modernc.org/cc@v1.0.1/v2/testdata/_sqlite/ext/fts5/fts5_expr.c (about) 1 /* 2 ** 2014 May 31 3 ** 4 ** The author disclaims copyright to this source code. In place of 5 ** a legal notice, here is a blessing: 6 ** 7 ** May you do good and not evil. 8 ** May you find forgiveness for yourself and forgive others. 9 ** May you share freely, never taking more than you give. 10 ** 11 ****************************************************************************** 12 ** 13 */ 14 15 16 17 #include "fts5Int.h" 18 #include "fts5parse.h" 19 20 /* 21 ** All token types in the generated fts5parse.h file are greater than 0. 22 */ 23 #define FTS5_EOF 0 24 25 #define FTS5_LARGEST_INT64 (0xffffffff|(((i64)0x7fffffff)<<32)) 26 27 typedef struct Fts5ExprTerm Fts5ExprTerm; 28 29 /* 30 ** Functions generated by lemon from fts5parse.y. 31 */ 32 void *sqlite3Fts5ParserAlloc(void *(*mallocProc)(u64)); 33 void sqlite3Fts5ParserFree(void*, void (*freeProc)(void*)); 34 void sqlite3Fts5Parser(void*, int, Fts5Token, Fts5Parse*); 35 #ifndef NDEBUG 36 #include <stdio.h> 37 void sqlite3Fts5ParserTrace(FILE*, char*); 38 #endif 39 40 41 struct Fts5Expr { 42 Fts5Index *pIndex; 43 Fts5Config *pConfig; 44 Fts5ExprNode *pRoot; 45 int bDesc; /* Iterate in descending rowid order */ 46 int nPhrase; /* Number of phrases in expression */ 47 Fts5ExprPhrase **apExprPhrase; /* Pointers to phrase objects */ 48 }; 49 50 /* 51 ** eType: 52 ** Expression node type. Always one of: 53 ** 54 ** FTS5_AND (nChild, apChild valid) 55 ** FTS5_OR (nChild, apChild valid) 56 ** FTS5_NOT (nChild, apChild valid) 57 ** FTS5_STRING (pNear valid) 58 ** FTS5_TERM (pNear valid) 59 */ 60 struct Fts5ExprNode { 61 int eType; /* Node type */ 62 int bEof; /* True at EOF */ 63 int bNomatch; /* True if entry is not a match */ 64 65 /* Next method for this node. */ 66 int (*xNext)(Fts5Expr*, Fts5ExprNode*, int, i64); 67 68 i64 iRowid; /* Current rowid */ 69 Fts5ExprNearset *pNear; /* For FTS5_STRING - cluster of phrases */ 70 71 /* Child nodes. For a NOT node, this array always contains 2 entries. For 72 ** AND or OR nodes, it contains 2 or more entries. */ 73 int nChild; /* Number of child nodes */ 74 Fts5ExprNode *apChild[1]; /* Array of child nodes */ 75 }; 76 77 #define Fts5NodeIsString(p) ((p)->eType==FTS5_TERM || (p)->eType==FTS5_STRING) 78 79 /* 80 ** Invoke the xNext method of an Fts5ExprNode object. This macro should be 81 ** used as if it has the same signature as the xNext() methods themselves. 82 */ 83 #define fts5ExprNodeNext(a,b,c,d) (b)->xNext((a), (b), (c), (d)) 84 85 /* 86 ** An instance of the following structure represents a single search term 87 ** or term prefix. 88 */ 89 struct Fts5ExprTerm { 90 int bPrefix; /* True for a prefix term */ 91 char *zTerm; /* nul-terminated term */ 92 Fts5IndexIter *pIter; /* Iterator for this term */ 93 Fts5ExprTerm *pSynonym; /* Pointer to first in list of synonyms */ 94 }; 95 96 /* 97 ** A phrase. One or more terms that must appear in a contiguous sequence 98 ** within a document for it to match. 99 */ 100 struct Fts5ExprPhrase { 101 Fts5ExprNode *pNode; /* FTS5_STRING node this phrase is part of */ 102 Fts5Buffer poslist; /* Current position list */ 103 int nTerm; /* Number of entries in aTerm[] */ 104 Fts5ExprTerm aTerm[1]; /* Terms that make up this phrase */ 105 }; 106 107 /* 108 ** One or more phrases that must appear within a certain token distance of 109 ** each other within each matching document. 110 */ 111 struct Fts5ExprNearset { 112 int nNear; /* NEAR parameter */ 113 Fts5Colset *pColset; /* Columns to search (NULL -> all columns) */ 114 int nPhrase; /* Number of entries in aPhrase[] array */ 115 Fts5ExprPhrase *apPhrase[1]; /* Array of phrase pointers */ 116 }; 117 118 119 /* 120 ** Parse context. 121 */ 122 struct Fts5Parse { 123 Fts5Config *pConfig; 124 char *zErr; 125 int rc; 126 int nPhrase; /* Size of apPhrase array */ 127 Fts5ExprPhrase **apPhrase; /* Array of all phrases */ 128 Fts5ExprNode *pExpr; /* Result of a successful parse */ 129 }; 130 131 void sqlite3Fts5ParseError(Fts5Parse *pParse, const char *zFmt, ...){ 132 va_list ap; 133 va_start(ap, zFmt); 134 if( pParse->rc==SQLITE_OK ){ 135 pParse->zErr = sqlite3_vmprintf(zFmt, ap); 136 pParse->rc = SQLITE_ERROR; 137 } 138 va_end(ap); 139 } 140 141 static int fts5ExprIsspace(char t){ 142 return t==' ' || t=='\t' || t=='\n' || t=='\r'; 143 } 144 145 /* 146 ** Read the first token from the nul-terminated string at *pz. 147 */ 148 static int fts5ExprGetToken( 149 Fts5Parse *pParse, 150 const char **pz, /* IN/OUT: Pointer into buffer */ 151 Fts5Token *pToken 152 ){ 153 const char *z = *pz; 154 int tok; 155 156 /* Skip past any whitespace */ 157 while( fts5ExprIsspace(*z) ) z++; 158 159 pToken->p = z; 160 pToken->n = 1; 161 switch( *z ){ 162 case '(': tok = FTS5_LP; break; 163 case ')': tok = FTS5_RP; break; 164 case '{': tok = FTS5_LCP; break; 165 case '}': tok = FTS5_RCP; break; 166 case ':': tok = FTS5_COLON; break; 167 case ',': tok = FTS5_COMMA; break; 168 case '+': tok = FTS5_PLUS; break; 169 case '*': tok = FTS5_STAR; break; 170 case '-': tok = FTS5_MINUS; break; 171 case '\0': tok = FTS5_EOF; break; 172 173 case '"': { 174 const char *z2; 175 tok = FTS5_STRING; 176 177 for(z2=&z[1]; 1; z2++){ 178 if( z2[0]=='"' ){ 179 z2++; 180 if( z2[0]!='"' ) break; 181 } 182 if( z2[0]=='\0' ){ 183 sqlite3Fts5ParseError(pParse, "unterminated string"); 184 return FTS5_EOF; 185 } 186 } 187 pToken->n = (z2 - z); 188 break; 189 } 190 191 default: { 192 const char *z2; 193 if( sqlite3Fts5IsBareword(z[0])==0 ){ 194 sqlite3Fts5ParseError(pParse, "fts5: syntax error near \"%.1s\"", z); 195 return FTS5_EOF; 196 } 197 tok = FTS5_STRING; 198 for(z2=&z[1]; sqlite3Fts5IsBareword(*z2); z2++); 199 pToken->n = (z2 - z); 200 if( pToken->n==2 && memcmp(pToken->p, "OR", 2)==0 ) tok = FTS5_OR; 201 if( pToken->n==3 && memcmp(pToken->p, "NOT", 3)==0 ) tok = FTS5_NOT; 202 if( pToken->n==3 && memcmp(pToken->p, "AND", 3)==0 ) tok = FTS5_AND; 203 break; 204 } 205 } 206 207 *pz = &pToken->p[pToken->n]; 208 return tok; 209 } 210 211 static void *fts5ParseAlloc(u64 t){ return sqlite3_malloc((int)t); } 212 static void fts5ParseFree(void *p){ sqlite3_free(p); } 213 214 int sqlite3Fts5ExprNew( 215 Fts5Config *pConfig, /* FTS5 Configuration */ 216 int iCol, 217 const char *zExpr, /* Expression text */ 218 Fts5Expr **ppNew, 219 char **pzErr 220 ){ 221 Fts5Parse sParse; 222 Fts5Token token; 223 const char *z = zExpr; 224 int t; /* Next token type */ 225 void *pEngine; 226 Fts5Expr *pNew; 227 228 *ppNew = 0; 229 *pzErr = 0; 230 memset(&sParse, 0, sizeof(sParse)); 231 pEngine = sqlite3Fts5ParserAlloc(fts5ParseAlloc); 232 if( pEngine==0 ){ return SQLITE_NOMEM; } 233 sParse.pConfig = pConfig; 234 235 do { 236 t = fts5ExprGetToken(&sParse, &z, &token); 237 sqlite3Fts5Parser(pEngine, t, token, &sParse); 238 }while( sParse.rc==SQLITE_OK && t!=FTS5_EOF ); 239 sqlite3Fts5ParserFree(pEngine, fts5ParseFree); 240 241 /* If the LHS of the MATCH expression was a user column, apply the 242 ** implicit column-filter. */ 243 if( iCol<pConfig->nCol && sParse.pExpr && sParse.rc==SQLITE_OK ){ 244 int n = sizeof(Fts5Colset); 245 Fts5Colset *pColset = (Fts5Colset*)sqlite3Fts5MallocZero(&sParse.rc, n); 246 if( pColset ){ 247 pColset->nCol = 1; 248 pColset->aiCol[0] = iCol; 249 sqlite3Fts5ParseSetColset(&sParse, sParse.pExpr, pColset); 250 } 251 } 252 253 assert( sParse.rc!=SQLITE_OK || sParse.zErr==0 ); 254 if( sParse.rc==SQLITE_OK ){ 255 *ppNew = pNew = sqlite3_malloc(sizeof(Fts5Expr)); 256 if( pNew==0 ){ 257 sParse.rc = SQLITE_NOMEM; 258 sqlite3Fts5ParseNodeFree(sParse.pExpr); 259 }else{ 260 if( !sParse.pExpr ){ 261 const int nByte = sizeof(Fts5ExprNode); 262 pNew->pRoot = (Fts5ExprNode*)sqlite3Fts5MallocZero(&sParse.rc, nByte); 263 if( pNew->pRoot ){ 264 pNew->pRoot->bEof = 1; 265 } 266 }else{ 267 pNew->pRoot = sParse.pExpr; 268 } 269 pNew->pIndex = 0; 270 pNew->pConfig = pConfig; 271 pNew->apExprPhrase = sParse.apPhrase; 272 pNew->nPhrase = sParse.nPhrase; 273 sParse.apPhrase = 0; 274 } 275 }else{ 276 sqlite3Fts5ParseNodeFree(sParse.pExpr); 277 } 278 279 sqlite3_free(sParse.apPhrase); 280 *pzErr = sParse.zErr; 281 return sParse.rc; 282 } 283 284 /* 285 ** Free the expression node object passed as the only argument. 286 */ 287 void sqlite3Fts5ParseNodeFree(Fts5ExprNode *p){ 288 if( p ){ 289 int i; 290 for(i=0; i<p->nChild; i++){ 291 sqlite3Fts5ParseNodeFree(p->apChild[i]); 292 } 293 sqlite3Fts5ParseNearsetFree(p->pNear); 294 sqlite3_free(p); 295 } 296 } 297 298 /* 299 ** Free the expression object passed as the only argument. 300 */ 301 void sqlite3Fts5ExprFree(Fts5Expr *p){ 302 if( p ){ 303 sqlite3Fts5ParseNodeFree(p->pRoot); 304 sqlite3_free(p->apExprPhrase); 305 sqlite3_free(p); 306 } 307 } 308 309 /* 310 ** Argument pTerm must be a synonym iterator. Return the current rowid 311 ** that it points to. 312 */ 313 static i64 fts5ExprSynonymRowid(Fts5ExprTerm *pTerm, int bDesc, int *pbEof){ 314 i64 iRet = 0; 315 int bRetValid = 0; 316 Fts5ExprTerm *p; 317 318 assert( pTerm->pSynonym ); 319 assert( bDesc==0 || bDesc==1 ); 320 for(p=pTerm; p; p=p->pSynonym){ 321 if( 0==sqlite3Fts5IterEof(p->pIter) ){ 322 i64 iRowid = p->pIter->iRowid; 323 if( bRetValid==0 || (bDesc!=(iRowid<iRet)) ){ 324 iRet = iRowid; 325 bRetValid = 1; 326 } 327 } 328 } 329 330 if( pbEof && bRetValid==0 ) *pbEof = 1; 331 return iRet; 332 } 333 334 /* 335 ** Argument pTerm must be a synonym iterator. 336 */ 337 static int fts5ExprSynonymList( 338 Fts5ExprTerm *pTerm, 339 i64 iRowid, 340 Fts5Buffer *pBuf, /* Use this buffer for space if required */ 341 u8 **pa, int *pn 342 ){ 343 Fts5PoslistReader aStatic[4]; 344 Fts5PoslistReader *aIter = aStatic; 345 int nIter = 0; 346 int nAlloc = 4; 347 int rc = SQLITE_OK; 348 Fts5ExprTerm *p; 349 350 assert( pTerm->pSynonym ); 351 for(p=pTerm; p; p=p->pSynonym){ 352 Fts5IndexIter *pIter = p->pIter; 353 if( sqlite3Fts5IterEof(pIter)==0 && pIter->iRowid==iRowid ){ 354 if( pIter->nData==0 ) continue; 355 if( nIter==nAlloc ){ 356 int nByte = sizeof(Fts5PoslistReader) * nAlloc * 2; 357 Fts5PoslistReader *aNew = (Fts5PoslistReader*)sqlite3_malloc(nByte); 358 if( aNew==0 ){ 359 rc = SQLITE_NOMEM; 360 goto synonym_poslist_out; 361 } 362 memcpy(aNew, aIter, sizeof(Fts5PoslistReader) * nIter); 363 nAlloc = nAlloc*2; 364 if( aIter!=aStatic ) sqlite3_free(aIter); 365 aIter = aNew; 366 } 367 sqlite3Fts5PoslistReaderInit(pIter->pData, pIter->nData, &aIter[nIter]); 368 assert( aIter[nIter].bEof==0 ); 369 nIter++; 370 } 371 } 372 373 if( nIter==1 ){ 374 *pa = (u8*)aIter[0].a; 375 *pn = aIter[0].n; 376 }else{ 377 Fts5PoslistWriter writer = {0}; 378 i64 iPrev = -1; 379 fts5BufferZero(pBuf); 380 while( 1 ){ 381 int i; 382 i64 iMin = FTS5_LARGEST_INT64; 383 for(i=0; i<nIter; i++){ 384 if( aIter[i].bEof==0 ){ 385 if( aIter[i].iPos==iPrev ){ 386 if( sqlite3Fts5PoslistReaderNext(&aIter[i]) ) continue; 387 } 388 if( aIter[i].iPos<iMin ){ 389 iMin = aIter[i].iPos; 390 } 391 } 392 } 393 if( iMin==FTS5_LARGEST_INT64 || rc!=SQLITE_OK ) break; 394 rc = sqlite3Fts5PoslistWriterAppend(pBuf, &writer, iMin); 395 iPrev = iMin; 396 } 397 if( rc==SQLITE_OK ){ 398 *pa = pBuf->p; 399 *pn = pBuf->n; 400 } 401 } 402 403 synonym_poslist_out: 404 if( aIter!=aStatic ) sqlite3_free(aIter); 405 return rc; 406 } 407 408 409 /* 410 ** All individual term iterators in pPhrase are guaranteed to be valid and 411 ** pointing to the same rowid when this function is called. This function 412 ** checks if the current rowid really is a match, and if so populates 413 ** the pPhrase->poslist buffer accordingly. Output parameter *pbMatch 414 ** is set to true if this is really a match, or false otherwise. 415 ** 416 ** SQLITE_OK is returned if an error occurs, or an SQLite error code 417 ** otherwise. It is not considered an error code if the current rowid is 418 ** not a match. 419 */ 420 static int fts5ExprPhraseIsMatch( 421 Fts5ExprNode *pNode, /* Node pPhrase belongs to */ 422 Fts5ExprPhrase *pPhrase, /* Phrase object to initialize */ 423 int *pbMatch /* OUT: Set to true if really a match */ 424 ){ 425 Fts5PoslistWriter writer = {0}; 426 Fts5PoslistReader aStatic[4]; 427 Fts5PoslistReader *aIter = aStatic; 428 int i; 429 int rc = SQLITE_OK; 430 431 fts5BufferZero(&pPhrase->poslist); 432 433 /* If the aStatic[] array is not large enough, allocate a large array 434 ** using sqlite3_malloc(). This approach could be improved upon. */ 435 if( pPhrase->nTerm>ArraySize(aStatic) ){ 436 int nByte = sizeof(Fts5PoslistReader) * pPhrase->nTerm; 437 aIter = (Fts5PoslistReader*)sqlite3_malloc(nByte); 438 if( !aIter ) return SQLITE_NOMEM; 439 } 440 memset(aIter, 0, sizeof(Fts5PoslistReader) * pPhrase->nTerm); 441 442 /* Initialize a term iterator for each term in the phrase */ 443 for(i=0; i<pPhrase->nTerm; i++){ 444 Fts5ExprTerm *pTerm = &pPhrase->aTerm[i]; 445 int n = 0; 446 int bFlag = 0; 447 u8 *a = 0; 448 if( pTerm->pSynonym ){ 449 Fts5Buffer buf = {0, 0, 0}; 450 rc = fts5ExprSynonymList(pTerm, pNode->iRowid, &buf, &a, &n); 451 if( rc ){ 452 sqlite3_free(a); 453 goto ismatch_out; 454 } 455 if( a==buf.p ) bFlag = 1; 456 }else{ 457 a = (u8*)pTerm->pIter->pData; 458 n = pTerm->pIter->nData; 459 } 460 sqlite3Fts5PoslistReaderInit(a, n, &aIter[i]); 461 aIter[i].bFlag = (u8)bFlag; 462 if( aIter[i].bEof ) goto ismatch_out; 463 } 464 465 while( 1 ){ 466 int bMatch; 467 i64 iPos = aIter[0].iPos; 468 do { 469 bMatch = 1; 470 for(i=0; i<pPhrase->nTerm; i++){ 471 Fts5PoslistReader *pPos = &aIter[i]; 472 i64 iAdj = iPos + i; 473 if( pPos->iPos!=iAdj ){ 474 bMatch = 0; 475 while( pPos->iPos<iAdj ){ 476 if( sqlite3Fts5PoslistReaderNext(pPos) ) goto ismatch_out; 477 } 478 if( pPos->iPos>iAdj ) iPos = pPos->iPos-i; 479 } 480 } 481 }while( bMatch==0 ); 482 483 /* Append position iPos to the output */ 484 rc = sqlite3Fts5PoslistWriterAppend(&pPhrase->poslist, &writer, iPos); 485 if( rc!=SQLITE_OK ) goto ismatch_out; 486 487 for(i=0; i<pPhrase->nTerm; i++){ 488 if( sqlite3Fts5PoslistReaderNext(&aIter[i]) ) goto ismatch_out; 489 } 490 } 491 492 ismatch_out: 493 *pbMatch = (pPhrase->poslist.n>0); 494 for(i=0; i<pPhrase->nTerm; i++){ 495 if( aIter[i].bFlag ) sqlite3_free((u8*)aIter[i].a); 496 } 497 if( aIter!=aStatic ) sqlite3_free(aIter); 498 return rc; 499 } 500 501 typedef struct Fts5LookaheadReader Fts5LookaheadReader; 502 struct Fts5LookaheadReader { 503 const u8 *a; /* Buffer containing position list */ 504 int n; /* Size of buffer a[] in bytes */ 505 int i; /* Current offset in position list */ 506 i64 iPos; /* Current position */ 507 i64 iLookahead; /* Next position */ 508 }; 509 510 #define FTS5_LOOKAHEAD_EOF (((i64)1) << 62) 511 512 static int fts5LookaheadReaderNext(Fts5LookaheadReader *p){ 513 p->iPos = p->iLookahead; 514 if( sqlite3Fts5PoslistNext64(p->a, p->n, &p->i, &p->iLookahead) ){ 515 p->iLookahead = FTS5_LOOKAHEAD_EOF; 516 } 517 return (p->iPos==FTS5_LOOKAHEAD_EOF); 518 } 519 520 static int fts5LookaheadReaderInit( 521 const u8 *a, int n, /* Buffer to read position list from */ 522 Fts5LookaheadReader *p /* Iterator object to initialize */ 523 ){ 524 memset(p, 0, sizeof(Fts5LookaheadReader)); 525 p->a = a; 526 p->n = n; 527 fts5LookaheadReaderNext(p); 528 return fts5LookaheadReaderNext(p); 529 } 530 531 typedef struct Fts5NearTrimmer Fts5NearTrimmer; 532 struct Fts5NearTrimmer { 533 Fts5LookaheadReader reader; /* Input iterator */ 534 Fts5PoslistWriter writer; /* Writer context */ 535 Fts5Buffer *pOut; /* Output poslist */ 536 }; 537 538 /* 539 ** The near-set object passed as the first argument contains more than 540 ** one phrase. All phrases currently point to the same row. The 541 ** Fts5ExprPhrase.poslist buffers are populated accordingly. This function 542 ** tests if the current row contains instances of each phrase sufficiently 543 ** close together to meet the NEAR constraint. Non-zero is returned if it 544 ** does, or zero otherwise. 545 ** 546 ** If in/out parameter (*pRc) is set to other than SQLITE_OK when this 547 ** function is called, it is a no-op. Or, if an error (e.g. SQLITE_NOMEM) 548 ** occurs within this function (*pRc) is set accordingly before returning. 549 ** The return value is undefined in both these cases. 550 ** 551 ** If no error occurs and non-zero (a match) is returned, the position-list 552 ** of each phrase object is edited to contain only those entries that 553 ** meet the constraint before returning. 554 */ 555 static int fts5ExprNearIsMatch(int *pRc, Fts5ExprNearset *pNear){ 556 Fts5NearTrimmer aStatic[4]; 557 Fts5NearTrimmer *a = aStatic; 558 Fts5ExprPhrase **apPhrase = pNear->apPhrase; 559 560 int i; 561 int rc = *pRc; 562 int bMatch; 563 564 assert( pNear->nPhrase>1 ); 565 566 /* If the aStatic[] array is not large enough, allocate a large array 567 ** using sqlite3_malloc(). This approach could be improved upon. */ 568 if( pNear->nPhrase>ArraySize(aStatic) ){ 569 int nByte = sizeof(Fts5NearTrimmer) * pNear->nPhrase; 570 a = (Fts5NearTrimmer*)sqlite3Fts5MallocZero(&rc, nByte); 571 }else{ 572 memset(aStatic, 0, sizeof(aStatic)); 573 } 574 if( rc!=SQLITE_OK ){ 575 *pRc = rc; 576 return 0; 577 } 578 579 /* Initialize a lookahead iterator for each phrase. After passing the 580 ** buffer and buffer size to the lookaside-reader init function, zero 581 ** the phrase poslist buffer. The new poslist for the phrase (containing 582 ** the same entries as the original with some entries removed on account 583 ** of the NEAR constraint) is written over the original even as it is 584 ** being read. This is safe as the entries for the new poslist are a 585 ** subset of the old, so it is not possible for data yet to be read to 586 ** be overwritten. */ 587 for(i=0; i<pNear->nPhrase; i++){ 588 Fts5Buffer *pPoslist = &apPhrase[i]->poslist; 589 fts5LookaheadReaderInit(pPoslist->p, pPoslist->n, &a[i].reader); 590 pPoslist->n = 0; 591 a[i].pOut = pPoslist; 592 } 593 594 while( 1 ){ 595 int iAdv; 596 i64 iMin; 597 i64 iMax; 598 599 /* This block advances the phrase iterators until they point to a set of 600 ** entries that together comprise a match. */ 601 iMax = a[0].reader.iPos; 602 do { 603 bMatch = 1; 604 for(i=0; i<pNear->nPhrase; i++){ 605 Fts5LookaheadReader *pPos = &a[i].reader; 606 iMin = iMax - pNear->apPhrase[i]->nTerm - pNear->nNear; 607 if( pPos->iPos<iMin || pPos->iPos>iMax ){ 608 bMatch = 0; 609 while( pPos->iPos<iMin ){ 610 if( fts5LookaheadReaderNext(pPos) ) goto ismatch_out; 611 } 612 if( pPos->iPos>iMax ) iMax = pPos->iPos; 613 } 614 } 615 }while( bMatch==0 ); 616 617 /* Add an entry to each output position list */ 618 for(i=0; i<pNear->nPhrase; i++){ 619 i64 iPos = a[i].reader.iPos; 620 Fts5PoslistWriter *pWriter = &a[i].writer; 621 if( a[i].pOut->n==0 || iPos!=pWriter->iPrev ){ 622 sqlite3Fts5PoslistWriterAppend(a[i].pOut, pWriter, iPos); 623 } 624 } 625 626 iAdv = 0; 627 iMin = a[0].reader.iLookahead; 628 for(i=0; i<pNear->nPhrase; i++){ 629 if( a[i].reader.iLookahead < iMin ){ 630 iMin = a[i].reader.iLookahead; 631 iAdv = i; 632 } 633 } 634 if( fts5LookaheadReaderNext(&a[iAdv].reader) ) goto ismatch_out; 635 } 636 637 ismatch_out: { 638 int bRet = a[0].pOut->n>0; 639 *pRc = rc; 640 if( a!=aStatic ) sqlite3_free(a); 641 return bRet; 642 } 643 } 644 645 /* 646 ** Advance iterator pIter until it points to a value equal to or laster 647 ** than the initial value of *piLast. If this means the iterator points 648 ** to a value laster than *piLast, update *piLast to the new lastest value. 649 ** 650 ** If the iterator reaches EOF, set *pbEof to true before returning. If 651 ** an error occurs, set *pRc to an error code. If either *pbEof or *pRc 652 ** are set, return a non-zero value. Otherwise, return zero. 653 */ 654 static int fts5ExprAdvanceto( 655 Fts5IndexIter *pIter, /* Iterator to advance */ 656 int bDesc, /* True if iterator is "rowid DESC" */ 657 i64 *piLast, /* IN/OUT: Lastest rowid seen so far */ 658 int *pRc, /* OUT: Error code */ 659 int *pbEof /* OUT: Set to true if EOF */ 660 ){ 661 i64 iLast = *piLast; 662 i64 iRowid; 663 664 iRowid = pIter->iRowid; 665 if( (bDesc==0 && iLast>iRowid) || (bDesc && iLast<iRowid) ){ 666 int rc = sqlite3Fts5IterNextFrom(pIter, iLast); 667 if( rc || sqlite3Fts5IterEof(pIter) ){ 668 *pRc = rc; 669 *pbEof = 1; 670 return 1; 671 } 672 iRowid = pIter->iRowid; 673 assert( (bDesc==0 && iRowid>=iLast) || (bDesc==1 && iRowid<=iLast) ); 674 } 675 *piLast = iRowid; 676 677 return 0; 678 } 679 680 static int fts5ExprSynonymAdvanceto( 681 Fts5ExprTerm *pTerm, /* Term iterator to advance */ 682 int bDesc, /* True if iterator is "rowid DESC" */ 683 i64 *piLast, /* IN/OUT: Lastest rowid seen so far */ 684 int *pRc /* OUT: Error code */ 685 ){ 686 int rc = SQLITE_OK; 687 i64 iLast = *piLast; 688 Fts5ExprTerm *p; 689 int bEof = 0; 690 691 for(p=pTerm; rc==SQLITE_OK && p; p=p->pSynonym){ 692 if( sqlite3Fts5IterEof(p->pIter)==0 ){ 693 i64 iRowid = p->pIter->iRowid; 694 if( (bDesc==0 && iLast>iRowid) || (bDesc && iLast<iRowid) ){ 695 rc = sqlite3Fts5IterNextFrom(p->pIter, iLast); 696 } 697 } 698 } 699 700 if( rc!=SQLITE_OK ){ 701 *pRc = rc; 702 bEof = 1; 703 }else{ 704 *piLast = fts5ExprSynonymRowid(pTerm, bDesc, &bEof); 705 } 706 return bEof; 707 } 708 709 710 static int fts5ExprNearTest( 711 int *pRc, 712 Fts5Expr *pExpr, /* Expression that pNear is a part of */ 713 Fts5ExprNode *pNode /* The "NEAR" node (FTS5_STRING) */ 714 ){ 715 Fts5ExprNearset *pNear = pNode->pNear; 716 int rc = *pRc; 717 718 if( pExpr->pConfig->eDetail!=FTS5_DETAIL_FULL ){ 719 Fts5ExprTerm *pTerm; 720 Fts5ExprPhrase *pPhrase = pNear->apPhrase[0]; 721 pPhrase->poslist.n = 0; 722 for(pTerm=&pPhrase->aTerm[0]; pTerm; pTerm=pTerm->pSynonym){ 723 Fts5IndexIter *pIter = pTerm->pIter; 724 if( sqlite3Fts5IterEof(pIter)==0 ){ 725 if( pIter->iRowid==pNode->iRowid && pIter->nData>0 ){ 726 pPhrase->poslist.n = 1; 727 } 728 } 729 } 730 return pPhrase->poslist.n; 731 }else{ 732 int i; 733 734 /* Check that each phrase in the nearset matches the current row. 735 ** Populate the pPhrase->poslist buffers at the same time. If any 736 ** phrase is not a match, break out of the loop early. */ 737 for(i=0; rc==SQLITE_OK && i<pNear->nPhrase; i++){ 738 Fts5ExprPhrase *pPhrase = pNear->apPhrase[i]; 739 if( pPhrase->nTerm>1 || pPhrase->aTerm[0].pSynonym || pNear->pColset ){ 740 int bMatch = 0; 741 rc = fts5ExprPhraseIsMatch(pNode, pPhrase, &bMatch); 742 if( bMatch==0 ) break; 743 }else{ 744 Fts5IndexIter *pIter = pPhrase->aTerm[0].pIter; 745 fts5BufferSet(&rc, &pPhrase->poslist, pIter->nData, pIter->pData); 746 } 747 } 748 749 *pRc = rc; 750 if( i==pNear->nPhrase && (i==1 || fts5ExprNearIsMatch(pRc, pNear)) ){ 751 return 1; 752 } 753 return 0; 754 } 755 } 756 757 758 /* 759 ** Initialize all term iterators in the pNear object. If any term is found 760 ** to match no documents at all, return immediately without initializing any 761 ** further iterators. 762 ** 763 ** If an error occurs, return an SQLite error code. Otherwise, return 764 ** SQLITE_OK. It is not considered an error if some term matches zero 765 ** documents. 766 */ 767 static int fts5ExprNearInitAll( 768 Fts5Expr *pExpr, 769 Fts5ExprNode *pNode 770 ){ 771 Fts5ExprNearset *pNear = pNode->pNear; 772 int i; 773 774 assert( pNode->bNomatch==0 ); 775 for(i=0; i<pNear->nPhrase; i++){ 776 Fts5ExprPhrase *pPhrase = pNear->apPhrase[i]; 777 if( pPhrase->nTerm==0 ){ 778 pNode->bEof = 1; 779 return SQLITE_OK; 780 }else{ 781 int j; 782 for(j=0; j<pPhrase->nTerm; j++){ 783 Fts5ExprTerm *pTerm = &pPhrase->aTerm[j]; 784 Fts5ExprTerm *p; 785 int bHit = 0; 786 787 for(p=pTerm; p; p=p->pSynonym){ 788 int rc; 789 if( p->pIter ){ 790 sqlite3Fts5IterClose(p->pIter); 791 p->pIter = 0; 792 } 793 rc = sqlite3Fts5IndexQuery( 794 pExpr->pIndex, p->zTerm, (int)strlen(p->zTerm), 795 (pTerm->bPrefix ? FTS5INDEX_QUERY_PREFIX : 0) | 796 (pExpr->bDesc ? FTS5INDEX_QUERY_DESC : 0), 797 pNear->pColset, 798 &p->pIter 799 ); 800 assert( (rc==SQLITE_OK)==(p->pIter!=0) ); 801 if( rc!=SQLITE_OK ) return rc; 802 if( 0==sqlite3Fts5IterEof(p->pIter) ){ 803 bHit = 1; 804 } 805 } 806 807 if( bHit==0 ){ 808 pNode->bEof = 1; 809 return SQLITE_OK; 810 } 811 } 812 } 813 } 814 815 pNode->bEof = 0; 816 return SQLITE_OK; 817 } 818 819 /* 820 ** If pExpr is an ASC iterator, this function returns a value with the 821 ** same sign as: 822 ** 823 ** (iLhs - iRhs) 824 ** 825 ** Otherwise, if this is a DESC iterator, the opposite is returned: 826 ** 827 ** (iRhs - iLhs) 828 */ 829 static int fts5RowidCmp( 830 Fts5Expr *pExpr, 831 i64 iLhs, 832 i64 iRhs 833 ){ 834 assert( pExpr->bDesc==0 || pExpr->bDesc==1 ); 835 if( pExpr->bDesc==0 ){ 836 if( iLhs<iRhs ) return -1; 837 return (iLhs > iRhs); 838 }else{ 839 if( iLhs>iRhs ) return -1; 840 return (iLhs < iRhs); 841 } 842 } 843 844 static void fts5ExprSetEof(Fts5ExprNode *pNode){ 845 int i; 846 pNode->bEof = 1; 847 pNode->bNomatch = 0; 848 for(i=0; i<pNode->nChild; i++){ 849 fts5ExprSetEof(pNode->apChild[i]); 850 } 851 } 852 853 static void fts5ExprNodeZeroPoslist(Fts5ExprNode *pNode){ 854 if( pNode->eType==FTS5_STRING || pNode->eType==FTS5_TERM ){ 855 Fts5ExprNearset *pNear = pNode->pNear; 856 int i; 857 for(i=0; i<pNear->nPhrase; i++){ 858 Fts5ExprPhrase *pPhrase = pNear->apPhrase[i]; 859 pPhrase->poslist.n = 0; 860 } 861 }else{ 862 int i; 863 for(i=0; i<pNode->nChild; i++){ 864 fts5ExprNodeZeroPoslist(pNode->apChild[i]); 865 } 866 } 867 } 868 869 870 871 /* 872 ** Compare the values currently indicated by the two nodes as follows: 873 ** 874 ** res = (*p1) - (*p2) 875 ** 876 ** Nodes that point to values that come later in the iteration order are 877 ** considered to be larger. Nodes at EOF are the largest of all. 878 ** 879 ** This means that if the iteration order is ASC, then numerically larger 880 ** rowids are considered larger. Or if it is the default DESC, numerically 881 ** smaller rowids are larger. 882 */ 883 static int fts5NodeCompare( 884 Fts5Expr *pExpr, 885 Fts5ExprNode *p1, 886 Fts5ExprNode *p2 887 ){ 888 if( p2->bEof ) return -1; 889 if( p1->bEof ) return +1; 890 return fts5RowidCmp(pExpr, p1->iRowid, p2->iRowid); 891 } 892 893 /* 894 ** All individual term iterators in pNear are guaranteed to be valid when 895 ** this function is called. This function checks if all term iterators 896 ** point to the same rowid, and if not, advances them until they do. 897 ** If an EOF is reached before this happens, *pbEof is set to true before 898 ** returning. 899 ** 900 ** SQLITE_OK is returned if an error occurs, or an SQLite error code 901 ** otherwise. It is not considered an error code if an iterator reaches 902 ** EOF. 903 */ 904 static int fts5ExprNodeTest_STRING( 905 Fts5Expr *pExpr, /* Expression pPhrase belongs to */ 906 Fts5ExprNode *pNode 907 ){ 908 Fts5ExprNearset *pNear = pNode->pNear; 909 Fts5ExprPhrase *pLeft = pNear->apPhrase[0]; 910 int rc = SQLITE_OK; 911 i64 iLast; /* Lastest rowid any iterator points to */ 912 int i, j; /* Phrase and token index, respectively */ 913 int bMatch; /* True if all terms are at the same rowid */ 914 const int bDesc = pExpr->bDesc; 915 916 /* Check that this node should not be FTS5_TERM */ 917 assert( pNear->nPhrase>1 918 || pNear->apPhrase[0]->nTerm>1 919 || pNear->apPhrase[0]->aTerm[0].pSynonym 920 ); 921 922 /* Initialize iLast, the "lastest" rowid any iterator points to. If the 923 ** iterator skips through rowids in the default ascending order, this means 924 ** the maximum rowid. Or, if the iterator is "ORDER BY rowid DESC", then it 925 ** means the minimum rowid. */ 926 if( pLeft->aTerm[0].pSynonym ){ 927 iLast = fts5ExprSynonymRowid(&pLeft->aTerm[0], bDesc, 0); 928 }else{ 929 iLast = pLeft->aTerm[0].pIter->iRowid; 930 } 931 932 do { 933 bMatch = 1; 934 for(i=0; i<pNear->nPhrase; i++){ 935 Fts5ExprPhrase *pPhrase = pNear->apPhrase[i]; 936 for(j=0; j<pPhrase->nTerm; j++){ 937 Fts5ExprTerm *pTerm = &pPhrase->aTerm[j]; 938 if( pTerm->pSynonym ){ 939 i64 iRowid = fts5ExprSynonymRowid(pTerm, bDesc, 0); 940 if( iRowid==iLast ) continue; 941 bMatch = 0; 942 if( fts5ExprSynonymAdvanceto(pTerm, bDesc, &iLast, &rc) ){ 943 pNode->bNomatch = 0; 944 pNode->bEof = 1; 945 return rc; 946 } 947 }else{ 948 Fts5IndexIter *pIter = pPhrase->aTerm[j].pIter; 949 if( pIter->iRowid==iLast || pIter->bEof ) continue; 950 bMatch = 0; 951 if( fts5ExprAdvanceto(pIter, bDesc, &iLast, &rc, &pNode->bEof) ){ 952 return rc; 953 } 954 } 955 } 956 } 957 }while( bMatch==0 ); 958 959 pNode->iRowid = iLast; 960 pNode->bNomatch = ((0==fts5ExprNearTest(&rc, pExpr, pNode)) && rc==SQLITE_OK); 961 assert( pNode->bEof==0 || pNode->bNomatch==0 ); 962 963 return rc; 964 } 965 966 /* 967 ** Advance the first term iterator in the first phrase of pNear. Set output 968 ** variable *pbEof to true if it reaches EOF or if an error occurs. 969 ** 970 ** Return SQLITE_OK if successful, or an SQLite error code if an error 971 ** occurs. 972 */ 973 static int fts5ExprNodeNext_STRING( 974 Fts5Expr *pExpr, /* Expression pPhrase belongs to */ 975 Fts5ExprNode *pNode, /* FTS5_STRING or FTS5_TERM node */ 976 int bFromValid, 977 i64 iFrom 978 ){ 979 Fts5ExprTerm *pTerm = &pNode->pNear->apPhrase[0]->aTerm[0]; 980 int rc = SQLITE_OK; 981 982 pNode->bNomatch = 0; 983 if( pTerm->pSynonym ){ 984 int bEof = 1; 985 Fts5ExprTerm *p; 986 987 /* Find the firstest rowid any synonym points to. */ 988 i64 iRowid = fts5ExprSynonymRowid(pTerm, pExpr->bDesc, 0); 989 990 /* Advance each iterator that currently points to iRowid. Or, if iFrom 991 ** is valid - each iterator that points to a rowid before iFrom. */ 992 for(p=pTerm; p; p=p->pSynonym){ 993 if( sqlite3Fts5IterEof(p->pIter)==0 ){ 994 i64 ii = p->pIter->iRowid; 995 if( ii==iRowid 996 || (bFromValid && ii!=iFrom && (ii>iFrom)==pExpr->bDesc) 997 ){ 998 if( bFromValid ){ 999 rc = sqlite3Fts5IterNextFrom(p->pIter, iFrom); 1000 }else{ 1001 rc = sqlite3Fts5IterNext(p->pIter); 1002 } 1003 if( rc!=SQLITE_OK ) break; 1004 if( sqlite3Fts5IterEof(p->pIter)==0 ){ 1005 bEof = 0; 1006 } 1007 }else{ 1008 bEof = 0; 1009 } 1010 } 1011 } 1012 1013 /* Set the EOF flag if either all synonym iterators are at EOF or an 1014 ** error has occurred. */ 1015 pNode->bEof = (rc || bEof); 1016 }else{ 1017 Fts5IndexIter *pIter = pTerm->pIter; 1018 1019 assert( Fts5NodeIsString(pNode) ); 1020 if( bFromValid ){ 1021 rc = sqlite3Fts5IterNextFrom(pIter, iFrom); 1022 }else{ 1023 rc = sqlite3Fts5IterNext(pIter); 1024 } 1025 1026 pNode->bEof = (rc || sqlite3Fts5IterEof(pIter)); 1027 } 1028 1029 if( pNode->bEof==0 ){ 1030 assert( rc==SQLITE_OK ); 1031 rc = fts5ExprNodeTest_STRING(pExpr, pNode); 1032 } 1033 1034 return rc; 1035 } 1036 1037 1038 static int fts5ExprNodeTest_TERM( 1039 Fts5Expr *pExpr, /* Expression that pNear is a part of */ 1040 Fts5ExprNode *pNode /* The "NEAR" node (FTS5_TERM) */ 1041 ){ 1042 /* As this "NEAR" object is actually a single phrase that consists 1043 ** of a single term only, grab pointers into the poslist managed by the 1044 ** fts5_index.c iterator object. This is much faster than synthesizing 1045 ** a new poslist the way we have to for more complicated phrase or NEAR 1046 ** expressions. */ 1047 Fts5ExprPhrase *pPhrase = pNode->pNear->apPhrase[0]; 1048 Fts5IndexIter *pIter = pPhrase->aTerm[0].pIter; 1049 1050 assert( pNode->eType==FTS5_TERM ); 1051 assert( pNode->pNear->nPhrase==1 && pPhrase->nTerm==1 ); 1052 assert( pPhrase->aTerm[0].pSynonym==0 ); 1053 1054 pPhrase->poslist.n = pIter->nData; 1055 if( pExpr->pConfig->eDetail==FTS5_DETAIL_FULL ){ 1056 pPhrase->poslist.p = (u8*)pIter->pData; 1057 } 1058 pNode->iRowid = pIter->iRowid; 1059 pNode->bNomatch = (pPhrase->poslist.n==0); 1060 return SQLITE_OK; 1061 } 1062 1063 /* 1064 ** xNext() method for a node of type FTS5_TERM. 1065 */ 1066 static int fts5ExprNodeNext_TERM( 1067 Fts5Expr *pExpr, 1068 Fts5ExprNode *pNode, 1069 int bFromValid, 1070 i64 iFrom 1071 ){ 1072 int rc; 1073 Fts5IndexIter *pIter = pNode->pNear->apPhrase[0]->aTerm[0].pIter; 1074 1075 assert( pNode->bEof==0 ); 1076 if( bFromValid ){ 1077 rc = sqlite3Fts5IterNextFrom(pIter, iFrom); 1078 }else{ 1079 rc = sqlite3Fts5IterNext(pIter); 1080 } 1081 if( rc==SQLITE_OK && sqlite3Fts5IterEof(pIter)==0 ){ 1082 rc = fts5ExprNodeTest_TERM(pExpr, pNode); 1083 }else{ 1084 pNode->bEof = 1; 1085 pNode->bNomatch = 0; 1086 } 1087 return rc; 1088 } 1089 1090 static void fts5ExprNodeTest_OR( 1091 Fts5Expr *pExpr, /* Expression of which pNode is a part */ 1092 Fts5ExprNode *pNode /* Expression node to test */ 1093 ){ 1094 Fts5ExprNode *pNext = pNode->apChild[0]; 1095 int i; 1096 1097 for(i=1; i<pNode->nChild; i++){ 1098 Fts5ExprNode *pChild = pNode->apChild[i]; 1099 int cmp = fts5NodeCompare(pExpr, pNext, pChild); 1100 if( cmp>0 || (cmp==0 && pChild->bNomatch==0) ){ 1101 pNext = pChild; 1102 } 1103 } 1104 pNode->iRowid = pNext->iRowid; 1105 pNode->bEof = pNext->bEof; 1106 pNode->bNomatch = pNext->bNomatch; 1107 } 1108 1109 static int fts5ExprNodeNext_OR( 1110 Fts5Expr *pExpr, 1111 Fts5ExprNode *pNode, 1112 int bFromValid, 1113 i64 iFrom 1114 ){ 1115 int i; 1116 i64 iLast = pNode->iRowid; 1117 1118 for(i=0; i<pNode->nChild; i++){ 1119 Fts5ExprNode *p1 = pNode->apChild[i]; 1120 assert( p1->bEof || fts5RowidCmp(pExpr, p1->iRowid, iLast)>=0 ); 1121 if( p1->bEof==0 ){ 1122 if( (p1->iRowid==iLast) 1123 || (bFromValid && fts5RowidCmp(pExpr, p1->iRowid, iFrom)<0) 1124 ){ 1125 int rc = fts5ExprNodeNext(pExpr, p1, bFromValid, iFrom); 1126 if( rc!=SQLITE_OK ){ 1127 pNode->bNomatch = 0; 1128 return rc; 1129 } 1130 } 1131 } 1132 } 1133 1134 fts5ExprNodeTest_OR(pExpr, pNode); 1135 return SQLITE_OK; 1136 } 1137 1138 /* 1139 ** Argument pNode is an FTS5_AND node. 1140 */ 1141 static int fts5ExprNodeTest_AND( 1142 Fts5Expr *pExpr, /* Expression pPhrase belongs to */ 1143 Fts5ExprNode *pAnd /* FTS5_AND node to advance */ 1144 ){ 1145 int iChild; 1146 i64 iLast = pAnd->iRowid; 1147 int rc = SQLITE_OK; 1148 int bMatch; 1149 1150 assert( pAnd->bEof==0 ); 1151 do { 1152 pAnd->bNomatch = 0; 1153 bMatch = 1; 1154 for(iChild=0; iChild<pAnd->nChild; iChild++){ 1155 Fts5ExprNode *pChild = pAnd->apChild[iChild]; 1156 int cmp = fts5RowidCmp(pExpr, iLast, pChild->iRowid); 1157 if( cmp>0 ){ 1158 /* Advance pChild until it points to iLast or laster */ 1159 rc = fts5ExprNodeNext(pExpr, pChild, 1, iLast); 1160 if( rc!=SQLITE_OK ){ 1161 pAnd->bNomatch = 0; 1162 return rc; 1163 } 1164 } 1165 1166 /* If the child node is now at EOF, so is the parent AND node. Otherwise, 1167 ** the child node is guaranteed to have advanced at least as far as 1168 ** rowid iLast. So if it is not at exactly iLast, pChild->iRowid is the 1169 ** new lastest rowid seen so far. */ 1170 assert( pChild->bEof || fts5RowidCmp(pExpr, iLast, pChild->iRowid)<=0 ); 1171 if( pChild->bEof ){ 1172 fts5ExprSetEof(pAnd); 1173 bMatch = 1; 1174 break; 1175 }else if( iLast!=pChild->iRowid ){ 1176 bMatch = 0; 1177 iLast = pChild->iRowid; 1178 } 1179 1180 if( pChild->bNomatch ){ 1181 pAnd->bNomatch = 1; 1182 } 1183 } 1184 }while( bMatch==0 ); 1185 1186 if( pAnd->bNomatch && pAnd!=pExpr->pRoot ){ 1187 fts5ExprNodeZeroPoslist(pAnd); 1188 } 1189 pAnd->iRowid = iLast; 1190 return SQLITE_OK; 1191 } 1192 1193 static int fts5ExprNodeNext_AND( 1194 Fts5Expr *pExpr, 1195 Fts5ExprNode *pNode, 1196 int bFromValid, 1197 i64 iFrom 1198 ){ 1199 int rc = fts5ExprNodeNext(pExpr, pNode->apChild[0], bFromValid, iFrom); 1200 if( rc==SQLITE_OK ){ 1201 rc = fts5ExprNodeTest_AND(pExpr, pNode); 1202 }else{ 1203 pNode->bNomatch = 0; 1204 } 1205 return rc; 1206 } 1207 1208 static int fts5ExprNodeTest_NOT( 1209 Fts5Expr *pExpr, /* Expression pPhrase belongs to */ 1210 Fts5ExprNode *pNode /* FTS5_NOT node to advance */ 1211 ){ 1212 int rc = SQLITE_OK; 1213 Fts5ExprNode *p1 = pNode->apChild[0]; 1214 Fts5ExprNode *p2 = pNode->apChild[1]; 1215 assert( pNode->nChild==2 ); 1216 1217 while( rc==SQLITE_OK && p1->bEof==0 ){ 1218 int cmp = fts5NodeCompare(pExpr, p1, p2); 1219 if( cmp>0 ){ 1220 rc = fts5ExprNodeNext(pExpr, p2, 1, p1->iRowid); 1221 cmp = fts5NodeCompare(pExpr, p1, p2); 1222 } 1223 assert( rc!=SQLITE_OK || cmp<=0 ); 1224 if( cmp || p2->bNomatch ) break; 1225 rc = fts5ExprNodeNext(pExpr, p1, 0, 0); 1226 } 1227 pNode->bEof = p1->bEof; 1228 pNode->bNomatch = p1->bNomatch; 1229 pNode->iRowid = p1->iRowid; 1230 if( p1->bEof ){ 1231 fts5ExprNodeZeroPoslist(p2); 1232 } 1233 return rc; 1234 } 1235 1236 static int fts5ExprNodeNext_NOT( 1237 Fts5Expr *pExpr, 1238 Fts5ExprNode *pNode, 1239 int bFromValid, 1240 i64 iFrom 1241 ){ 1242 int rc = fts5ExprNodeNext(pExpr, pNode->apChild[0], bFromValid, iFrom); 1243 if( rc==SQLITE_OK ){ 1244 rc = fts5ExprNodeTest_NOT(pExpr, pNode); 1245 } 1246 if( rc!=SQLITE_OK ){ 1247 pNode->bNomatch = 0; 1248 } 1249 return rc; 1250 } 1251 1252 /* 1253 ** If pNode currently points to a match, this function returns SQLITE_OK 1254 ** without modifying it. Otherwise, pNode is advanced until it does point 1255 ** to a match or EOF is reached. 1256 */ 1257 static int fts5ExprNodeTest( 1258 Fts5Expr *pExpr, /* Expression of which pNode is a part */ 1259 Fts5ExprNode *pNode /* Expression node to test */ 1260 ){ 1261 int rc = SQLITE_OK; 1262 if( pNode->bEof==0 ){ 1263 switch( pNode->eType ){ 1264 1265 case FTS5_STRING: { 1266 rc = fts5ExprNodeTest_STRING(pExpr, pNode); 1267 break; 1268 } 1269 1270 case FTS5_TERM: { 1271 rc = fts5ExprNodeTest_TERM(pExpr, pNode); 1272 break; 1273 } 1274 1275 case FTS5_AND: { 1276 rc = fts5ExprNodeTest_AND(pExpr, pNode); 1277 break; 1278 } 1279 1280 case FTS5_OR: { 1281 fts5ExprNodeTest_OR(pExpr, pNode); 1282 break; 1283 } 1284 1285 default: assert( pNode->eType==FTS5_NOT ); { 1286 rc = fts5ExprNodeTest_NOT(pExpr, pNode); 1287 break; 1288 } 1289 } 1290 } 1291 return rc; 1292 } 1293 1294 1295 /* 1296 ** Set node pNode, which is part of expression pExpr, to point to the first 1297 ** match. If there are no matches, set the Node.bEof flag to indicate EOF. 1298 ** 1299 ** Return an SQLite error code if an error occurs, or SQLITE_OK otherwise. 1300 ** It is not an error if there are no matches. 1301 */ 1302 static int fts5ExprNodeFirst(Fts5Expr *pExpr, Fts5ExprNode *pNode){ 1303 int rc = SQLITE_OK; 1304 pNode->bEof = 0; 1305 pNode->bNomatch = 0; 1306 1307 if( Fts5NodeIsString(pNode) ){ 1308 /* Initialize all term iterators in the NEAR object. */ 1309 rc = fts5ExprNearInitAll(pExpr, pNode); 1310 }else if( pNode->xNext==0 ){ 1311 pNode->bEof = 1; 1312 }else{ 1313 int i; 1314 int nEof = 0; 1315 for(i=0; i<pNode->nChild && rc==SQLITE_OK; i++){ 1316 Fts5ExprNode *pChild = pNode->apChild[i]; 1317 rc = fts5ExprNodeFirst(pExpr, pNode->apChild[i]); 1318 assert( pChild->bEof==0 || pChild->bEof==1 ); 1319 nEof += pChild->bEof; 1320 } 1321 pNode->iRowid = pNode->apChild[0]->iRowid; 1322 1323 switch( pNode->eType ){ 1324 case FTS5_AND: 1325 if( nEof>0 ) fts5ExprSetEof(pNode); 1326 break; 1327 1328 case FTS5_OR: 1329 if( pNode->nChild==nEof ) fts5ExprSetEof(pNode); 1330 break; 1331 1332 default: 1333 assert( pNode->eType==FTS5_NOT ); 1334 pNode->bEof = pNode->apChild[0]->bEof; 1335 break; 1336 } 1337 } 1338 1339 if( rc==SQLITE_OK ){ 1340 rc = fts5ExprNodeTest(pExpr, pNode); 1341 } 1342 return rc; 1343 } 1344 1345 1346 /* 1347 ** Begin iterating through the set of documents in index pIdx matched by 1348 ** the MATCH expression passed as the first argument. If the "bDesc" 1349 ** parameter is passed a non-zero value, iteration is in descending rowid 1350 ** order. Or, if it is zero, in ascending order. 1351 ** 1352 ** If iterating in ascending rowid order (bDesc==0), the first document 1353 ** visited is that with the smallest rowid that is larger than or equal 1354 ** to parameter iFirst. Or, if iterating in ascending order (bDesc==1), 1355 ** then the first document visited must have a rowid smaller than or 1356 ** equal to iFirst. 1357 ** 1358 ** Return SQLITE_OK if successful, or an SQLite error code otherwise. It 1359 ** is not considered an error if the query does not match any documents. 1360 */ 1361 int sqlite3Fts5ExprFirst(Fts5Expr *p, Fts5Index *pIdx, i64 iFirst, int bDesc){ 1362 Fts5ExprNode *pRoot = p->pRoot; 1363 int rc; /* Return code */ 1364 1365 p->pIndex = pIdx; 1366 p->bDesc = bDesc; 1367 rc = fts5ExprNodeFirst(p, pRoot); 1368 1369 /* If not at EOF but the current rowid occurs earlier than iFirst in 1370 ** the iteration order, move to document iFirst or later. */ 1371 if( rc==SQLITE_OK 1372 && 0==pRoot->bEof 1373 && fts5RowidCmp(p, pRoot->iRowid, iFirst)<0 1374 ){ 1375 rc = fts5ExprNodeNext(p, pRoot, 1, iFirst); 1376 } 1377 1378 /* If the iterator is not at a real match, skip forward until it is. */ 1379 while( pRoot->bNomatch ){ 1380 assert( pRoot->bEof==0 && rc==SQLITE_OK ); 1381 rc = fts5ExprNodeNext(p, pRoot, 0, 0); 1382 } 1383 return rc; 1384 } 1385 1386 /* 1387 ** Move to the next document 1388 ** 1389 ** Return SQLITE_OK if successful, or an SQLite error code otherwise. It 1390 ** is not considered an error if the query does not match any documents. 1391 */ 1392 int sqlite3Fts5ExprNext(Fts5Expr *p, i64 iLast){ 1393 int rc; 1394 Fts5ExprNode *pRoot = p->pRoot; 1395 assert( pRoot->bEof==0 && pRoot->bNomatch==0 ); 1396 do { 1397 rc = fts5ExprNodeNext(p, pRoot, 0, 0); 1398 assert( pRoot->bNomatch==0 || (rc==SQLITE_OK && pRoot->bEof==0) ); 1399 }while( pRoot->bNomatch ); 1400 if( fts5RowidCmp(p, pRoot->iRowid, iLast)>0 ){ 1401 pRoot->bEof = 1; 1402 } 1403 return rc; 1404 } 1405 1406 int sqlite3Fts5ExprEof(Fts5Expr *p){ 1407 return p->pRoot->bEof; 1408 } 1409 1410 i64 sqlite3Fts5ExprRowid(Fts5Expr *p){ 1411 return p->pRoot->iRowid; 1412 } 1413 1414 static int fts5ParseStringFromToken(Fts5Token *pToken, char **pz){ 1415 int rc = SQLITE_OK; 1416 *pz = sqlite3Fts5Strndup(&rc, pToken->p, pToken->n); 1417 return rc; 1418 } 1419 1420 /* 1421 ** Free the phrase object passed as the only argument. 1422 */ 1423 static void fts5ExprPhraseFree(Fts5ExprPhrase *pPhrase){ 1424 if( pPhrase ){ 1425 int i; 1426 for(i=0; i<pPhrase->nTerm; i++){ 1427 Fts5ExprTerm *pSyn; 1428 Fts5ExprTerm *pNext; 1429 Fts5ExprTerm *pTerm = &pPhrase->aTerm[i]; 1430 sqlite3_free(pTerm->zTerm); 1431 sqlite3Fts5IterClose(pTerm->pIter); 1432 for(pSyn=pTerm->pSynonym; pSyn; pSyn=pNext){ 1433 pNext = pSyn->pSynonym; 1434 sqlite3Fts5IterClose(pSyn->pIter); 1435 fts5BufferFree((Fts5Buffer*)&pSyn[1]); 1436 sqlite3_free(pSyn); 1437 } 1438 } 1439 if( pPhrase->poslist.nSpace>0 ) fts5BufferFree(&pPhrase->poslist); 1440 sqlite3_free(pPhrase); 1441 } 1442 } 1443 1444 /* 1445 ** If argument pNear is NULL, then a new Fts5ExprNearset object is allocated 1446 ** and populated with pPhrase. Or, if pNear is not NULL, phrase pPhrase is 1447 ** appended to it and the results returned. 1448 ** 1449 ** If an OOM error occurs, both the pNear and pPhrase objects are freed and 1450 ** NULL returned. 1451 */ 1452 Fts5ExprNearset *sqlite3Fts5ParseNearset( 1453 Fts5Parse *pParse, /* Parse context */ 1454 Fts5ExprNearset *pNear, /* Existing nearset, or NULL */ 1455 Fts5ExprPhrase *pPhrase /* Recently parsed phrase */ 1456 ){ 1457 const int SZALLOC = 8; 1458 Fts5ExprNearset *pRet = 0; 1459 1460 if( pParse->rc==SQLITE_OK ){ 1461 if( pPhrase==0 ){ 1462 return pNear; 1463 } 1464 if( pNear==0 ){ 1465 int nByte = sizeof(Fts5ExprNearset) + SZALLOC * sizeof(Fts5ExprPhrase*); 1466 pRet = sqlite3_malloc(nByte); 1467 if( pRet==0 ){ 1468 pParse->rc = SQLITE_NOMEM; 1469 }else{ 1470 memset(pRet, 0, nByte); 1471 } 1472 }else if( (pNear->nPhrase % SZALLOC)==0 ){ 1473 int nNew = pNear->nPhrase + SZALLOC; 1474 int nByte = sizeof(Fts5ExprNearset) + nNew * sizeof(Fts5ExprPhrase*); 1475 1476 pRet = (Fts5ExprNearset*)sqlite3_realloc(pNear, nByte); 1477 if( pRet==0 ){ 1478 pParse->rc = SQLITE_NOMEM; 1479 } 1480 }else{ 1481 pRet = pNear; 1482 } 1483 } 1484 1485 if( pRet==0 ){ 1486 assert( pParse->rc!=SQLITE_OK ); 1487 sqlite3Fts5ParseNearsetFree(pNear); 1488 sqlite3Fts5ParsePhraseFree(pPhrase); 1489 }else{ 1490 if( pRet->nPhrase>0 ){ 1491 Fts5ExprPhrase *pLast = pRet->apPhrase[pRet->nPhrase-1]; 1492 assert( pLast==pParse->apPhrase[pParse->nPhrase-2] ); 1493 if( pPhrase->nTerm==0 ){ 1494 fts5ExprPhraseFree(pPhrase); 1495 pRet->nPhrase--; 1496 pParse->nPhrase--; 1497 pPhrase = pLast; 1498 }else if( pLast->nTerm==0 ){ 1499 fts5ExprPhraseFree(pLast); 1500 pParse->apPhrase[pParse->nPhrase-2] = pPhrase; 1501 pParse->nPhrase--; 1502 pRet->nPhrase--; 1503 } 1504 } 1505 pRet->apPhrase[pRet->nPhrase++] = pPhrase; 1506 } 1507 return pRet; 1508 } 1509 1510 typedef struct TokenCtx TokenCtx; 1511 struct TokenCtx { 1512 Fts5ExprPhrase *pPhrase; 1513 int rc; 1514 }; 1515 1516 /* 1517 ** Callback for tokenizing terms used by ParseTerm(). 1518 */ 1519 static int fts5ParseTokenize( 1520 void *pContext, /* Pointer to Fts5InsertCtx object */ 1521 int tflags, /* Mask of FTS5_TOKEN_* flags */ 1522 const char *pToken, /* Buffer containing token */ 1523 int nToken, /* Size of token in bytes */ 1524 int iUnused1, /* Start offset of token */ 1525 int iUnused2 /* End offset of token */ 1526 ){ 1527 int rc = SQLITE_OK; 1528 const int SZALLOC = 8; 1529 TokenCtx *pCtx = (TokenCtx*)pContext; 1530 Fts5ExprPhrase *pPhrase = pCtx->pPhrase; 1531 1532 UNUSED_PARAM2(iUnused1, iUnused2); 1533 1534 /* If an error has already occurred, this is a no-op */ 1535 if( pCtx->rc!=SQLITE_OK ) return pCtx->rc; 1536 if( nToken>FTS5_MAX_TOKEN_SIZE ) nToken = FTS5_MAX_TOKEN_SIZE; 1537 1538 if( pPhrase && pPhrase->nTerm>0 && (tflags & FTS5_TOKEN_COLOCATED) ){ 1539 Fts5ExprTerm *pSyn; 1540 int nByte = sizeof(Fts5ExprTerm) + sizeof(Fts5Buffer) + nToken+1; 1541 pSyn = (Fts5ExprTerm*)sqlite3_malloc(nByte); 1542 if( pSyn==0 ){ 1543 rc = SQLITE_NOMEM; 1544 }else{ 1545 memset(pSyn, 0, nByte); 1546 pSyn->zTerm = ((char*)pSyn) + sizeof(Fts5ExprTerm) + sizeof(Fts5Buffer); 1547 memcpy(pSyn->zTerm, pToken, nToken); 1548 pSyn->pSynonym = pPhrase->aTerm[pPhrase->nTerm-1].pSynonym; 1549 pPhrase->aTerm[pPhrase->nTerm-1].pSynonym = pSyn; 1550 } 1551 }else{ 1552 Fts5ExprTerm *pTerm; 1553 if( pPhrase==0 || (pPhrase->nTerm % SZALLOC)==0 ){ 1554 Fts5ExprPhrase *pNew; 1555 int nNew = SZALLOC + (pPhrase ? pPhrase->nTerm : 0); 1556 1557 pNew = (Fts5ExprPhrase*)sqlite3_realloc(pPhrase, 1558 sizeof(Fts5ExprPhrase) + sizeof(Fts5ExprTerm) * nNew 1559 ); 1560 if( pNew==0 ){ 1561 rc = SQLITE_NOMEM; 1562 }else{ 1563 if( pPhrase==0 ) memset(pNew, 0, sizeof(Fts5ExprPhrase)); 1564 pCtx->pPhrase = pPhrase = pNew; 1565 pNew->nTerm = nNew - SZALLOC; 1566 } 1567 } 1568 1569 if( rc==SQLITE_OK ){ 1570 pTerm = &pPhrase->aTerm[pPhrase->nTerm++]; 1571 memset(pTerm, 0, sizeof(Fts5ExprTerm)); 1572 pTerm->zTerm = sqlite3Fts5Strndup(&rc, pToken, nToken); 1573 } 1574 } 1575 1576 pCtx->rc = rc; 1577 return rc; 1578 } 1579 1580 1581 /* 1582 ** Free the phrase object passed as the only argument. 1583 */ 1584 void sqlite3Fts5ParsePhraseFree(Fts5ExprPhrase *pPhrase){ 1585 fts5ExprPhraseFree(pPhrase); 1586 } 1587 1588 /* 1589 ** Free the phrase object passed as the second argument. 1590 */ 1591 void sqlite3Fts5ParseNearsetFree(Fts5ExprNearset *pNear){ 1592 if( pNear ){ 1593 int i; 1594 for(i=0; i<pNear->nPhrase; i++){ 1595 fts5ExprPhraseFree(pNear->apPhrase[i]); 1596 } 1597 sqlite3_free(pNear->pColset); 1598 sqlite3_free(pNear); 1599 } 1600 } 1601 1602 void sqlite3Fts5ParseFinished(Fts5Parse *pParse, Fts5ExprNode *p){ 1603 assert( pParse->pExpr==0 ); 1604 pParse->pExpr = p; 1605 } 1606 1607 /* 1608 ** This function is called by the parser to process a string token. The 1609 ** string may or may not be quoted. In any case it is tokenized and a 1610 ** phrase object consisting of all tokens returned. 1611 */ 1612 Fts5ExprPhrase *sqlite3Fts5ParseTerm( 1613 Fts5Parse *pParse, /* Parse context */ 1614 Fts5ExprPhrase *pAppend, /* Phrase to append to */ 1615 Fts5Token *pToken, /* String to tokenize */ 1616 int bPrefix /* True if there is a trailing "*" */ 1617 ){ 1618 Fts5Config *pConfig = pParse->pConfig; 1619 TokenCtx sCtx; /* Context object passed to callback */ 1620 int rc; /* Tokenize return code */ 1621 char *z = 0; 1622 1623 memset(&sCtx, 0, sizeof(TokenCtx)); 1624 sCtx.pPhrase = pAppend; 1625 1626 rc = fts5ParseStringFromToken(pToken, &z); 1627 if( rc==SQLITE_OK ){ 1628 int flags = FTS5_TOKENIZE_QUERY | (bPrefix ? FTS5_TOKENIZE_PREFIX : 0); 1629 int n; 1630 sqlite3Fts5Dequote(z); 1631 n = (int)strlen(z); 1632 rc = sqlite3Fts5Tokenize(pConfig, flags, z, n, &sCtx, fts5ParseTokenize); 1633 } 1634 sqlite3_free(z); 1635 if( rc || (rc = sCtx.rc) ){ 1636 pParse->rc = rc; 1637 fts5ExprPhraseFree(sCtx.pPhrase); 1638 sCtx.pPhrase = 0; 1639 }else{ 1640 1641 if( pAppend==0 ){ 1642 if( (pParse->nPhrase % 8)==0 ){ 1643 int nByte = sizeof(Fts5ExprPhrase*) * (pParse->nPhrase + 8); 1644 Fts5ExprPhrase **apNew; 1645 apNew = (Fts5ExprPhrase**)sqlite3_realloc(pParse->apPhrase, nByte); 1646 if( apNew==0 ){ 1647 pParse->rc = SQLITE_NOMEM; 1648 fts5ExprPhraseFree(sCtx.pPhrase); 1649 return 0; 1650 } 1651 pParse->apPhrase = apNew; 1652 } 1653 pParse->nPhrase++; 1654 } 1655 1656 if( sCtx.pPhrase==0 ){ 1657 /* This happens when parsing a token or quoted phrase that contains 1658 ** no token characters at all. (e.g ... MATCH '""'). */ 1659 sCtx.pPhrase = sqlite3Fts5MallocZero(&pParse->rc, sizeof(Fts5ExprPhrase)); 1660 }else if( sCtx.pPhrase->nTerm ){ 1661 sCtx.pPhrase->aTerm[sCtx.pPhrase->nTerm-1].bPrefix = bPrefix; 1662 } 1663 pParse->apPhrase[pParse->nPhrase-1] = sCtx.pPhrase; 1664 } 1665 1666 return sCtx.pPhrase; 1667 } 1668 1669 /* 1670 ** Create a new FTS5 expression by cloning phrase iPhrase of the 1671 ** expression passed as the second argument. 1672 */ 1673 int sqlite3Fts5ExprClonePhrase( 1674 Fts5Expr *pExpr, 1675 int iPhrase, 1676 Fts5Expr **ppNew 1677 ){ 1678 int rc = SQLITE_OK; /* Return code */ 1679 Fts5ExprPhrase *pOrig; /* The phrase extracted from pExpr */ 1680 Fts5Expr *pNew = 0; /* Expression to return via *ppNew */ 1681 TokenCtx sCtx = {0,0}; /* Context object for fts5ParseTokenize */ 1682 1683 pOrig = pExpr->apExprPhrase[iPhrase]; 1684 pNew = (Fts5Expr*)sqlite3Fts5MallocZero(&rc, sizeof(Fts5Expr)); 1685 if( rc==SQLITE_OK ){ 1686 pNew->apExprPhrase = (Fts5ExprPhrase**)sqlite3Fts5MallocZero(&rc, 1687 sizeof(Fts5ExprPhrase*)); 1688 } 1689 if( rc==SQLITE_OK ){ 1690 pNew->pRoot = (Fts5ExprNode*)sqlite3Fts5MallocZero(&rc, 1691 sizeof(Fts5ExprNode)); 1692 } 1693 if( rc==SQLITE_OK ){ 1694 pNew->pRoot->pNear = (Fts5ExprNearset*)sqlite3Fts5MallocZero(&rc, 1695 sizeof(Fts5ExprNearset) + sizeof(Fts5ExprPhrase*)); 1696 } 1697 if( rc==SQLITE_OK ){ 1698 Fts5Colset *pColsetOrig = pOrig->pNode->pNear->pColset; 1699 if( pColsetOrig ){ 1700 int nByte = sizeof(Fts5Colset) + (pColsetOrig->nCol-1) * sizeof(int); 1701 Fts5Colset *pColset = (Fts5Colset*)sqlite3Fts5MallocZero(&rc, nByte); 1702 if( pColset ){ 1703 memcpy(pColset, pColsetOrig, nByte); 1704 } 1705 pNew->pRoot->pNear->pColset = pColset; 1706 } 1707 } 1708 1709 if( pOrig->nTerm ){ 1710 int i; /* Used to iterate through phrase terms */ 1711 for(i=0; rc==SQLITE_OK && i<pOrig->nTerm; i++){ 1712 int tflags = 0; 1713 Fts5ExprTerm *p; 1714 for(p=&pOrig->aTerm[i]; p && rc==SQLITE_OK; p=p->pSynonym){ 1715 const char *zTerm = p->zTerm; 1716 rc = fts5ParseTokenize((void*)&sCtx, tflags, zTerm, (int)strlen(zTerm), 1717 0, 0); 1718 tflags = FTS5_TOKEN_COLOCATED; 1719 } 1720 if( rc==SQLITE_OK ){ 1721 sCtx.pPhrase->aTerm[i].bPrefix = pOrig->aTerm[i].bPrefix; 1722 } 1723 } 1724 }else{ 1725 /* This happens when parsing a token or quoted phrase that contains 1726 ** no token characters at all. (e.g ... MATCH '""'). */ 1727 sCtx.pPhrase = sqlite3Fts5MallocZero(&rc, sizeof(Fts5ExprPhrase)); 1728 } 1729 1730 if( rc==SQLITE_OK ){ 1731 /* All the allocations succeeded. Put the expression object together. */ 1732 pNew->pIndex = pExpr->pIndex; 1733 pNew->pConfig = pExpr->pConfig; 1734 pNew->nPhrase = 1; 1735 pNew->apExprPhrase[0] = sCtx.pPhrase; 1736 pNew->pRoot->pNear->apPhrase[0] = sCtx.pPhrase; 1737 pNew->pRoot->pNear->nPhrase = 1; 1738 sCtx.pPhrase->pNode = pNew->pRoot; 1739 1740 if( pOrig->nTerm==1 && pOrig->aTerm[0].pSynonym==0 ){ 1741 pNew->pRoot->eType = FTS5_TERM; 1742 pNew->pRoot->xNext = fts5ExprNodeNext_TERM; 1743 }else{ 1744 pNew->pRoot->eType = FTS5_STRING; 1745 pNew->pRoot->xNext = fts5ExprNodeNext_STRING; 1746 } 1747 }else{ 1748 sqlite3Fts5ExprFree(pNew); 1749 fts5ExprPhraseFree(sCtx.pPhrase); 1750 pNew = 0; 1751 } 1752 1753 *ppNew = pNew; 1754 return rc; 1755 } 1756 1757 1758 /* 1759 ** Token pTok has appeared in a MATCH expression where the NEAR operator 1760 ** is expected. If token pTok does not contain "NEAR", store an error 1761 ** in the pParse object. 1762 */ 1763 void sqlite3Fts5ParseNear(Fts5Parse *pParse, Fts5Token *pTok){ 1764 if( pTok->n!=4 || memcmp("NEAR", pTok->p, 4) ){ 1765 sqlite3Fts5ParseError( 1766 pParse, "fts5: syntax error near \"%.*s\"", pTok->n, pTok->p 1767 ); 1768 } 1769 } 1770 1771 void sqlite3Fts5ParseSetDistance( 1772 Fts5Parse *pParse, 1773 Fts5ExprNearset *pNear, 1774 Fts5Token *p 1775 ){ 1776 if( pNear ){ 1777 int nNear = 0; 1778 int i; 1779 if( p->n ){ 1780 for(i=0; i<p->n; i++){ 1781 char c = (char)p->p[i]; 1782 if( c<'0' || c>'9' ){ 1783 sqlite3Fts5ParseError( 1784 pParse, "expected integer, got \"%.*s\"", p->n, p->p 1785 ); 1786 return; 1787 } 1788 nNear = nNear * 10 + (p->p[i] - '0'); 1789 } 1790 }else{ 1791 nNear = FTS5_DEFAULT_NEARDIST; 1792 } 1793 pNear->nNear = nNear; 1794 } 1795 } 1796 1797 /* 1798 ** The second argument passed to this function may be NULL, or it may be 1799 ** an existing Fts5Colset object. This function returns a pointer to 1800 ** a new colset object containing the contents of (p) with new value column 1801 ** number iCol appended. 1802 ** 1803 ** If an OOM error occurs, store an error code in pParse and return NULL. 1804 ** The old colset object (if any) is not freed in this case. 1805 */ 1806 static Fts5Colset *fts5ParseColset( 1807 Fts5Parse *pParse, /* Store SQLITE_NOMEM here if required */ 1808 Fts5Colset *p, /* Existing colset object */ 1809 int iCol /* New column to add to colset object */ 1810 ){ 1811 int nCol = p ? p->nCol : 0; /* Num. columns already in colset object */ 1812 Fts5Colset *pNew; /* New colset object to return */ 1813 1814 assert( pParse->rc==SQLITE_OK ); 1815 assert( iCol>=0 && iCol<pParse->pConfig->nCol ); 1816 1817 pNew = sqlite3_realloc(p, sizeof(Fts5Colset) + sizeof(int)*nCol); 1818 if( pNew==0 ){ 1819 pParse->rc = SQLITE_NOMEM; 1820 }else{ 1821 int *aiCol = pNew->aiCol; 1822 int i, j; 1823 for(i=0; i<nCol; i++){ 1824 if( aiCol[i]==iCol ) return pNew; 1825 if( aiCol[i]>iCol ) break; 1826 } 1827 for(j=nCol; j>i; j--){ 1828 aiCol[j] = aiCol[j-1]; 1829 } 1830 aiCol[i] = iCol; 1831 pNew->nCol = nCol+1; 1832 1833 #ifndef NDEBUG 1834 /* Check that the array is in order and contains no duplicate entries. */ 1835 for(i=1; i<pNew->nCol; i++) assert( pNew->aiCol[i]>pNew->aiCol[i-1] ); 1836 #endif 1837 } 1838 1839 return pNew; 1840 } 1841 1842 /* 1843 ** Allocate and return an Fts5Colset object specifying the inverse of 1844 ** the colset passed as the second argument. Free the colset passed 1845 ** as the second argument before returning. 1846 */ 1847 Fts5Colset *sqlite3Fts5ParseColsetInvert(Fts5Parse *pParse, Fts5Colset *p){ 1848 Fts5Colset *pRet; 1849 int nCol = pParse->pConfig->nCol; 1850 1851 pRet = (Fts5Colset*)sqlite3Fts5MallocZero(&pParse->rc, 1852 sizeof(Fts5Colset) + sizeof(int)*nCol 1853 ); 1854 if( pRet ){ 1855 int i; 1856 int iOld = 0; 1857 for(i=0; i<nCol; i++){ 1858 if( iOld>=p->nCol || p->aiCol[iOld]!=i ){ 1859 pRet->aiCol[pRet->nCol++] = i; 1860 }else{ 1861 iOld++; 1862 } 1863 } 1864 } 1865 1866 sqlite3_free(p); 1867 return pRet; 1868 } 1869 1870 Fts5Colset *sqlite3Fts5ParseColset( 1871 Fts5Parse *pParse, /* Store SQLITE_NOMEM here if required */ 1872 Fts5Colset *pColset, /* Existing colset object */ 1873 Fts5Token *p 1874 ){ 1875 Fts5Colset *pRet = 0; 1876 int iCol; 1877 char *z; /* Dequoted copy of token p */ 1878 1879 z = sqlite3Fts5Strndup(&pParse->rc, p->p, p->n); 1880 if( pParse->rc==SQLITE_OK ){ 1881 Fts5Config *pConfig = pParse->pConfig; 1882 sqlite3Fts5Dequote(z); 1883 for(iCol=0; iCol<pConfig->nCol; iCol++){ 1884 if( 0==sqlite3_stricmp(pConfig->azCol[iCol], z) ) break; 1885 } 1886 if( iCol==pConfig->nCol ){ 1887 sqlite3Fts5ParseError(pParse, "no such column: %s", z); 1888 }else{ 1889 pRet = fts5ParseColset(pParse, pColset, iCol); 1890 } 1891 sqlite3_free(z); 1892 } 1893 1894 if( pRet==0 ){ 1895 assert( pParse->rc!=SQLITE_OK ); 1896 sqlite3_free(pColset); 1897 } 1898 1899 return pRet; 1900 } 1901 1902 /* 1903 ** If argument pOrig is NULL, or if (*pRc) is set to anything other than 1904 ** SQLITE_OK when this function is called, NULL is returned. 1905 ** 1906 ** Otherwise, a copy of (*pOrig) is made into memory obtained from 1907 ** sqlite3Fts5MallocZero() and a pointer to it returned. If the allocation 1908 ** fails, (*pRc) is set to SQLITE_NOMEM and NULL is returned. 1909 */ 1910 static Fts5Colset *fts5CloneColset(int *pRc, Fts5Colset *pOrig){ 1911 Fts5Colset *pRet; 1912 if( pOrig ){ 1913 int nByte = sizeof(Fts5Colset) + (pOrig->nCol-1) * sizeof(int); 1914 pRet = (Fts5Colset*)sqlite3Fts5MallocZero(pRc, nByte); 1915 if( pRet ){ 1916 memcpy(pRet, pOrig, nByte); 1917 } 1918 }else{ 1919 pRet = 0; 1920 } 1921 return pRet; 1922 } 1923 1924 /* 1925 ** Remove from colset pColset any columns that are not also in colset pMerge. 1926 */ 1927 static void fts5MergeColset(Fts5Colset *pColset, Fts5Colset *pMerge){ 1928 int iIn = 0; /* Next input in pColset */ 1929 int iMerge = 0; /* Next input in pMerge */ 1930 int iOut = 0; /* Next output slot in pColset */ 1931 1932 while( iIn<pColset->nCol && iMerge<pMerge->nCol ){ 1933 int iDiff = pColset->aiCol[iIn] - pMerge->aiCol[iMerge]; 1934 if( iDiff==0 ){ 1935 pColset->aiCol[iOut++] = pMerge->aiCol[iMerge]; 1936 iMerge++; 1937 iIn++; 1938 }else if( iDiff>0 ){ 1939 iMerge++; 1940 }else{ 1941 iIn++; 1942 } 1943 } 1944 pColset->nCol = iOut; 1945 } 1946 1947 /* 1948 ** Recursively apply colset pColset to expression node pNode and all of 1949 ** its decendents. If (*ppFree) is not NULL, it contains a spare copy 1950 ** of pColset. This function may use the spare copy and set (*ppFree) to 1951 ** zero, or it may create copies of pColset using fts5CloneColset(). 1952 */ 1953 static void fts5ParseSetColset( 1954 Fts5Parse *pParse, 1955 Fts5ExprNode *pNode, 1956 Fts5Colset *pColset, 1957 Fts5Colset **ppFree 1958 ){ 1959 if( pParse->rc==SQLITE_OK ){ 1960 assert( pNode->eType==FTS5_TERM || pNode->eType==FTS5_STRING 1961 || pNode->eType==FTS5_AND || pNode->eType==FTS5_OR 1962 || pNode->eType==FTS5_NOT || pNode->eType==FTS5_EOF 1963 ); 1964 if( pNode->eType==FTS5_STRING || pNode->eType==FTS5_TERM ){ 1965 Fts5ExprNearset *pNear = pNode->pNear; 1966 if( pNear->pColset ){ 1967 fts5MergeColset(pNear->pColset, pColset); 1968 if( pNear->pColset->nCol==0 ){ 1969 pNode->eType = FTS5_EOF; 1970 pNode->xNext = 0; 1971 } 1972 }else if( *ppFree ){ 1973 pNear->pColset = pColset; 1974 *ppFree = 0; 1975 }else{ 1976 pNear->pColset = fts5CloneColset(&pParse->rc, pColset); 1977 } 1978 }else{ 1979 int i; 1980 assert( pNode->eType!=FTS5_EOF || pNode->nChild==0 ); 1981 for(i=0; i<pNode->nChild; i++){ 1982 fts5ParseSetColset(pParse, pNode->apChild[i], pColset, ppFree); 1983 } 1984 } 1985 } 1986 } 1987 1988 /* 1989 ** Apply colset pColset to expression node pExpr and all of its descendents. 1990 */ 1991 void sqlite3Fts5ParseSetColset( 1992 Fts5Parse *pParse, 1993 Fts5ExprNode *pExpr, 1994 Fts5Colset *pColset 1995 ){ 1996 Fts5Colset *pFree = pColset; 1997 if( pParse->pConfig->eDetail==FTS5_DETAIL_NONE ){ 1998 pParse->rc = SQLITE_ERROR; 1999 pParse->zErr = sqlite3_mprintf( 2000 "fts5: column queries are not supported (detail=none)" 2001 ); 2002 }else{ 2003 fts5ParseSetColset(pParse, pExpr, pColset, &pFree); 2004 } 2005 sqlite3_free(pFree); 2006 } 2007 2008 static void fts5ExprAssignXNext(Fts5ExprNode *pNode){ 2009 switch( pNode->eType ){ 2010 case FTS5_STRING: { 2011 Fts5ExprNearset *pNear = pNode->pNear; 2012 if( pNear->nPhrase==1 && pNear->apPhrase[0]->nTerm==1 2013 && pNear->apPhrase[0]->aTerm[0].pSynonym==0 2014 ){ 2015 pNode->eType = FTS5_TERM; 2016 pNode->xNext = fts5ExprNodeNext_TERM; 2017 }else{ 2018 pNode->xNext = fts5ExprNodeNext_STRING; 2019 } 2020 break; 2021 }; 2022 2023 case FTS5_OR: { 2024 pNode->xNext = fts5ExprNodeNext_OR; 2025 break; 2026 }; 2027 2028 case FTS5_AND: { 2029 pNode->xNext = fts5ExprNodeNext_AND; 2030 break; 2031 }; 2032 2033 default: assert( pNode->eType==FTS5_NOT ); { 2034 pNode->xNext = fts5ExprNodeNext_NOT; 2035 break; 2036 }; 2037 } 2038 } 2039 2040 static void fts5ExprAddChildren(Fts5ExprNode *p, Fts5ExprNode *pSub){ 2041 if( p->eType!=FTS5_NOT && pSub->eType==p->eType ){ 2042 int nByte = sizeof(Fts5ExprNode*) * pSub->nChild; 2043 memcpy(&p->apChild[p->nChild], pSub->apChild, nByte); 2044 p->nChild += pSub->nChild; 2045 sqlite3_free(pSub); 2046 }else{ 2047 p->apChild[p->nChild++] = pSub; 2048 } 2049 } 2050 2051 /* 2052 ** Allocate and return a new expression object. If anything goes wrong (i.e. 2053 ** OOM error), leave an error code in pParse and return NULL. 2054 */ 2055 Fts5ExprNode *sqlite3Fts5ParseNode( 2056 Fts5Parse *pParse, /* Parse context */ 2057 int eType, /* FTS5_STRING, AND, OR or NOT */ 2058 Fts5ExprNode *pLeft, /* Left hand child expression */ 2059 Fts5ExprNode *pRight, /* Right hand child expression */ 2060 Fts5ExprNearset *pNear /* For STRING expressions, the near cluster */ 2061 ){ 2062 Fts5ExprNode *pRet = 0; 2063 2064 if( pParse->rc==SQLITE_OK ){ 2065 int nChild = 0; /* Number of children of returned node */ 2066 int nByte; /* Bytes of space to allocate for this node */ 2067 2068 assert( (eType!=FTS5_STRING && !pNear) 2069 || (eType==FTS5_STRING && !pLeft && !pRight) 2070 ); 2071 if( eType==FTS5_STRING && pNear==0 ) return 0; 2072 if( eType!=FTS5_STRING && pLeft==0 ) return pRight; 2073 if( eType!=FTS5_STRING && pRight==0 ) return pLeft; 2074 2075 if( eType==FTS5_NOT ){ 2076 nChild = 2; 2077 }else if( eType==FTS5_AND || eType==FTS5_OR ){ 2078 nChild = 2; 2079 if( pLeft->eType==eType ) nChild += pLeft->nChild-1; 2080 if( pRight->eType==eType ) nChild += pRight->nChild-1; 2081 } 2082 2083 nByte = sizeof(Fts5ExprNode) + sizeof(Fts5ExprNode*)*(nChild-1); 2084 pRet = (Fts5ExprNode*)sqlite3Fts5MallocZero(&pParse->rc, nByte); 2085 2086 if( pRet ){ 2087 pRet->eType = eType; 2088 pRet->pNear = pNear; 2089 fts5ExprAssignXNext(pRet); 2090 if( eType==FTS5_STRING ){ 2091 int iPhrase; 2092 for(iPhrase=0; iPhrase<pNear->nPhrase; iPhrase++){ 2093 pNear->apPhrase[iPhrase]->pNode = pRet; 2094 if( pNear->apPhrase[iPhrase]->nTerm==0 ){ 2095 pRet->xNext = 0; 2096 pRet->eType = FTS5_EOF; 2097 } 2098 } 2099 2100 if( pParse->pConfig->eDetail!=FTS5_DETAIL_FULL 2101 && (pNear->nPhrase!=1 || pNear->apPhrase[0]->nTerm>1) 2102 ){ 2103 assert( pParse->rc==SQLITE_OK ); 2104 pParse->rc = SQLITE_ERROR; 2105 assert( pParse->zErr==0 ); 2106 pParse->zErr = sqlite3_mprintf( 2107 "fts5: %s queries are not supported (detail!=full)", 2108 pNear->nPhrase==1 ? "phrase": "NEAR" 2109 ); 2110 sqlite3_free(pRet); 2111 pRet = 0; 2112 } 2113 2114 }else{ 2115 fts5ExprAddChildren(pRet, pLeft); 2116 fts5ExprAddChildren(pRet, pRight); 2117 } 2118 } 2119 } 2120 2121 if( pRet==0 ){ 2122 assert( pParse->rc!=SQLITE_OK ); 2123 sqlite3Fts5ParseNodeFree(pLeft); 2124 sqlite3Fts5ParseNodeFree(pRight); 2125 sqlite3Fts5ParseNearsetFree(pNear); 2126 } 2127 return pRet; 2128 } 2129 2130 Fts5ExprNode *sqlite3Fts5ParseImplicitAnd( 2131 Fts5Parse *pParse, /* Parse context */ 2132 Fts5ExprNode *pLeft, /* Left hand child expression */ 2133 Fts5ExprNode *pRight /* Right hand child expression */ 2134 ){ 2135 Fts5ExprNode *pRet = 0; 2136 Fts5ExprNode *pPrev; 2137 2138 if( pParse->rc ){ 2139 sqlite3Fts5ParseNodeFree(pLeft); 2140 sqlite3Fts5ParseNodeFree(pRight); 2141 }else{ 2142 2143 assert( pLeft->eType==FTS5_STRING 2144 || pLeft->eType==FTS5_TERM 2145 || pLeft->eType==FTS5_EOF 2146 || pLeft->eType==FTS5_AND 2147 ); 2148 assert( pRight->eType==FTS5_STRING 2149 || pRight->eType==FTS5_TERM 2150 || pRight->eType==FTS5_EOF 2151 ); 2152 2153 if( pLeft->eType==FTS5_AND ){ 2154 pPrev = pLeft->apChild[pLeft->nChild-1]; 2155 }else{ 2156 pPrev = pLeft; 2157 } 2158 assert( pPrev->eType==FTS5_STRING 2159 || pPrev->eType==FTS5_TERM 2160 || pPrev->eType==FTS5_EOF 2161 ); 2162 2163 if( pRight->eType==FTS5_EOF ){ 2164 assert( pParse->apPhrase[pParse->nPhrase-1]==pRight->pNear->apPhrase[0] ); 2165 sqlite3Fts5ParseNodeFree(pRight); 2166 pRet = pLeft; 2167 pParse->nPhrase--; 2168 } 2169 else if( pPrev->eType==FTS5_EOF ){ 2170 Fts5ExprPhrase **ap; 2171 2172 if( pPrev==pLeft ){ 2173 pRet = pRight; 2174 }else{ 2175 pLeft->apChild[pLeft->nChild-1] = pRight; 2176 pRet = pLeft; 2177 } 2178 2179 ap = &pParse->apPhrase[pParse->nPhrase-1-pRight->pNear->nPhrase]; 2180 assert( ap[0]==pPrev->pNear->apPhrase[0] ); 2181 memmove(ap, &ap[1], sizeof(Fts5ExprPhrase*)*pRight->pNear->nPhrase); 2182 pParse->nPhrase--; 2183 2184 sqlite3Fts5ParseNodeFree(pPrev); 2185 } 2186 else{ 2187 pRet = sqlite3Fts5ParseNode(pParse, FTS5_AND, pLeft, pRight, 0); 2188 } 2189 } 2190 2191 return pRet; 2192 } 2193 2194 static char *fts5ExprTermPrint(Fts5ExprTerm *pTerm){ 2195 int nByte = 0; 2196 Fts5ExprTerm *p; 2197 char *zQuoted; 2198 2199 /* Determine the maximum amount of space required. */ 2200 for(p=pTerm; p; p=p->pSynonym){ 2201 nByte += (int)strlen(pTerm->zTerm) * 2 + 3 + 2; 2202 } 2203 zQuoted = sqlite3_malloc(nByte); 2204 2205 if( zQuoted ){ 2206 int i = 0; 2207 for(p=pTerm; p; p=p->pSynonym){ 2208 char *zIn = p->zTerm; 2209 zQuoted[i++] = '"'; 2210 while( *zIn ){ 2211 if( *zIn=='"' ) zQuoted[i++] = '"'; 2212 zQuoted[i++] = *zIn++; 2213 } 2214 zQuoted[i++] = '"'; 2215 if( p->pSynonym ) zQuoted[i++] = '|'; 2216 } 2217 if( pTerm->bPrefix ){ 2218 zQuoted[i++] = ' '; 2219 zQuoted[i++] = '*'; 2220 } 2221 zQuoted[i++] = '\0'; 2222 } 2223 return zQuoted; 2224 } 2225 2226 static char *fts5PrintfAppend(char *zApp, const char *zFmt, ...){ 2227 char *zNew; 2228 va_list ap; 2229 va_start(ap, zFmt); 2230 zNew = sqlite3_vmprintf(zFmt, ap); 2231 va_end(ap); 2232 if( zApp && zNew ){ 2233 char *zNew2 = sqlite3_mprintf("%s%s", zApp, zNew); 2234 sqlite3_free(zNew); 2235 zNew = zNew2; 2236 } 2237 sqlite3_free(zApp); 2238 return zNew; 2239 } 2240 2241 /* 2242 ** Compose a tcl-readable representation of expression pExpr. Return a 2243 ** pointer to a buffer containing that representation. It is the 2244 ** responsibility of the caller to at some point free the buffer using 2245 ** sqlite3_free(). 2246 */ 2247 static char *fts5ExprPrintTcl( 2248 Fts5Config *pConfig, 2249 const char *zNearsetCmd, 2250 Fts5ExprNode *pExpr 2251 ){ 2252 char *zRet = 0; 2253 if( pExpr->eType==FTS5_STRING || pExpr->eType==FTS5_TERM ){ 2254 Fts5ExprNearset *pNear = pExpr->pNear; 2255 int i; 2256 int iTerm; 2257 2258 zRet = fts5PrintfAppend(zRet, "%s ", zNearsetCmd); 2259 if( zRet==0 ) return 0; 2260 if( pNear->pColset ){ 2261 int *aiCol = pNear->pColset->aiCol; 2262 int nCol = pNear->pColset->nCol; 2263 if( nCol==1 ){ 2264 zRet = fts5PrintfAppend(zRet, "-col %d ", aiCol[0]); 2265 }else{ 2266 zRet = fts5PrintfAppend(zRet, "-col {%d", aiCol[0]); 2267 for(i=1; i<pNear->pColset->nCol; i++){ 2268 zRet = fts5PrintfAppend(zRet, " %d", aiCol[i]); 2269 } 2270 zRet = fts5PrintfAppend(zRet, "} "); 2271 } 2272 if( zRet==0 ) return 0; 2273 } 2274 2275 if( pNear->nPhrase>1 ){ 2276 zRet = fts5PrintfAppend(zRet, "-near %d ", pNear->nNear); 2277 if( zRet==0 ) return 0; 2278 } 2279 2280 zRet = fts5PrintfAppend(zRet, "--"); 2281 if( zRet==0 ) return 0; 2282 2283 for(i=0; i<pNear->nPhrase; i++){ 2284 Fts5ExprPhrase *pPhrase = pNear->apPhrase[i]; 2285 2286 zRet = fts5PrintfAppend(zRet, " {"); 2287 for(iTerm=0; zRet && iTerm<pPhrase->nTerm; iTerm++){ 2288 char *zTerm = pPhrase->aTerm[iTerm].zTerm; 2289 zRet = fts5PrintfAppend(zRet, "%s%s", iTerm==0?"":" ", zTerm); 2290 if( pPhrase->aTerm[iTerm].bPrefix ){ 2291 zRet = fts5PrintfAppend(zRet, "*"); 2292 } 2293 } 2294 2295 if( zRet ) zRet = fts5PrintfAppend(zRet, "}"); 2296 if( zRet==0 ) return 0; 2297 } 2298 2299 }else{ 2300 char const *zOp = 0; 2301 int i; 2302 switch( pExpr->eType ){ 2303 case FTS5_AND: zOp = "AND"; break; 2304 case FTS5_NOT: zOp = "NOT"; break; 2305 default: 2306 assert( pExpr->eType==FTS5_OR ); 2307 zOp = "OR"; 2308 break; 2309 } 2310 2311 zRet = sqlite3_mprintf("%s", zOp); 2312 for(i=0; zRet && i<pExpr->nChild; i++){ 2313 char *z = fts5ExprPrintTcl(pConfig, zNearsetCmd, pExpr->apChild[i]); 2314 if( !z ){ 2315 sqlite3_free(zRet); 2316 zRet = 0; 2317 }else{ 2318 zRet = fts5PrintfAppend(zRet, " [%z]", z); 2319 } 2320 } 2321 } 2322 2323 return zRet; 2324 } 2325 2326 static char *fts5ExprPrint(Fts5Config *pConfig, Fts5ExprNode *pExpr){ 2327 char *zRet = 0; 2328 if( pExpr->eType==0 ){ 2329 return sqlite3_mprintf("\"\""); 2330 }else 2331 if( pExpr->eType==FTS5_STRING || pExpr->eType==FTS5_TERM ){ 2332 Fts5ExprNearset *pNear = pExpr->pNear; 2333 int i; 2334 int iTerm; 2335 2336 if( pNear->pColset ){ 2337 int iCol = pNear->pColset->aiCol[0]; 2338 zRet = fts5PrintfAppend(zRet, "%s : ", pConfig->azCol[iCol]); 2339 if( zRet==0 ) return 0; 2340 } 2341 2342 if( pNear->nPhrase>1 ){ 2343 zRet = fts5PrintfAppend(zRet, "NEAR("); 2344 if( zRet==0 ) return 0; 2345 } 2346 2347 for(i=0; i<pNear->nPhrase; i++){ 2348 Fts5ExprPhrase *pPhrase = pNear->apPhrase[i]; 2349 if( i!=0 ){ 2350 zRet = fts5PrintfAppend(zRet, " "); 2351 if( zRet==0 ) return 0; 2352 } 2353 for(iTerm=0; iTerm<pPhrase->nTerm; iTerm++){ 2354 char *zTerm = fts5ExprTermPrint(&pPhrase->aTerm[iTerm]); 2355 if( zTerm ){ 2356 zRet = fts5PrintfAppend(zRet, "%s%s", iTerm==0?"":" + ", zTerm); 2357 sqlite3_free(zTerm); 2358 } 2359 if( zTerm==0 || zRet==0 ){ 2360 sqlite3_free(zRet); 2361 return 0; 2362 } 2363 } 2364 } 2365 2366 if( pNear->nPhrase>1 ){ 2367 zRet = fts5PrintfAppend(zRet, ", %d)", pNear->nNear); 2368 if( zRet==0 ) return 0; 2369 } 2370 2371 }else{ 2372 char const *zOp = 0; 2373 int i; 2374 2375 switch( pExpr->eType ){ 2376 case FTS5_AND: zOp = " AND "; break; 2377 case FTS5_NOT: zOp = " NOT "; break; 2378 default: 2379 assert( pExpr->eType==FTS5_OR ); 2380 zOp = " OR "; 2381 break; 2382 } 2383 2384 for(i=0; i<pExpr->nChild; i++){ 2385 char *z = fts5ExprPrint(pConfig, pExpr->apChild[i]); 2386 if( z==0 ){ 2387 sqlite3_free(zRet); 2388 zRet = 0; 2389 }else{ 2390 int e = pExpr->apChild[i]->eType; 2391 int b = (e!=FTS5_STRING && e!=FTS5_TERM && e!=FTS5_EOF); 2392 zRet = fts5PrintfAppend(zRet, "%s%s%z%s", 2393 (i==0 ? "" : zOp), 2394 (b?"(":""), z, (b?")":"") 2395 ); 2396 } 2397 if( zRet==0 ) break; 2398 } 2399 } 2400 2401 return zRet; 2402 } 2403 2404 /* 2405 ** The implementation of user-defined scalar functions fts5_expr() (bTcl==0) 2406 ** and fts5_expr_tcl() (bTcl!=0). 2407 */ 2408 static void fts5ExprFunction( 2409 sqlite3_context *pCtx, /* Function call context */ 2410 int nArg, /* Number of args */ 2411 sqlite3_value **apVal, /* Function arguments */ 2412 int bTcl 2413 ){ 2414 Fts5Global *pGlobal = (Fts5Global*)sqlite3_user_data(pCtx); 2415 sqlite3 *db = sqlite3_context_db_handle(pCtx); 2416 const char *zExpr = 0; 2417 char *zErr = 0; 2418 Fts5Expr *pExpr = 0; 2419 int rc; 2420 int i; 2421 2422 const char **azConfig; /* Array of arguments for Fts5Config */ 2423 const char *zNearsetCmd = "nearset"; 2424 int nConfig; /* Size of azConfig[] */ 2425 Fts5Config *pConfig = 0; 2426 int iArg = 1; 2427 2428 if( nArg<1 ){ 2429 zErr = sqlite3_mprintf("wrong number of arguments to function %s", 2430 bTcl ? "fts5_expr_tcl" : "fts5_expr" 2431 ); 2432 sqlite3_result_error(pCtx, zErr, -1); 2433 sqlite3_free(zErr); 2434 return; 2435 } 2436 2437 if( bTcl && nArg>1 ){ 2438 zNearsetCmd = (const char*)sqlite3_value_text(apVal[1]); 2439 iArg = 2; 2440 } 2441 2442 nConfig = 3 + (nArg-iArg); 2443 azConfig = (const char**)sqlite3_malloc(sizeof(char*) * nConfig); 2444 if( azConfig==0 ){ 2445 sqlite3_result_error_nomem(pCtx); 2446 return; 2447 } 2448 azConfig[0] = 0; 2449 azConfig[1] = "main"; 2450 azConfig[2] = "tbl"; 2451 for(i=3; iArg<nArg; iArg++){ 2452 azConfig[i++] = (const char*)sqlite3_value_text(apVal[iArg]); 2453 } 2454 2455 zExpr = (const char*)sqlite3_value_text(apVal[0]); 2456 2457 rc = sqlite3Fts5ConfigParse(pGlobal, db, nConfig, azConfig, &pConfig, &zErr); 2458 if( rc==SQLITE_OK ){ 2459 rc = sqlite3Fts5ExprNew(pConfig, pConfig->nCol, zExpr, &pExpr, &zErr); 2460 } 2461 if( rc==SQLITE_OK ){ 2462 char *zText; 2463 if( pExpr->pRoot->xNext==0 ){ 2464 zText = sqlite3_mprintf(""); 2465 }else if( bTcl ){ 2466 zText = fts5ExprPrintTcl(pConfig, zNearsetCmd, pExpr->pRoot); 2467 }else{ 2468 zText = fts5ExprPrint(pConfig, pExpr->pRoot); 2469 } 2470 if( zText==0 ){ 2471 rc = SQLITE_NOMEM; 2472 }else{ 2473 sqlite3_result_text(pCtx, zText, -1, SQLITE_TRANSIENT); 2474 sqlite3_free(zText); 2475 } 2476 } 2477 2478 if( rc!=SQLITE_OK ){ 2479 if( zErr ){ 2480 sqlite3_result_error(pCtx, zErr, -1); 2481 sqlite3_free(zErr); 2482 }else{ 2483 sqlite3_result_error_code(pCtx, rc); 2484 } 2485 } 2486 sqlite3_free((void *)azConfig); 2487 sqlite3Fts5ConfigFree(pConfig); 2488 sqlite3Fts5ExprFree(pExpr); 2489 } 2490 2491 static void fts5ExprFunctionHr( 2492 sqlite3_context *pCtx, /* Function call context */ 2493 int nArg, /* Number of args */ 2494 sqlite3_value **apVal /* Function arguments */ 2495 ){ 2496 fts5ExprFunction(pCtx, nArg, apVal, 0); 2497 } 2498 static void fts5ExprFunctionTcl( 2499 sqlite3_context *pCtx, /* Function call context */ 2500 int nArg, /* Number of args */ 2501 sqlite3_value **apVal /* Function arguments */ 2502 ){ 2503 fts5ExprFunction(pCtx, nArg, apVal, 1); 2504 } 2505 2506 /* 2507 ** The implementation of an SQLite user-defined-function that accepts a 2508 ** single integer as an argument. If the integer is an alpha-numeric 2509 ** unicode code point, 1 is returned. Otherwise 0. 2510 */ 2511 static void fts5ExprIsAlnum( 2512 sqlite3_context *pCtx, /* Function call context */ 2513 int nArg, /* Number of args */ 2514 sqlite3_value **apVal /* Function arguments */ 2515 ){ 2516 int iCode; 2517 if( nArg!=1 ){ 2518 sqlite3_result_error(pCtx, 2519 "wrong number of arguments to function fts5_isalnum", -1 2520 ); 2521 return; 2522 } 2523 iCode = sqlite3_value_int(apVal[0]); 2524 sqlite3_result_int(pCtx, sqlite3Fts5UnicodeIsalnum(iCode)); 2525 } 2526 2527 static void fts5ExprFold( 2528 sqlite3_context *pCtx, /* Function call context */ 2529 int nArg, /* Number of args */ 2530 sqlite3_value **apVal /* Function arguments */ 2531 ){ 2532 if( nArg!=1 && nArg!=2 ){ 2533 sqlite3_result_error(pCtx, 2534 "wrong number of arguments to function fts5_fold", -1 2535 ); 2536 }else{ 2537 int iCode; 2538 int bRemoveDiacritics = 0; 2539 iCode = sqlite3_value_int(apVal[0]); 2540 if( nArg==2 ) bRemoveDiacritics = sqlite3_value_int(apVal[1]); 2541 sqlite3_result_int(pCtx, sqlite3Fts5UnicodeFold(iCode, bRemoveDiacritics)); 2542 } 2543 } 2544 2545 /* 2546 ** This is called during initialization to register the fts5_expr() scalar 2547 ** UDF with the SQLite handle passed as the only argument. 2548 */ 2549 int sqlite3Fts5ExprInit(Fts5Global *pGlobal, sqlite3 *db){ 2550 struct Fts5ExprFunc { 2551 const char *z; 2552 void (*x)(sqlite3_context*,int,sqlite3_value**); 2553 } aFunc[] = { 2554 { "fts5_expr", fts5ExprFunctionHr }, 2555 { "fts5_expr_tcl", fts5ExprFunctionTcl }, 2556 { "fts5_isalnum", fts5ExprIsAlnum }, 2557 { "fts5_fold", fts5ExprFold }, 2558 }; 2559 int i; 2560 int rc = SQLITE_OK; 2561 void *pCtx = (void*)pGlobal; 2562 2563 for(i=0; rc==SQLITE_OK && i<ArraySize(aFunc); i++){ 2564 struct Fts5ExprFunc *p = &aFunc[i]; 2565 rc = sqlite3_create_function(db, p->z, -1, SQLITE_UTF8, pCtx, p->x, 0, 0); 2566 } 2567 2568 /* Avoid a warning indicating that sqlite3Fts5ParserTrace() is unused */ 2569 #ifndef NDEBUG 2570 (void)sqlite3Fts5ParserTrace; 2571 #endif 2572 2573 return rc; 2574 } 2575 2576 /* 2577 ** Return the number of phrases in expression pExpr. 2578 */ 2579 int sqlite3Fts5ExprPhraseCount(Fts5Expr *pExpr){ 2580 return (pExpr ? pExpr->nPhrase : 0); 2581 } 2582 2583 /* 2584 ** Return the number of terms in the iPhrase'th phrase in pExpr. 2585 */ 2586 int sqlite3Fts5ExprPhraseSize(Fts5Expr *pExpr, int iPhrase){ 2587 if( iPhrase<0 || iPhrase>=pExpr->nPhrase ) return 0; 2588 return pExpr->apExprPhrase[iPhrase]->nTerm; 2589 } 2590 2591 /* 2592 ** This function is used to access the current position list for phrase 2593 ** iPhrase. 2594 */ 2595 int sqlite3Fts5ExprPoslist(Fts5Expr *pExpr, int iPhrase, const u8 **pa){ 2596 int nRet; 2597 Fts5ExprPhrase *pPhrase = pExpr->apExprPhrase[iPhrase]; 2598 Fts5ExprNode *pNode = pPhrase->pNode; 2599 if( pNode->bEof==0 && pNode->iRowid==pExpr->pRoot->iRowid ){ 2600 *pa = pPhrase->poslist.p; 2601 nRet = pPhrase->poslist.n; 2602 }else{ 2603 *pa = 0; 2604 nRet = 0; 2605 } 2606 return nRet; 2607 } 2608 2609 struct Fts5PoslistPopulator { 2610 Fts5PoslistWriter writer; 2611 int bOk; /* True if ok to populate */ 2612 int bMiss; 2613 }; 2614 2615 Fts5PoslistPopulator *sqlite3Fts5ExprClearPoslists(Fts5Expr *pExpr, int bLive){ 2616 Fts5PoslistPopulator *pRet; 2617 pRet = sqlite3_malloc(sizeof(Fts5PoslistPopulator)*pExpr->nPhrase); 2618 if( pRet ){ 2619 int i; 2620 memset(pRet, 0, sizeof(Fts5PoslistPopulator)*pExpr->nPhrase); 2621 for(i=0; i<pExpr->nPhrase; i++){ 2622 Fts5Buffer *pBuf = &pExpr->apExprPhrase[i]->poslist; 2623 Fts5ExprNode *pNode = pExpr->apExprPhrase[i]->pNode; 2624 assert( pExpr->apExprPhrase[i]->nTerm==1 ); 2625 if( bLive && 2626 (pBuf->n==0 || pNode->iRowid!=pExpr->pRoot->iRowid || pNode->bEof) 2627 ){ 2628 pRet[i].bMiss = 1; 2629 }else{ 2630 pBuf->n = 0; 2631 } 2632 } 2633 } 2634 return pRet; 2635 } 2636 2637 struct Fts5ExprCtx { 2638 Fts5Expr *pExpr; 2639 Fts5PoslistPopulator *aPopulator; 2640 i64 iOff; 2641 }; 2642 typedef struct Fts5ExprCtx Fts5ExprCtx; 2643 2644 /* 2645 ** TODO: Make this more efficient! 2646 */ 2647 static int fts5ExprColsetTest(Fts5Colset *pColset, int iCol){ 2648 int i; 2649 for(i=0; i<pColset->nCol; i++){ 2650 if( pColset->aiCol[i]==iCol ) return 1; 2651 } 2652 return 0; 2653 } 2654 2655 static int fts5ExprPopulatePoslistsCb( 2656 void *pCtx, /* Copy of 2nd argument to xTokenize() */ 2657 int tflags, /* Mask of FTS5_TOKEN_* flags */ 2658 const char *pToken, /* Pointer to buffer containing token */ 2659 int nToken, /* Size of token in bytes */ 2660 int iUnused1, /* Byte offset of token within input text */ 2661 int iUnused2 /* Byte offset of end of token within input text */ 2662 ){ 2663 Fts5ExprCtx *p = (Fts5ExprCtx*)pCtx; 2664 Fts5Expr *pExpr = p->pExpr; 2665 int i; 2666 2667 UNUSED_PARAM2(iUnused1, iUnused2); 2668 2669 if( nToken>FTS5_MAX_TOKEN_SIZE ) nToken = FTS5_MAX_TOKEN_SIZE; 2670 if( (tflags & FTS5_TOKEN_COLOCATED)==0 ) p->iOff++; 2671 for(i=0; i<pExpr->nPhrase; i++){ 2672 Fts5ExprTerm *pTerm; 2673 if( p->aPopulator[i].bOk==0 ) continue; 2674 for(pTerm=&pExpr->apExprPhrase[i]->aTerm[0]; pTerm; pTerm=pTerm->pSynonym){ 2675 int nTerm = (int)strlen(pTerm->zTerm); 2676 if( (nTerm==nToken || (nTerm<nToken && pTerm->bPrefix)) 2677 && memcmp(pTerm->zTerm, pToken, nTerm)==0 2678 ){ 2679 int rc = sqlite3Fts5PoslistWriterAppend( 2680 &pExpr->apExprPhrase[i]->poslist, &p->aPopulator[i].writer, p->iOff 2681 ); 2682 if( rc ) return rc; 2683 break; 2684 } 2685 } 2686 } 2687 return SQLITE_OK; 2688 } 2689 2690 int sqlite3Fts5ExprPopulatePoslists( 2691 Fts5Config *pConfig, 2692 Fts5Expr *pExpr, 2693 Fts5PoslistPopulator *aPopulator, 2694 int iCol, 2695 const char *z, int n 2696 ){ 2697 int i; 2698 Fts5ExprCtx sCtx; 2699 sCtx.pExpr = pExpr; 2700 sCtx.aPopulator = aPopulator; 2701 sCtx.iOff = (((i64)iCol) << 32) - 1; 2702 2703 for(i=0; i<pExpr->nPhrase; i++){ 2704 Fts5ExprNode *pNode = pExpr->apExprPhrase[i]->pNode; 2705 Fts5Colset *pColset = pNode->pNear->pColset; 2706 if( (pColset && 0==fts5ExprColsetTest(pColset, iCol)) 2707 || aPopulator[i].bMiss 2708 ){ 2709 aPopulator[i].bOk = 0; 2710 }else{ 2711 aPopulator[i].bOk = 1; 2712 } 2713 } 2714 2715 return sqlite3Fts5Tokenize(pConfig, 2716 FTS5_TOKENIZE_DOCUMENT, z, n, (void*)&sCtx, fts5ExprPopulatePoslistsCb 2717 ); 2718 } 2719 2720 static void fts5ExprClearPoslists(Fts5ExprNode *pNode){ 2721 if( pNode->eType==FTS5_TERM || pNode->eType==FTS5_STRING ){ 2722 pNode->pNear->apPhrase[0]->poslist.n = 0; 2723 }else{ 2724 int i; 2725 for(i=0; i<pNode->nChild; i++){ 2726 fts5ExprClearPoslists(pNode->apChild[i]); 2727 } 2728 } 2729 } 2730 2731 static int fts5ExprCheckPoslists(Fts5ExprNode *pNode, i64 iRowid){ 2732 pNode->iRowid = iRowid; 2733 pNode->bEof = 0; 2734 switch( pNode->eType ){ 2735 case FTS5_TERM: 2736 case FTS5_STRING: 2737 return (pNode->pNear->apPhrase[0]->poslist.n>0); 2738 2739 case FTS5_AND: { 2740 int i; 2741 for(i=0; i<pNode->nChild; i++){ 2742 if( fts5ExprCheckPoslists(pNode->apChild[i], iRowid)==0 ){ 2743 fts5ExprClearPoslists(pNode); 2744 return 0; 2745 } 2746 } 2747 break; 2748 } 2749 2750 case FTS5_OR: { 2751 int i; 2752 int bRet = 0; 2753 for(i=0; i<pNode->nChild; i++){ 2754 if( fts5ExprCheckPoslists(pNode->apChild[i], iRowid) ){ 2755 bRet = 1; 2756 } 2757 } 2758 return bRet; 2759 } 2760 2761 default: { 2762 assert( pNode->eType==FTS5_NOT ); 2763 if( 0==fts5ExprCheckPoslists(pNode->apChild[0], iRowid) 2764 || 0!=fts5ExprCheckPoslists(pNode->apChild[1], iRowid) 2765 ){ 2766 fts5ExprClearPoslists(pNode); 2767 return 0; 2768 } 2769 break; 2770 } 2771 } 2772 return 1; 2773 } 2774 2775 void sqlite3Fts5ExprCheckPoslists(Fts5Expr *pExpr, i64 iRowid){ 2776 fts5ExprCheckPoslists(pExpr->pRoot, iRowid); 2777 } 2778 2779 /* 2780 ** This function is only called for detail=columns tables. 2781 */ 2782 int sqlite3Fts5ExprPhraseCollist( 2783 Fts5Expr *pExpr, 2784 int iPhrase, 2785 const u8 **ppCollist, 2786 int *pnCollist 2787 ){ 2788 Fts5ExprPhrase *pPhrase = pExpr->apExprPhrase[iPhrase]; 2789 Fts5ExprNode *pNode = pPhrase->pNode; 2790 int rc = SQLITE_OK; 2791 2792 assert( iPhrase>=0 && iPhrase<pExpr->nPhrase ); 2793 assert( pExpr->pConfig->eDetail==FTS5_DETAIL_COLUMNS ); 2794 2795 if( pNode->bEof==0 2796 && pNode->iRowid==pExpr->pRoot->iRowid 2797 && pPhrase->poslist.n>0 2798 ){ 2799 Fts5ExprTerm *pTerm = &pPhrase->aTerm[0]; 2800 if( pTerm->pSynonym ){ 2801 Fts5Buffer *pBuf = (Fts5Buffer*)&pTerm->pSynonym[1]; 2802 rc = fts5ExprSynonymList( 2803 pTerm, pNode->iRowid, pBuf, (u8**)ppCollist, pnCollist 2804 ); 2805 }else{ 2806 *ppCollist = pPhrase->aTerm[0].pIter->pData; 2807 *pnCollist = pPhrase->aTerm[0].pIter->nData; 2808 } 2809 }else{ 2810 *ppCollist = 0; 2811 *pnCollist = 0; 2812 } 2813 2814 return rc; 2815 } 2816