5280 n = fts3GetVarint32Safe(pData+1, &nTerm, nData-1); | |
5281 if( !n || nTerm<0 || nTerm>nData-1-n ) return SQLITE_CORRUPT_BKPT; | |
5282 dataBufferInit(&pReader->term, nTerm); | |
5283 dataBufferReplace(&pReader->term, pData+1+n, nTerm); | |
5284 | |
5285 /* Position after the first term. */ | |
5286 pReader->pData = pData+1+n+nTerm; | |
5287 pReader->nData = nData-1-n-nTerm; | |
5288 return SQLITE_OK; | |
5289 } | |
5290 | |
5291 /* Step the reader forward to the next term. */ | |
5292 static int leafReaderStep(LeafReader *pReader){ | |
5293 int n, nData, nPrefix, nSuffix; | |
5294 assert( !leafReaderAtEnd(pReader) ); | |
5295 | |
5296 /* Skip previous entry's data block. */ | |
5297 n = fts3GetVarint32Safe(pReader->pData, &nData, pReader->nData); | |
5298 if( !n || nData<0 || nData>pReader->nData-n ) return SQLITE_CORRUPT_BKPT; | |
5299 pReader->pData += n+nData; | |
5300 pReader->nData -= n+nData; | |
5301 | |
5302 if( !leafReaderAtEnd(pReader) ){ | |
5303 /* Construct the new term using a prefix from the old term plus a | |
5304 ** suffix from the leaf data. | |
5305 */ | |
5306 n = fts3GetVarint32Safe(pReader->pData, &nPrefix, pReader->nData); | |
5307 if( !n ) return SQLITE_CORRUPT_BKPT; | |
5308 pReader->nData -= n; | |
5309 pReader->pData += n; | |
5310 n = fts3GetVarint32Safe(pReader->pData, &nSuffix, pReader->nData); | |
5311 if( !n ) return SQLITE_CORRUPT_BKPT; | |
5312 pReader->nData -= n; | |
5313 pReader->pData += n; | |
5314 if( nSuffix<0 || nSuffix>pReader->nData ) return SQLITE_CORRUPT_BKPT; | |
5315 if( nPrefix<0 || nPrefix>pReader->term.nData ) return SQLITE_CORRUPT_BKPT; | |
5316 pReader->term.nData = nPrefix; | |
5317 dataBufferAppend(&pReader->term, pReader->pData, nSuffix); | |
5318 | |
5319 pReader->pData += nSuffix; | |
5320 pReader->nData -= nSuffix; | |
5321 } | |
5322 return SQLITE_OK; | |
5323 } | |
5324 | |
5325 /* strcmp-style comparison of pReader's current term against pTerm. | |
5326 ** If isPrefix, equality means equal through nTerm bytes. | |
5327 */ | |
5328 static int leafReaderTermCmp(LeafReader *pReader, | |
5329 const char *pTerm, int nTerm, int isPrefix){ | |
5330 int c, n = pReader->term.nData<nTerm ? pReader->term.nData : nTerm; | |
5331 if( n==0 ){ | |
5332 if( pReader->term.nData>0 ) return -1; | |
5333 if(nTerm>0 ) return 1; | |
5334 return 0; | |
5335 } | |
5336 | |
5337 c = memcmp(pReader->term.pData, pTerm, n); | |
5338 if( c!=0 ) return c; | |
5339 if( isPrefix && n==nTerm ) return 0; | |
5340 return pReader->term.nData - nTerm; | |
5341 } | |
5342 | |
5343 | |
5344 /****************************************************************/ | |
5345 /* LeavesReader wraps LeafReader to allow iterating over the entire | |
5346 ** leaf layer of the tree. | |
5347 */ | |
5348 typedef struct LeavesReader { | |
5349 int idx; /* Index within the segment. */ | |
5350 | |
5351 sqlite3_stmt *pStmt; /* Statement we're streaming leaves from. */ | |
5352 int eof; /* we've seen SQLITE_DONE from pStmt. */ | |
5353 | |
5354 LeafReader leafReader; /* reader for the current leaf. */ | |
5355 DataBuffer rootData; /* root data for inline. */ | |
5356 } LeavesReader; | |
5357 | |
5358 /* Access the current term. */ | |
5359 static int leavesReaderTermBytes(LeavesReader *pReader){ | |
5360 assert( !pReader->eof ); | |
5361 return leafReaderTermBytes(&pReader->leafReader); | |
5362 } | |
5363 static const char *leavesReaderTerm(LeavesReader *pReader){ | |
5364 assert( !pReader->eof ); | |
5365 return leafReaderTerm(&pReader->leafReader); | |
5366 } | |
5367 | |
5368 /* Access the doclist data for the current term. */ | |
5369 static int leavesReaderDataBytes(LeavesReader *pReader){ | |
5370 assert( !pReader->eof ); | |
5371 return leafReaderDataBytes(&pReader->leafReader); | |
5372 } | |
5373 static const char *leavesReaderData(LeavesReader *pReader){ | |
5374 assert( !pReader->eof ); | |
5375 return leafReaderData(&pReader->leafReader); | |
5376 } | |
5377 | |
5378 static int leavesReaderAtEnd(LeavesReader *pReader){ | |
5379 return pReader->eof; | |
5380 } | |
5381 | |
5382 /* loadSegmentLeaves() may not read all the way to SQLITE_DONE, thus | |
5383 ** leaving the statement handle open, which locks the table. | |
5384 */ | |
5385 /* TODO(shess) This "solution" is not satisfactory. Really, there | |
5386 ** should be check-in function for all statement handles which | |
5387 ** arranges to call sqlite3_reset(). This most likely will require | |
5388 ** modification to control flow all over the place, though, so for now | |
5389 ** just punt. | |
5390 ** | |
5391 ** Note the the current system assumes that segment merges will run to | |
5392 ** completion, which is why this particular probably hasn't arisen in | |
5393 ** this case. Probably a brittle assumption. | |
5394 */ | |
5395 static int leavesReaderReset(LeavesReader *pReader){ | |
5396 return sqlite3_reset(pReader->pStmt); | |
5397 } | |
5398 | |
5399 static void leavesReaderDestroy(LeavesReader *pReader){ | |
5400 /* If idx is -1, that means we're using a non-cached statement | |
5401 ** handle in the optimize() case, so we need to release it. | |
5402 */ | |
5403 if( pReader->pStmt!=NULL && pReader->idx==-1 ){ | |
5404 sqlite3_finalize(pReader->pStmt); | |
5405 } | |
5406 leafReaderDestroy(&pReader->leafReader); | |
5407 dataBufferDestroy(&pReader->rootData); | |
5408 SCRAMBLE(pReader); | |
5409 } | |
5410 | |
5411 /* Initialize pReader with the given root data (if iStartBlockid==0 | |
5412 ** the leaf data was entirely contained in the root), or from the | |
5413 ** stream of blocks between iStartBlockid and iEndBlockid, inclusive. | |
5414 */ | |
5415 static int leavesReaderInit(fulltext_vtab *v, | |
5416 int idx, | |
5417 sqlite_int64 iStartBlockid, | |
5418 sqlite_int64 iEndBlockid, | |
5419 const char *pRootData, int nRootData, | |
5420 LeavesReader *pReader){ | |
5421 CLEAR(pReader); | |
5422 pReader->idx = idx; | |
5423 | |
5424 dataBufferInit(&pReader->rootData, 0); | |
5425 if( iStartBlockid==0 ){ | |
5426 int rc; | |
5427 /* Corrupt if this can't be a leaf node. */ | |
5428 if( pRootData==NULL || nRootData<1 || pRootData[0]!='\0' ){ | |
5429 return SQLITE_CORRUPT_BKPT; | |
5430 } | |
5431 /* Entire leaf level fit in root data. */ | |
5432 dataBufferReplace(&pReader->rootData, pRootData, nRootData); | |
5433 rc = leafReaderInit(pReader->rootData.pData, pReader->rootData.nData, | |
5434 &pReader->leafReader); | |
5435 if( rc!=SQLITE_OK ){ | |
5436 dataBufferDestroy(&pReader->rootData); | |
5437 return rc; | |
5438 } | |
5439 }else{ | |
5440 sqlite3_stmt *s; | |
5441 int rc = sql_get_leaf_statement(v, idx, &s); | |
5442 if( rc!=SQLITE_OK ) return rc; | |
5443 | |
5444 rc = sqlite3_bind_int64(s, 1, iStartBlockid); | |
5445 if( rc!=SQLITE_OK ) goto err; | |
5446 | |
5447 rc = sqlite3_bind_int64(s, 2, iEndBlockid); | |
5448 if( rc!=SQLITE_OK ) goto err; | |
5449 | |
5450 rc = sqlite3_step(s); | |
5451 | |
5452 /* Corrupt if interior node referenced missing leaf node. */ | |
5453 if( rc==SQLITE_DONE ){ | |
5454 rc = SQLITE_CORRUPT_BKPT; | |
5455 goto err; | |
5456 } | |
5457 | |
5458 if( rc!=SQLITE_ROW ) goto err; | |
5459 rc = SQLITE_OK; | |
5460 | |
5461 /* Corrupt if leaf data isn't a blob. */ | |
5462 if( sqlite3_column_type(s, 0)!=SQLITE_BLOB ){ | |
5463 rc = SQLITE_CORRUPT_BKPT; | |
5464 }else{ | |
5465 const char *pLeafData = sqlite3_column_blob(s, 0); | |
5466 int nLeafData = sqlite3_column_bytes(s, 0); | |
5467 | |
5468 /* Corrupt if this can't be a leaf node. */ | |
5469 if( pLeafData==NULL || nLeafData<1 || pLeafData[0]!='\0' ){ | |
5470 rc = SQLITE_CORRUPT_BKPT; | |
5471 }else{ | |
5472 rc = leafReaderInit(pLeafData, nLeafData, &pReader->leafReader); | |
5473 } | |
5474 } | |
5475 | |
5476 err: | |
5477 if( rc!=SQLITE_OK ){ | |
5478 if( idx==-1 ){ | |
5479 sqlite3_finalize(s); | |
5480 }else{ | |
5481 sqlite3_reset(s); | |
5482 } | |
5483 return rc; | |
5484 } | |
5485 | |
5486 pReader->pStmt = s; | |
5487 } | |
5488 return SQLITE_OK; | |
5489 } | |
5490 | |
5491 /* Step the current leaf forward to the next term. If we reach the | |
5492 ** end of the current leaf, step forward to the next leaf block. | |
5493 */ | |
5494 static int leavesReaderStep(fulltext_vtab *v, LeavesReader *pReader){ | |
5495 int rc; | |
5496 assert( !leavesReaderAtEnd(pReader) ); | |
5497 rc = leafReaderStep(&pReader->leafReader); | |
5498 if( rc!=SQLITE_OK ) return rc; | |
5499 | |
5500 if( leafReaderAtEnd(&pReader->leafReader) ){ | |
5501 if( pReader->rootData.pData ){ | |
5502 pReader->eof = 1; | |
5503 return SQLITE_OK; | |
5504 } | |
5505 rc = sqlite3_step(pReader->pStmt); | |
5506 if( rc!=SQLITE_ROW ){ | |
5507 pReader->eof = 1; | |
5508 return rc==SQLITE_DONE ? SQLITE_OK : rc; | |
5509 } | |
5510 | |
5511 /* Corrupt if leaf data isn't a blob. */ | |
5512 if( sqlite3_column_type(pReader->pStmt, 0)!=SQLITE_BLOB ){ | |
5513 return SQLITE_CORRUPT_BKPT; | |
5514 }else{ | |
5515 LeafReader tmp; | |
5516 const char *pLeafData = sqlite3_column_blob(pReader->pStmt, 0); | |
5517 int nLeafData = sqlite3_column_bytes(pReader->pStmt, 0); | |
5518 | |
5519 /* Corrupt if this can't be a leaf node. */ | |
5520 if( pLeafData==NULL || nLeafData<1 || pLeafData[0]!='\0' ){ | |
5521 return SQLITE_CORRUPT_BKPT; | |
5522 } | |
5523 | |
5524 rc = leafReaderInit(pLeafData, nLeafData, &tmp); | |
5525 if( rc!=SQLITE_OK ) return rc; | |
5526 leafReaderDestroy(&pReader->leafReader); | |
5527 pReader->leafReader = tmp; | |
5528 } | |
5529 } | |
5530 return SQLITE_OK; | |
5531 } | |
5532 | |
5533 /* Order LeavesReaders by their term, ignoring idx. Readers at eof | |
5534 ** always sort to the end. | |
5535 */ | |
5536 static int leavesReaderTermCmp(LeavesReader *lr1, LeavesReader *lr2){ | |
5537 if( leavesReaderAtEnd(lr1) ){ | |
5538 if( leavesReaderAtEnd(lr2) ) return 0; | |
5539 return 1; | |
5540 } | |
5541 if( leavesReaderAtEnd(lr2) ) return -1; | |
5542 | |
5543 return leafReaderTermCmp(&lr1->leafReader, | |
5544 leavesReaderTerm(lr2), leavesReaderTermBytes(lr2), | |
5545 0); | |
5546 } | |
5547 | |
5548 /* Similar to leavesReaderTermCmp(), with additional ordering by idx | |
5549 ** so that older segments sort before newer segments. | |
5550 */ | |
5551 static int leavesReaderCmp(LeavesReader *lr1, LeavesReader *lr2){ | |
5552 int c = leavesReaderTermCmp(lr1, lr2); | |
5553 if( c!=0 ) return c; | |
5554 return lr1->idx-lr2->idx; | |
5555 } | |
5556 | |
5557 /* Assume that pLr[1]..pLr[nLr] are sorted. Bubble pLr[0] into its | |
5558 ** sorted position. | |
5559 */ | |
5560 static void leavesReaderReorder(LeavesReader *pLr, int nLr){ | |
5561 while( nLr>1 && leavesReaderCmp(pLr, pLr+1)>0 ){ | |
5562 LeavesReader tmp = pLr[0]; | |
5563 pLr[0] = pLr[1]; | |
5564 pLr[1] = tmp; | |
5565 nLr--; | |
5566 pLr++; | |
5567 } | |
5568 } | |
5569 | |
5570 /* Initializes pReaders with the segments from level iLevel, returning | |
5571 ** the number of segments in *piReaders. Leaves pReaders in sorted | |
5572 ** order. | |
5573 */ | |
5574 static int leavesReadersInit(fulltext_vtab *v, int iLevel, | |
5575 LeavesReader *pReaders, int *piReaders){ | |
5576 sqlite3_stmt *s; | |
5577 int i, rc = sql_get_statement(v, SEGDIR_SELECT_LEVEL_STMT, &s); | |
5578 if( rc!=SQLITE_OK ) return rc; | |
5579 | |
5580 rc = sqlite3_bind_int(s, 1, iLevel); | |
5581 if( rc!=SQLITE_OK ) return rc; | |
5582 | |
5583 i = 0; | |
5584 while( (rc = sqlite3_step(s))==SQLITE_ROW ){ | |
5585 sqlite_int64 iStart = sqlite3_column_int64(s, 0); | |
5586 sqlite_int64 iEnd = sqlite3_column_int64(s, 1); | |
5587 const char *pRootData = sqlite3_column_blob(s, 2); | |
5588 int nRootData = sqlite3_column_bytes(s, 2); | |
5589 sqlite_int64 iIndex = sqlite3_column_int64(s, 3); | |
5590 | |
5591 /* Corrupt if we get back different types than we stored. */ | |
5592 /* Also corrupt if the index is not sequential starting at 0. */ | |
5593 if( sqlite3_column_type(s, 0)!=SQLITE_INTEGER || | |
5594 sqlite3_column_type(s, 1)!=SQLITE_INTEGER || | |
5595 sqlite3_column_type(s, 2)!=SQLITE_BLOB || | |
5596 i!=iIndex || | |
5597 i>=MERGE_COUNT ){ | |
5598 rc = SQLITE_CORRUPT_BKPT; | |
5599 break; | |
5600 } | |
5601 | |
5602 rc = leavesReaderInit(v, i, iStart, iEnd, pRootData, nRootData, | |
5603 &pReaders[i]); | |
5604 if( rc!=SQLITE_OK ) break; | |
5605 | |
5606 i++; | |
5607 } | |
5608 if( rc!=SQLITE_DONE ){ | |
5609 while( i-->0 ){ | |
5610 leavesReaderDestroy(&pReaders[i]); | |
5611 } | |
5612 sqlite3_reset(s); /* So we don't leave a lock. */ | |
5613 return rc; | |
5614 } | |
5615 | |
5616 *piReaders = i; | |
5617 | |
5618 /* Leave our results sorted by term, then age. */ | |
5619 while( i-- ){ | |
5620 leavesReaderReorder(pReaders+i, *piReaders-i); | |
5621 } | |
5622 return SQLITE_OK; | |
5623 } | |
5624 | |
5625 /* Merge doclists from pReaders[nReaders] into a single doclist, which | |
5626 ** is written to pWriter. Assumes pReaders is ordered oldest to | |
5627 ** newest. | |
5628 */ | |
5629 /* TODO(shess) Consider putting this inline in segmentMerge(). */ | |
5630 static int leavesReadersMerge(fulltext_vtab *v, | |
5631 LeavesReader *pReaders, int nReaders, | |
5632 LeafWriter *pWriter){ | |
5633 DLReader dlReaders[MERGE_COUNT]; | |
5634 const char *pTerm = leavesReaderTerm(pReaders); | |
5635 int i, nTerm = leavesReaderTermBytes(pReaders); | |
5636 int rc; | |
5637 | |
5638 assert( nReaders<=MERGE_COUNT ); | |
5639 | |
5640 for(i=0; i<nReaders; i++){ | |
5641 const char *pData = leavesReaderData(pReaders+i); | |
5642 if( pData==NULL ){ | |
5643 rc = SQLITE_CORRUPT_BKPT; | |
5644 break; | |
5645 } | |
5646 rc = dlrInit(&dlReaders[i], DL_DEFAULT, | |
5647 pData, | |
5648 leavesReaderDataBytes(pReaders+i)); | |
5649 if( rc!=SQLITE_OK ) break; | |
5650 } | |
5651 if( rc!=SQLITE_OK ){ | |
5652 while( i-->0 ){ | |
5653 dlrDestroy(&dlReaders[i]); | |
5654 } | |
5655 return rc; | |
5656 } | |
5657 | |
5658 return leafWriterStepMerge(v, pWriter, pTerm, nTerm, dlReaders, nReaders); | |
5659 } | |
5660 | |
5661 /* Forward ref due to mutual recursion with segdirNextIndex(). */ | |
5662 static int segmentMerge(fulltext_vtab *v, int iLevel); | |
5663 | |
5664 /* Put the next available index at iLevel into *pidx. If iLevel | |
5665 ** already has MERGE_COUNT segments, they are merged to a higher | |
5666 ** level to make room. | |
5667 */ | |
5668 static int segdirNextIndex(fulltext_vtab *v, int iLevel, int *pidx){ | |
5669 int rc = segdir_max_index(v, iLevel, pidx); | |
5670 if( rc==SQLITE_DONE ){ /* No segments at iLevel. */ | |
5671 *pidx = 0; | |
5672 }else if( rc==SQLITE_ROW ){ | |
5673 if( *pidx==(MERGE_COUNT-1) ){ | |
5674 rc = segmentMerge(v, iLevel); | |
5675 if( rc!=SQLITE_OK ) return rc; | |
5676 *pidx = 0; | |
5677 }else{ | |
5678 (*pidx)++; | |
5679 } | |
5680 }else{ | |
5681 return rc; | |
5682 } | |
5683 return SQLITE_OK; | |
5684 } | |
5685 | |
5686 /* Merge MERGE_COUNT segments at iLevel into a new segment at | |
5687 ** iLevel+1. If iLevel+1 is already full of segments, those will be | |
5688 ** merged to make room. | |
5689 */ | |
5690 static int segmentMerge(fulltext_vtab *v, int iLevel){ | |
5691 LeafWriter writer; | |
5692 LeavesReader lrs[MERGE_COUNT]; | |
5693 int i, rc, idx = 0; | |
5694 | |
5695 /* Determine the next available segment index at the next level, | |
5696 ** merging as necessary. | |
5697 */ | |
5698 rc = segdirNextIndex(v, iLevel+1, &idx); | |
5699 if( rc!=SQLITE_OK ) return rc; | |
5700 | |
5701 /* TODO(shess) This assumes that we'll always see exactly | |
5702 ** MERGE_COUNT segments to merge at a given level. That will be | |
5703 ** broken if we allow the developer to request preemptive or | |
5704 ** deferred merging. | |
5705 */ | |
5706 memset(&lrs, '\0', sizeof(lrs)); | |
5707 rc = leavesReadersInit(v, iLevel, lrs, &i); | |
5708 if( rc!=SQLITE_OK ) return rc; | |
5709 | |
5710 leafWriterInit(iLevel+1, idx, &writer); | |
5711 | |
5712 if( i!=MERGE_COUNT ){ | |
5713 rc = SQLITE_CORRUPT_BKPT; | |
5714 goto err; | |
5715 } | |
5716 | |
5717 /* Since leavesReaderReorder() pushes readers at eof to the end, | |
5718 ** when the first reader is empty, all will be empty. | |
5719 */ | |
5720 while( !leavesReaderAtEnd(lrs) ){ | |
5721 /* Figure out how many readers share their next term. */ | |
5722 for(i=1; i<MERGE_COUNT && !leavesReaderAtEnd(lrs+i); i++){ | |
5723 if( 0!=leavesReaderTermCmp(lrs, lrs+i) ) break; | |
5724 } | |
5725 | |
5726 rc = leavesReadersMerge(v, lrs, i, &writer); | |
5727 if( rc!=SQLITE_OK ) goto err; | |
5728 | |
5729 /* Step forward those that were merged. */ | |
5730 while( i-->0 ){ | |
5731 rc = leavesReaderStep(v, lrs+i); | |
5732 if( rc!=SQLITE_OK ) goto err; | |
5733 | |
5734 /* Reorder by term, then by age. */ | |
5735 leavesReaderReorder(lrs+i, MERGE_COUNT-i); | |
5736 } | |
5737 } | |
5738 | |
5739 for(i=0; i<MERGE_COUNT; i++){ | |
5740 leavesReaderDestroy(&lrs[i]); | |
5741 } | |
5742 | |
5743 rc = leafWriterFinalize(v, &writer); | |
5744 leafWriterDestroy(&writer); | |
5745 if( rc!=SQLITE_OK ) return rc; | |
5746 | |
5747 /* Delete the merged segment data. */ | |
5748 return segdir_delete(v, iLevel); | |
5749 | |
5750 err: | |
5751 for(i=0; i<MERGE_COUNT; i++){ | |
5752 leavesReaderDestroy(&lrs[i]); | |
5753 } | |
5754 leafWriterDestroy(&writer); | |
5755 return rc; | |
5756 } | |
5757 | |
5758 /* Accumulate the union of *acc and *pData into *acc. */ | |
5759 static int docListAccumulateUnion(DataBuffer *acc, | |
5760 const char *pData, int nData) { | |
5761 DataBuffer tmp = *acc; | |
5762 int rc; | |
5763 dataBufferInit(acc, tmp.nData+nData); | |
5764 rc = docListUnion(tmp.pData, tmp.nData, pData, nData, acc); | |
5765 dataBufferDestroy(&tmp); | |
5766 return rc; | |
5767 } | |
5768 | |
5769 /* TODO(shess) It might be interesting to explore different merge | |
5770 ** strategies, here. For instance, since this is a sorted merge, we | |
5771 ** could easily merge many doclists in parallel. With some | |
5772 ** comprehension of the storage format, we could merge all of the | |
5773 ** doclists within a leaf node directly from the leaf node's storage. | |
5774 ** It may be worthwhile to merge smaller doclists before larger | |
5775 ** doclists, since they can be traversed more quickly - but the | |
5776 ** results may have less overlap, making them more expensive in a | |
5777 ** different way. | |
5778 */ | |
5779 | |
5780 /* Scan pReader for pTerm/nTerm, and merge the term's doclist over | |
5781 ** *out (any doclists with duplicate docids overwrite those in *out). | |
5782 ** Internal function for loadSegmentLeaf(). | |
5783 */ | |
5784 static int loadSegmentLeavesInt(fulltext_vtab *v, LeavesReader *pReader, | |
5785 const char *pTerm, int nTerm, int isPrefix, | |
5786 DataBuffer *out){ | |
5787 /* doclist data is accumulated into pBuffers similar to how one does | |
5788 ** increment in binary arithmetic. If index 0 is empty, the data is | |
5789 ** stored there. If there is data there, it is merged and the | |
5790 ** results carried into position 1, with further merge-and-carry | |
5791 ** until an empty position is found. | |
5792 */ | |
5793 DataBuffer *pBuffers = NULL; | |
5794 int nBuffers = 0, nMaxBuffers = 0, rc; | |
5795 | |
5796 assert( nTerm>0 ); | |
5797 | |
5798 for(rc=SQLITE_OK; rc==SQLITE_OK && !leavesReaderAtEnd(pReader); | |
5799 rc=leavesReaderStep(v, pReader)){ | |
5800 /* TODO(shess) Really want leavesReaderTermCmp(), but that name is | |
5801 ** already taken to compare the terms of two LeavesReaders. Think | |
5802 ** on a better name. [Meanwhile, break encapsulation rather than | |
5803 ** use a confusing name.] | |
5804 */ | |
5805 int c = leafReaderTermCmp(&pReader->leafReader, pTerm, nTerm, isPrefix); | |
5806 if( c>0 ) break; /* Past any possible matches. */ | |
5807 if( c==0 ){ | |
5808 int iBuffer, nData; | |
5809 const char *pData = leavesReaderData(pReader); | |
5810 if( pData==NULL ){ | |
5811 rc = SQLITE_CORRUPT_BKPT; | |
5812 break; | |
5813 } | |
5814 nData = leavesReaderDataBytes(pReader); | |
5815 | |
5816 /* Find the first empty buffer. */ | |
5817 for(iBuffer=0; iBuffer<nBuffers; ++iBuffer){ | |
5818 if( 0==pBuffers[iBuffer].nData ) break; | |
5819 } | |
5820 | |
5821 /* Out of buffers, add an empty one. */ | |
5822 if( iBuffer==nBuffers ){ | |
5823 if( nBuffers==nMaxBuffers ){ | |
5824 DataBuffer *p; | |
5825 nMaxBuffers += 20; | |
5826 | |
5827 /* Manual realloc so we can handle NULL appropriately. */ | |
5828 p = sqlite3_malloc(nMaxBuffers*sizeof(*pBuffers)); | |
5829 if( p==NULL ){ | |
5830 rc = SQLITE_NOMEM; | |
5831 break; | |
5832 } | |
5833 | |
5834 if( nBuffers>0 ){ | |
5835 assert(pBuffers!=NULL); | |
5836 memcpy(p, pBuffers, nBuffers*sizeof(*pBuffers)); | |
5837 sqlite3_free(pBuffers); | |
5838 } | |
5839 pBuffers = p; | |
5840 } | |
5841 dataBufferInit(&(pBuffers[nBuffers]), 0); | |
5842 nBuffers++; | |
5843 } | |
5844 | |
5845 /* At this point, must have an empty at iBuffer. */ | |
5846 assert(iBuffer<nBuffers && pBuffers[iBuffer].nData==0); | |
5847 | |
5848 /* If empty was first buffer, no need for merge logic. */ | |
5849 if( iBuffer==0 ){ | |
5850 dataBufferReplace(&(pBuffers[0]), pData, nData); | |
5851 }else{ | |
5852 /* pAcc is the empty buffer the merged data will end up in. */ | |
5853 DataBuffer *pAcc = &(pBuffers[iBuffer]); | |
5854 DataBuffer *p = &(pBuffers[0]); | |
5855 | |
5856 /* Handle position 0 specially to avoid need to prime pAcc | |
5857 ** with pData/nData. | |
5858 */ | |
5859 dataBufferSwap(p, pAcc); | |
5860 rc = docListAccumulateUnion(pAcc, pData, nData); | |
5861 if( rc!=SQLITE_OK ) goto err; | |
5862 | |
5863 /* Accumulate remaining doclists into pAcc. */ | |
5864 for(++p; p<pAcc; ++p){ | |
5865 rc = docListAccumulateUnion(pAcc, p->pData, p->nData); | |
5866 if( rc!=SQLITE_OK ) goto err; | |
5867 | |
5868 /* dataBufferReset() could allow a large doclist to blow up | |
5869 ** our memory requirements. | |
5870 */ | |
5871 if( p->nCapacity<1024 ){ | |
5872 dataBufferReset(p); | |
5873 }else{ | |
5874 dataBufferDestroy(p); | |
5875 dataBufferInit(p, 0); | |
5876 } | |
5877 } | |
5878 } | |
5879 } | |
5880 } | |
5881 | |
5882 /* Union all the doclists together into *out. */ | |
5883 /* TODO(shess) What if *out is big? Sigh. */ | |
5884 if( rc==SQLITE_OK && nBuffers>0 ){ | |
5885 int iBuffer; | |
5886 for(iBuffer=0; iBuffer<nBuffers; ++iBuffer){ | |
5887 if( pBuffers[iBuffer].nData>0 ){ | |
5888 if( out->nData==0 ){ | |
5889 dataBufferSwap(out, &(pBuffers[iBuffer])); | |
5890 }else{ | |
5891 rc = docListAccumulateUnion(out, pBuffers[iBuffer].pData, | |
5892 pBuffers[iBuffer].nData); | |
5893 if( rc!=SQLITE_OK ) break; | |
5894 } | |
5895 } | |
5896 } | |
5897 } | |
5898 | |
5899 err: | |
5900 while( nBuffers-- ){ | |
5901 dataBufferDestroy(&(pBuffers[nBuffers])); | |
5902 } | |
5903 if( pBuffers!=NULL ) sqlite3_free(pBuffers); | |
5904 | |
5905 return rc; | |
5906 } | |
5907 | |
5908 /* Call loadSegmentLeavesInt() with pData/nData as input. */ | |
5909 static int loadSegmentLeaf(fulltext_vtab *v, const char *pData, int nData, | |
5910 const char *pTerm, int nTerm, int isPrefix, | |
5911 DataBuffer *out){ | |
5912 LeavesReader reader; | |
5913 int rc; | |
5914 | |
5915 assert( nData>1 ); | |
5916 assert( *pData=='\0' ); | |
5917 rc = leavesReaderInit(v, 0, 0, 0, pData, nData, &reader); | |
5918 if( rc!=SQLITE_OK ) return rc; | |
5919 | |
5920 rc = loadSegmentLeavesInt(v, &reader, pTerm, nTerm, isPrefix, out); | |
5921 leavesReaderReset(&reader); | |
5922 leavesReaderDestroy(&reader); | |
5923 return rc; | |
5924 } | |
5925 | |
5926 /* Call loadSegmentLeavesInt() with the leaf nodes from iStartLeaf to | |
5927 ** iEndLeaf (inclusive) as input, and merge the resulting doclist into | |
5928 ** out. | |
5929 */ | |
5930 static int loadSegmentLeaves(fulltext_vtab *v, | |
5931 sqlite_int64 iStartLeaf, sqlite_int64 iEndLeaf, | |
5932 const char *pTerm, int nTerm, int isPrefix, | |
5933 DataBuffer *out){ | |
5934 int rc; | |
5935 LeavesReader reader; | |
5936 | |
5937 assert( iStartLeaf<=iEndLeaf ); | |
5938 rc = leavesReaderInit(v, 0, iStartLeaf, iEndLeaf, NULL, 0, &reader); | |
5939 if( rc!=SQLITE_OK ) return rc; | |
5940 | |
5941 rc = loadSegmentLeavesInt(v, &reader, pTerm, nTerm, isPrefix, out); | |
5942 leavesReaderReset(&reader); | |
5943 leavesReaderDestroy(&reader); | |
5944 return rc; | |
5945 } | |
5946 | |
5947 /* Taking pData/nData as an interior node, find the sequence of child | |
5948 ** nodes which could include pTerm/nTerm/isPrefix. Note that the | |
5949 ** interior node terms logically come between the blocks, so there is | |
5950 ** one more blockid than there are terms (that block contains terms >= | |
5951 ** the last interior-node term). | |
5952 */ | |
5953 /* TODO(shess) The calling code may already know that the end child is | |
5954 ** not worth calculating, because the end may be in a later sibling | |
5955 ** node. Consider whether breaking symmetry is worthwhile. I suspect | |
5956 ** it is not worthwhile. | |
5957 */ | |
5958 static int getChildrenContaining(const char *pData, int nData, | |
5959 const char *pTerm, int nTerm, int isPrefix, | |
5960 sqlite_int64 *piStartChild, | |
5961 sqlite_int64 *piEndChild){ | |
5962 InteriorReader reader; | |
5963 int rc; | |
5964 | |
5965 assert( nData>1 ); | |
5966 assert( *pData!='\0' ); | |
5967 rc = interiorReaderInit(pData, nData, &reader); | |
5968 if( rc!=SQLITE_OK ) return rc; | |
5969 | |
5970 /* Scan for the first child which could contain pTerm/nTerm. */ | |
5971 while( !interiorReaderAtEnd(&reader) ){ | |
5972 if( interiorReaderTermCmp(&reader, pTerm, nTerm, 0)>0 ) break; | |
5973 rc = interiorReaderStep(&reader); | |
5974 if( rc!=SQLITE_OK ){ | |
5975 interiorReaderDestroy(&reader); | |
5976 return rc; | |
5977 } | |
5978 } | |
5979 *piStartChild = interiorReaderCurrentBlockid(&reader); | |
5980 | |
5981 /* Keep scanning to find a term greater than our term, using prefix | |
5982 ** comparison if indicated. If isPrefix is false, this will be the | |
5983 ** same blockid as the starting block. | |
5984 */ | |
5985 while( !interiorReaderAtEnd(&reader) ){ | |
5986 if( interiorReaderTermCmp(&reader, pTerm, nTerm, isPrefix)>0 ) break; | |
5987 rc = interiorReaderStep(&reader); | |
5988 if( rc!=SQLITE_OK ){ | |
5989 interiorReaderDestroy(&reader); | |
5990 return rc; | |
5991 } | |
5992 } | |
5993 *piEndChild = interiorReaderCurrentBlockid(&reader); | |
5994 | |
5995 interiorReaderDestroy(&reader); | |
5996 | |
5997 /* Children must ascend, and if !prefix, both must be the same. */ | |
5998 assert( *piEndChild>=*piStartChild ); | |
5999 assert( isPrefix || *piStartChild==*piEndChild ); | |
6000 return rc; | |
6001 } | |
6002 | |
6003 /* Read block at iBlockid and pass it with other params to | |
6004 ** getChildrenContaining(). | |
6005 */ | |
6006 static int loadAndGetChildrenContaining( | |
6007 fulltext_vtab *v, | |
6008 sqlite_int64 iBlockid, | |
6009 const char *pTerm, int nTerm, int isPrefix, | |
6010 sqlite_int64 *piStartChild, sqlite_int64 *piEndChild | |
6011 ){ | |
6012 sqlite3_stmt *s = NULL; | |
6013 int rc; | |
6014 | |
6015 assert( iBlockid!=0 ); | |
6016 assert( pTerm!=NULL ); | |
6017 assert( nTerm!=0 ); /* TODO(shess) Why not allow this? */ | |
6018 assert( piStartChild!=NULL ); | |
6019 assert( piEndChild!=NULL ); | |
6020 | |
6021 rc = sql_get_statement(v, BLOCK_SELECT_STMT, &s); | |
6022 if( rc!=SQLITE_OK ) return rc; | |
6023 | |
6024 rc = sqlite3_bind_int64(s, 1, iBlockid); | |
6025 if( rc!=SQLITE_OK ) return rc; | |
6026 | |
6027 rc = sqlite3_step(s); | |
6028 /* Corrupt if interior node references missing child node. */ | |
6029 if( rc==SQLITE_DONE ) return SQLITE_CORRUPT_BKPT; | |
6030 if( rc!=SQLITE_ROW ) return rc; | |
6031 | |
6032 /* Corrupt if child node isn't a blob. */ | |
6033 if( sqlite3_column_type(s, 0)!=SQLITE_BLOB ){ | |
6034 sqlite3_reset(s); /* So we don't leave a lock. */ | |
6035 return SQLITE_CORRUPT_BKPT; | |
6036 }else{ | |
6037 const char *pData = sqlite3_column_blob(s, 0); | |
6038 int nData = sqlite3_column_bytes(s, 0); | |
6039 | |
6040 /* Corrupt if child is not a valid interior node. */ | |
6041 if( pData==NULL || nData<1 || pData[0]=='\0' ){ | |
6042 sqlite3_reset(s); /* So we don't leave a lock. */ | |
6043 return SQLITE_CORRUPT_BKPT; | |
6044 } | |
6045 | |
6046 rc = getChildrenContaining(pData, nData, pTerm, nTerm, | |
6047 isPrefix, piStartChild, piEndChild); | |
6048 if( rc!=SQLITE_OK ){ | |
6049 sqlite3_reset(s); | |
6050 return rc; | |
6051 } | |
6052 } | |
6053 | |
6054 /* We expect only one row. We must execute another sqlite3_step() | |
6055 * to complete the iteration; otherwise the table will remain | |
6056 * locked. */ | |
6057 rc = sqlite3_step(s); | |
6058 if( rc==SQLITE_ROW ) return SQLITE_ERROR; | |
6059 if( rc!=SQLITE_DONE ) return rc; | |
6060 | |
6061 return SQLITE_OK; | |
6062 } | |
6063 | |
6064 /* Traverse the tree represented by pData[nData] looking for | |
6065 ** pTerm[nTerm], placing its doclist into *out. This is internal to | |
6066 ** loadSegment() to make error-handling cleaner. | |
6067 */ | |
6068 static int loadSegmentInt(fulltext_vtab *v, const char *pData, int nData, | |
6069 sqlite_int64 iLeavesEnd, | |
6070 const char *pTerm, int nTerm, int isPrefix, | |
6071 DataBuffer *out){ | |
6072 /* Special case where root is a leaf. */ | |
6073 if( *pData=='\0' ){ | |
6074 return loadSegmentLeaf(v, pData, nData, pTerm, nTerm, isPrefix, out); | |
6075 }else{ | |
6076 int rc; | |
6077 sqlite_int64 iStartChild, iEndChild; | |
6078 | |
6079 /* Process pData as an interior node, then loop down the tree | |
6080 ** until we find the set of leaf nodes to scan for the term. | |
6081 */ | |
6082 rc = getChildrenContaining(pData, nData, pTerm, nTerm, isPrefix, | |
6083 &iStartChild, &iEndChild); | |
6084 if( rc!=SQLITE_OK ) return rc; | |
6085 while( iStartChild>iLeavesEnd ){ | |
6086 sqlite_int64 iNextStart, iNextEnd; | |
6087 rc = loadAndGetChildrenContaining(v, iStartChild, pTerm, nTerm, isPrefix, | |
6088 &iNextStart, &iNextEnd); | |
6089 if( rc!=SQLITE_OK ) return rc; | |
6090 | |
6091 /* If we've branched, follow the end branch, too. */ | |
6092 if( iStartChild!=iEndChild ){ | |
6093 sqlite_int64 iDummy; | |
6094 rc = loadAndGetChildrenContaining(v, iEndChild, pTerm, nTerm, isPrefix, | |
6095 &iDummy, &iNextEnd); | |
6096 if( rc!=SQLITE_OK ) return rc; | |
6097 } | |
6098 | |
6099 assert( iNextStart<=iNextEnd ); | |
6100 iStartChild = iNextStart; | |
6101 iEndChild = iNextEnd; | |
6102 } | |
6103 assert( iStartChild<=iLeavesEnd ); | |
6104 assert( iEndChild<=iLeavesEnd ); | |
6105 | |
6106 /* Scan through the leaf segments for doclists. */ | |
6107 return loadSegmentLeaves(v, iStartChild, iEndChild, | |
6108 pTerm, nTerm, isPrefix, out); | |
6109 } | |
6110 } | |
6111 | |
6112 /* Call loadSegmentInt() to collect the doclist for pTerm/nTerm, then | |
6113 ** merge its doclist over *out (any duplicate doclists read from the | |
6114 ** segment rooted at pData will overwrite those in *out). | |
6115 */ | |
6116 /* TODO(shess) Consider changing this to determine the depth of the | |
6117 ** leaves using either the first characters of interior nodes (when | |
6118 ** ==1, we're one level above the leaves), or the first character of | |
6119 ** the root (which will describe the height of the tree directly). | |
6120 ** Either feels somewhat tricky to me. | |
6121 */ | |
6122 /* TODO(shess) The current merge is likely to be slow for large | |
6123 ** doclists (though it should process from newest/smallest to | |
6124 ** oldest/largest, so it may not be that bad). It might be useful to | |
6125 ** modify things to allow for N-way merging. This could either be | |
6126 ** within a segment, with pairwise merges across segments, or across | |
6127 ** all segments at once. | |
6128 */ | |
6129 static int loadSegment(fulltext_vtab *v, const char *pData, int nData, | |
6130 sqlite_int64 iLeavesEnd, | |
6131 const char *pTerm, int nTerm, int isPrefix, | |
6132 DataBuffer *out){ | |
6133 DataBuffer result; | |
6134 int rc; | |
6135 | |
6136 /* Corrupt if segment root can't be valid. */ | |
6137 if( pData==NULL || nData<1 ) return SQLITE_CORRUPT_BKPT; | |
6138 | |
6139 /* This code should never be called with buffered updates. */ | |
6140 assert( v->nPendingData<0 ); | |
6141 | |
6142 dataBufferInit(&result, 0); | |
6143 rc = loadSegmentInt(v, pData, nData, iLeavesEnd, | |
6144 pTerm, nTerm, isPrefix, &result); | |
6145 if( rc==SQLITE_OK && result.nData>0 ){ | |
6146 if( out->nData==0 ){ | |
6147 DataBuffer tmp = *out; | |
6148 *out = result; | |
6149 result = tmp; | |
6150 }else{ | |
6151 DataBuffer merged; | |
6152 DLReader readers[2]; | |
6153 | |
6154 rc = dlrInit(&readers[0], DL_DEFAULT, out->pData, out->nData); | |
6155 if( rc==SQLITE_OK ){ | |
6156 rc = dlrInit(&readers[1], DL_DEFAULT, result.pData, result.nData); | |
6157 if( rc==SQLITE_OK ){ | |
6158 dataBufferInit(&merged, out->nData+result.nData); | |
6159 rc = docListMerge(&merged, readers, 2); | |
6160 dataBufferDestroy(out); | |
6161 *out = merged; | |
6162 dlrDestroy(&readers[1]); | |
6163 } | |
6164 dlrDestroy(&readers[0]); | |
6165 } | |
6166 } | |
6167 } | |
6168 | |
6169 dataBufferDestroy(&result); | |
6170 return rc; | |
6171 } | |
6172 | |
6173 /* Scan the database and merge together the posting lists for the term | |
6174 ** into *out. | |
6175 */ | |
6176 static int termSelect( | |
6177 fulltext_vtab *v, | |
6178 int iColumn, | |
6179 const char *pTerm, int nTerm, /* Term to query for */ | |
6180 int isPrefix, /* True for a prefix search */ | |
6181 DocListType iType, | |
6182 DataBuffer *out /* Write results here */ | |
6183 ){ | |
6184 DataBuffer doclist; | |
6185 sqlite3_stmt *s; | |
6186 int rc = sql_get_statement(v, SEGDIR_SELECT_ALL_STMT, &s); | |
6187 if( rc!=SQLITE_OK ) return rc; | |
6188 | |
6189 /* This code should never be called with buffered updates. */ | |
6190 assert( v->nPendingData<0 ); | |
6191 | |
6192 dataBufferInit(&doclist, 0); | |
6193 dataBufferInit(out, 0); | |
6194 | |
6195 /* Traverse the segments from oldest to newest so that newer doclist | |
6196 ** elements for given docids overwrite older elements. | |
6197 */ | |
6198 while( (rc = sqlite3_step(s))==SQLITE_ROW ){ | |
6199 const char *pData = sqlite3_column_blob(s, 2); | |
6200 const int nData = sqlite3_column_bytes(s, 2); | |
6201 const sqlite_int64 iLeavesEnd = sqlite3_column_int64(s, 1); | |
6202 | |
6203 /* Corrupt if we get back different types than we stored. */ | |
6204 if( sqlite3_column_type(s, 1)!=SQLITE_INTEGER || | |
6205 sqlite3_column_type(s, 2)!=SQLITE_BLOB ){ | |
6206 rc = SQLITE_CORRUPT_BKPT; | |
6207 goto err; | |
6208 } | |
6209 | |
6210 rc = loadSegment(v, pData, nData, iLeavesEnd, pTerm, nTerm, isPrefix, | |
6211 &doclist); | |
6212 if( rc!=SQLITE_OK ) goto err; | |
6213 } | |
6214 if( rc==SQLITE_DONE ){ | |
6215 rc = SQLITE_OK; | |
6216 if( doclist.nData!=0 ){ | |
6217 /* TODO(shess) The old term_select_all() code applied the column | |
6218 ** restrict as we merged segments, leading to smaller buffers. | |
6219 ** This is probably worthwhile to bring back, once the new storage | |
6220 ** system is checked in. | |
6221 */ | |
6222 if( iColumn==v->nColumn) iColumn = -1; | |
6223 rc = docListTrim(DL_DEFAULT, doclist.pData, doclist.nData, | |
6224 iColumn, iType, out); | |
6225 } | |
6226 } | |
6227 | |
6228 err: | |
6229 sqlite3_reset(s); /* So we don't leave a lock. */ | |
6230 dataBufferDestroy(&doclist); | |
6231 return rc; | |
6232 } | |
6233 | |
6234 /****************************************************************/ | |
6235 /* Used to hold hashtable data for sorting. */ | |
6236 typedef struct TermData { | |
6237 const char *pTerm; | |
6238 int nTerm; | |
6239 DLCollector *pCollector; | |
6240 } TermData; | |
6241 | |
6242 /* Orders TermData elements in strcmp fashion ( <0 for less-than, 0 | |
6243 ** for equal, >0 for greater-than). | |
6244 */ | |
6245 static int termDataCmp(const void *av, const void *bv){ | |
6246 const TermData *a = (const TermData *)av; | |
6247 const TermData *b = (const TermData *)bv; | |
6248 int n = a->nTerm<b->nTerm ? a->nTerm : b->nTerm; | |
6249 int c = memcmp(a->pTerm, b->pTerm, n); | |
6250 if( c!=0 ) return c; | |
6251 return a->nTerm-b->nTerm; | |
6252 } | |
6253 | |
6254 /* Order pTerms data by term, then write a new level 0 segment using | |
6255 ** LeafWriter. | |
6256 */ | |
6257 static int writeZeroSegment(fulltext_vtab *v, fts3Hash *pTerms){ | |
6258 fts3HashElem *e; | |
6259 int idx, rc, i, n; | |
6260 TermData *pData; | |
6261 LeafWriter writer; | |
6262 DataBuffer dl; | |
6263 | |
6264 /* Determine the next index at level 0, merging as necessary. */ | |
6265 rc = segdirNextIndex(v, 0, &idx); | |
6266 if( rc!=SQLITE_OK ) return rc; | |
6267 | |
6268 n = fts3HashCount(pTerms); | |
6269 pData = sqlite3_malloc(n*sizeof(TermData)); | |
6270 | |
6271 for(i = 0, e = fts3HashFirst(pTerms); e; i++, e = fts3HashNext(e)){ | |
6272 assert( i<n ); | |
6273 pData[i].pTerm = fts3HashKey(e); | |
6274 pData[i].nTerm = fts3HashKeysize(e); | |
6275 pData[i].pCollector = fts3HashData(e); | |
6276 } | |
6277 assert( i==n ); | |
6278 | |
6279 /* TODO(shess) Should we allow user-defined collation sequences, | |
6280 ** here? I think we only need that once we support prefix searches. | |
6281 */ | |
6282 if( n>1 ) qsort(pData, n, sizeof(*pData), termDataCmp); | |
6283 | |
6284 /* TODO(shess) Refactor so that we can write directly to the segment | |
6285 ** DataBuffer, as happens for segment merges. | |
6286 */ | |
6287 leafWriterInit(0, idx, &writer); | |
6288 dataBufferInit(&dl, 0); | |
6289 for(i=0; i<n; i++){ | |
6290 dataBufferReset(&dl); | |
6291 dlcAddDoclist(pData[i].pCollector, &dl); | |
6292 rc = leafWriterStep(v, &writer, | |
6293 pData[i].pTerm, pData[i].nTerm, dl.pData, dl.nData); | |
6294 if( rc!=SQLITE_OK ) goto err; | |
6295 } | |
6296 rc = leafWriterFinalize(v, &writer); | |
6297 | |
6298 err: | |
6299 dataBufferDestroy(&dl); | |
6300 sqlite3_free(pData); | |
6301 leafWriterDestroy(&writer); | |
6302 return rc; | |
6303 } | |
6304 | |
6305 /* If pendingTerms has data, free it. */ | |
6306 static int clearPendingTerms(fulltext_vtab *v){ | |
6307 if( v->nPendingData>=0 ){ | |
6308 fts3HashElem *e; | |
6309 for(e=fts3HashFirst(&v->pendingTerms); e; e=fts3HashNext(e)){ | |
6310 dlcDelete(fts3HashData(e)); | |
6311 } | |
6312 fts3HashClear(&v->pendingTerms); | |
6313 v->nPendingData = -1; | |
6314 } | |
6315 return SQLITE_OK; | |
6316 } | |
6317 | |
6318 /* If pendingTerms has data, flush it to a level-zero segment, and | |
6319 ** free it. | |
6320 */ | |
6321 static int flushPendingTerms(fulltext_vtab *v){ | |
6322 if( v->nPendingData>=0 ){ | |
6323 int rc = writeZeroSegment(v, &v->pendingTerms); | |
6324 if( rc==SQLITE_OK ) clearPendingTerms(v); | |
6325 return rc; | |
6326 } | |
6327 return SQLITE_OK; | |
6328 } | |
6329 | |
6330 /* If pendingTerms is "too big", or docid is out of order, flush it. | |
6331 ** Regardless, be certain that pendingTerms is initialized for use. | |
6332 */ | |
6333 static int initPendingTerms(fulltext_vtab *v, sqlite_int64 iDocid){ | |
6334 /* TODO(shess) Explore whether partially flushing the buffer on | |
6335 ** forced-flush would provide better performance. I suspect that if | |
6336 ** we ordered the doclists by size and flushed the largest until the | |
6337 ** buffer was half empty, that would let the less frequent terms | |
6338 ** generate longer doclists. | |
6339 */ | |
6340 if( iDocid<=v->iPrevDocid || v->nPendingData>kPendingThreshold ){ | |
6341 int rc = flushPendingTerms(v); | |
6342 if( rc!=SQLITE_OK ) return rc; | |
6343 } | |
6344 if( v->nPendingData<0 ){ | |
6345 fts3HashInit(&v->pendingTerms, FTS3_HASH_STRING, 1); | |
6346 v->nPendingData = 0; | |
6347 } | |
6348 v->iPrevDocid = iDocid; | |
6349 return SQLITE_OK; | |
6350 } | |
6351 | |
6352 /* This function implements the xUpdate callback; it is the top-level entry | |
6353 * point for inserting, deleting or updating a row in a full-text table. */ | |
6354 static int fulltextUpdate(sqlite3_vtab *pVtab, int nArg, sqlite3_value **ppArg, | |
6355 sqlite_int64 *pRowid){ | |
6356 fulltext_vtab *v = (fulltext_vtab *) pVtab; | |
6357 int rc; | |
6358 | |
6359 FTSTRACE(("FTS3 Update %p\n", pVtab)); | |
6360 | |
6361 if( nArg<2 ){ | |
6362 rc = index_delete(v, sqlite3_value_int64(ppArg[0])); | |
6363 if( rc==SQLITE_OK ){ | |
6364 /* If we just deleted the last row in the table, clear out the | |
6365 ** index data. | |
6366 */ | |
6367 rc = content_exists(v); | |
6368 if( rc==SQLITE_ROW ){ | |
6369 rc = SQLITE_OK; | |
6370 }else if( rc==SQLITE_DONE ){ | |
6371 /* Clear the pending terms so we don't flush a useless level-0 | |
6372 ** segment when the transaction closes. | |
6373 */ | |
6374 rc = clearPendingTerms(v); | |
6375 if( rc==SQLITE_OK ){ | |
6376 rc = segdir_delete_all(v); | |
6377 } | |
6378 } | |
6379 } | |
6380 } else if( sqlite3_value_type(ppArg[0]) != SQLITE_NULL ){ | |
6381 /* An update: | |
6382 * ppArg[0] = old rowid | |
6383 * ppArg[1] = new rowid | |
6384 * ppArg[2..2+v->nColumn-1] = values | |
6385 * ppArg[2+v->nColumn] = value for magic column (we ignore this) | |
6386 * ppArg[2+v->nColumn+1] = value for docid | |
6387 */ | |
6388 sqlite_int64 rowid = sqlite3_value_int64(ppArg[0]); | |
6389 if( sqlite3_value_type(ppArg[1]) != SQLITE_INTEGER || | |
6390 sqlite3_value_int64(ppArg[1]) != rowid ){ | |
6391 rc = SQLITE_ERROR; /* we don't allow changing the rowid */ | |
6392 }else if( sqlite3_value_type(ppArg[2+v->nColumn+1]) != SQLITE_INTEGER || | |
6393 sqlite3_value_int64(ppArg[2+v->nColumn+1]) != rowid ){ | |
6394 rc = SQLITE_ERROR; /* we don't allow changing the docid */ | |
6395 }else{ | |
6396 assert( nArg==2+v->nColumn+2); | |
6397 rc = index_update(v, rowid, &ppArg[2]); | |
6398 } | |
6399 } else { | |
6400 /* An insert: | |
6401 * ppArg[1] = requested rowid | |
6402 * ppArg[2..2+v->nColumn-1] = values | |
6403 * ppArg[2+v->nColumn] = value for magic column (we ignore this) | |
6404 * ppArg[2+v->nColumn+1] = value for docid | |
6405 */ | |
6406 sqlite3_value *pRequestDocid = ppArg[2+v->nColumn+1]; | |
6407 assert( nArg==2+v->nColumn+2); | |
6408 if( SQLITE_NULL != sqlite3_value_type(pRequestDocid) && | |
6409 SQLITE_NULL != sqlite3_value_type(ppArg[1]) ){ | |
6410 /* TODO(shess) Consider allowing this to work if the values are | |
6411 ** identical. I'm inclined to discourage that usage, though, | |
6412 ** given that both rowid and docid are special columns. Better | |
6413 ** would be to define one or the other as the default winner, | |
6414 ** but should it be fts3-centric (docid) or SQLite-centric | |
6415 ** (rowid)? | |
6416 */ | |
6417 rc = SQLITE_ERROR; | |
6418 }else{ | |
6419 if( SQLITE_NULL == sqlite3_value_type(pRequestDocid) ){ | |
6420 pRequestDocid = ppArg[1]; | |
6421 } | |
6422 rc = index_insert(v, pRequestDocid, &ppArg[2], pRowid); | |
6423 } | |
6424 } | |
6425 | |
6426 return rc; | |
6427 } | |
6428 | |
6429 static int fulltextSync(sqlite3_vtab *pVtab){ | |
6430 FTSTRACE(("FTS3 xSync()\n")); | |
6431 return flushPendingTerms((fulltext_vtab *)pVtab); | |
6432 } | |
6433 | |
6434 static int fulltextBegin(sqlite3_vtab *pVtab){ | |
6435 fulltext_vtab *v = (fulltext_vtab *) pVtab; | |
6436 FTSTRACE(("FTS3 xBegin()\n")); | |
6437 | |
6438 /* Any buffered updates should have been cleared by the previous | |
6439 ** transaction. | |
6440 */ | |
6441 assert( v->nPendingData<0 ); | |
6442 return clearPendingTerms(v); | |
6443 } | |
6444 | |
6445 static int fulltextCommit(sqlite3_vtab *pVtab){ | |
6446 fulltext_vtab *v = (fulltext_vtab *) pVtab; | |
6447 FTSTRACE(("FTS3 xCommit()\n")); | |
6448 | |
6449 /* Buffered updates should have been cleared by fulltextSync(). */ | |
6450 assert( v->nPendingData<0 ); | |
6451 return clearPendingTerms(v); | |
6452 } | |
6453 | |
6454 static int fulltextRollback(sqlite3_vtab *pVtab){ | |
6455 FTSTRACE(("FTS3 xRollback()\n")); | |
6456 return clearPendingTerms((fulltext_vtab *)pVtab); | |
6457 } | |
6458 | |
6459 /* | |
6460 ** Implementation of the snippet() function for FTS3 | |
6461 */ | |
6462 static void snippetFunc( | |
6463 sqlite3_context *pContext, | |
6464 int argc, | |
6465 sqlite3_value **argv | |
6466 ){ | |
6467 fulltext_cursor *pCursor; | |
6468 if( argc<1 ) return; | |
6469 if( sqlite3_value_type(argv[0])!=SQLITE_BLOB || | |
6470 sqlite3_value_bytes(argv[0])!=sizeof(pCursor) ){ | |
6471 sqlite3_result_error(pContext, "illegal first argument to html_snippet",-1); | |
6472 }else{ | |
6473 const char *zStart = "<b>"; | |
6474 const char *zEnd = "</b>"; | |
6475 const char *zEllipsis = "<b>...</b>"; | |
6476 memcpy(&pCursor, sqlite3_value_blob(argv[0]), sizeof(pCursor)); | |
6477 if( argc>=2 ){ | |
6478 zStart = (const char*)sqlite3_value_text(argv[1]); | |
6479 if( argc>=3 ){ | |
6480 zEnd = (const char*)sqlite3_value_text(argv[2]); | |
6481 if( argc>=4 ){ | |
6482 zEllipsis = (const char*)sqlite3_value_text(argv[3]); | |
6483 } | |
6484 } | |
6485 } | |
6486 snippetAllOffsets(pCursor); | |
6487 snippetText(pCursor, zStart, zEnd, zEllipsis); | |
6488 sqlite3_result_text(pContext, pCursor->snippet.zSnippet, | |
6489 pCursor->snippet.nSnippet, SQLITE_STATIC); | |
6490 } | |
6491 } | |
6492 | |
6493 /* | |
6494 ** Implementation of the offsets() function for FTS3 | |
6495 */ | |
6496 static void snippetOffsetsFunc( | |
6497 sqlite3_context *pContext, | |
6498 int argc, | |
6499 sqlite3_value **argv | |
6500 ){ | |
6501 fulltext_cursor *pCursor; | |
6502 if( argc<1 ) return; | |
6503 if( sqlite3_value_type(argv[0])!=SQLITE_BLOB || | |
6504 sqlite3_value_bytes(argv[0])!=sizeof(pCursor) ){ | |
6505 sqlite3_result_error(pContext, "illegal first argument to offsets",-1); | |
6506 }else{ | |
6507 memcpy(&pCursor, sqlite3_value_blob(argv[0]), sizeof(pCursor)); | |
6508 snippetAllOffsets(pCursor); | |
6509 snippetOffsetText(&pCursor->snippet); | |
6510 sqlite3_result_text(pContext, | |
6511 pCursor->snippet.zOffset, pCursor->snippet.nOffset, | |
6512 SQLITE_STATIC); | |
6513 } | |
6514 } | |
6515 | |
6516 /* OptLeavesReader is nearly identical to LeavesReader, except that | |
6517 ** where LeavesReader is geared towards the merging of complete | |
6518 ** segment levels (with exactly MERGE_COUNT segments), OptLeavesReader | |
6519 ** is geared towards implementation of the optimize() function, and | |
6520 ** can merge all segments simultaneously. This version may be | |
6521 ** somewhat less efficient than LeavesReader because it merges into an | |
6522 ** accumulator rather than doing an N-way merge, but since segment | |
6523 ** size grows exponentially (so segment count logrithmically) this is | |
6524 ** probably not an immediate problem. | |
6525 */ | |
6526 /* TODO(shess): Prove that assertion, or extend the merge code to | |
6527 ** merge tree fashion (like the prefix-searching code does). | |
6528 */ | |
6529 /* TODO(shess): OptLeavesReader and LeavesReader could probably be | |
6530 ** merged with little or no loss of performance for LeavesReader. The | |
6531 ** merged code would need to handle >MERGE_COUNT segments, and would | |
6532 ** also need to be able to optionally optimize away deletes. | |
6533 */ | |
6534 typedef struct OptLeavesReader { | |
6535 /* Segment number, to order readers by age. */ | |
6536 int segment; | |
6537 LeavesReader reader; | |
6538 } OptLeavesReader; | |
6539 | |
6540 static int optLeavesReaderAtEnd(OptLeavesReader *pReader){ | |
6541 return leavesReaderAtEnd(&pReader->reader); | |
6542 } | |
6543 static int optLeavesReaderTermBytes(OptLeavesReader *pReader){ | |
6544 return leavesReaderTermBytes(&pReader->reader); | |
6545 } | |
6546 static const char *optLeavesReaderData(OptLeavesReader *pReader){ | |
6547 return leavesReaderData(&pReader->reader); | |
6548 } | |
6549 static int optLeavesReaderDataBytes(OptLeavesReader *pReader){ | |
6550 return leavesReaderDataBytes(&pReader->reader); | |
6551 } | |
6552 static const char *optLeavesReaderTerm(OptLeavesReader *pReader){ | |
6553 return leavesReaderTerm(&pReader->reader); | |
6554 } | |
6555 static int optLeavesReaderStep(fulltext_vtab *v, OptLeavesReader *pReader){ | |
6556 return leavesReaderStep(v, &pReader->reader); | |
6557 } | |
6558 static int optLeavesReaderTermCmp(OptLeavesReader *lr1, OptLeavesReader *lr2){ | |
6559 return leavesReaderTermCmp(&lr1->reader, &lr2->reader); | |
6560 } | |
6561 /* Order by term ascending, segment ascending (oldest to newest), with | |
6562 ** exhausted readers to the end. | |
6563 */ | |
6564 static int optLeavesReaderCmp(OptLeavesReader *lr1, OptLeavesReader *lr2){ | |
6565 int c = optLeavesReaderTermCmp(lr1, lr2); | |
6566 if( c!=0 ) return c; | |
6567 return lr1->segment-lr2->segment; | |
6568 } | |
6569 /* Bubble pLr[0] to appropriate place in pLr[1..nLr-1]. Assumes that | |
6570 ** pLr[1..nLr-1] is already sorted. | |
6571 */ | |
6572 static void optLeavesReaderReorder(OptLeavesReader *pLr, int nLr){ | |
6573 while( nLr>1 && optLeavesReaderCmp(pLr, pLr+1)>0 ){ | |
6574 OptLeavesReader tmp = pLr[0]; | |
6575 pLr[0] = pLr[1]; | |
6576 pLr[1] = tmp; | |
6577 nLr--; | |
6578 pLr++; | |
6579 } | |
6580 } | |
6581 | |
6582 /* optimize() helper function. Put the readers in order and iterate | |
6583 ** through them, merging doclists for matching terms into pWriter. | |
6584 ** Returns SQLITE_OK on success, or the SQLite error code which | |
6585 ** prevented success. | |
6586 */ | |
6587 static int optimizeInternal(fulltext_vtab *v, | |
6588 OptLeavesReader *readers, int nReaders, | |
6589 LeafWriter *pWriter){ | |
6590 int i, rc = SQLITE_OK; | |
6591 DataBuffer doclist, merged, tmp; | |
6592 const char *pData; | |
6593 | |
6594 /* Order the readers. */ | |
6595 i = nReaders; | |
6596 while( i-- > 0 ){ | |
6597 optLeavesReaderReorder(&readers[i], nReaders-i); | |
6598 } | |
6599 | |
6600 dataBufferInit(&doclist, LEAF_MAX); | |
6601 dataBufferInit(&merged, LEAF_MAX); | |
6602 | |
6603 /* Exhausted readers bubble to the end, so when the first reader is | |
6604 ** at eof, all are at eof. | |
6605 */ | |
6606 while( !optLeavesReaderAtEnd(&readers[0]) ){ | |
6607 | |
6608 /* Figure out how many readers share the next term. */ | |
6609 for(i=1; i<nReaders && !optLeavesReaderAtEnd(&readers[i]); i++){ | |
6610 if( 0!=optLeavesReaderTermCmp(&readers[0], &readers[i]) ) break; | |
6611 } | |
6612 | |
6613 pData = optLeavesReaderData(&readers[0]); | |
6614 if( pData==NULL ){ | |
6615 rc = SQLITE_CORRUPT_BKPT; | |
6616 break; | |
6617 } | |
6618 | |
6619 /* Special-case for no merge. */ | |
6620 if( i==1 ){ | |
6621 /* Trim deletions from the doclist. */ | |
6622 dataBufferReset(&merged); | |
6623 rc = docListTrim(DL_DEFAULT, pData, | |
6624 optLeavesReaderDataBytes(&readers[0]), | |
6625 -1, DL_DEFAULT, &merged); | |
6626 if( rc!=SQLITE_OK ) break; | |
6627 }else{ | |
6628 DLReader dlReaders[MERGE_COUNT]; | |
6629 int iReader, nReaders; | |
6630 | |
6631 /* Prime the pipeline with the first reader's doclist. After | |
6632 ** one pass index 0 will reference the accumulated doclist. | |
6633 */ | |
6634 rc = dlrInit(&dlReaders[0], DL_DEFAULT, | |
6635 pData, | |
6636 optLeavesReaderDataBytes(&readers[0])); | |
6637 if( rc!=SQLITE_OK ) break; | |
6638 iReader = 1; | |
6639 | |
6640 assert( iReader<i ); /* Must execute the loop at least once. */ | |
6641 while( iReader<i ){ | |
6642 /* Merge 16 inputs per pass. */ | |
6643 for( nReaders=1; iReader<i && nReaders<MERGE_COUNT; | |
6644 iReader++, nReaders++ ){ | |
6645 pData = optLeavesReaderData(&readers[iReader]); | |
6646 if( pData==NULL ){ | |
6647 rc = SQLITE_CORRUPT_BKPT; | |
6648 break; | |
6649 } | |
6650 rc = dlrInit(&dlReaders[nReaders], DL_DEFAULT, pData, | |
6651 optLeavesReaderDataBytes(&readers[iReader])); | |
6652 if( rc!=SQLITE_OK ) break; | |
6653 } | |
6654 | |
6655 /* Merge doclists and swap result into accumulator. */ | |
6656 if( rc==SQLITE_OK ){ | |
6657 dataBufferReset(&merged); | |
6658 rc = docListMerge(&merged, dlReaders, nReaders); | |
6659 tmp = merged; | |
6660 merged = doclist; | |
6661 doclist = tmp; | |
6662 } | |
6663 | |
6664 while( nReaders-- > 0 ){ | |
6665 dlrDestroy(&dlReaders[nReaders]); | |
6666 } | |
6667 | |
6668 if( rc!=SQLITE_OK ) goto err; | |
6669 | |
6670 /* Accumulated doclist to reader 0 for next pass. */ | |
6671 rc = dlrInit(&dlReaders[0], DL_DEFAULT, doclist.pData, doclist.nData); | |
6672 if( rc!=SQLITE_OK ) goto err; | |
6673 } | |
6674 | |
6675 /* Destroy reader that was left in the pipeline. */ | |
6676 dlrDestroy(&dlReaders[0]); | |
6677 | |
6678 /* Trim deletions from the doclist. */ | |
6679 dataBufferReset(&merged); | |
6680 rc = docListTrim(DL_DEFAULT, doclist.pData, doclist.nData, | |
6681 -1, DL_DEFAULT, &merged); | |
6682 if( rc!=SQLITE_OK ) goto err; | |
6683 } | |
6684 | |
6685 /* Only pass doclists with hits (skip if all hits deleted). */ | |
6686 if( merged.nData>0 ){ | |
6687 rc = leafWriterStep(v, pWriter, | |
6688 optLeavesReaderTerm(&readers[0]), | |
6689 optLeavesReaderTermBytes(&readers[0]), | |
6690 merged.pData, merged.nData); | |
6691 if( rc!=SQLITE_OK ) goto err; | |
6692 } | |
6693 | |
6694 /* Step merged readers to next term and reorder. */ | |
6695 while( i-- > 0 ){ | |
6696 rc = optLeavesReaderStep(v, &readers[i]); | |
6697 if( rc!=SQLITE_OK ) goto err; | |
6698 | |
6699 optLeavesReaderReorder(&readers[i], nReaders-i); | |
6700 } | |
6701 } | |
6702 | |
6703 err: | |
6704 dataBufferDestroy(&doclist); | |
6705 dataBufferDestroy(&merged); | |
6706 return rc; | |
6707 } | |
6708 | |
6709 /* Implement optimize() function for FTS3. optimize(t) merges all | |
6710 ** segments in the fts index into a single segment. 't' is the magic | |
6711 ** table-named column. | |
6712 */ | |
6713 static void optimizeFunc(sqlite3_context *pContext, | |
6714 int argc, sqlite3_value **argv){ | |
6715 fulltext_cursor *pCursor; | |
6716 if( argc>1 ){ | |
6717 sqlite3_result_error(pContext, "excess arguments to optimize()",-1); | |
6718 }else if( sqlite3_value_type(argv[0])!=SQLITE_BLOB || | |
6719 sqlite3_value_bytes(argv[0])!=sizeof(pCursor) ){ | |
6720 sqlite3_result_error(pContext, "illegal first argument to optimize",-1); | |
6721 }else{ | |
6722 fulltext_vtab *v; | |
6723 int i, rc, iMaxLevel; | |
6724 OptLeavesReader *readers; | |
6725 int nReaders; | |
6726 LeafWriter writer; | |
6727 sqlite3_stmt *s; | |
6728 | |
6729 memcpy(&pCursor, sqlite3_value_blob(argv[0]), sizeof(pCursor)); | |
6730 v = cursor_vtab(pCursor); | |
6731 | |
6732 /* Flush any buffered updates before optimizing. */ | |
6733 rc = flushPendingTerms(v); | |
6734 if( rc!=SQLITE_OK ) goto err; | |
6735 | |
6736 rc = segdir_count(v, &nReaders, &iMaxLevel); | |
6737 if( rc!=SQLITE_OK ) goto err; | |
6738 if( nReaders==0 || nReaders==1 ){ | |
6739 sqlite3_result_text(pContext, "Index already optimal", -1, | |
6740 SQLITE_STATIC); | |
6741 return; | |
6742 } | |
6743 | |
6744 rc = sql_get_statement(v, SEGDIR_SELECT_ALL_STMT, &s); | |
6745 if( rc!=SQLITE_OK ) goto err; | |
6746 | |
6747 readers = sqlite3_malloc(nReaders*sizeof(readers[0])); | |
6748 if( readers==NULL ) goto err; | |
6749 | |
6750 /* Note that there will already be a segment at this position | |
6751 ** until we call segdir_delete() on iMaxLevel. | |
6752 */ | |
6753 leafWriterInit(iMaxLevel, 0, &writer); | |
6754 | |
6755 i = 0; | |
6756 while( (rc = sqlite3_step(s))==SQLITE_ROW ){ | |
6757 sqlite_int64 iStart = sqlite3_column_int64(s, 0); | |
6758 sqlite_int64 iEnd = sqlite3_column_int64(s, 1); | |
6759 const char *pRootData = sqlite3_column_blob(s, 2); | |
6760 int nRootData = sqlite3_column_bytes(s, 2); | |
6761 | |
6762 /* Corrupt if we get back different types than we stored. */ | |
6763 if( sqlite3_column_type(s, 0)!=SQLITE_INTEGER || | |
6764 sqlite3_column_type(s, 1)!=SQLITE_INTEGER || | |
6765 sqlite3_column_type(s, 2)!=SQLITE_BLOB ){ | |
6766 rc = SQLITE_CORRUPT_BKPT; | |
6767 break; | |
6768 } | |
6769 | |
6770 assert( i<nReaders ); | |
6771 rc = leavesReaderInit(v, -1, iStart, iEnd, pRootData, nRootData, | |
6772 &readers[i].reader); | |
6773 if( rc!=SQLITE_OK ) break; | |
6774 | |
6775 readers[i].segment = i; | |
6776 i++; | |
6777 } | |
6778 | |
6779 /* If we managed to successfully read them all, optimize them. */ | |
6780 if( rc==SQLITE_DONE ){ | |
6781 assert( i==nReaders ); | |
6782 rc = optimizeInternal(v, readers, nReaders, &writer); | |
6783 }else{ | |
6784 sqlite3_reset(s); /* So we don't leave a lock. */ | |
6785 } | |
6786 | |
6787 while( i-- > 0 ){ | |
6788 leavesReaderDestroy(&readers[i].reader); | |
6789 } | |
6790 sqlite3_free(readers); | |
6791 | |
6792 /* If we've successfully gotten to here, delete the old segments | |
6793 ** and flush the interior structure of the new segment. | |
6794 */ | |
6795 if( rc==SQLITE_OK ){ | |
6796 for( i=0; i<=iMaxLevel; i++ ){ | |
6797 rc = segdir_delete(v, i); | |
6798 if( rc!=SQLITE_OK ) break; | |
6799 } | |
6800 | |
6801 if( rc==SQLITE_OK ) rc = leafWriterFinalize(v, &writer); | |
6802 } | |
6803 | |
6804 leafWriterDestroy(&writer); | |
6805 | |
6806 if( rc!=SQLITE_OK ) goto err; | |
6807 | |
6808 sqlite3_result_text(pContext, "Index optimized", -1, SQLITE_STATIC); | |
6809 return; | |
6810 | |
6811 /* TODO(shess): Error-handling needs to be improved along the | |
6812 ** lines of the dump_ functions. | |
6813 */ | |
6814 err: | |
6815 { | |
6816 char buf[512]; | |
6817 sqlite3_snprintf(sizeof(buf), buf, "Error in optimize: %s", | |
6818 sqlite3_errmsg(sqlite3_context_db_handle(pContext))); | |
6819 sqlite3_result_error(pContext, buf, -1); | |
6820 } | |
6821 } | |
6822 } | |
6823 | |
6824 #ifdef SQLITE_TEST | |
6825 /* Generate an error of the form "<prefix>: <msg>". If msg is NULL, | |
6826 ** pull the error from the context's db handle. | |
6827 */ | |
6828 static void generateError(sqlite3_context *pContext, | |
6829 const char *prefix, const char *msg){ | |
6830 char buf[512]; | |
6831 if( msg==NULL ) msg = sqlite3_errmsg(sqlite3_context_db_handle(pContext)); | |
6832 sqlite3_snprintf(sizeof(buf), buf, "%s: %s", prefix, msg); | |
6833 sqlite3_result_error(pContext, buf, -1); | |
6834 } | |
6835 | |
6836 /* Helper function to collect the set of terms in the segment into | |
6837 ** pTerms. The segment is defined by the leaf nodes between | |
6838 ** iStartBlockid and iEndBlockid, inclusive, or by the contents of | |
6839 ** pRootData if iStartBlockid is 0 (in which case the entire segment | |
6840 ** fit in a leaf). | |
6841 */ | |
6842 static int collectSegmentTerms(fulltext_vtab *v, sqlite3_stmt *s, | |
6843 fts3Hash *pTerms){ | |
6844 const sqlite_int64 iStartBlockid = sqlite3_column_int64(s, 0); | |
6845 const sqlite_int64 iEndBlockid = sqlite3_column_int64(s, 1); | |
6846 const char *pRootData = sqlite3_column_blob(s, 2); | |
6847 const int nRootData = sqlite3_column_bytes(s, 2); | |
6848 int rc; | |
6849 LeavesReader reader; | |
6850 | |
6851 /* Corrupt if we get back different types than we stored. */ | |
6852 if( sqlite3_column_type(s, 0)!=SQLITE_INTEGER || | |
6853 sqlite3_column_type(s, 1)!=SQLITE_INTEGER || | |
6854 sqlite3_column_type(s, 2)!=SQLITE_BLOB ){ | |
6855 return SQLITE_CORRUPT_BKPT; | |
6856 } | |
6857 | |
6858 rc = leavesReaderInit(v, 0, iStartBlockid, iEndBlockid, | |
6859 pRootData, nRootData, &reader); | |
6860 if( rc!=SQLITE_OK ) return rc; | |
6861 | |
6862 while( rc==SQLITE_OK && !leavesReaderAtEnd(&reader) ){ | |
6863 const char *pTerm = leavesReaderTerm(&reader); | |
6864 const int nTerm = leavesReaderTermBytes(&reader); | |
6865 void *oldValue = sqlite3Fts3HashFind(pTerms, pTerm, nTerm); | |
6866 void *newValue = (void *)((char *)oldValue+1); | |
6867 | |
6868 /* From the comment before sqlite3Fts3HashInsert in fts3_hash.c, | |
6869 ** the data value passed is returned in case of malloc failure. | |
6870 */ | |
6871 if( newValue==sqlite3Fts3HashInsert(pTerms, pTerm, nTerm, newValue) ){ | |
6872 rc = SQLITE_NOMEM; | |
6873 }else{ | |
6874 rc = leavesReaderStep(v, &reader); | |
6875 } | |
6876 } | |
6877 | |
6878 leavesReaderDestroy(&reader); | |
6879 return rc; | |
6880 } | |
6881 | |
6882 /* Helper function to build the result string for dump_terms(). */ | |
6883 static int generateTermsResult(sqlite3_context *pContext, fts3Hash *pTerms){ | |
6884 int iTerm, nTerms, nResultBytes, iByte; | |
6885 char *result; | |
6886 TermData *pData; | |
6887 fts3HashElem *e; | |
6888 | |
6889 /* Iterate pTerms to generate an array of terms in pData for | |
6890 ** sorting. | |
6891 */ | |
6892 nTerms = fts3HashCount(pTerms); | |
6893 assert( nTerms>0 ); | |
6894 pData = sqlite3_malloc(nTerms*sizeof(TermData)); | |
6895 if( pData==NULL ) return SQLITE_NOMEM; | |
6896 | |
6897 nResultBytes = 0; | |
6898 for(iTerm = 0, e = fts3HashFirst(pTerms); e; iTerm++, e = fts3HashNext(e)){ | |
6899 nResultBytes += fts3HashKeysize(e)+1; /* Term plus trailing space */ | |
6900 assert( iTerm<nTerms ); | |
6901 pData[iTerm].pTerm = fts3HashKey(e); | |
6902 pData[iTerm].nTerm = fts3HashKeysize(e); | |
6903 pData[iTerm].pCollector = fts3HashData(e); /* unused */ | |
6904 } | |
6905 assert( iTerm==nTerms ); | |
6906 | |
6907 assert( nResultBytes>0 ); /* nTerms>0, nResultsBytes must be, too. */ | |
6908 result = sqlite3_malloc(nResultBytes); | |
6909 if( result==NULL ){ | |
6910 sqlite3_free(pData); | |
6911 return SQLITE_NOMEM; | |
6912 } | |
6913 | |
6914 if( nTerms>1 ) qsort(pData, nTerms, sizeof(*pData), termDataCmp); | |
6915 | |
6916 /* Read the terms in order to build the result. */ | |
6917 iByte = 0; | |
6918 for(iTerm=0; iTerm<nTerms; ++iTerm){ | |
6919 memcpy(result+iByte, pData[iTerm].pTerm, pData[iTerm].nTerm); | |
6920 iByte += pData[iTerm].nTerm; | |
6921 result[iByte++] = ' '; | |
6922 } | |
6923 assert( iByte==nResultBytes ); | |
6924 assert( result[nResultBytes-1]==' ' ); | |
6925 result[nResultBytes-1] = '\0'; | |
6926 | |
6927 /* Passes away ownership of result. */ | |
6928 sqlite3_result_text(pContext, result, nResultBytes-1, sqlite3_free); | |
6929 sqlite3_free(pData); | |
6930 return SQLITE_OK; | |
6931 } | |
6932 | |
6933 /* Implements dump_terms() for use in inspecting the fts3 index from | |
6934 ** tests. TEXT result containing the ordered list of terms joined by | |
6935 ** spaces. dump_terms(t, level, idx) dumps the terms for the segment | |
6936 ** specified by level, idx (in %_segdir), while dump_terms(t) dumps | |
6937 ** all terms in the index. In both cases t is the fts table's magic | |
6938 ** table-named column. | |
6939 */ | |
6940 static void dumpTermsFunc( | |
6941 sqlite3_context *pContext, | |
6942 int argc, sqlite3_value **argv | |
6943 ){ | |
6944 fulltext_cursor *pCursor; | |
6945 if( argc!=3 && argc!=1 ){ | |
6946 generateError(pContext, "dump_terms", "incorrect arguments"); | |
6947 }else if( sqlite3_value_type(argv[0])!=SQLITE_BLOB || | |
6948 sqlite3_value_bytes(argv[0])!=sizeof(pCursor) ){ | |
6949 generateError(pContext, "dump_terms", "illegal first argument"); | |
6950 }else{ | |
6951 fulltext_vtab *v; | |
6952 fts3Hash terms; | |
6953 sqlite3_stmt *s = NULL; | |
6954 int rc; | |
6955 | |
6956 memcpy(&pCursor, sqlite3_value_blob(argv[0]), sizeof(pCursor)); | |
6957 v = cursor_vtab(pCursor); | |
6958 | |
6959 /* If passed only the cursor column, get all segments. Otherwise | |
6960 ** get the segment described by the following two arguments. | |
6961 */ | |
6962 if( argc==1 ){ | |
6963 rc = sql_get_statement(v, SEGDIR_SELECT_ALL_STMT, &s); | |
6964 }else{ | |
6965 rc = sql_get_statement(v, SEGDIR_SELECT_SEGMENT_STMT, &s); | |
6966 if( rc==SQLITE_OK ){ | |
6967 rc = sqlite3_bind_int(s, 1, sqlite3_value_int(argv[1])); | |
6968 if( rc==SQLITE_OK ){ | |
6969 rc = sqlite3_bind_int(s, 2, sqlite3_value_int(argv[2])); | |
6970 } | |
6971 } | |
6972 } | |
6973 | |
6974 if( rc!=SQLITE_OK ){ | |
6975 generateError(pContext, "dump_terms", NULL); | |
6976 return; | |
6977 } | |
6978 | |
6979 /* Collect the terms for each segment. */ | |
6980 sqlite3Fts3HashInit(&terms, FTS3_HASH_STRING, 1); | |
6981 while( (rc = sqlite3_step(s))==SQLITE_ROW ){ | |
6982 rc = collectSegmentTerms(v, s, &terms); | |
6983 if( rc!=SQLITE_OK ) break; | |
6984 } | |
6985 | |
6986 if( rc!=SQLITE_DONE ){ | |
6987 sqlite3_reset(s); | |
6988 generateError(pContext, "dump_terms", NULL); | |
6989 }else{ | |
6990 const int nTerms = fts3HashCount(&terms); | |
6991 if( nTerms>0 ){ | |
6992 rc = generateTermsResult(pContext, &terms); | |
6993 if( rc==SQLITE_NOMEM ){ | |
6994 generateError(pContext, "dump_terms", "out of memory"); | |
6995 }else{ | |
6996 assert( rc==SQLITE_OK ); | |
6997 } | |
6998 }else if( argc==3 ){ | |
6999 /* The specific segment asked for could not be found. */ | |
7000 generateError(pContext, "dump_terms", "segment not found"); | |
7001 }else{ | |
7002 /* No segments found. */ | |
7003 /* TODO(shess): It should be impossible to reach this. This | |
7004 ** case can only happen for an empty table, in which case | |
7005 ** SQLite has no rows to call this function on. | |
7006 */ | |
7007 sqlite3_result_null(pContext); | |
7008 } | |
7009 } | |
7010 sqlite3Fts3HashClear(&terms); | |
7011 } | |
7012 } | |
7013 | |
7014 /* Expand the DL_DEFAULT doclist in pData into a text result in | |
7015 ** pContext. | |
7016 */ | |
7017 static void createDoclistResult(sqlite3_context *pContext, | |
7018 const char *pData, int nData){ | |
7019 DataBuffer dump; | |
7020 DLReader dlReader; | |
7021 int rc; | |
7022 | |
7023 assert( pData!=NULL && nData>0 ); | |
7024 | |
7025 rc = dlrInit(&dlReader, DL_DEFAULT, pData, nData); | |
7026 if( rc!=SQLITE_OK ) return rc; | |
7027 dataBufferInit(&dump, 0); | |
7028 for( ; rc==SQLITE_OK && !dlrAtEnd(&dlReader); rc = dlrStep(&dlReader) ){ | |
7029 char buf[256]; | |
7030 PLReader plReader; | |
7031 | |
7032 rc = plrInit(&plReader, &dlReader); | |
7033 if( rc!=SQLITE_OK ) break; | |
7034 if( DL_DEFAULT==DL_DOCIDS || plrAtEnd(&plReader) ){ | |
7035 sqlite3_snprintf(sizeof(buf), buf, "[%lld] ", dlrDocid(&dlReader)); | |
7036 dataBufferAppend(&dump, buf, strlen(buf)); | |
7037 }else{ | |
7038 int iColumn = plrColumn(&plReader); | |
7039 | |
7040 sqlite3_snprintf(sizeof(buf), buf, "[%lld %d[", | |
7041 dlrDocid(&dlReader), iColumn); | |
7042 dataBufferAppend(&dump, buf, strlen(buf)); | |
7043 | |
7044 for( ; !plrAtEnd(&plReader); rc = plrStep(&plReader) ){ | |
7045 if( rc!=SQLITE_OK ) break; | |
7046 if( plrColumn(&plReader)!=iColumn ){ | |
7047 iColumn = plrColumn(&plReader); | |
7048 sqlite3_snprintf(sizeof(buf), buf, "] %d[", iColumn); | |
7049 assert( dump.nData>0 ); | |
7050 dump.nData--; /* Overwrite trailing space. */ | |
7051 assert( dump.pData[dump.nData]==' '); | |
7052 dataBufferAppend(&dump, buf, strlen(buf)); | |
7053 } | |
7054 if( DL_DEFAULT==DL_POSITIONS_OFFSETS ){ | |
7055 sqlite3_snprintf(sizeof(buf), buf, "%d,%d,%d ", | |
7056 plrPosition(&plReader), | |
7057 plrStartOffset(&plReader), plrEndOffset(&plReader)); | |
7058 }else if( DL_DEFAULT==DL_POSITIONS ){ | |
7059 sqlite3_snprintf(sizeof(buf), buf, "%d ", plrPosition(&plReader)); | |
7060 }else{ | |
7061 assert( NULL=="Unhandled DL_DEFAULT value"); | |
7062 } | |
7063 dataBufferAppend(&dump, buf, strlen(buf)); | |
7064 } | |
7065 plrDestroy(&plReader); | |
7066 if( rc!= SQLITE_OK ) break; | |
7067 | |
7068 assert( dump.nData>0 ); | |
7069 dump.nData--; /* Overwrite trailing space. */ | |
7070 assert( dump.pData[dump.nData]==' '); | |
7071 dataBufferAppend(&dump, "]] ", 3); | |
7072 } | |
7073 } | |
7074 dlrDestroy(&dlReader); | |
7075 if( rc!=SQLITE_OK ){ | |
7076 dataBufferDestroy(&dump); | |
7077 return rc; | |
7078 } | |
7079 | |
7080 assert( dump.nData>0 ); | |
7081 dump.nData--; /* Overwrite trailing space. */ | |
7082 assert( dump.pData[dump.nData]==' '); | |
7083 dump.pData[dump.nData] = '\0'; | |
7084 assert( dump.nData>0 ); | |
7085 | |
7086 /* Passes ownership of dump's buffer to pContext. */ | |
7087 sqlite3_result_text(pContext, dump.pData, dump.nData, sqlite3_free); | |
7088 dump.pData = NULL; | |
7089 dump.nData = dump.nCapacity = 0; | |
7090 return SQLITE_OK; | |
7091 } | |
7092 | |
7093 /* Implements dump_doclist() for use in inspecting the fts3 index from | |
7094 ** tests. TEXT result containing a string representation of the | |
7095 ** doclist for the indicated term. dump_doclist(t, term, level, idx) | |
7096 ** dumps the doclist for term from the segment specified by level, idx | |
7097 ** (in %_segdir), while dump_doclist(t, term) dumps the logical | |
7098 ** doclist for the term across all segments. The per-segment doclist | |
7099 ** can contain deletions, while the full-index doclist will not | |
7100 ** (deletions are omitted). | |
7101 ** | |
7102 ** Result formats differ with the setting of DL_DEFAULTS. Examples: | |
7103 ** | |
7104 ** DL_DOCIDS: [1] [3] [7] | |
7105 ** DL_POSITIONS: [1 0[0 4] 1[17]] [3 1[5]] | |
7106 ** DL_POSITIONS_OFFSETS: [1 0[0,0,3 4,23,26] 1[17,102,105]] [3 1[5,20,23]] | |
7107 ** | |
7108 ** In each case the number after the outer '[' is the docid. In the | |
7109 ** latter two cases, the number before the inner '[' is the column | |
7110 ** associated with the values within. For DL_POSITIONS the numbers | |
7111 ** within are the positions, for DL_POSITIONS_OFFSETS they are the | |
7112 ** position, the start offset, and the end offset. | |
7113 */ | |
7114 static void dumpDoclistFunc( | |
7115 sqlite3_context *pContext, | |
7116 int argc, sqlite3_value **argv | |
7117 ){ | |
7118 fulltext_cursor *pCursor; | |
7119 if( argc!=2 && argc!=4 ){ | |
7120 generateError(pContext, "dump_doclist", "incorrect arguments"); | |
7121 }else if( sqlite3_value_type(argv[0])!=SQLITE_BLOB || | |
7122 sqlite3_value_bytes(argv[0])!=sizeof(pCursor) ){ | |
7123 generateError(pContext, "dump_doclist", "illegal first argument"); | |
7124 }else if( sqlite3_value_text(argv[1])==NULL || | |
7125 sqlite3_value_text(argv[1])[0]=='\0' ){ | |
7126 generateError(pContext, "dump_doclist", "empty second argument"); | |
7127 }else{ | |
7128 const char *pTerm = (const char *)sqlite3_value_text(argv[1]); | |
7129 const int nTerm = strlen(pTerm); | |
7130 fulltext_vtab *v; | |
7131 int rc; | |
7132 DataBuffer doclist; | |
7133 | |
7134 memcpy(&pCursor, sqlite3_value_blob(argv[0]), sizeof(pCursor)); | |
7135 v = cursor_vtab(pCursor); | |
7136 | |
7137 dataBufferInit(&doclist, 0); | |
7138 | |
7139 /* termSelect() yields the same logical doclist that queries are | |
7140 ** run against. | |
7141 */ | |
7142 if( argc==2 ){ | |
7143 rc = termSelect(v, v->nColumn, pTerm, nTerm, 0, DL_DEFAULT, &doclist); | |
7144 }else{ | |
7145 sqlite3_stmt *s = NULL; | |
7146 | |
7147 /* Get our specific segment's information. */ | |
7148 rc = sql_get_statement(v, SEGDIR_SELECT_SEGMENT_STMT, &s); | |
7149 if( rc==SQLITE_OK ){ | |
7150 rc = sqlite3_bind_int(s, 1, sqlite3_value_int(argv[2])); | |
7151 if( rc==SQLITE_OK ){ | |
7152 rc = sqlite3_bind_int(s, 2, sqlite3_value_int(argv[3])); | |
7153 } | |
7154 } | |
7155 | |
7156 if( rc==SQLITE_OK ){ | |
7157 rc = sqlite3_step(s); | |
7158 | |
7159 if( rc==SQLITE_DONE ){ | |
7160 dataBufferDestroy(&doclist); | |
7161 generateError(pContext, "dump_doclist", "segment not found"); | |
7162 return; | |
7163 } | |
7164 | |
7165 /* Found a segment, load it into doclist. */ | |
7166 if( rc==SQLITE_ROW ){ | |
7167 const sqlite_int64 iLeavesEnd = sqlite3_column_int64(s, 1); | |
7168 const char *pData = sqlite3_column_blob(s, 2); | |
7169 const int nData = sqlite3_column_bytes(s, 2); | |
7170 | |
7171 /* loadSegment() is used by termSelect() to load each | |
7172 ** segment's data. | |
7173 */ | |
7174 rc = loadSegment(v, pData, nData, iLeavesEnd, pTerm, nTerm, 0, | |
7175 &doclist); | |
7176 if( rc==SQLITE_OK ){ | |
7177 rc = sqlite3_step(s); | |
7178 | |
7179 /* Should not have more than one matching segment. */ | |
7180 if( rc!=SQLITE_DONE ){ | |
7181 sqlite3_reset(s); | |
7182 dataBufferDestroy(&doclist); | |
7183 generateError(pContext, "dump_doclist", "invalid segdir"); | |
7184 return; | |
7185 } | |
7186 rc = SQLITE_OK; | |
7187 } | |
7188 } | |
7189 } | |
7190 | |
7191 sqlite3_reset(s); | |
7192 } | |
7193 | |
7194 if( rc==SQLITE_OK ){ | |
7195 if( doclist.nData>0 ){ | |
7196 createDoclistResult(pContext, doclist.pData, doclist.nData); | |
7197 }else{ | |
7198 /* TODO(shess): This can happen if the term is not present, or | |
7199 ** if all instances of the term have been deleted and this is | |
7200 ** an all-index dump. It may be interesting to distinguish | |
7201 ** these cases. | |
7202 */ | |
7203 sqlite3_result_text(pContext, "", 0, SQLITE_STATIC); | |
7204 } | |
7205 }else if( rc==SQLITE_NOMEM ){ | |
7206 /* Handle out-of-memory cases specially because if they are | |
7207 ** generated in fts3 code they may not be reflected in the db | |
7208 ** handle. | |
7209 */ | |
7210 /* TODO(shess): Handle this more comprehensively. | |
7211 ** sqlite3ErrStr() has what I need, but is internal. | |
7212 */ | |
7213 generateError(pContext, "dump_doclist", "out of memory"); | |
7214 }else{ | |
7215 generateError(pContext, "dump_doclist", NULL); | |
7216 } | |
7217 | |
7218 dataBufferDestroy(&doclist); | |
7219 } | |
7220 } | |
7221 #endif | |
7222 | 3471 |
7223 /* | 3472 /* |
7224 ** This routine implements the xFindFunction method for the FTS3 | 3473 ** This routine implements the xFindFunction method for the FTS3 |
7225 ** virtual table. | 3474 ** virtual table. |
7226 */ | 3475 */ |
7227 static int fulltextFindFunction( | 3476 static int fts3FindFunctionMethod( |
7228 sqlite3_vtab *pVtab, | 3477 sqlite3_vtab *pVtab, /* Virtual table handle */ |
7229 int nArg, | 3478 int nArg, /* Number of SQL function arguments */ |
7230 const char *zName, | 3479 const char *zName, /* Name of SQL function */ |
7231 void (**pxFunc)(sqlite3_context*,int,sqlite3_value**), | 3480 void (**pxFunc)(sqlite3_context*,int,sqlite3_value**), /* OUT: Result */ |
7232 void **ppArg | 3481 void **ppArg /* Unused */ |
7233 ){ | 3482 ){ |
7234 if( strcmp(zName,"snippet")==0 ){ | 3483 struct Overloaded { |
7235 *pxFunc = snippetFunc; | 3484 const char *zName; |
7236 return 1; | 3485 void (*xFunc)(sqlite3_context*,int,sqlite3_value**); |
7237 }else if( strcmp(zName,"offsets")==0 ){ | 3486 } aOverload[] = { |
7238 *pxFunc = snippetOffsetsFunc; | 3487 { "snippet", fts3SnippetFunc }, |
7239 return 1; | 3488 { "offsets", fts3OffsetsFunc }, |
7240 }else if( strcmp(zName,"optimize")==0 ){ | 3489 { "optimize", fts3OptimizeFunc }, |
7241 *pxFunc = optimizeFunc; | 3490 { "matchinfo", fts3MatchinfoFunc }, |
7242 return 1; | 3491 }; |
7243 #ifdef SQLITE_TEST | 3492 int i; /* Iterator variable */ |
7244 /* NOTE(shess): These functions are present only for testing | 3493 |
7245 ** purposes. No particular effort is made to optimize their | 3494 UNUSED_PARAMETER(pVtab); |
7246 ** execution or how they build their results. | 3495 UNUSED_PARAMETER(nArg); |
7247 */ | 3496 UNUSED_PARAMETER(ppArg); |
7248 }else if( strcmp(zName,"dump_terms")==0 ){ | 3497 |
7249 /* fprintf(stderr, "Found dump_terms\n"); */ | 3498 for(i=0; i<SizeofArray(aOverload); i++){ |
7250 *pxFunc = dumpTermsFunc; | 3499 if( strcmp(zName, aOverload[i].zName)==0 ){ |
7251 return 1; | 3500 *pxFunc = aOverload[i].xFunc; |
7252 }else if( strcmp(zName,"dump_doclist")==0 ){ | 3501 return 1; |
7253 /* fprintf(stderr, "Found dump_doclist\n"); */ | 3502 } |
7254 *pxFunc = dumpDoclistFunc; | 3503 } |
7255 return 1; | 3504 |
7256 #endif | 3505 /* No function of the specified name was found. Return 0. */ |
7257 } | |
7258 return 0; | 3506 return 0; |
7259 } | 3507 } |
7260 | 3508 |
7261 /* | 3509 /* |
7262 ** Rename an fts3 table. | 3510 ** Implementation of FTS3 xRename method. Rename an fts3 table. |
7263 */ | 3511 */ |
7264 static int fulltextRename( | 3512 static int fts3RenameMethod( |
7265 sqlite3_vtab *pVtab, | 3513 sqlite3_vtab *pVtab, /* Virtual table handle */ |
7266 const char *zName | 3514 const char *zName /* New name of table */ |
7267 ){ | 3515 ){ |
7268 fulltext_vtab *p = (fulltext_vtab *)pVtab; | 3516 Fts3Table *p = (Fts3Table *)pVtab; |
7269 int rc = SQLITE_NOMEM; | 3517 sqlite3 *db = p->db; /* Database connection */ |
7270 char *zSql = sqlite3_mprintf( | 3518 int rc; /* Return Code */ |
7271 "ALTER TABLE %Q.'%q_content' RENAME TO '%q_content';" | 3519 |
7272 "ALTER TABLE %Q.'%q_segments' RENAME TO '%q_segments';" | 3520 rc = sqlite3Fts3PendingTermsFlush(p); |
7273 "ALTER TABLE %Q.'%q_segdir' RENAME TO '%q_segdir';" | 3521 if( rc!=SQLITE_OK ){ |
7274 , p->zDb, p->zName, zName | 3522 return rc; |
7275 , p->zDb, p->zName, zName | 3523 } |
7276 , p->zDb, p->zName, zName | 3524 |
| 3525 fts3DbExec(&rc, db, |
| 3526 "ALTER TABLE %Q.'%q_content' RENAME TO '%q_content';", |
| 3527 p->zDb, p->zName, zName |
7277 ); | 3528 ); |
7278 if( zSql ){ | 3529 if( p->bHasDocsize ){ |
7279 rc = sqlite3_exec(p->db, zSql, 0, 0, 0); | 3530 fts3DbExec(&rc, db, |
7280 sqlite3_free(zSql); | 3531 "ALTER TABLE %Q.'%q_docsize' RENAME TO '%q_docsize';", |
7281 } | 3532 p->zDb, p->zName, zName |
| 3533 ); |
| 3534 } |
| 3535 if( p->bHasStat ){ |
| 3536 fts3DbExec(&rc, db, |
| 3537 "ALTER TABLE %Q.'%q_stat' RENAME TO '%q_stat';", |
| 3538 p->zDb, p->zName, zName |
| 3539 ); |
| 3540 } |
| 3541 fts3DbExec(&rc, db, |
| 3542 "ALTER TABLE %Q.'%q_segments' RENAME TO '%q_segments';", |
| 3543 p->zDb, p->zName, zName |
| 3544 ); |
| 3545 fts3DbExec(&rc, db, |
| 3546 "ALTER TABLE %Q.'%q_segdir' RENAME TO '%q_segdir';", |
| 3547 p->zDb, p->zName, zName |
| 3548 ); |
7282 return rc; | 3549 return rc; |
7283 } | 3550 } |
7284 | 3551 |
7285 static const sqlite3_module fts3Module = { | 3552 static const sqlite3_module fts3Module = { |
7286 /* iVersion */ 0, | 3553 /* iVersion */ 0, |
7287 /* xCreate */ fulltextCreate, | 3554 /* xCreate */ fts3CreateMethod, |
7288 /* xConnect */ fulltextConnect, | 3555 /* xConnect */ fts3ConnectMethod, |
7289 /* xBestIndex */ fulltextBestIndex, | 3556 /* xBestIndex */ fts3BestIndexMethod, |
7290 /* xDisconnect */ fulltextDisconnect, | 3557 /* xDisconnect */ fts3DisconnectMethod, |
7291 /* xDestroy */ fulltextDestroy, | 3558 /* xDestroy */ fts3DestroyMethod, |
7292 /* xOpen */ fulltextOpen, | 3559 /* xOpen */ fts3OpenMethod, |
7293 /* xClose */ fulltextClose, | 3560 /* xClose */ fts3CloseMethod, |
7294 /* xFilter */ fulltextFilter, | 3561 /* xFilter */ fts3FilterMethod, |
7295 /* xNext */ fulltextNext, | 3562 /* xNext */ fts3NextMethod, |
7296 /* xEof */ fulltextEof, | 3563 /* xEof */ fts3EofMethod, |
7297 /* xColumn */ fulltextColumn, | 3564 /* xColumn */ fts3ColumnMethod, |
7298 /* xRowid */ fulltextRowid, | 3565 /* xRowid */ fts3RowidMethod, |
7299 /* xUpdate */ fulltextUpdate, | 3566 /* xUpdate */ fts3UpdateMethod, |
7300 /* xBegin */ fulltextBegin, | 3567 /* xBegin */ fts3BeginMethod, |
7301 /* xSync */ fulltextSync, | 3568 /* xSync */ fts3SyncMethod, |
7302 /* xCommit */ fulltextCommit, | 3569 /* xCommit */ fts3CommitMethod, |
7303 /* xRollback */ fulltextRollback, | 3570 /* xRollback */ fts3RollbackMethod, |
7304 /* xFindFunction */ fulltextFindFunction, | 3571 /* xFindFunction */ fts3FindFunctionMethod, |
7305 /* xRename */ fulltextRename, | 3572 /* xRename */ fts3RenameMethod, |
7306 }; | 3573 }; |
7307 | 3574 |
| 3575 /* |
| 3576 ** This function is registered as the module destructor (called when an |
| 3577 ** FTS3 enabled database connection is closed). It frees the memory |
| 3578 ** allocated for the tokenizer hash table. |
| 3579 */ |
7308 static void hashDestroy(void *p){ | 3580 static void hashDestroy(void *p){ |
7309 fts3Hash *pHash = (fts3Hash *)p; | 3581 Fts3Hash *pHash = (Fts3Hash *)p; |
7310 sqlite3Fts3HashClear(pHash); | 3582 sqlite3Fts3HashClear(pHash); |
7311 sqlite3_free(pHash); | 3583 sqlite3_free(pHash); |
7312 } | 3584 } |
7313 | 3585 |
7314 /* | 3586 /* |
7315 ** The fts3 built-in tokenizers - "simple" and "porter" - are implemented | 3587 ** The fts3 built-in tokenizers - "simple", "porter" and "icu"- are |
7316 ** in files fts3_tokenizer1.c and fts3_porter.c respectively. The following | 3588 ** implemented in files fts3_tokenizer1.c, fts3_porter.c and fts3_icu.c |
7317 ** two forward declarations are for functions declared in these files | 3589 ** respectively. The following three forward declarations are for functions |
7318 ** used to retrieve the respective implementations. | 3590 ** declared in these files used to retrieve the respective implementations. |
7319 ** | 3591 ** |
7320 ** Calling sqlite3Fts3SimpleTokenizerModule() sets the value pointed | 3592 ** Calling sqlite3Fts3SimpleTokenizerModule() sets the value pointed |
7321 ** to by the argument to point a the "simple" tokenizer implementation. | 3593 ** to by the argument to point to the "simple" tokenizer implementation. |
7322 ** Function ...PorterTokenizerModule() sets *pModule to point to the | 3594 ** And so on. |
7323 ** porter tokenizer/stemmer implementation. | |
7324 */ | 3595 */ |
7325 void sqlite3Fts3SimpleTokenizerModule(sqlite3_tokenizer_module const**ppModule); | 3596 void sqlite3Fts3SimpleTokenizerModule(sqlite3_tokenizer_module const**ppModule); |
7326 void sqlite3Fts3PorterTokenizerModule(sqlite3_tokenizer_module const**ppModule); | 3597 void sqlite3Fts3PorterTokenizerModule(sqlite3_tokenizer_module const**ppModule); |
| 3598 #ifdef SQLITE_ENABLE_ICU |
7327 void sqlite3Fts3IcuTokenizerModule(sqlite3_tokenizer_module const**ppModule); | 3599 void sqlite3Fts3IcuTokenizerModule(sqlite3_tokenizer_module const**ppModule); |
7328 | 3600 #endif |
7329 int sqlite3Fts3InitHashTable(sqlite3 *, fts3Hash *, const char *); | |
7330 | 3601 |
7331 /* | 3602 /* |
7332 ** Initialise the fts3 extension. If this extension is built as part | 3603 ** Initialise the fts3 extension. If this extension is built as part |
7333 ** of the sqlite library, then this function is called directly by | 3604 ** of the sqlite library, then this function is called directly by |
7334 ** SQLite. If fts3 is built as a dynamically loadable extension, this | 3605 ** SQLite. If fts3 is built as a dynamically loadable extension, this |
7335 ** function is called by the sqlite3_extension_init() entry point. | 3606 ** function is called by the sqlite3_extension_init() entry point. |
7336 */ | 3607 */ |
7337 int sqlite3Fts3Init(sqlite3 *db){ | 3608 int sqlite3Fts3Init(sqlite3 *db){ |
7338 int rc = SQLITE_OK; | 3609 int rc = SQLITE_OK; |
7339 fts3Hash *pHash = 0; | 3610 Fts3Hash *pHash = 0; |
7340 const sqlite3_tokenizer_module *pSimple = 0; | 3611 const sqlite3_tokenizer_module *pSimple = 0; |
7341 const sqlite3_tokenizer_module *pPorter = 0; | 3612 const sqlite3_tokenizer_module *pPorter = 0; |
| 3613 |
| 3614 #ifdef SQLITE_ENABLE_ICU |
7342 const sqlite3_tokenizer_module *pIcu = 0; | 3615 const sqlite3_tokenizer_module *pIcu = 0; |
| 3616 sqlite3Fts3IcuTokenizerModule(&pIcu); |
| 3617 #endif |
| 3618 |
| 3619 rc = sqlite3Fts3InitAux(db); |
| 3620 if( rc!=SQLITE_OK ) return rc; |
7343 | 3621 |
7344 sqlite3Fts3SimpleTokenizerModule(&pSimple); | 3622 sqlite3Fts3SimpleTokenizerModule(&pSimple); |
7345 sqlite3Fts3PorterTokenizerModule(&pPorter); | 3623 sqlite3Fts3PorterTokenizerModule(&pPorter); |
7346 #ifdef SQLITE_ENABLE_ICU | |
7347 sqlite3Fts3IcuTokenizerModule(&pIcu); | |
7348 #endif | |
7349 | 3624 |
7350 /* Allocate and initialise the hash-table used to store tokenizers. */ | 3625 /* Allocate and initialise the hash-table used to store tokenizers. */ |
7351 pHash = sqlite3_malloc(sizeof(fts3Hash)); | 3626 pHash = sqlite3_malloc(sizeof(Fts3Hash)); |
7352 if( !pHash ){ | 3627 if( !pHash ){ |
7353 rc = SQLITE_NOMEM; | 3628 rc = SQLITE_NOMEM; |
7354 }else{ | 3629 }else{ |
7355 sqlite3Fts3HashInit(pHash, FTS3_HASH_STRING, 1); | 3630 sqlite3Fts3HashInit(pHash, FTS3_HASH_STRING, 1); |
7356 } | 3631 } |
7357 | 3632 |
7358 /* Load the built-in tokenizers into the hash table */ | 3633 /* Load the built-in tokenizers into the hash table */ |
7359 if( rc==SQLITE_OK ){ | 3634 if( rc==SQLITE_OK ){ |
7360 if( sqlite3Fts3HashInsert(pHash, "simple", 7, (void *)pSimple) | 3635 if( sqlite3Fts3HashInsert(pHash, "simple", 7, (void *)pSimple) |
7361 || sqlite3Fts3HashInsert(pHash, "porter", 7, (void *)pPorter) | 3636 || sqlite3Fts3HashInsert(pHash, "porter", 7, (void *)pPorter) |
| 3637 #ifdef SQLITE_ENABLE_ICU |
7362 || (pIcu && sqlite3Fts3HashInsert(pHash, "icu", 4, (void *)pIcu)) | 3638 || (pIcu && sqlite3Fts3HashInsert(pHash, "icu", 4, (void *)pIcu)) |
| 3639 #endif |
7363 ){ | 3640 ){ |
7364 rc = SQLITE_NOMEM; | 3641 rc = SQLITE_NOMEM; |
7365 } | 3642 } |
7366 } | 3643 } |
7367 | 3644 |
7368 #ifdef SQLITE_TEST | 3645 #ifdef SQLITE_TEST |
7369 sqlite3Fts3ExprInitTestInterface(db); | 3646 if( rc==SQLITE_OK ){ |
| 3647 rc = sqlite3Fts3ExprInitTestInterface(db); |
| 3648 } |
7370 #endif | 3649 #endif |
7371 | 3650 |
7372 /* Create the virtual table wrapper around the hash-table and overload | 3651 /* Create the virtual table wrapper around the hash-table and overload |
7373 ** the two scalar functions. If this is successful, register the | 3652 ** the two scalar functions. If this is successful, register the |
7374 ** module with sqlite. | 3653 ** module with sqlite. |
7375 */ | 3654 */ |
7376 if( SQLITE_OK==rc | 3655 if( SQLITE_OK==rc |
7377 #if CHROMIUM_FTS3_CHANGES && !SQLITE_TEST | 3656 #if CHROMIUM_FTS3_CHANGES && !SQLITE_TEST |
7378 /* fts3_tokenizer() disabled for security reasons. */ | 3657 /* fts3_tokenizer() disabled for security reasons. */ |
7379 #else | 3658 #else |
7380 && SQLITE_OK==(rc = sqlite3Fts3InitHashTable(db, pHash, "fts3_tokenizer")) | 3659 && SQLITE_OK==(rc = sqlite3Fts3InitHashTable(db, pHash, "fts3_tokenizer")) |
7381 #endif | 3660 #endif |
7382 && SQLITE_OK==(rc = sqlite3_overload_function(db, "snippet", -1)) | 3661 && SQLITE_OK==(rc = sqlite3_overload_function(db, "snippet", -1)) |
7383 && SQLITE_OK==(rc = sqlite3_overload_function(db, "offsets", -1)) | 3662 && SQLITE_OK==(rc = sqlite3_overload_function(db, "offsets", 1)) |
7384 && SQLITE_OK==(rc = sqlite3_overload_function(db, "optimize", -1)) | 3663 && SQLITE_OK==(rc = sqlite3_overload_function(db, "matchinfo", 1)) |
7385 #ifdef SQLITE_TEST | 3664 && SQLITE_OK==(rc = sqlite3_overload_function(db, "matchinfo", 2)) |
7386 && SQLITE_OK==(rc = sqlite3_overload_function(db, "dump_terms", -1)) | 3665 && SQLITE_OK==(rc = sqlite3_overload_function(db, "optimize", 1)) |
7387 && SQLITE_OK==(rc = sqlite3_overload_function(db, "dump_doclist", -1)) | |
7388 #endif | |
7389 ){ | 3666 ){ |
7390 return sqlite3_create_module_v2( | 3667 rc = sqlite3_create_module_v2( |
7391 db, "fts3", &fts3Module, (void *)pHash, hashDestroy | 3668 db, "fts3", &fts3Module, (void *)pHash, hashDestroy |
7392 ); | 3669 ); |
| 3670 #if CHROMIUM_FTS3_CHANGES && !SQLITE_TEST |
| 3671 /* Disable fts4 pending review. */ |
| 3672 #else |
| 3673 if( rc==SQLITE_OK ){ |
| 3674 rc = sqlite3_create_module_v2( |
| 3675 db, "fts4", &fts3Module, (void *)pHash, 0 |
| 3676 ); |
| 3677 } |
| 3678 #endif |
| 3679 return rc; |
7393 } | 3680 } |
7394 | 3681 |
7395 /* An error has occurred. Delete the hash table and return the error code. */ | 3682 /* An error has occurred. Delete the hash table and return the error code. */ |
7396 assert( rc!=SQLITE_OK ); | 3683 assert( rc!=SQLITE_OK ); |
7397 if( pHash ){ | 3684 if( pHash ){ |
7398 sqlite3Fts3HashClear(pHash); | 3685 sqlite3Fts3HashClear(pHash); |
7399 sqlite3_free(pHash); | 3686 sqlite3_free(pHash); |
7400 } | 3687 } |
7401 return rc; | 3688 return rc; |
7402 } | 3689 } |
7403 | 3690 |
7404 #if !SQLITE_CORE | 3691 #if !SQLITE_CORE |
7405 int sqlite3_extension_init( | 3692 int sqlite3_extension_init( |
7406 sqlite3 *db, | 3693 sqlite3 *db, |
7407 char **pzErrMsg, | 3694 char **pzErrMsg, |
7408 const sqlite3_api_routines *pApi | 3695 const sqlite3_api_routines *pApi |
7409 ){ | 3696 ){ |
7410 SQLITE_EXTENSION_INIT2(pApi) | 3697 SQLITE_EXTENSION_INIT2(pApi) |
7411 return sqlite3Fts3Init(db); | 3698 return sqlite3Fts3Init(db); |
7412 } | 3699 } |
7413 #endif | 3700 #endif |
7414 | 3701 |
7415 #endif /* !defined(SQLITE_CORE) || defined(SQLITE_ENABLE_FTS3) */ | 3702 #endif |
OLD | NEW |