1 /* 2 ** 2001 September 15 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 ** This module contains C code that generates VDBE code used to process 13 ** the WHERE clause of SQL statements. This module is reponsible for 14 ** generating the code that loops through a table looking for applicable 15 ** rows. Indices are selected and used to speed the search when doing 16 ** so is applicable. Because this module is responsible for selecting 17 ** indices, you might also think of this module as the "query optimizer". 18 ** 19 ** $Id: where.c,v 1.298 2008/04/10 13:33:18 drh Exp $ 20 */ 21 #include "sqliteInt.h" 22 23 /* 24 ** The number of bits in a Bitmask. "BMS" means "BitMask Size". 25 */ 26 #define BMS (sizeof(Bitmask)*8) 27 28 /* 29 ** Trace output macros 30 */ 31 #if defined(SQLITE_TEST) || defined(SQLITE_DEBUG) 32 int sqlite3WhereTrace = 0; 33 # define WHERETRACE(X) if(sqlite3WhereTrace) sqlite3DebugPrintf X 34 #else 35 # define WHERETRACE(X) 36 #endif 37 38 /* Forward reference 39 */ 40 typedef struct WhereClause WhereClause; 41 typedef struct ExprMaskSet ExprMaskSet; 42 43 /* 44 ** The query generator uses an array of instances of this structure to 45 ** help it analyze the subexpressions of the WHERE clause. Each WHERE 46 ** clause subexpression is separated from the others by an AND operator. 47 ** 48 ** All WhereTerms are collected into a single WhereClause structure. 49 ** The following identity holds: 50 ** 51 ** WhereTerm.pWC->a[WhereTerm.idx] == WhereTerm 52 ** 53 ** When a term is of the form: 54 ** 55 ** X <op> <expr> 56 ** 57 ** where X is a column name and <op> is one of certain operators, 58 ** then WhereTerm.leftCursor and WhereTerm.leftColumn record the 59 ** cursor number and column number for X. WhereTerm.operator records 60 ** the <op> using a bitmask encoding defined by WO_xxx below. The 61 ** use of a bitmask encoding for the operator allows us to search 62 ** quickly for terms that match any of several different operators. 63 ** 64 ** prereqRight and prereqAll record sets of cursor numbers, 65 ** but they do so indirectly. A single ExprMaskSet structure translates 66 ** cursor number into bits and the translated bit is stored in the prereq 67 ** fields. The translation is used in order to maximize the number of 68 ** bits that will fit in a Bitmask. The VDBE cursor numbers might be 69 ** spread out over the non-negative integers. For example, the cursor 70 ** numbers might be 3, 8, 9, 10, 20, 23, 41, and 45. The ExprMaskSet 71 ** translates these sparse cursor numbers into consecutive integers 72 ** beginning with 0 in order to make the best possible use of the available 73 ** bits in the Bitmask. So, in the example above, the cursor numbers 74 ** would be mapped into integers 0 through 7. 75 */ 76 typedef struct WhereTerm WhereTerm; 77 struct WhereTerm { 78 Expr *pExpr; /* Pointer to the subexpression */ 79 i16 iParent; /* Disable pWC->a[iParent] when this term disabled */ 80 i16 leftCursor; /* Cursor number of X in "X <op> <expr>" */ 81 i16 leftColumn; /* Column number of X in "X <op> <expr>" */ 82 u16 eOperator; /* A WO_xx value describing <op> */ 83 u8 flags; /* Bit flags. See below */ 84 u8 nChild; /* Number of children that must disable us */ 85 WhereClause *pWC; /* The clause this term is part of */ 86 Bitmask prereqRight; /* Bitmask of tables used by pRight */ 87 Bitmask prereqAll; /* Bitmask of tables referenced by p */ 88 }; 89 90 /* 91 ** Allowed values of WhereTerm.flags 92 */ 93 #define TERM_DYNAMIC 0x01 /* Need to call sqlite3ExprDelete(pExpr) */ 94 #define TERM_VIRTUAL 0x02 /* Added by the optimizer. Do not code */ 95 #define TERM_CODED 0x04 /* This term is already coded */ 96 #define TERM_COPIED 0x08 /* Has a child */ 97 #define TERM_OR_OK 0x10 /* Used during OR-clause processing */ 98 99 /* 100 ** An instance of the following structure holds all information about a 101 ** WHERE clause. Mostly this is a container for one or more WhereTerms. 102 */ 103 struct WhereClause { 104 Parse *pParse; /* The parser context */ 105 ExprMaskSet *pMaskSet; /* Mapping of table indices to bitmasks */ 106 int nTerm; /* Number of terms */ 107 int nSlot; /* Number of entries in a[] */ 108 WhereTerm *a; /* Each a[] describes a term of the WHERE cluase */ 109 WhereTerm aStatic[10]; /* Initial static space for a[] */ 110 }; 111 112 /* 113 ** An instance of the following structure keeps track of a mapping 114 ** between VDBE cursor numbers and bits of the bitmasks in WhereTerm. 115 ** 116 ** The VDBE cursor numbers are small integers contained in 117 ** SrcList_item.iCursor and Expr.iTable fields. For any given WHERE 118 ** clause, the cursor numbers might not begin with 0 and they might 119 ** contain gaps in the numbering sequence. But we want to make maximum 120 ** use of the bits in our bitmasks. This structure provides a mapping 121 ** from the sparse cursor numbers into consecutive integers beginning 122 ** with 0. 123 ** 124 ** If ExprMaskSet.ix[A]==B it means that The A-th bit of a Bitmask 125 ** corresponds VDBE cursor number B. The A-th bit of a bitmask is 1<<A. 126 ** 127 ** For example, if the WHERE clause expression used these VDBE 128 ** cursors: 4, 5, 8, 29, 57, 73. Then the ExprMaskSet structure 129 ** would map those cursor numbers into bits 0 through 5. 130 ** 131 ** Note that the mapping is not necessarily ordered. In the example 132 ** above, the mapping might go like this: 4->3, 5->1, 8->2, 29->0, 133 ** 57->5, 73->4. Or one of 719 other combinations might be used. It 134 ** does not really matter. What is important is that sparse cursor 135 ** numbers all get mapped into bit numbers that begin with 0 and contain 136 ** no gaps. 137 */ 138 struct ExprMaskSet { 139 int n; /* Number of assigned cursor values */ 140 int ix[sizeof(Bitmask)*8]; /* Cursor assigned to each bit */ 141 }; 142 143 144 /* 145 ** Bitmasks for the operators that indices are able to exploit. An 146 ** OR-ed combination of these values can be used when searching for 147 ** terms in the where clause. 148 */ 149 #define WO_IN 1 150 #define WO_EQ 2 151 #define WO_LT (WO_EQ<<(TK_LT-TK_EQ)) 152 #define WO_LE (WO_EQ<<(TK_LE-TK_EQ)) 153 #define WO_GT (WO_EQ<<(TK_GT-TK_EQ)) 154 #define WO_GE (WO_EQ<<(TK_GE-TK_EQ)) 155 #define WO_MATCH 64 156 #define WO_ISNULL 128 157 158 /* 159 ** Value for flags returned by bestIndex(). 160 ** 161 ** The least significant byte is reserved as a mask for WO_ values above. 162 ** The WhereLevel.flags field is usually set to WO_IN|WO_EQ|WO_ISNULL. 163 ** But if the table is the right table of a left join, WhereLevel.flags 164 ** is set to WO_IN|WO_EQ. The WhereLevel.flags field can then be used as 165 ** the "op" parameter to findTerm when we are resolving equality constraints. 166 ** ISNULL constraints will then not be used on the right table of a left 167 ** join. Tickets #2177 and #2189. 168 */ 169 #define WHERE_ROWID_EQ 0x000100 /* rowid=EXPR or rowid IN (...) */ 170 #define WHERE_ROWID_RANGE 0x000200 /* rowid<EXPR and/or rowid>EXPR */ 171 #define WHERE_COLUMN_EQ 0x001000 /* x=EXPR or x IN (...) */ 172 #define WHERE_COLUMN_RANGE 0x002000 /* x<EXPR and/or x>EXPR */ 173 #define WHERE_COLUMN_IN 0x004000 /* x IN (...) */ 174 #define WHERE_TOP_LIMIT 0x010000 /* x<EXPR or x<=EXPR constraint */ 175 #define WHERE_BTM_LIMIT 0x020000 /* x>EXPR or x>=EXPR constraint */ 176 #define WHERE_IDX_ONLY 0x080000 /* Use index only - omit table */ 177 #define WHERE_ORDERBY 0x100000 /* Output will appear in correct order */ 178 #define WHERE_REVERSE 0x200000 /* Scan in reverse order */ 179 #define WHERE_UNIQUE 0x400000 /* Selects no more than one row */ 180 #define WHERE_VIRTUALTABLE 0x800000 /* Use virtual-table processing */ 181 182 /* 183 ** Initialize a preallocated WhereClause structure. 184 */ 185 static void whereClauseInit( 186 WhereClause *pWC, /* The WhereClause to be initialized */ 187 Parse *pParse, /* The parsing context */ 188 ExprMaskSet *pMaskSet /* Mapping from table indices to bitmasks */ 189 ){ 190 pWC->pParse = pParse; 191 pWC->pMaskSet = pMaskSet; 192 pWC->nTerm = 0; 193 pWC->nSlot = ArraySize(pWC->aStatic); 194 pWC->a = pWC->aStatic; 195 } 196 197 /* 198 ** Deallocate a WhereClause structure. The WhereClause structure 199 ** itself is not freed. This routine is the inverse of whereClauseInit(). 200 */ 201 static void whereClauseClear(WhereClause *pWC){ 202 int i; 203 WhereTerm *a; 204 for(i=pWC->nTerm-1, a=pWC->a; i>=0; i--, a++){ 205 if( a->flags & TERM_DYNAMIC ){ 206 sqlite3ExprDelete(a->pExpr); 207 } 208 } 209 if( pWC->a!=pWC->aStatic ){ 210 sqlite3_free(pWC->a); 211 } 212 } 213 214 /* 215 ** Add a new entries to the WhereClause structure. Increase the allocated 216 ** space as necessary. 217 ** 218 ** If the flags argument includes TERM_DYNAMIC, then responsibility 219 ** for freeing the expression p is assumed by the WhereClause object. 220 ** 221 ** WARNING: This routine might reallocate the space used to store 222 ** WhereTerms. All pointers to WhereTerms should be invalided after 223 ** calling this routine. Such pointers may be reinitialized by referencing 224 ** the pWC->a[] array. 225 */ 226 static int whereClauseInsert(WhereClause *pWC, Expr *p, int flags){ 227 WhereTerm *pTerm; 228 int idx; 229 if( pWC->nTerm>=pWC->nSlot ){ 230 WhereTerm *pOld = pWC->a; 231 pWC->a = sqlite3_malloc( sizeof(pWC->a[0])*pWC->nSlot*2 ); 232 if( pWC->a==0 ){ 233 pWC->pParse->db->mallocFailed = 1; 234 if( flags & TERM_DYNAMIC ){ 235 sqlite3ExprDelete(p); 236 } 237 pWC->a = pOld; 238 return 0; 239 } 240 memcpy(pWC->a, pOld, sizeof(pWC->a[0])*pWC->nTerm); 241 if( pOld!=pWC->aStatic ){ 242 sqlite3_free(pOld); 243 } 244 pWC->nSlot *= 2; 245 } 246 pTerm = &pWC->a[idx = pWC->nTerm]; 247 pWC->nTerm++; 248 pTerm->pExpr = p; 249 pTerm->flags = flags; 250 pTerm->pWC = pWC; 251 pTerm->iParent = -1; 252 return idx; 253 } 254 255 /* 256 ** This routine identifies subexpressions in the WHERE clause where 257 ** each subexpression is separated by the AND operator or some other 258 ** operator specified in the op parameter. The WhereClause structure 259 ** is filled with pointers to subexpressions. For example: 260 ** 261 ** WHERE a=='hello' AND coalesce(b,11)<10 AND (c+12!=d OR c==22) 262 ** \________/ \_______________/ \________________/ 263 ** slot[0] slot[1] slot[2] 264 ** 265 ** The original WHERE clause in pExpr is unaltered. All this routine 266 ** does is make slot[] entries point to substructure within pExpr. 267 ** 268 ** In the previous sentence and in the diagram, "slot[]" refers to 269 ** the WhereClause.a[] array. This array grows as needed to contain 270 ** all terms of the WHERE clause. 271 */ 272 static void whereSplit(WhereClause *pWC, Expr *pExpr, int op){ 273 if( pExpr==0 ) return; 274 if( pExpr->op!=op ){ 275 whereClauseInsert(pWC, pExpr, 0); 276 }else{ 277 whereSplit(pWC, pExpr->pLeft, op); 278 whereSplit(pWC, pExpr->pRight, op); 279 } 280 } 281 282 /* 283 ** Initialize an expression mask set 284 */ 285 #define initMaskSet(P) memset(P, 0, sizeof(*P)) 286 287 /* 288 ** Return the bitmask for the given cursor number. Return 0 if 289 ** iCursor is not in the set. 290 */ 291 static Bitmask getMask(ExprMaskSet *pMaskSet, int iCursor){ 292 int i; 293 for(i=0; i<pMaskSet->n; i++){ 294 if( pMaskSet->ix[i]==iCursor ){ 295 return ((Bitmask)1)<<i; 296 } 297 } 298 return 0; 299 } 300 301 /* 302 ** Create a new mask for cursor iCursor. 303 ** 304 ** There is one cursor per table in the FROM clause. The number of 305 ** tables in the FROM clause is limited by a test early in the 306 ** sqlite3WhereBegin() routine. So we know that the pMaskSet->ix[] 307 ** array will never overflow. 308 */ 309 static void createMask(ExprMaskSet *pMaskSet, int iCursor){ 310 assert( pMaskSet->n < ArraySize(pMaskSet->ix) ); 311 pMaskSet->ix[pMaskSet->n++] = iCursor; 312 } 313 314 /* 315 ** This routine walks (recursively) an expression tree and generates 316 ** a bitmask indicating which tables are used in that expression 317 ** tree. 318 ** 319 ** In order for this routine to work, the calling function must have 320 ** previously invoked sqlite3ExprResolveNames() on the expression. See 321 ** the header comment on that routine for additional information. 322 ** The sqlite3ExprResolveNames() routines looks for column names and 323 ** sets their opcodes to TK_COLUMN and their Expr.iTable fields to 324 ** the VDBE cursor number of the table. This routine just has to 325 ** translate the cursor numbers into bitmask values and OR all 326 ** the bitmasks together. 327 */ 328 static Bitmask exprListTableUsage(ExprMaskSet*, ExprList*); 329 static Bitmask exprSelectTableUsage(ExprMaskSet*, Select*); 330 static Bitmask exprTableUsage(ExprMaskSet *pMaskSet, Expr *p){ 331 Bitmask mask = 0; 332 if( p==0 ) return 0; 333 if( p->op==TK_COLUMN ){ 334 mask = getMask(pMaskSet, p->iTable); 335 return mask; 336 } 337 mask = exprTableUsage(pMaskSet, p->pRight); 338 mask |= exprTableUsage(pMaskSet, p->pLeft); 339 mask |= exprListTableUsage(pMaskSet, p->pList); 340 mask |= exprSelectTableUsage(pMaskSet, p->pSelect); 341 return mask; 342 } 343 static Bitmask exprListTableUsage(ExprMaskSet *pMaskSet, ExprList *pList){ 344 int i; 345 Bitmask mask = 0; 346 if( pList ){ 347 for(i=0; i<pList->nExpr; i++){ 348 mask |= exprTableUsage(pMaskSet, pList->a[i].pExpr); 349 } 350 } 351 return mask; 352 } 353 static Bitmask exprSelectTableUsage(ExprMaskSet *pMaskSet, Select *pS){ 354 Bitmask mask = 0; 355 while( pS ){ 356 mask |= exprListTableUsage(pMaskSet, pS->pEList); 357 mask |= exprListTableUsage(pMaskSet, pS->pGroupBy); 358 mask |= exprListTableUsage(pMaskSet, pS->pOrderBy); 359 mask |= exprTableUsage(pMaskSet, pS->pWhere); 360 mask |= exprTableUsage(pMaskSet, pS->pHaving); 361 pS = pS->pPrior; 362 } 363 return mask; 364 } 365 366 /* 367 ** Return TRUE if the given operator is one of the operators that is 368 ** allowed for an indexable WHERE clause term. The allowed operators are 369 ** "=", "<", ">", "<=", ">=", and "IN". 370 */ 371 static int allowedOp(int op){ 372 assert( TK_GT>TK_EQ && TK_GT<TK_GE ); 373 assert( TK_LT>TK_EQ && TK_LT<TK_GE ); 374 assert( TK_LE>TK_EQ && TK_LE<TK_GE ); 375 assert( TK_GE==TK_EQ+4 ); 376 return op==TK_IN || (op>=TK_EQ && op<=TK_GE) || op==TK_ISNULL; 377 } 378 379 /* 380 ** Swap two objects of type T. 381 */ 382 #define SWAP(TYPE,A,B) {TYPE t=A; A=B; B=t;} 383 384 /* 385 ** Commute a comparision operator. Expressions of the form "X op Y" 386 ** are converted into "Y op X". 387 ** 388 ** If a collation sequence is associated with either the left or right 389 ** side of the comparison, it remains associated with the same side after 390 ** the commutation. So "Y collate NOCASE op X" becomes 391 ** "X collate NOCASE op Y". This is because any collation sequence on 392 ** the left hand side of a comparison overrides any collation sequence 393 ** attached to the right. For the same reason the EP_ExpCollate flag 394 ** is not commuted. 395 */ 396 static void exprCommute(Expr *pExpr){ 397 u16 expRight = (pExpr->pRight->flags & EP_ExpCollate); 398 u16 expLeft = (pExpr->pLeft->flags & EP_ExpCollate); 399 assert( allowedOp(pExpr->op) && pExpr->op!=TK_IN ); 400 SWAP(CollSeq*,pExpr->pRight->pColl,pExpr->pLeft->pColl); 401 pExpr->pRight->flags = (pExpr->pRight->flags & ~EP_ExpCollate) | expLeft; 402 pExpr->pLeft->flags = (pExpr->pLeft->flags & ~EP_ExpCollate) | expRight; 403 SWAP(Expr*,pExpr->pRight,pExpr->pLeft); 404 if( pExpr->op>=TK_GT ){ 405 assert( TK_LT==TK_GT+2 ); 406 assert( TK_GE==TK_LE+2 ); 407 assert( TK_GT>TK_EQ ); 408 assert( TK_GT<TK_LE ); 409 assert( pExpr->op>=TK_GT && pExpr->op<=TK_GE ); 410 pExpr->op = ((pExpr->op-TK_GT)^2)+TK_GT; 411 } 412 } 413 414 /* 415 ** Translate from TK_xx operator to WO_xx bitmask. 416 */ 417 static int operatorMask(int op){ 418 int c; 419 assert( allowedOp(op) ); 420 if( op==TK_IN ){ 421 c = WO_IN; 422 }else if( op==TK_ISNULL ){ 423 c = WO_ISNULL; 424 }else{ 425 c = WO_EQ<<(op-TK_EQ); 426 } 427 assert( op!=TK_ISNULL || c==WO_ISNULL ); 428 assert( op!=TK_IN || c==WO_IN ); 429 assert( op!=TK_EQ || c==WO_EQ ); 430 assert( op!=TK_LT || c==WO_LT ); 431 assert( op!=TK_LE || c==WO_LE ); 432 assert( op!=TK_GT || c==WO_GT ); 433 assert( op!=TK_GE || c==WO_GE ); 434 return c; 435 } 436 437 /* 438 ** Search for a term in the WHERE clause that is of the form "X <op> <expr>" 439 ** where X is a reference to the iColumn of table iCur and <op> is one of 440 ** the WO_xx operator codes specified by the op parameter. 441 ** Return a pointer to the term. Return 0 if not found. 442 */ 443 static WhereTerm *findTerm( 444 WhereClause *pWC, /* The WHERE clause to be searched */ 445 int iCur, /* Cursor number of LHS */ 446 int iColumn, /* Column number of LHS */ 447 Bitmask notReady, /* RHS must not overlap with this mask */ 448 u16 op, /* Mask of WO_xx values describing operator */ 449 Index *pIdx /* Must be compatible with this index, if not NULL */ 450 ){ 451 WhereTerm *pTerm; 452 int k; 453 for(pTerm=pWC->a, k=pWC->nTerm; k; k--, pTerm++){ 454 if( pTerm->leftCursor==iCur 455 && (pTerm->prereqRight & notReady)==0 456 && pTerm->leftColumn==iColumn 457 && (pTerm->eOperator & op)!=0 458 ){ 459 if( iCur>=0 && pIdx && pTerm->eOperator!=WO_ISNULL ){ 460 Expr *pX = pTerm->pExpr; 461 CollSeq *pColl; 462 char idxaff; 463 int j; 464 Parse *pParse = pWC->pParse; 465 466 idxaff = pIdx->pTable->aCol[iColumn].affinity; 467 if( !sqlite3IndexAffinityOk(pX, idxaff) ) continue; 468 469 /* Figure out the collation sequence required from an index for 470 ** it to be useful for optimising expression pX. Store this 471 ** value in variable pColl. 472 */ 473 assert(pX->pLeft); 474 pColl = sqlite3BinaryCompareCollSeq(pParse, pX->pLeft, pX->pRight); 475 if( !pColl ){ 476 pColl = pParse->db->pDfltColl; 477 } 478 479 for(j=0; j<pIdx->nColumn && pIdx->aiColumn[j]!=iColumn; j++){} 480 assert( j<pIdx->nColumn ); 481 if( sqlite3StrICmp(pColl->zName, pIdx->azColl[j]) ) continue; 482 } 483 return pTerm; 484 } 485 } 486 return 0; 487 } 488 489 /* Forward reference */ 490 static void exprAnalyze(SrcList*, WhereClause*, int); 491 492 /* 493 ** Call exprAnalyze on all terms in a WHERE clause. 494 ** 495 ** 496 */ 497 static void exprAnalyzeAll( 498 SrcList *pTabList, /* the FROM clause */ 499 WhereClause *pWC /* the WHERE clause to be analyzed */ 500 ){ 501 int i; 502 for(i=pWC->nTerm-1; i>=0; i--){ 503 exprAnalyze(pTabList, pWC, i); 504 } 505 } 506 507 #ifndef SQLITE_OMIT_LIKE_OPTIMIZATION 508 /* 509 ** Check to see if the given expression is a LIKE or GLOB operator that 510 ** can be optimized using inequality constraints. Return TRUE if it is 511 ** so and false if not. 512 ** 513 ** In order for the operator to be optimizible, the RHS must be a string 514 ** literal that does not begin with a wildcard. 515 */ 516 static int isLikeOrGlob( 517 sqlite3 *db, /* The database */ 518 Expr *pExpr, /* Test this expression */ 519 int *pnPattern, /* Number of non-wildcard prefix characters */ 520 int *pisComplete, /* True if the only wildcard is % in the last character */ 521 int *pnoCase /* True if uppercase is equivalent to lowercase */ 522 ){ 523 const char *z; 524 Expr *pRight, *pLeft; 525 ExprList *pList; 526 int c, cnt; 527 char wc[3]; 528 CollSeq *pColl; 529 530 if( !sqlite3IsLikeFunction(db, pExpr, pnoCase, wc) ){ 531 return 0; 532 } 533 #ifdef SQLITE_EBCDIC 534 if( *pnoCase ) return 0; 535 #endif 536 pList = pExpr->pList; 537 pRight = pList->a[0].pExpr; 538 if( pRight->op!=TK_STRING 539 && (pRight->op!=TK_REGISTER || pRight->iColumn!=TK_STRING) ){ 540 return 0; 541 } 542 pLeft = pList->a[1].pExpr; 543 if( pLeft->op!=TK_COLUMN ){ 544 return 0; 545 } 546 pColl = pLeft->pColl; 547 assert( pColl!=0 || pLeft->iColumn==-1 ); 548 if( pColl==0 ){ 549 /* No collation is defined for the ROWID. Use the default. */ 550 pColl = db->pDfltColl; 551 } 552 if( (pColl->type!=SQLITE_COLL_BINARY || *pnoCase) && 553 (pColl->type!=SQLITE_COLL_NOCASE || !*pnoCase) ){ 554 return 0; 555 } 556 sqlite3DequoteExpr(db, pRight); 557 z = (char *)pRight->token.z; 558 cnt = 0; 559 if( z ){ 560 while( (c=z[cnt])!=0 && c!=wc[0] && c!=wc[1] && c!=wc[2] ){ cnt++; } 561 } 562 if( cnt==0 || 255==(u8)z[cnt] ){ 563 return 0; 564 } 565 *pisComplete = z[cnt]==wc[0] && z[cnt+1]==0; 566 *pnPattern = cnt; 567 return 1; 568 } 569 #endif /* SQLITE_OMIT_LIKE_OPTIMIZATION */ 570 571 572 #ifndef SQLITE_OMIT_VIRTUALTABLE 573 /* 574 ** Check to see if the given expression is of the form 575 ** 576 ** column MATCH expr 577 ** 578 ** If it is then return TRUE. If not, return FALSE. 579 */ 580 static int isMatchOfColumn( 581 Expr *pExpr /* Test this expression */ 582 ){ 583 ExprList *pList; 584 585 if( pExpr->op!=TK_FUNCTION ){ 586 return 0; 587 } 588 if( pExpr->token.n!=5 || 589 sqlite3StrNICmp((const char*)pExpr->token.z,"match",5)!=0 ){ 590 return 0; 591 } 592 pList = pExpr->pList; 593 if( pList->nExpr!=2 ){ 594 return 0; 595 } 596 if( pList->a[1].pExpr->op != TK_COLUMN ){ 597 return 0; 598 } 599 return 1; 600 } 601 #endif /* SQLITE_OMIT_VIRTUALTABLE */ 602 603 /* 604 ** If the pBase expression originated in the ON or USING clause of 605 ** a join, then transfer the appropriate markings over to derived. 606 */ 607 static void transferJoinMarkings(Expr *pDerived, Expr *pBase){ 608 pDerived->flags |= pBase->flags & EP_FromJoin; 609 pDerived->iRightJoinTable = pBase->iRightJoinTable; 610 } 611 612 #if !defined(SQLITE_OMIT_OR_OPTIMIZATION) && !defined(SQLITE_OMIT_SUBQUERY) 613 /* 614 ** Return TRUE if the given term of an OR clause can be converted 615 ** into an IN clause. The iCursor and iColumn define the left-hand 616 ** side of the IN clause. 617 ** 618 ** The context is that we have multiple OR-connected equality terms 619 ** like this: 620 ** 621 ** a=<expr1> OR a=<expr2> OR b=<expr3> OR ... 622 ** 623 ** The pOrTerm input to this routine corresponds to a single term of 624 ** this OR clause. In order for the term to be a condidate for 625 ** conversion to an IN operator, the following must be true: 626 ** 627 ** * The left-hand side of the term must be the column which 628 ** is identified by iCursor and iColumn. 629 ** 630 ** * If the right-hand side is also a column, then the affinities 631 ** of both right and left sides must be such that no type 632 ** conversions are required on the right. (Ticket #2249) 633 ** 634 ** If both of these conditions are true, then return true. Otherwise 635 ** return false. 636 */ 637 static int orTermIsOptCandidate(WhereTerm *pOrTerm, int iCursor, int iColumn){ 638 int affLeft, affRight; 639 assert( pOrTerm->eOperator==WO_EQ ); 640 if( pOrTerm->leftCursor!=iCursor ){ 641 return 0; 642 } 643 if( pOrTerm->leftColumn!=iColumn ){ 644 return 0; 645 } 646 affRight = sqlite3ExprAffinity(pOrTerm->pExpr->pRight); 647 if( affRight==0 ){ 648 return 1; 649 } 650 affLeft = sqlite3ExprAffinity(pOrTerm->pExpr->pLeft); 651 if( affRight!=affLeft ){ 652 return 0; 653 } 654 return 1; 655 } 656 657 /* 658 ** Return true if the given term of an OR clause can be ignored during 659 ** a check to make sure all OR terms are candidates for optimization. 660 ** In other words, return true if a call to the orTermIsOptCandidate() 661 ** above returned false but it is not necessary to disqualify the 662 ** optimization. 663 ** 664 ** Suppose the original OR phrase was this: 665 ** 666 ** a=4 OR a=11 OR a=b 667 ** 668 ** During analysis, the third term gets flipped around and duplicate 669 ** so that we are left with this: 670 ** 671 ** a=4 OR a=11 OR a=b OR b=a 672 ** 673 ** Since the last two terms are duplicates, only one of them 674 ** has to qualify in order for the whole phrase to qualify. When 675 ** this routine is called, we know that pOrTerm did not qualify. 676 ** This routine merely checks to see if pOrTerm has a duplicate that 677 ** might qualify. If there is a duplicate that has not yet been 678 ** disqualified, then return true. If there are no duplicates, or 679 ** the duplicate has also been disqualifed, return false. 680 */ 681 static int orTermHasOkDuplicate(WhereClause *pOr, WhereTerm *pOrTerm){ 682 if( pOrTerm->flags & TERM_COPIED ){ 683 /* This is the original term. The duplicate is to the left had 684 ** has not yet been analyzed and thus has not yet been disqualified. */ 685 return 1; 686 } 687 if( (pOrTerm->flags & TERM_VIRTUAL)!=0 688 && (pOr->a[pOrTerm->iParent].flags & TERM_OR_OK)!=0 ){ 689 /* This is a duplicate term. The original qualified so this one 690 ** does not have to. */ 691 return 1; 692 } 693 /* This is either a singleton term or else it is a duplicate for 694 ** which the original did not qualify. Either way we are done for. */ 695 return 0; 696 } 697 #endif /* !SQLITE_OMIT_OR_OPTIMIZATION && !SQLITE_OMIT_SUBQUERY */ 698 699 /* 700 ** The input to this routine is an WhereTerm structure with only the 701 ** "pExpr" field filled in. The job of this routine is to analyze the 702 ** subexpression and populate all the other fields of the WhereTerm 703 ** structure. 704 ** 705 ** If the expression is of the form "<expr> <op> X" it gets commuted 706 ** to the standard form of "X <op> <expr>". If the expression is of 707 ** the form "X <op> Y" where both X and Y are columns, then the original 708 ** expression is unchanged and a new virtual expression of the form 709 ** "Y <op> X" is added to the WHERE clause and analyzed separately. 710 */ 711 static void exprAnalyze( 712 SrcList *pSrc, /* the FROM clause */ 713 WhereClause *pWC, /* the WHERE clause */ 714 int idxTerm /* Index of the term to be analyzed */ 715 ){ 716 WhereTerm *pTerm; 717 ExprMaskSet *pMaskSet; 718 Expr *pExpr; 719 Bitmask prereqLeft; 720 Bitmask prereqAll; 721 int nPattern; 722 int isComplete; 723 int noCase; 724 int op; 725 Parse *pParse = pWC->pParse; 726 sqlite3 *db = pParse->db; 727 728 if( db->mallocFailed ){ 729 return; 730 } 731 pTerm = &pWC->a[idxTerm]; 732 pMaskSet = pWC->pMaskSet; 733 pExpr = pTerm->pExpr; 734 prereqLeft = exprTableUsage(pMaskSet, pExpr->pLeft); 735 op = pExpr->op; 736 if( op==TK_IN ){ 737 assert( pExpr->pRight==0 ); 738 pTerm->prereqRight = exprListTableUsage(pMaskSet, pExpr->pList) 739 | exprSelectTableUsage(pMaskSet, pExpr->pSelect); 740 }else if( op==TK_ISNULL ){ 741 pTerm->prereqRight = 0; 742 }else{ 743 pTerm->prereqRight = exprTableUsage(pMaskSet, pExpr->pRight); 744 } 745 prereqAll = exprTableUsage(pMaskSet, pExpr); 746 if( ExprHasProperty(pExpr, EP_FromJoin) ){ 747 Bitmask x = getMask(pMaskSet, pExpr->iRightJoinTable); 748 prereqAll |= x; 749 pTerm->prereqRight |= x-1; /* ON clause terms may not be used with an index 750 ** on left table of a LEFT JOIN. Ticket #3015 */ 751 } 752 pTerm->prereqAll = prereqAll; 753 pTerm->leftCursor = -1; 754 pTerm->iParent = -1; 755 pTerm->eOperator = 0; 756 if( allowedOp(op) && (pTerm->prereqRight & prereqLeft)==0 ){ 757 Expr *pLeft = pExpr->pLeft; 758 Expr *pRight = pExpr->pRight; 759 if( pLeft->op==TK_COLUMN ){ 760 pTerm->leftCursor = pLeft->iTable; 761 pTerm->leftColumn = pLeft->iColumn; 762 pTerm->eOperator = operatorMask(op); 763 } 764 if( pRight && pRight->op==TK_COLUMN ){ 765 WhereTerm *pNew; 766 Expr *pDup; 767 if( pTerm->leftCursor>=0 ){ 768 int idxNew; 769 pDup = sqlite3ExprDup(db, pExpr); 770 if( db->mallocFailed ){ 771 sqlite3ExprDelete(pDup); 772 return; 773 } 774 idxNew = whereClauseInsert(pWC, pDup, TERM_VIRTUAL|TERM_DYNAMIC); 775 if( idxNew==0 ) return; 776 pNew = &pWC->a[idxNew]; 777 pNew->iParent = idxTerm; 778 pTerm = &pWC->a[idxTerm]; 779 pTerm->nChild = 1; 780 pTerm->flags |= TERM_COPIED; 781 }else{ 782 pDup = pExpr; 783 pNew = pTerm; 784 } 785 exprCommute(pDup); 786 pLeft = pDup->pLeft; 787 pNew->leftCursor = pLeft->iTable; 788 pNew->leftColumn = pLeft->iColumn; 789 pNew->prereqRight = prereqLeft; 790 pNew->prereqAll = prereqAll; 791 pNew->eOperator = operatorMask(pDup->op); 792 } 793 } 794 795 #ifndef SQLITE_OMIT_BETWEEN_OPTIMIZATION 796 /* If a term is the BETWEEN operator, create two new virtual terms 797 ** that define the range that the BETWEEN implements. 798 */ 799 else if( pExpr->op==TK_BETWEEN ){ 800 ExprList *pList = pExpr->pList; 801 int i; 802 static const u8 ops[] = {TK_GE, TK_LE}; 803 assert( pList!=0 ); 804 assert( pList->nExpr==2 ); 805 for(i=0; i<2; i++){ 806 Expr *pNewExpr; 807 int idxNew; 808 pNewExpr = sqlite3Expr(db, ops[i], sqlite3ExprDup(db, pExpr->pLeft), 809 sqlite3ExprDup(db, pList->a[i].pExpr), 0); 810 idxNew = whereClauseInsert(pWC, pNewExpr, TERM_VIRTUAL|TERM_DYNAMIC); 811 exprAnalyze(pSrc, pWC, idxNew); 812 pTerm = &pWC->a[idxTerm]; 813 pWC->a[idxNew].iParent = idxTerm; 814 } 815 pTerm->nChild = 2; 816 } 817 #endif /* SQLITE_OMIT_BETWEEN_OPTIMIZATION */ 818 819 #if !defined(SQLITE_OMIT_OR_OPTIMIZATION) && !defined(SQLITE_OMIT_SUBQUERY) 820 /* Attempt to convert OR-connected terms into an IN operator so that 821 ** they can make use of indices. Example: 822 ** 823 ** x = expr1 OR expr2 = x OR x = expr3 824 ** 825 ** is converted into 826 ** 827 ** x IN (expr1,expr2,expr3) 828 ** 829 ** This optimization must be omitted if OMIT_SUBQUERY is defined because 830 ** the compiler for the the IN operator is part of sub-queries. 831 */ 832 else if( pExpr->op==TK_OR ){ 833 int ok; 834 int i, j; 835 int iColumn, iCursor; 836 WhereClause sOr; 837 WhereTerm *pOrTerm; 838 839 assert( (pTerm->flags & TERM_DYNAMIC)==0 ); 840 whereClauseInit(&sOr, pWC->pParse, pMaskSet); 841 whereSplit(&sOr, pExpr, TK_OR); 842 exprAnalyzeAll(pSrc, &sOr); 843 assert( sOr.nTerm>=2 ); 844 j = 0; 845 if( db->mallocFailed ) goto or_not_possible; 846 do{ 847 assert( j<sOr.nTerm ); 848 iColumn = sOr.a[j].leftColumn; 849 iCursor = sOr.a[j].leftCursor; 850 ok = iCursor>=0; 851 for(i=sOr.nTerm-1, pOrTerm=sOr.a; i>=0 && ok; i--, pOrTerm++){ 852 if( pOrTerm->eOperator!=WO_EQ ){ 853 goto or_not_possible; 854 } 855 if( orTermIsOptCandidate(pOrTerm, iCursor, iColumn) ){ 856 pOrTerm->flags |= TERM_OR_OK; 857 }else if( orTermHasOkDuplicate(&sOr, pOrTerm) ){ 858 pOrTerm->flags &= ~TERM_OR_OK; 859 }else{ 860 ok = 0; 861 } 862 } 863 }while( !ok && (sOr.a[j++].flags & TERM_COPIED)!=0 && j<2 ); 864 if( ok ){ 865 ExprList *pList = 0; 866 Expr *pNew, *pDup; 867 Expr *pLeft = 0; 868 for(i=sOr.nTerm-1, pOrTerm=sOr.a; i>=0 && ok; i--, pOrTerm++){ 869 if( (pOrTerm->flags & TERM_OR_OK)==0 ) continue; 870 pDup = sqlite3ExprDup(db, pOrTerm->pExpr->pRight); 871 pList = sqlite3ExprListAppend(pWC->pParse, pList, pDup, 0); 872 pLeft = pOrTerm->pExpr->pLeft; 873 } 874 assert( pLeft!=0 ); 875 pDup = sqlite3ExprDup(db, pLeft); 876 pNew = sqlite3Expr(db, TK_IN, pDup, 0, 0); 877 if( pNew ){ 878 int idxNew; 879 transferJoinMarkings(pNew, pExpr); 880 pNew->pList = pList; 881 idxNew = whereClauseInsert(pWC, pNew, TERM_VIRTUAL|TERM_DYNAMIC); 882 exprAnalyze(pSrc, pWC, idxNew); 883 pTerm = &pWC->a[idxTerm]; 884 pWC->a[idxNew].iParent = idxTerm; 885 pTerm->nChild = 1; 886 }else{ 887 sqlite3ExprListDelete(pList); 888 } 889 } 890 or_not_possible: 891 whereClauseClear(&sOr); 892 } 893 #endif /* SQLITE_OMIT_OR_OPTIMIZATION */ 894 895 #ifndef SQLITE_OMIT_LIKE_OPTIMIZATION 896 /* Add constraints to reduce the search space on a LIKE or GLOB 897 ** operator. 898 ** 899 ** A like pattern of the form "x LIKE 'abc%'" is changed into constraints 900 ** 901 ** x>='abc' AND x<'abd' AND x LIKE 'abc%' 902 ** 903 ** The last character of the prefix "abc" is incremented to form the 904 ** termination condidtion "abd". This trick of incrementing the last 905 ** is not 255 and if the character set is not EBCDIC. 906 */ 907 if( isLikeOrGlob(db, pExpr, &nPattern, &isComplete, &noCase) ){ 908 Expr *pLeft, *pRight; 909 Expr *pStr1, *pStr2; 910 Expr *pNewExpr1, *pNewExpr2; 911 int idxNew1, idxNew2; 912 913 pLeft = pExpr->pList->a[1].pExpr; 914 pRight = pExpr->pList->a[0].pExpr; 915 pStr1 = sqlite3PExpr(pParse, TK_STRING, 0, 0, 0); 916 if( pStr1 ){ 917 sqlite3TokenCopy(db, &pStr1->token, &pRight->token); 918 pStr1->token.n = nPattern; 919 pStr1->flags = EP_Dequoted; 920 } 921 pStr2 = sqlite3ExprDup(db, pStr1); 922 if( !db->mallocFailed ){ 923 u8 c, *pC; 924 assert( pStr2->token.dyn ); 925 pC = (u8*)&pStr2->token.z[nPattern-1]; 926 c = *pC; 927 if( noCase ) c = sqlite3UpperToLower[c]; 928 *pC = c + 1; 929 } 930 pNewExpr1 = sqlite3PExpr(pParse, TK_GE, sqlite3ExprDup(db,pLeft), pStr1, 0); 931 idxNew1 = whereClauseInsert(pWC, pNewExpr1, TERM_VIRTUAL|TERM_DYNAMIC); 932 exprAnalyze(pSrc, pWC, idxNew1); 933 pNewExpr2 = sqlite3PExpr(pParse, TK_LT, sqlite3ExprDup(db,pLeft), pStr2, 0); 934 idxNew2 = whereClauseInsert(pWC, pNewExpr2, TERM_VIRTUAL|TERM_DYNAMIC); 935 exprAnalyze(pSrc, pWC, idxNew2); 936 pTerm = &pWC->a[idxTerm]; 937 if( isComplete ){ 938 pWC->a[idxNew1].iParent = idxTerm; 939 pWC->a[idxNew2].iParent = idxTerm; 940 pTerm->nChild = 2; 941 } 942 } 943 #endif /* SQLITE_OMIT_LIKE_OPTIMIZATION */ 944 945 #ifndef SQLITE_OMIT_VIRTUALTABLE 946 /* Add a WO_MATCH auxiliary term to the constraint set if the 947 ** current expression is of the form: column MATCH expr. 948 ** This information is used by the xBestIndex methods of 949 ** virtual tables. The native query optimizer does not attempt 950 ** to do anything with MATCH functions. 951 */ 952 if( isMatchOfColumn(pExpr) ){ 953 int idxNew; 954 Expr *pRight, *pLeft; 955 WhereTerm *pNewTerm; 956 Bitmask prereqColumn, prereqExpr; 957 958 pRight = pExpr->pList->a[0].pExpr; 959 pLeft = pExpr->pList->a[1].pExpr; 960 prereqExpr = exprTableUsage(pMaskSet, pRight); 961 prereqColumn = exprTableUsage(pMaskSet, pLeft); 962 if( (prereqExpr & prereqColumn)==0 ){ 963 Expr *pNewExpr; 964 pNewExpr = sqlite3Expr(db, TK_MATCH, 0, sqlite3ExprDup(db, pRight), 0); 965 idxNew = whereClauseInsert(pWC, pNewExpr, TERM_VIRTUAL|TERM_DYNAMIC); 966 pNewTerm = &pWC->a[idxNew]; 967 pNewTerm->prereqRight = prereqExpr; 968 pNewTerm->leftCursor = pLeft->iTable; 969 pNewTerm->leftColumn = pLeft->iColumn; 970 pNewTerm->eOperator = WO_MATCH; 971 pNewTerm->iParent = idxTerm; 972 pTerm = &pWC->a[idxTerm]; 973 pTerm->nChild = 1; 974 pTerm->flags |= TERM_COPIED; 975 pNewTerm->prereqAll = pTerm->prereqAll; 976 } 977 } 978 #endif /* SQLITE_OMIT_VIRTUALTABLE */ 979 } 980 981 /* 982 ** Return TRUE if any of the expressions in pList->a[iFirst...] contain 983 ** a reference to any table other than the iBase table. 984 */ 985 static int referencesOtherTables( 986 ExprList *pList, /* Search expressions in ths list */ 987 ExprMaskSet *pMaskSet, /* Mapping from tables to bitmaps */ 988 int iFirst, /* Be searching with the iFirst-th expression */ 989 int iBase /* Ignore references to this table */ 990 ){ 991 Bitmask allowed = ~getMask(pMaskSet, iBase); 992 while( iFirst<pList->nExpr ){ 993 if( (exprTableUsage(pMaskSet, pList->a[iFirst++].pExpr)&allowed)!=0 ){ 994 return 1; 995 } 996 } 997 return 0; 998 } 999 1000 1001 /* 1002 ** This routine decides if pIdx can be used to satisfy the ORDER BY 1003 ** clause. If it can, it returns 1. If pIdx cannot satisfy the 1004 ** ORDER BY clause, this routine returns 0. 1005 ** 1006 ** pOrderBy is an ORDER BY clause from a SELECT statement. pTab is the 1007 ** left-most table in the FROM clause of that same SELECT statement and 1008 ** the table has a cursor number of "base". pIdx is an index on pTab. 1009 ** 1010 ** nEqCol is the number of columns of pIdx that are used as equality 1011 ** constraints. Any of these columns may be missing from the ORDER BY 1012 ** clause and the match can still be a success. 1013 ** 1014 ** All terms of the ORDER BY that match against the index must be either 1015 ** ASC or DESC. (Terms of the ORDER BY clause past the end of a UNIQUE 1016 ** index do not need to satisfy this constraint.) The *pbRev value is 1017 ** set to 1 if the ORDER BY clause is all DESC and it is set to 0 if 1018 ** the ORDER BY clause is all ASC. 1019 */ 1020 static int isSortingIndex( 1021 Parse *pParse, /* Parsing context */ 1022 ExprMaskSet *pMaskSet, /* Mapping from table indices to bitmaps */ 1023 Index *pIdx, /* The index we are testing */ 1024 int base, /* Cursor number for the table to be sorted */ 1025 ExprList *pOrderBy, /* The ORDER BY clause */ 1026 int nEqCol, /* Number of index columns with == constraints */ 1027 int *pbRev /* Set to 1 if ORDER BY is DESC */ 1028 ){ 1029 int i, j; /* Loop counters */ 1030 int sortOrder = 0; /* XOR of index and ORDER BY sort direction */ 1031 int nTerm; /* Number of ORDER BY terms */ 1032 struct ExprList_item *pTerm; /* A term of the ORDER BY clause */ 1033 sqlite3 *db = pParse->db; 1034 1035 assert( pOrderBy!=0 ); 1036 nTerm = pOrderBy->nExpr; 1037 assert( nTerm>0 ); 1038 1039 /* Match terms of the ORDER BY clause against columns of 1040 ** the index. 1041 ** 1042 ** Note that indices have pIdx->nColumn regular columns plus 1043 ** one additional column containing the rowid. The rowid column 1044 ** of the index is also allowed to match against the ORDER BY 1045 ** clause. 1046 */ 1047 for(i=j=0, pTerm=pOrderBy->a; j<nTerm && i<=pIdx->nColumn; i++){ 1048 Expr *pExpr; /* The expression of the ORDER BY pTerm */ 1049 CollSeq *pColl; /* The collating sequence of pExpr */ 1050 int termSortOrder; /* Sort order for this term */ 1051 int iColumn; /* The i-th column of the index. -1 for rowid */ 1052 int iSortOrder; /* 1 for DESC, 0 for ASC on the i-th index term */ 1053 const char *zColl; /* Name of the collating sequence for i-th index term */ 1054 1055 pExpr = pTerm->pExpr; 1056 if( pExpr->op!=TK_COLUMN || pExpr->iTable!=base ){ 1057 /* Can not use an index sort on anything that is not a column in the 1058 ** left-most table of the FROM clause */ 1059 break; 1060 } 1061 pColl = sqlite3ExprCollSeq(pParse, pExpr); 1062 if( !pColl ){ 1063 pColl = db->pDfltColl; 1064 } 1065 if( i<pIdx->nColumn ){ 1066 iColumn = pIdx->aiColumn[i]; 1067 if( iColumn==pIdx->pTable->iPKey ){ 1068 iColumn = -1; 1069 } 1070 iSortOrder = pIdx->aSortOrder[i]; 1071 zColl = pIdx->azColl[i]; 1072 }else{ 1073 iColumn = -1; 1074 iSortOrder = 0; 1075 zColl = pColl->zName; 1076 } 1077 if( pExpr->iColumn!=iColumn || sqlite3StrICmp(pColl->zName, zColl) ){ 1078 /* Term j of the ORDER BY clause does not match column i of the index */ 1079 if( i<nEqCol ){ 1080 /* If an index column that is constrained by == fails to match an 1081 ** ORDER BY term, that is OK. Just ignore that column of the index 1082 */ 1083 continue; 1084 }else{ 1085 /* If an index column fails to match and is not constrained by == 1086 ** then the index cannot satisfy the ORDER BY constraint. 1087 */ 1088 return 0; 1089 } 1090 } 1091 assert( pIdx->aSortOrder!=0 ); 1092 assert( pTerm->sortOrder==0 || pTerm->sortOrder==1 ); 1093 assert( iSortOrder==0 || iSortOrder==1 ); 1094 termSortOrder = iSortOrder ^ pTerm->sortOrder; 1095 if( i>nEqCol ){ 1096 if( termSortOrder!=sortOrder ){ 1097 /* Indices can only be used if all ORDER BY terms past the 1098 ** equality constraints are all either DESC or ASC. */ 1099 return 0; 1100 } 1101 }else{ 1102 sortOrder = termSortOrder; 1103 } 1104 j++; 1105 pTerm++; 1106 if( iColumn<0 && !referencesOtherTables(pOrderBy, pMaskSet, j, base) ){ 1107 /* If the indexed column is the primary key and everything matches 1108 ** so far and none of the ORDER BY terms to the right reference other 1109 ** tables in the join, then we are assured that the index can be used 1110 ** to sort because the primary key is unique and so none of the other 1111 ** columns will make any difference 1112 */ 1113 j = nTerm; 1114 } 1115 } 1116 1117 *pbRev = sortOrder!=0; 1118 if( j>=nTerm ){ 1119 /* All terms of the ORDER BY clause are covered by this index so 1120 ** this index can be used for sorting. */ 1121 return 1; 1122 } 1123 if( pIdx->onError!=OE_None && i==pIdx->nColumn 1124 && !referencesOtherTables(pOrderBy, pMaskSet, j, base) ){ 1125 /* All terms of this index match some prefix of the ORDER BY clause 1126 ** and the index is UNIQUE and no terms on the tail of the ORDER BY 1127 ** clause reference other tables in a join. If this is all true then 1128 ** the order by clause is superfluous. */ 1129 return 1; 1130 } 1131 return 0; 1132 } 1133 1134 /* 1135 ** Check table to see if the ORDER BY clause in pOrderBy can be satisfied 1136 ** by sorting in order of ROWID. Return true if so and set *pbRev to be 1137 ** true for reverse ROWID and false for forward ROWID order. 1138 */ 1139 static int sortableByRowid( 1140 int base, /* Cursor number for table to be sorted */ 1141 ExprList *pOrderBy, /* The ORDER BY clause */ 1142 ExprMaskSet *pMaskSet, /* Mapping from tables to bitmaps */ 1143 int *pbRev /* Set to 1 if ORDER BY is DESC */ 1144 ){ 1145 Expr *p; 1146 1147 assert( pOrderBy!=0 ); 1148 assert( pOrderBy->nExpr>0 ); 1149 p = pOrderBy->a[0].pExpr; 1150 if( p->op==TK_COLUMN && p->iTable==base && p->iColumn==-1 1151 && !referencesOtherTables(pOrderBy, pMaskSet, 1, base) ){ 1152 *pbRev = pOrderBy->a[0].sortOrder; 1153 return 1; 1154 } 1155 return 0; 1156 } 1157 1158 /* 1159 ** Prepare a crude estimate of the logarithm of the input value. 1160 ** The results need not be exact. This is only used for estimating 1161 ** the total cost of performing operatings with O(logN) or O(NlogN) 1162 ** complexity. Because N is just a guess, it is no great tragedy if 1163 ** logN is a little off. 1164 */ 1165 static double estLog(double N){ 1166 double logN = 1; 1167 double x = 10; 1168 while( N>x ){ 1169 logN += 1; 1170 x *= 10; 1171 } 1172 return logN; 1173 } 1174 1175 /* 1176 ** Two routines for printing the content of an sqlite3_index_info 1177 ** structure. Used for testing and debugging only. If neither 1178 ** SQLITE_TEST or SQLITE_DEBUG are defined, then these routines 1179 ** are no-ops. 1180 */ 1181 #if !defined(SQLITE_OMIT_VIRTUALTABLE) && defined(SQLITE_DEBUG) 1182 static void TRACE_IDX_INPUTS(sqlite3_index_info *p){ 1183 int i; 1184 if( !sqlite3WhereTrace ) return; 1185 for(i=0; i<p->nConstraint; i++){ 1186 sqlite3DebugPrintf(" constraint[%d]: col=%d termid=%d op=%d usabled=%d\n", 1187 i, 1188 p->aConstraint[i].iColumn, 1189 p->aConstraint[i].iTermOffset, 1190 p->aConstraint[i].op, 1191 p->aConstraint[i].usable); 1192 } 1193 for(i=0; i<p->nOrderBy; i++){ 1194 sqlite3DebugPrintf(" orderby[%d]: col=%d desc=%d\n", 1195 i, 1196 p->aOrderBy[i].iColumn, 1197 p->aOrderBy[i].desc); 1198 } 1199 } 1200 static void TRACE_IDX_OUTPUTS(sqlite3_index_info *p){ 1201 int i; 1202 if( !sqlite3WhereTrace ) return; 1203 for(i=0; i<p->nConstraint; i++){ 1204 sqlite3DebugPrintf(" usage[%d]: argvIdx=%d omit=%d\n", 1205 i, 1206 p->aConstraintUsage[i].argvIndex, 1207 p->aConstraintUsage[i].omit); 1208 } 1209 sqlite3DebugPrintf(" idxNum=%d\n", p->idxNum); 1210 sqlite3DebugPrintf(" idxStr=%s\n", p->idxStr); 1211 sqlite3DebugPrintf(" orderByConsumed=%d\n", p->orderByConsumed); 1212 sqlite3DebugPrintf(" estimatedCost=%g\n", p->estimatedCost); 1213 } 1214 #else 1215 #define TRACE_IDX_INPUTS(A) 1216 #define TRACE_IDX_OUTPUTS(A) 1217 #endif 1218 1219 #ifndef SQLITE_OMIT_VIRTUALTABLE 1220 /* 1221 ** Compute the best index for a virtual table. 1222 ** 1223 ** The best index is computed by the xBestIndex method of the virtual 1224 ** table module. This routine is really just a wrapper that sets up 1225 ** the sqlite3_index_info structure that is used to communicate with 1226 ** xBestIndex. 1227 ** 1228 ** In a join, this routine might be called multiple times for the 1229 ** same virtual table. The sqlite3_index_info structure is created 1230 ** and initialized on the first invocation and reused on all subsequent 1231 ** invocations. The sqlite3_index_info structure is also used when 1232 ** code is generated to access the virtual table. The whereInfoDelete() 1233 ** routine takes care of freeing the sqlite3_index_info structure after 1234 ** everybody has finished with it. 1235 */ 1236 static double bestVirtualIndex( 1237 Parse *pParse, /* The parsing context */ 1238 WhereClause *pWC, /* The WHERE clause */ 1239 struct SrcList_item *pSrc, /* The FROM clause term to search */ 1240 Bitmask notReady, /* Mask of cursors that are not available */ 1241 ExprList *pOrderBy, /* The order by clause */ 1242 int orderByUsable, /* True if we can potential sort */ 1243 sqlite3_index_info **ppIdxInfo /* Index information passed to xBestIndex */ 1244 ){ 1245 Table *pTab = pSrc->pTab; 1246 sqlite3_index_info *pIdxInfo; 1247 struct sqlite3_index_constraint *pIdxCons; 1248 struct sqlite3_index_orderby *pIdxOrderBy; 1249 struct sqlite3_index_constraint_usage *pUsage; 1250 WhereTerm *pTerm; 1251 int i, j; 1252 int nOrderBy; 1253 int rc; 1254 1255 /* If the sqlite3_index_info structure has not been previously 1256 ** allocated and initialized for this virtual table, then allocate 1257 ** and initialize it now 1258 */ 1259 pIdxInfo = *ppIdxInfo; 1260 if( pIdxInfo==0 ){ 1261 WhereTerm *pTerm; 1262 int nTerm; 1263 WHERETRACE(("Recomputing index info for %s...\n", pTab->zName)); 1264 1265 /* Count the number of possible WHERE clause constraints referring 1266 ** to this virtual table */ 1267 for(i=nTerm=0, pTerm=pWC->a; i<pWC->nTerm; i++, pTerm++){ 1268 if( pTerm->leftCursor != pSrc->iCursor ) continue; 1269 if( pTerm->eOperator==WO_IN ) continue; 1270 if( pTerm->eOperator==WO_ISNULL ) continue; 1271 nTerm++; 1272 } 1273 1274 /* If the ORDER BY clause contains only columns in the current 1275 ** virtual table then allocate space for the aOrderBy part of 1276 ** the sqlite3_index_info structure. 1277 */ 1278 nOrderBy = 0; 1279 if( pOrderBy ){ 1280 for(i=0; i<pOrderBy->nExpr; i++){ 1281 Expr *pExpr = pOrderBy->a[i].pExpr; 1282 if( pExpr->op!=TK_COLUMN || pExpr->iTable!=pSrc->iCursor ) break; 1283 } 1284 if( i==pOrderBy->nExpr ){ 1285 nOrderBy = pOrderBy->nExpr; 1286 } 1287 } 1288 1289 /* Allocate the sqlite3_index_info structure 1290 */ 1291 pIdxInfo = sqlite3DbMallocZero(pParse->db, sizeof(*pIdxInfo) 1292 + (sizeof(*pIdxCons) + sizeof(*pUsage))*nTerm 1293 + sizeof(*pIdxOrderBy)*nOrderBy ); 1294 if( pIdxInfo==0 ){ 1295 sqlite3ErrorMsg(pParse, "out of memory"); 1296 return 0.0; 1297 } 1298 *ppIdxInfo = pIdxInfo; 1299 1300 /* Initialize the structure. The sqlite3_index_info structure contains 1301 ** many fields that are declared "const" to prevent xBestIndex from 1302 ** changing them. We have to do some funky casting in order to 1303 ** initialize those fields. 1304 */ 1305 pIdxCons = (struct sqlite3_index_constraint*)&pIdxInfo[1]; 1306 pIdxOrderBy = (struct sqlite3_index_orderby*)&pIdxCons[nTerm]; 1307 pUsage = (struct sqlite3_index_constraint_usage*)&pIdxOrderBy[nOrderBy]; 1308 *(int*)&pIdxInfo->nConstraint = nTerm; 1309 *(int*)&pIdxInfo->nOrderBy = nOrderBy; 1310 *(struct sqlite3_index_constraint**)&pIdxInfo->aConstraint = pIdxCons; 1311 *(struct sqlite3_index_orderby**)&pIdxInfo->aOrderBy = pIdxOrderBy; 1312 *(struct sqlite3_index_constraint_usage**)&pIdxInfo->aConstraintUsage = 1313 pUsage; 1314 1315 for(i=j=0, pTerm=pWC->a; i<pWC->nTerm; i++, pTerm++){ 1316 if( pTerm->leftCursor != pSrc->iCursor ) continue; 1317 if( pTerm->eOperator==WO_IN ) continue; 1318 if( pTerm->eOperator==WO_ISNULL ) continue; 1319 pIdxCons[j].iColumn = pTerm->leftColumn; 1320 pIdxCons[j].iTermOffset = i; 1321 pIdxCons[j].op = pTerm->eOperator; 1322 /* The direct assignment in the previous line is possible only because 1323 ** the WO_ and SQLITE_INDEX_CONSTRAINT_ codes are identical. The 1324 ** following asserts verify this fact. */ 1325 assert( WO_EQ==SQLITE_INDEX_CONSTRAINT_EQ ); 1326 assert( WO_LT==SQLITE_INDEX_CONSTRAINT_LT ); 1327 assert( WO_LE==SQLITE_INDEX_CONSTRAINT_LE ); 1328 assert( WO_GT==SQLITE_INDEX_CONSTRAINT_GT ); 1329 assert( WO_GE==SQLITE_INDEX_CONSTRAINT_GE ); 1330 assert( WO_MATCH==SQLITE_INDEX_CONSTRAINT_MATCH ); 1331 assert( pTerm->eOperator & (WO_EQ|WO_LT|WO_LE|WO_GT|WO_GE|WO_MATCH) ); 1332 j++; 1333 } 1334 for(i=0; i<nOrderBy; i++){ 1335 Expr *pExpr = pOrderBy->a[i].pExpr; 1336 pIdxOrderBy[i].iColumn = pExpr->iColumn; 1337 pIdxOrderBy[i].desc = pOrderBy->a[i].sortOrder; 1338 } 1339 } 1340 1341 /* At this point, the sqlite3_index_info structure that pIdxInfo points 1342 ** to will have been initialized, either during the current invocation or 1343 ** during some prior invocation. Now we just have to customize the 1344 ** details of pIdxInfo for the current invocation and pass it to 1345 ** xBestIndex. 1346 */ 1347 1348 /* The module name must be defined. Also, by this point there must 1349 ** be a pointer to an sqlite3_vtab structure. Otherwise 1350 ** sqlite3ViewGetColumnNames() would have picked up the error. 1351 */ 1352 assert( pTab->azModuleArg && pTab->azModuleArg[0] ); 1353 assert( pTab->pVtab ); 1354 #if 0 1355 if( pTab->pVtab==0 ){ 1356 sqlite3ErrorMsg(pParse, "undefined module %s for table %s", 1357 pTab->azModuleArg[0], pTab->zName); 1358 return 0.0; 1359 } 1360 #endif 1361 1362 /* Set the aConstraint[].usable fields and initialize all 1363 ** output variables to zero. 1364 ** 1365 ** aConstraint[].usable is true for constraints where the right-hand 1366 ** side contains only references to tables to the left of the current 1367 ** table. In other words, if the constraint is of the form: 1368 ** 1369 ** column = expr 1370 ** 1371 ** and we are evaluating a join, then the constraint on column is 1372 ** only valid if all tables referenced in expr occur to the left 1373 ** of the table containing column. 1374 ** 1375 ** The aConstraints[] array contains entries for all constraints 1376 ** on the current table. That way we only have to compute it once 1377 ** even though we might try to pick the best index multiple times. 1378 ** For each attempt at picking an index, the order of tables in the 1379 ** join might be different so we have to recompute the usable flag 1380 ** each time. 1381 */ 1382 pIdxCons = *(struct sqlite3_index_constraint**)&pIdxInfo->aConstraint; 1383 pUsage = pIdxInfo->aConstraintUsage; 1384 for(i=0; i<pIdxInfo->nConstraint; i++, pIdxCons++){ 1385 j = pIdxCons->iTermOffset; 1386 pTerm = &pWC->a[j]; 1387 pIdxCons->usable = (pTerm->prereqRight & notReady)==0; 1388 } 1389 memset(pUsage, 0, sizeof(pUsage[0])*pIdxInfo->nConstraint); 1390 if( pIdxInfo->needToFreeIdxStr ){ 1391 sqlite3_free(pIdxInfo->idxStr); 1392 } 1393 pIdxInfo->idxStr = 0; 1394 pIdxInfo->idxNum = 0; 1395 pIdxInfo->needToFreeIdxStr = 0; 1396 pIdxInfo->orderByConsumed = 0; 1397 pIdxInfo->estimatedCost = SQLITE_BIG_DBL / 2.0; 1398 nOrderBy = pIdxInfo->nOrderBy; 1399 if( pIdxInfo->nOrderBy && !orderByUsable ){ 1400 *(int*)&pIdxInfo->nOrderBy = 0; 1401 } 1402 1403 (void)sqlite3SafetyOff(pParse->db); 1404 WHERETRACE(("xBestIndex for %s\n", pTab->zName)); 1405 TRACE_IDX_INPUTS(pIdxInfo); 1406 rc = pTab->pVtab->pModule->xBestIndex(pTab->pVtab, pIdxInfo); 1407 TRACE_IDX_OUTPUTS(pIdxInfo); 1408 (void)sqlite3SafetyOn(pParse->db); 1409 1410 for(i=0; i<pIdxInfo->nConstraint; i++){ 1411 if( !pIdxInfo->aConstraint[i].usable && pUsage[i].argvIndex>0 ){ 1412 sqlite3ErrorMsg(pParse, 1413 "table %s: xBestIndex returned an invalid plan", pTab->zName); 1414 return 0.0; 1415 } 1416 } 1417 1418 if( rc!=SQLITE_OK ){ 1419 if( rc==SQLITE_NOMEM ){ 1420 pParse->db->mallocFailed = 1; 1421 }else { 1422 sqlite3ErrorMsg(pParse, "%s", sqlite3ErrStr(rc)); 1423 } 1424 } 1425 *(int*)&pIdxInfo->nOrderBy = nOrderBy; 1426 1427 return pIdxInfo->estimatedCost; 1428 } 1429 #endif /* SQLITE_OMIT_VIRTUALTABLE */ 1430 1431 /* 1432 ** Find the best index for accessing a particular table. Return a pointer 1433 ** to the index, flags that describe how the index should be used, the 1434 ** number of equality constraints, and the "cost" for this index. 1435 ** 1436 ** The lowest cost index wins. The cost is an estimate of the amount of 1437 ** CPU and disk I/O need to process the request using the selected index. 1438 ** Factors that influence cost include: 1439 ** 1440 ** * The estimated number of rows that will be retrieved. (The 1441 ** fewer the better.) 1442 ** 1443 ** * Whether or not sorting must occur. 1444 ** 1445 ** * Whether or not there must be separate lookups in the 1446 ** index and in the main table. 1447 ** 1448 */ 1449 static double bestIndex( 1450 Parse *pParse, /* The parsing context */ 1451 WhereClause *pWC, /* The WHERE clause */ 1452 struct SrcList_item *pSrc, /* The FROM clause term to search */ 1453 Bitmask notReady, /* Mask of cursors that are not available */ 1454 ExprList *pOrderBy, /* The order by clause */ 1455 Index **ppIndex, /* Make *ppIndex point to the best index */ 1456 int *pFlags, /* Put flags describing this choice in *pFlags */ 1457 int *pnEq /* Put the number of == or IN constraints here */ 1458 ){ 1459 WhereTerm *pTerm; 1460 Index *bestIdx = 0; /* Index that gives the lowest cost */ 1461 double lowestCost; /* The cost of using bestIdx */ 1462 int bestFlags = 0; /* Flags associated with bestIdx */ 1463 int bestNEq = 0; /* Best value for nEq */ 1464 int iCur = pSrc->iCursor; /* The cursor of the table to be accessed */ 1465 Index *pProbe; /* An index we are evaluating */ 1466 int rev; /* True to scan in reverse order */ 1467 int flags; /* Flags associated with pProbe */ 1468 int nEq; /* Number of == or IN constraints */ 1469 int eqTermMask; /* Mask of valid equality operators */ 1470 double cost; /* Cost of using pProbe */ 1471 1472 WHERETRACE(("bestIndex: tbl=%s notReady=%x\n", pSrc->pTab->zName, notReady)); 1473 lowestCost = SQLITE_BIG_DBL; 1474 pProbe = pSrc->pTab->pIndex; 1475 1476 /* If the table has no indices and there are no terms in the where 1477 ** clause that refer to the ROWID, then we will never be able to do 1478 ** anything other than a full table scan on this table. We might as 1479 ** well put it first in the join order. That way, perhaps it can be 1480 ** referenced by other tables in the join. 1481 */ 1482 if( pProbe==0 && 1483 findTerm(pWC, iCur, -1, 0, WO_EQ|WO_IN|WO_LT|WO_LE|WO_GT|WO_GE,0)==0 && 1484 (pOrderBy==0 || !sortableByRowid(iCur, pOrderBy, pWC->pMaskSet, &rev)) ){ 1485 *pFlags = 0; 1486 *ppIndex = 0; 1487 *pnEq = 0; 1488 return 0.0; 1489 } 1490 1491 /* Check for a rowid=EXPR or rowid IN (...) constraints 1492 */ 1493 pTerm = findTerm(pWC, iCur, -1, notReady, WO_EQ|WO_IN, 0); 1494 if( pTerm ){ 1495 Expr *pExpr; 1496 *ppIndex = 0; 1497 bestFlags = WHERE_ROWID_EQ; 1498 if( pTerm->eOperator & WO_EQ ){ 1499 /* Rowid== is always the best pick. Look no further. Because only 1500 ** a single row is generated, output is always in sorted order */ 1501 *pFlags = WHERE_ROWID_EQ | WHERE_UNIQUE; 1502 *pnEq = 1; 1503 WHERETRACE(("... best is rowid\n")); 1504 return 0.0; 1505 }else if( (pExpr = pTerm->pExpr)->pList!=0 ){ 1506 /* Rowid IN (LIST): cost is NlogN where N is the number of list 1507 ** elements. */ 1508 lowestCost = pExpr->pList->nExpr; 1509 lowestCost *= estLog(lowestCost); 1510 }else{ 1511 /* Rowid IN (SELECT): cost is NlogN where N is the number of rows 1512 ** in the result of the inner select. We have no way to estimate 1513 ** that value so make a wild guess. */ 1514 lowestCost = 200; 1515 } 1516 WHERETRACE(("... rowid IN cost: %.9g\n", lowestCost)); 1517 } 1518 1519 /* Estimate the cost of a table scan. If we do not know how many 1520 ** entries are in the table, use 1 million as a guess. 1521 */ 1522 cost = pProbe ? pProbe->aiRowEst[0] : 1000000; 1523 WHERETRACE(("... table scan base cost: %.9g\n", cost)); 1524 flags = WHERE_ROWID_RANGE; 1525 1526 /* Check for constraints on a range of rowids in a table scan. 1527 */ 1528 pTerm = findTerm(pWC, iCur, -1, notReady, WO_LT|WO_LE|WO_GT|WO_GE, 0); 1529 if( pTerm ){ 1530 if( findTerm(pWC, iCur, -1, notReady, WO_LT|WO_LE, 0) ){ 1531 flags |= WHERE_TOP_LIMIT; 1532 cost /= 3; /* Guess that rowid<EXPR eliminates two-thirds or rows */ 1533 } 1534 if( findTerm(pWC, iCur, -1, notReady, WO_GT|WO_GE, 0) ){ 1535 flags |= WHERE_BTM_LIMIT; 1536 cost /= 3; /* Guess that rowid>EXPR eliminates two-thirds of rows */ 1537 } 1538 WHERETRACE(("... rowid range reduces cost to %.9g\n", cost)); 1539 }else{ 1540 flags = 0; 1541 } 1542 1543 /* If the table scan does not satisfy the ORDER BY clause, increase 1544 ** the cost by NlogN to cover the expense of sorting. */ 1545 if( pOrderBy ){ 1546 if( sortableByRowid(iCur, pOrderBy, pWC->pMaskSet, &rev) ){ 1547 flags |= WHERE_ORDERBY|WHERE_ROWID_RANGE; 1548 if( rev ){ 1549 flags |= WHERE_REVERSE; 1550 } 1551 }else{ 1552 cost += cost*estLog(cost); 1553 WHERETRACE(("... sorting increases cost to %.9g\n", cost)); 1554 } 1555 } 1556 if( cost<lowestCost ){ 1557 lowestCost = cost; 1558 bestFlags = flags; 1559 } 1560 1561 /* If the pSrc table is the right table of a LEFT JOIN then we may not 1562 ** use an index to satisfy IS NULL constraints on that table. This is 1563 ** because columns might end up being NULL if the table does not match - 1564 ** a circumstance which the index cannot help us discover. Ticket #2177. 1565 */ 1566 if( (pSrc->jointype & JT_LEFT)!=0 ){ 1567 eqTermMask = WO_EQ|WO_IN; 1568 }else{ 1569 eqTermMask = WO_EQ|WO_IN|WO_ISNULL; 1570 } 1571 1572 /* Look at each index. 1573 */ 1574 for(; pProbe; pProbe=pProbe->pNext){ 1575 int i; /* Loop counter */ 1576 double inMultiplier = 1; 1577 1578 WHERETRACE(("... index %s:\n", pProbe->zName)); 1579 1580 /* Count the number of columns in the index that are satisfied 1581 ** by x=EXPR constraints or x IN (...) constraints. 1582 */ 1583 flags = 0; 1584 for(i=0; i<pProbe->nColumn; i++){ 1585 int j = pProbe->aiColumn[i]; 1586 pTerm = findTerm(pWC, iCur, j, notReady, eqTermMask, pProbe); 1587 if( pTerm==0 ) break; 1588 flags |= WHERE_COLUMN_EQ; 1589 if( pTerm->eOperator & WO_IN ){ 1590 Expr *pExpr = pTerm->pExpr; 1591 flags |= WHERE_COLUMN_IN; 1592 if( pExpr->pSelect!=0 ){ 1593 inMultiplier *= 25; 1594 }else if( pExpr->pList!=0 ){ 1595 inMultiplier *= pExpr->pList->nExpr + 1; 1596 } 1597 } 1598 } 1599 cost = pProbe->aiRowEst[i] * inMultiplier * estLog(inMultiplier); 1600 nEq = i; 1601 if( pProbe->onError!=OE_None && (flags & WHERE_COLUMN_IN)==0 1602 && nEq==pProbe->nColumn ){ 1603 flags |= WHERE_UNIQUE; 1604 } 1605 WHERETRACE(("...... nEq=%d inMult=%.9g cost=%.9g\n",nEq,inMultiplier,cost)); 1606 1607 /* Look for range constraints 1608 */ 1609 if( nEq<pProbe->nColumn ){ 1610 int j = pProbe->aiColumn[nEq]; 1611 pTerm = findTerm(pWC, iCur, j, notReady, WO_LT|WO_LE|WO_GT|WO_GE, pProbe); 1612 if( pTerm ){ 1613 flags |= WHERE_COLUMN_RANGE; 1614 if( findTerm(pWC, iCur, j, notReady, WO_LT|WO_LE, pProbe) ){ 1615 flags |= WHERE_TOP_LIMIT; 1616 cost /= 3; 1617 } 1618 if( findTerm(pWC, iCur, j, notReady, WO_GT|WO_GE, pProbe) ){ 1619 flags |= WHERE_BTM_LIMIT; 1620 cost /= 3; 1621 } 1622 WHERETRACE(("...... range reduces cost to %.9g\n", cost)); 1623 } 1624 } 1625 1626 /* Add the additional cost of sorting if that is a factor. 1627 */ 1628 if( pOrderBy ){ 1629 if( (flags & WHERE_COLUMN_IN)==0 && 1630 isSortingIndex(pParse,pWC->pMaskSet,pProbe,iCur,pOrderBy,nEq,&rev) ){ 1631 if( flags==0 ){ 1632 flags = WHERE_COLUMN_RANGE; 1633 } 1634 flags |= WHERE_ORDERBY; 1635 if( rev ){ 1636 flags |= WHERE_REVERSE; 1637 } 1638 }else{ 1639 cost += cost*estLog(cost); 1640 WHERETRACE(("...... orderby increases cost to %.9g\n", cost)); 1641 } 1642 } 1643 1644 /* Check to see if we can get away with using just the index without 1645 ** ever reading the table. If that is the case, then halve the 1646 ** cost of this index. 1647 */ 1648 if( flags && pSrc->colUsed < (((Bitmask)1)<<(BMS-1)) ){ 1649 Bitmask m = pSrc->colUsed; 1650 int j; 1651 for(j=0; j<pProbe->nColumn; j++){ 1652 int x = pProbe->aiColumn[j]; 1653 if( x<BMS-1 ){ 1654 m &= ~(((Bitmask)1)<<x); 1655 } 1656 } 1657 if( m==0 ){ 1658 flags |= WHERE_IDX_ONLY; 1659 cost /= 2; 1660 WHERETRACE(("...... idx-only reduces cost to %.9g\n", cost)); 1661 } 1662 } 1663 1664 /* If this index has achieved the lowest cost so far, then use it. 1665 */ 1666 if( flags && cost < lowestCost ){ 1667 bestIdx = pProbe; 1668 lowestCost = cost; 1669 bestFlags = flags; 1670 bestNEq = nEq; 1671 } 1672 } 1673 1674 /* Report the best result 1675 */ 1676 *ppIndex = bestIdx; 1677 WHERETRACE(("best index is %s, cost=%.9g, flags=%x, nEq=%d\n", 1678 bestIdx ? bestIdx->zName : "(none)", lowestCost, bestFlags, bestNEq)); 1679 *pFlags = bestFlags | eqTermMask; 1680 *pnEq = bestNEq; 1681 return lowestCost; 1682 } 1683 1684 1685 /* 1686 ** Disable a term in the WHERE clause. Except, do not disable the term 1687 ** if it controls a LEFT OUTER JOIN and it did not originate in the ON 1688 ** or USING clause of that join. 1689 ** 1690 ** Consider the term t2.z='ok' in the following queries: 1691 ** 1692 ** (1) SELECT * FROM t1 LEFT JOIN t2 ON t1.a=t2.x WHERE t2.z='ok' 1693 ** (2) SELECT * FROM t1 LEFT JOIN t2 ON t1.a=t2.x AND t2.z='ok' 1694 ** (3) SELECT * FROM t1, t2 WHERE t1.a=t2.x AND t2.z='ok' 1695 ** 1696 ** The t2.z='ok' is disabled in the in (2) because it originates 1697 ** in the ON clause. The term is disabled in (3) because it is not part 1698 ** of a LEFT OUTER JOIN. In (1), the term is not disabled. 1699 ** 1700 ** Disabling a term causes that term to not be tested in the inner loop 1701 ** of the join. Disabling is an optimization. When terms are satisfied 1702 ** by indices, we disable them to prevent redundant tests in the inner 1703 ** loop. We would get the correct results if nothing were ever disabled, 1704 ** but joins might run a little slower. The trick is to disable as much 1705 ** as we can without disabling too much. If we disabled in (1), we'd get 1706 ** the wrong answer. See ticket #813. 1707 */ 1708 static void disableTerm(WhereLevel *pLevel, WhereTerm *pTerm){ 1709 if( pTerm 1710 && (pTerm->flags & TERM_CODED)==0 1711 && (pLevel->iLeftJoin==0 || ExprHasProperty(pTerm->pExpr, EP_FromJoin)) 1712 ){ 1713 pTerm->flags |= TERM_CODED; 1714 if( pTerm->iParent>=0 ){ 1715 WhereTerm *pOther = &pTerm->pWC->a[pTerm->iParent]; 1716 if( (--pOther->nChild)==0 ){ 1717 disableTerm(pLevel, pOther); 1718 } 1719 } 1720 } 1721 } 1722 1723 /* 1724 ** Generate code that builds a probe for an index. 1725 ** 1726 ** There should be nColumn values on the stack. The index 1727 ** to be probed is pIdx. Pop the values from the stack and 1728 ** replace them all with a single record that is the index 1729 ** problem. 1730 */ 1731 static void buildIndexProbe( 1732 Parse *pParse, /* Parsing and code generation context */ 1733 int nColumn, /* The number of columns to check for NULL */ 1734 Index *pIdx, /* Index that we will be searching */ 1735 int regSrc, /* Take values from this register */ 1736 int regDest /* Write the result into this register */ 1737 ){ 1738 Vdbe *v = pParse->pVdbe; 1739 assert( regSrc>0 ); 1740 assert( regDest>0 ); 1741 assert( v!=0 ); 1742 sqlite3VdbeAddOp3(v, OP_MakeRecord, regSrc, nColumn, regDest); 1743 sqlite3IndexAffinityStr(v, pIdx); 1744 sqlite3ExprCacheAffinityChange(pParse, regSrc, nColumn); 1745 } 1746 1747 1748 /* 1749 ** Generate code for a single equality term of the WHERE clause. An equality 1750 ** term can be either X=expr or X IN (...). pTerm is the term to be 1751 ** coded. 1752 ** 1753 ** The current value for the constraint is left in register iReg. 1754 ** 1755 ** For a constraint of the form X=expr, the expression is evaluated and its 1756 ** result is left on the stack. For constraints of the form X IN (...) 1757 ** this routine sets up a loop that will iterate over all values of X. 1758 */ 1759 static int codeEqualityTerm( 1760 Parse *pParse, /* The parsing context */ 1761 WhereTerm *pTerm, /* The term of the WHERE clause to be coded */ 1762 WhereLevel *pLevel, /* When level of the FROM clause we are working on */ 1763 int iTarget /* Attempt to leave results in this register */ 1764 ){ 1765 Expr *pX = pTerm->pExpr; 1766 Vdbe *v = pParse->pVdbe; 1767 int iReg; /* Register holding results */ 1768 1769 if( iTarget<=0 ){ 1770 iReg = iTarget = sqlite3GetTempReg(pParse); 1771 } 1772 if( pX->op==TK_EQ ){ 1773 iReg = sqlite3ExprCodeTarget(pParse, pX->pRight, iTarget); 1774 }else if( pX->op==TK_ISNULL ){ 1775 iReg = iTarget; 1776 sqlite3VdbeAddOp2(v, OP_Null, 0, iReg); 1777 #ifndef SQLITE_OMIT_SUBQUERY 1778 }else{ 1779 int eType; 1780 int iTab; 1781 struct InLoop *pIn; 1782 1783 assert( pX->op==TK_IN ); 1784 iReg = iTarget; 1785 eType = sqlite3FindInIndex(pParse, pX, 1); 1786 iTab = pX->iTable; 1787 sqlite3VdbeAddOp2(v, OP_Rewind, iTab, 0); 1788 VdbeComment((v, "%.*s", pX->span.n, pX->span.z)); 1789 if( pLevel->nIn==0 ){ 1790 pLevel->nxt = sqlite3VdbeMakeLabel(v); 1791 } 1792 pLevel->nIn++; 1793 pLevel->aInLoop = sqlite3DbReallocOrFree(pParse->db, pLevel->aInLoop, 1794 sizeof(pLevel->aInLoop[0])*pLevel->nIn); 1795 pIn = pLevel->aInLoop; 1796 if( pIn ){ 1797 pIn += pLevel->nIn - 1; 1798 pIn->iCur = iTab; 1799 if( eType==IN_INDEX_ROWID ){ 1800 pIn->topAddr = sqlite3VdbeAddOp2(v, OP_Rowid, iTab, iReg); 1801 }else{ 1802 pIn->topAddr = sqlite3VdbeAddOp3(v, OP_Column, iTab, 0, iReg); 1803 } 1804 sqlite3VdbeAddOp1(v, OP_IsNull, iReg); 1805 }else{ 1806 pLevel->nIn = 0; 1807 } 1808 #endif 1809 } 1810 disableTerm(pLevel, pTerm); 1811 return iReg; 1812 } 1813 1814 /* 1815 ** Generate code that will evaluate all == and IN constraints for an 1816 ** index. The values for all constraints are left on the stack. 1817 ** 1818 ** For example, consider table t1(a,b,c,d,e,f) with index i1(a,b,c). 1819 ** Suppose the WHERE clause is this: a==5 AND b IN (1,2,3) AND c>5 AND c<10 1820 ** The index has as many as three equality constraints, but in this 1821 ** example, the third "c" value is an inequality. So only two 1822 ** constraints are coded. This routine will generate code to evaluate 1823 ** a==5 and b IN (1,2,3). The current values for a and b will be left 1824 ** on the stack - a is the deepest and b the shallowest. 1825 ** 1826 ** In the example above nEq==2. But this subroutine works for any value 1827 ** of nEq including 0. If nEq==0, this routine is nearly a no-op. 1828 ** The only thing it does is allocate the pLevel->iMem memory cell. 1829 ** 1830 ** This routine always allocates at least one memory cell and puts 1831 ** the address of that memory cell in pLevel->iMem. The code that 1832 ** calls this routine will use pLevel->iMem to store the termination 1833 ** key value of the loop. If one or more IN operators appear, then 1834 ** this routine allocates an additional nEq memory cells for internal 1835 ** use. 1836 */ 1837 static int codeAllEqualityTerms( 1838 Parse *pParse, /* Parsing context */ 1839 WhereLevel *pLevel, /* Which nested loop of the FROM we are coding */ 1840 WhereClause *pWC, /* The WHERE clause */ 1841 Bitmask notReady, /* Which parts of FROM have not yet been coded */ 1842 int nExtraReg /* Number of extra registers to allocate */ 1843 ){ 1844 int nEq = pLevel->nEq; /* The number of == or IN constraints to code */ 1845 Vdbe *v = pParse->pVdbe; /* The virtual machine under construction */ 1846 Index *pIdx = pLevel->pIdx; /* The index being used for this loop */ 1847 int iCur = pLevel->iTabCur; /* The cursor of the table */ 1848 WhereTerm *pTerm; /* A single constraint term */ 1849 int j; /* Loop counter */ 1850 int regBase; /* Base register */ 1851 1852 /* Figure out how many memory cells we will need then allocate them. 1853 ** We always need at least one used to store the loop terminator 1854 ** value. If there are IN operators we'll need one for each == or 1855 ** IN constraint. 1856 */ 1857 pLevel->iMem = pParse->nMem + 1; 1858 regBase = pParse->nMem + 2; 1859 pParse->nMem += pLevel->nEq + 2 + nExtraReg; 1860 1861 /* Evaluate the equality constraints 1862 */ 1863 assert( pIdx->nColumn>=nEq ); 1864 for(j=0; j<nEq; j++){ 1865 int r1; 1866 int k = pIdx->aiColumn[j]; 1867 pTerm = findTerm(pWC, iCur, k, notReady, pLevel->flags, pIdx); 1868 if( pTerm==0 ) break; 1869 assert( (pTerm->flags & TERM_CODED)==0 ); 1870 r1 = codeEqualityTerm(pParse, pTerm, pLevel, regBase+j); 1871 if( r1!=regBase+j ){ 1872 sqlite3VdbeAddOp2(v, OP_SCopy, r1, regBase+j); 1873 } 1874 if( (pTerm->eOperator & (WO_ISNULL|WO_IN))==0 ){ 1875 sqlite3VdbeAddOp2(v, OP_IsNull, regBase+j, pLevel->brk); 1876 } 1877 } 1878 return regBase; 1879 } 1880 1881 #if defined(SQLITE_TEST) 1882 /* 1883 ** The following variable holds a text description of query plan generated 1884 ** by the most recent call to sqlite3WhereBegin(). Each call to WhereBegin 1885 ** overwrites the previous. This information is used for testing and 1886 ** analysis only. 1887 */ 1888 char sqlite3_query_plan[BMS*2*40]; /* Text of the join */ 1889 static int nQPlan = 0; /* Next free slow in _query_plan[] */ 1890 1891 #endif /* SQLITE_TEST */ 1892 1893 1894 /* 1895 ** Free a WhereInfo structure 1896 */ 1897 static void whereInfoFree(WhereInfo *pWInfo){ 1898 if( pWInfo ){ 1899 int i; 1900 for(i=0; i<pWInfo->nLevel; i++){ 1901 sqlite3_index_info *pInfo = pWInfo->a[i].pIdxInfo; 1902 if( pInfo ){ 1903 assert( pInfo->needToFreeIdxStr==0 ); 1904 sqlite3_free(pInfo); 1905 } 1906 } 1907 sqlite3_free(pWInfo); 1908 } 1909 } 1910 1911 1912 /* 1913 ** Generate the beginning of the loop used for WHERE clause processing. 1914 ** The return value is a pointer to an opaque structure that contains 1915 ** information needed to terminate the loop. Later, the calling routine 1916 ** should invoke sqlite3WhereEnd() with the return value of this function 1917 ** in order to complete the WHERE clause processing. 1918 ** 1919 ** If an error occurs, this routine returns NULL. 1920 ** 1921 ** The basic idea is to do a nested loop, one loop for each table in 1922 ** the FROM clause of a select. (INSERT and UPDATE statements are the 1923 ** same as a SELECT with only a single table in the FROM clause.) For 1924 ** example, if the SQL is this: 1925 ** 1926 ** SELECT * FROM t1, t2, t3 WHERE ...; 1927 ** 1928 ** Then the code generated is conceptually like the following: 1929 ** 1930 ** foreach row1 in t1 do \ Code generated 1931 ** foreach row2 in t2 do |-- by sqlite3WhereBegin() 1932 ** foreach row3 in t3 do / 1933 ** ... 1934 ** end \ Code generated 1935 ** end |-- by sqlite3WhereEnd() 1936 ** end / 1937 ** 1938 ** Note that the loops might not be nested in the order in which they 1939 ** appear in the FROM clause if a different order is better able to make 1940 ** use of indices. Note also that when the IN operator appears in 1941 ** the WHERE clause, it might result in additional nested loops for 1942 ** scanning through all values on the right-hand side of the IN. 1943 ** 1944 ** There are Btree cursors associated with each table. t1 uses cursor 1945 ** number pTabList->a[0].iCursor. t2 uses the cursor pTabList->a[1].iCursor. 1946 ** And so forth. This routine generates code to open those VDBE cursors 1947 ** and sqlite3WhereEnd() generates the code to close them. 1948 ** 1949 ** The code that sqlite3WhereBegin() generates leaves the cursors named 1950 ** in pTabList pointing at their appropriate entries. The [...] code 1951 ** can use OP_Column and OP_Rowid opcodes on these cursors to extract 1952 ** data from the various tables of the loop. 1953 ** 1954 ** If the WHERE clause is empty, the foreach loops must each scan their 1955 ** entire tables. Thus a three-way join is an O(N^3) operation. But if 1956 ** the tables have indices and there are terms in the WHERE clause that 1957 ** refer to those indices, a complete table scan can be avoided and the 1958 ** code will run much faster. Most of the work of this routine is checking 1959 ** to see if there are indices that can be used to speed up the loop. 1960 ** 1961 ** Terms of the WHERE clause are also used to limit which rows actually 1962 ** make it to the "..." in the middle of the loop. After each "foreach", 1963 ** terms of the WHERE clause that use only terms in that loop and outer 1964 ** loops are evaluated and if false a jump is made around all subsequent 1965 ** inner loops (or around the "..." if the test occurs within the inner- 1966 ** most loop) 1967 ** 1968 ** OUTER JOINS 1969 ** 1970 ** An outer join of tables t1 and t2 is conceptally coded as follows: 1971 ** 1972 ** foreach row1 in t1 do 1973 ** flag = 0 1974 ** foreach row2 in t2 do 1975 ** start: 1976 ** ... 1977 ** flag = 1 1978 ** end 1979 ** if flag==0 then 1980 ** move the row2 cursor to a null row 1981 ** goto start 1982 ** fi 1983 ** end 1984 ** 1985 ** ORDER BY CLAUSE PROCESSING 1986 ** 1987 ** *ppOrderBy is a pointer to the ORDER BY clause of a SELECT statement, 1988 ** if there is one. If there is no ORDER BY clause or if this routine 1989 ** is called from an UPDATE or DELETE statement, then ppOrderBy is NULL. 1990 ** 1991 ** If an index can be used so that the natural output order of the table 1992 ** scan is correct for the ORDER BY clause, then that index is used and 1993 ** *ppOrderBy is set to NULL. This is an optimization that prevents an 1994 ** unnecessary sort of the result set if an index appropriate for the 1995 ** ORDER BY clause already exists. 1996 ** 1997 ** If the where clause loops cannot be arranged to provide the correct 1998 ** output order, then the *ppOrderBy is unchanged. 1999 */ 2000 WhereInfo *sqlite3WhereBegin( 2001 Parse *pParse, /* The parser context */ 2002 SrcList *pTabList, /* A list of all tables to be scanned */ 2003 Expr *pWhere, /* The WHERE clause */ 2004 ExprList **ppOrderBy, /* An ORDER BY clause, or NULL */ 2005 u8 wflags /* One of the WHERE_* flags defined in sqliteInt.h */ 2006 ){ 2007 int i; /* Loop counter */ 2008 WhereInfo *pWInfo; /* Will become the return value of this function */ 2009 Vdbe *v = pParse->pVdbe; /* The virtual database engine */ 2010 int brk, cont = 0; /* Addresses used during code generation */ 2011 Bitmask notReady; /* Cursors that are not yet positioned */ 2012 WhereTerm *pTerm; /* A single term in the WHERE clause */ 2013 ExprMaskSet maskSet; /* The expression mask set */ 2014 WhereClause wc; /* The WHERE clause is divided into these terms */ 2015 struct SrcList_item *pTabItem; /* A single entry from pTabList */ 2016 WhereLevel *pLevel; /* A single level in the pWInfo list */ 2017 int iFrom; /* First unused FROM clause element */ 2018 int andFlags; /* AND-ed combination of all wc.a[].flags */ 2019 sqlite3 *db; /* Database connection */ 2020 ExprList *pOrderBy = 0; 2021 2022 /* The number of tables in the FROM clause is limited by the number of 2023 ** bits in a Bitmask 2024 */ 2025 if( pTabList->nSrc>BMS ){ 2026 sqlite3ErrorMsg(pParse, "at most %d tables in a join", BMS); 2027 return 0; 2028 } 2029 2030 if( ppOrderBy ){ 2031 pOrderBy = *ppOrderBy; 2032 } 2033 2034 /* Split the WHERE clause into separate subexpressions where each 2035 ** subexpression is separated by an AND operator. 2036 */ 2037 initMaskSet(&maskSet); 2038 whereClauseInit(&wc, pParse, &maskSet); 2039 sqlite3ExprCodeConstants(pParse, pWhere); 2040 whereSplit(&wc, pWhere, TK_AND); 2041 2042 /* Allocate and initialize the WhereInfo structure that will become the 2043 ** return value. 2044 */ 2045 db = pParse->db; 2046 pWInfo = sqlite3DbMallocZero(db, 2047 sizeof(WhereInfo) + pTabList->nSrc*sizeof(WhereLevel)); 2048 if( db->mallocFailed ){ 2049 goto whereBeginNoMem; 2050 } 2051 pWInfo->nLevel = pTabList->nSrc; 2052 pWInfo->pParse = pParse; 2053 pWInfo->pTabList = pTabList; 2054 pWInfo->iBreak = sqlite3VdbeMakeLabel(v); 2055 2056 /* Special case: a WHERE clause that is constant. Evaluate the 2057 ** expression and either jump over all of the code or fall thru. 2058 */ 2059 if( pWhere && (pTabList->nSrc==0 || sqlite3ExprIsConstantNotJoin(pWhere)) ){ 2060 sqlite3ExprIfFalse(pParse, pWhere, pWInfo->iBreak, SQLITE_JUMPIFNULL); 2061 pWhere = 0; 2062 } 2063 2064 /* Assign a bit from the bitmask to every term in the FROM clause. 2065 ** 2066 ** When assigning bitmask values to FROM clause cursors, it must be 2067 ** the case that if X is the bitmask for the N-th FROM clause term then 2068 ** the bitmask for all FROM clause terms to the left of the N-th term 2069 ** is (X-1). An expression from the ON clause of a LEFT JOIN can use 2070 ** its Expr.iRightJoinTable value to find the bitmask of the right table 2071 ** of the join. Subtracting one from the right table bitmask gives a 2072 ** bitmask for all tables to the left of the join. Knowing the bitmask 2073 ** for all tables to the left of a left join is important. Ticket #3015. 2074 */ 2075 for(i=0; i<pTabList->nSrc; i++){ 2076 createMask(&maskSet, pTabList->a[i].iCursor); 2077 } 2078 #ifndef NDEBUG 2079 { 2080 Bitmask toTheLeft = 0; 2081 for(i=0; i<pTabList->nSrc; i++){ 2082 Bitmask m = getMask(&maskSet, pTabList->a[i].iCursor); 2083 assert( (m-1)==toTheLeft ); 2084 toTheLeft |= m; 2085 } 2086 } 2087 #endif 2088 2089 /* Analyze all of the subexpressions. Note that exprAnalyze() might 2090 ** add new virtual terms onto the end of the WHERE clause. We do not 2091 ** want to analyze these virtual terms, so start analyzing at the end 2092 ** and work forward so that the added virtual terms are never processed. 2093 */ 2094 exprAnalyzeAll(pTabList, &wc); 2095 if( db->mallocFailed ){ 2096 goto whereBeginNoMem; 2097 } 2098 2099 /* Chose the best index to use for each table in the FROM clause. 2100 ** 2101 ** This loop fills in the following fields: 2102 ** 2103 ** pWInfo->a[].pIdx The index to use for this level of the loop. 2104 ** pWInfo->a[].flags WHERE_xxx flags associated with pIdx 2105 ** pWInfo->a[].nEq The number of == and IN constraints 2106 ** pWInfo->a[].iFrom When term of the FROM clause is being coded 2107 ** pWInfo->a[].iTabCur The VDBE cursor for the database table 2108 ** pWInfo->a[].iIdxCur The VDBE cursor for the index 2109 ** 2110 ** This loop also figures out the nesting order of tables in the FROM 2111 ** clause. 2112 */ 2113 notReady = ~(Bitmask)0; 2114 pTabItem = pTabList->a; 2115 pLevel = pWInfo->a; 2116 andFlags = ~0; 2117 WHERETRACE(("*** Optimizer Start ***\n")); 2118 for(i=iFrom=0, pLevel=pWInfo->a; i<pTabList->nSrc; i++, pLevel++){ 2119 Index *pIdx; /* Index for FROM table at pTabItem */ 2120 int flags; /* Flags asssociated with pIdx */ 2121 int nEq; /* Number of == or IN constraints */ 2122 double cost; /* The cost for pIdx */ 2123 int j; /* For looping over FROM tables */ 2124 Index *pBest = 0; /* The best index seen so far */ 2125 int bestFlags = 0; /* Flags associated with pBest */ 2126 int bestNEq = 0; /* nEq associated with pBest */ 2127 double lowestCost; /* Cost of the pBest */ 2128 int bestJ = 0; /* The value of j */ 2129 Bitmask m; /* Bitmask value for j or bestJ */ 2130 int once = 0; /* True when first table is seen */ 2131 sqlite3_index_info *pIndex; /* Current virtual index */ 2132 2133 lowestCost = SQLITE_BIG_DBL; 2134 for(j=iFrom, pTabItem=&pTabList->a[j]; j<pTabList->nSrc; j++, pTabItem++){ 2135 int doNotReorder; /* True if this table should not be reordered */ 2136 2137 doNotReorder = (pTabItem->jointype & (JT_LEFT|JT_CROSS))!=0; 2138 if( once && doNotReorder ) break; 2139 m = getMask(&maskSet, pTabItem->iCursor); 2140 if( (m & notReady)==0 ){ 2141 if( j==iFrom ) iFrom++; 2142 continue; 2143 } 2144 assert( pTabItem->pTab ); 2145 #ifndef SQLITE_OMIT_VIRTUALTABLE 2146 if( IsVirtual(pTabItem->pTab) ){ 2147 sqlite3_index_info **ppIdxInfo = &pWInfo->a[j].pIdxInfo; 2148 cost = bestVirtualIndex(pParse, &wc, pTabItem, notReady, 2149 ppOrderBy ? *ppOrderBy : 0, i==0, 2150 ppIdxInfo); 2151 flags = WHERE_VIRTUALTABLE; 2152 pIndex = *ppIdxInfo; 2153 if( pIndex && pIndex->orderByConsumed ){ 2154 flags = WHERE_VIRTUALTABLE | WHERE_ORDERBY; 2155 } 2156 pIdx = 0; 2157 nEq = 0; 2158 if( (SQLITE_BIG_DBL/2.0)<cost ){ 2159 /* The cost is not allowed to be larger than SQLITE_BIG_DBL (the 2160 ** inital value of lowestCost in this loop. If it is, then 2161 ** the (cost<lowestCost) test below will never be true and 2162 ** pLevel->pBestIdx never set. 2163 */ 2164 cost = (SQLITE_BIG_DBL/2.0); 2165 } 2166 }else 2167 #endif 2168 { 2169 cost = bestIndex(pParse, &wc, pTabItem, notReady, 2170 (i==0 && ppOrderBy) ? *ppOrderBy : 0, 2171 &pIdx, &flags, &nEq); 2172 pIndex = 0; 2173 } 2174 if( cost<lowestCost ){ 2175 once = 1; 2176 lowestCost = cost; 2177 pBest = pIdx; 2178 bestFlags = flags; 2179 bestNEq = nEq; 2180 bestJ = j; 2181 pLevel->pBestIdx = pIndex; 2182 } 2183 if( doNotReorder ) break; 2184 } 2185 WHERETRACE(("*** Optimizer choose table %d for loop %d\n", bestJ, 2186 pLevel-pWInfo->a)); 2187 if( (bestFlags & WHERE_ORDERBY)!=0 ){ 2188 *ppOrderBy = 0; 2189 } 2190 andFlags &= bestFlags; 2191 pLevel->flags = bestFlags; 2192 pLevel->pIdx = pBest; 2193 pLevel->nEq = bestNEq; 2194 pLevel->aInLoop = 0; 2195 pLevel->nIn = 0; 2196 if( pBest ){ 2197 pLevel->iIdxCur = pParse->nTab++; 2198 }else{ 2199 pLevel->iIdxCur = -1; 2200 } 2201 notReady &= ~getMask(&maskSet, pTabList->a[bestJ].iCursor); 2202 pLevel->iFrom = bestJ; 2203 } 2204 WHERETRACE(("*** Optimizer Finished ***\n")); 2205 2206 /* If the total query only selects a single row, then the ORDER BY 2207 ** clause is irrelevant. 2208 */ 2209 if( (andFlags & WHERE_UNIQUE)!=0 && ppOrderBy ){ 2210 *ppOrderBy = 0; 2211 } 2212 2213 /* If the caller is an UPDATE or DELETE statement that is requesting 2214 ** to use a one-pass algorithm, determine if this is appropriate. 2215 ** The one-pass algorithm only works if the WHERE clause constraints 2216 ** the statement to update a single row. 2217 */ 2218 assert( (wflags & WHERE_ONEPASS_DESIRED)==0 || pWInfo->nLevel==1 ); 2219 if( (wflags & WHERE_ONEPASS_DESIRED)!=0 && (andFlags & WHERE_UNIQUE)!=0 ){ 2220 pWInfo->okOnePass = 1; 2221 pWInfo->a[0].flags &= ~WHERE_IDX_ONLY; 2222 } 2223 2224 /* Open all tables in the pTabList and any indices selected for 2225 ** searching those tables. 2226 */ 2227 sqlite3CodeVerifySchema(pParse, -1); /* Insert the cookie verifier Goto */ 2228 for(i=0, pLevel=pWInfo->a; i<pTabList->nSrc; i++, pLevel++){ 2229 Table *pTab; /* Table to open */ 2230 Index *pIx; /* Index used to access pTab (if any) */ 2231 int iDb; /* Index of database containing table/index */ 2232 int iIdxCur = pLevel->iIdxCur; 2233 2234 #ifndef SQLITE_OMIT_EXPLAIN 2235 if( pParse->explain==2 ){ 2236 char *zMsg; 2237 struct SrcList_item *pItem = &pTabList->a[pLevel->iFrom]; 2238 zMsg = sqlite3MPrintf(db, "TABLE %s", pItem->zName); 2239 if( pItem->zAlias ){ 2240 zMsg = sqlite3MPrintf(db, "%z AS %s", zMsg, pItem->zAlias); 2241 } 2242 if( (pIx = pLevel->pIdx)!=0 ){ 2243 zMsg = sqlite3MPrintf(db, "%z WITH INDEX %s", zMsg, pIx->zName); 2244 }else if( pLevel->flags & (WHERE_ROWID_EQ|WHERE_ROWID_RANGE) ){ 2245 zMsg = sqlite3MPrintf(db, "%z USING PRIMARY KEY", zMsg); 2246 } 2247 #ifndef SQLITE_OMIT_VIRTUALTABLE 2248 else if( pLevel->pBestIdx ){ 2249 sqlite3_index_info *pBestIdx = pLevel->pBestIdx; 2250 zMsg = sqlite3MPrintf(db, "%z VIRTUAL TABLE INDEX %d:%s", zMsg, 2251 pBestIdx->idxNum, pBestIdx->idxStr); 2252 } 2253 #endif 2254 if( pLevel->flags & WHERE_ORDERBY ){ 2255 zMsg = sqlite3MPrintf(db, "%z ORDER BY", zMsg); 2256 } 2257 sqlite3VdbeAddOp4(v, OP_Explain, i, pLevel->iFrom, 0, zMsg, P4_DYNAMIC); 2258 } 2259 #endif /* SQLITE_OMIT_EXPLAIN */ 2260 pTabItem = &pTabList->a[pLevel->iFrom]; 2261 pTab = pTabItem->pTab; 2262 iDb = sqlite3SchemaToIndex(pParse->db, pTab->pSchema); 2263 if( pTab->isEphem || pTab->pSelect ) continue; 2264 #ifndef SQLITE_OMIT_VIRTUALTABLE 2265 if( pLevel->pBestIdx ){ 2266 int iCur = pTabItem->iCursor; 2267 sqlite3VdbeAddOp4(v, OP_VOpen, iCur, 0, 0, 2268 (const char*)pTab->pVtab, P4_VTAB); 2269 }else 2270 #endif 2271 if( (pLevel->flags & WHERE_IDX_ONLY)==0 ){ 2272 int op = pWInfo->okOnePass ? OP_OpenWrite : OP_OpenRead; 2273 sqlite3OpenTable(pParse, pTabItem->iCursor, iDb, pTab, op); 2274 if( !pWInfo->okOnePass && pTab->nCol<(sizeof(Bitmask)*8) ){ 2275 Bitmask b = pTabItem->colUsed; 2276 int n = 0; 2277 for(; b; b=b>>1, n++){} 2278 sqlite3VdbeChangeP2(v, sqlite3VdbeCurrentAddr(v)-2, n); 2279 assert( n<=pTab->nCol ); 2280 } 2281 }else{ 2282 sqlite3TableLock(pParse, iDb, pTab->tnum, 0, pTab->zName); 2283 } 2284 pLevel->iTabCur = pTabItem->iCursor; 2285 if( (pIx = pLevel->pIdx)!=0 ){ 2286 KeyInfo *pKey = sqlite3IndexKeyinfo(pParse, pIx); 2287 assert( pIx->pSchema==pTab->pSchema ); 2288 sqlite3VdbeAddOp2(v, OP_SetNumColumns, 0, pIx->nColumn+1); 2289 sqlite3VdbeAddOp4(v, OP_OpenRead, iIdxCur, pIx->tnum, iDb, 2290 (char*)pKey, P4_KEYINFO_HANDOFF); 2291 VdbeComment((v, "%s", pIx->zName)); 2292 } 2293 sqlite3CodeVerifySchema(pParse, iDb); 2294 } 2295 pWInfo->iTop = sqlite3VdbeCurrentAddr(v); 2296 2297 /* Generate the code to do the search. Each iteration of the for 2298 ** loop below generates code for a single nested loop of the VM 2299 ** program. 2300 */ 2301 notReady = ~(Bitmask)0; 2302 for(i=0, pLevel=pWInfo->a; i<pTabList->nSrc; i++, pLevel++){ 2303 int j; 2304 int iCur = pTabItem->iCursor; /* The VDBE cursor for the table */ 2305 Index *pIdx; /* The index we will be using */ 2306 int nxt; /* Where to jump to continue with the next IN case */ 2307 int iIdxCur; /* The VDBE cursor for the index */ 2308 int omitTable; /* True if we use the index only */ 2309 int bRev; /* True if we need to scan in reverse order */ 2310 2311 pTabItem = &pTabList->a[pLevel->iFrom]; 2312 iCur = pTabItem->iCursor; 2313 pIdx = pLevel->pIdx; 2314 iIdxCur = pLevel->iIdxCur; 2315 bRev = (pLevel->flags & WHERE_REVERSE)!=0; 2316 omitTable = (pLevel->flags & WHERE_IDX_ONLY)!=0; 2317 2318 /* Create labels for the "break" and "continue" instructions 2319 ** for the current loop. Jump to brk to break out of a loop. 2320 ** Jump to cont to go immediately to the next iteration of the 2321 ** loop. 2322 ** 2323 ** When there is an IN operator, we also have a "nxt" label that 2324 ** means to continue with the next IN value combination. When 2325 ** there are no IN operators in the constraints, the "nxt" label 2326 ** is the same as "brk". 2327 */ 2328 brk = pLevel->brk = pLevel->nxt = sqlite3VdbeMakeLabel(v); 2329 cont = pLevel->cont = sqlite3VdbeMakeLabel(v); 2330 2331 /* If this is the right table of a LEFT OUTER JOIN, allocate and 2332 ** initialize a memory cell that records if this table matches any 2333 ** row of the left table of the join. 2334 */ 2335 if( pLevel->iFrom>0 && (pTabItem[0].jointype & JT_LEFT)!=0 ){ 2336 pLevel->iLeftJoin = ++pParse->nMem; 2337 sqlite3VdbeAddOp2(v, OP_Integer, 0, pLevel->iLeftJoin); 2338 VdbeComment((v, "init LEFT JOIN no-match flag")); 2339 } 2340 2341 #ifndef SQLITE_OMIT_VIRTUALTABLE 2342 if( pLevel->pBestIdx ){ 2343 /* Case 0: The table is a virtual-table. Use the VFilter and VNext 2344 ** to access the data. 2345 */ 2346 int j; 2347 int iReg; /* P3 Value for OP_VFilter */ 2348 sqlite3_index_info *pBestIdx = pLevel->pBestIdx; 2349 int nConstraint = pBestIdx->nConstraint; 2350 struct sqlite3_index_constraint_usage *aUsage = 2351 pBestIdx->aConstraintUsage; 2352 const struct sqlite3_index_constraint *aConstraint = 2353 pBestIdx->aConstraint; 2354 2355 iReg = sqlite3GetTempRange(pParse, nConstraint+2); 2356 for(j=1; j<=nConstraint; j++){ 2357 int k; 2358 for(k=0; k<nConstraint; k++){ 2359 if( aUsage[k].argvIndex==j ){ 2360 int iTerm = aConstraint[k].iTermOffset; 2361 sqlite3ExprCode(pParse, wc.a[iTerm].pExpr->pRight, iReg+j+1); 2362 break; 2363 } 2364 } 2365 if( k==nConstraint ) break; 2366 } 2367 sqlite3VdbeAddOp2(v, OP_Integer, pBestIdx->idxNum, iReg); 2368 sqlite3VdbeAddOp2(v, OP_Integer, j-1, iReg+1); 2369 sqlite3VdbeAddOp4(v, OP_VFilter, iCur, brk, iReg, pBestIdx->idxStr, 2370 pBestIdx->needToFreeIdxStr ? P4_MPRINTF : P4_STATIC); 2371 sqlite3ReleaseTempRange(pParse, iReg, nConstraint+2); 2372 pBestIdx->needToFreeIdxStr = 0; 2373 for(j=0; j<pBestIdx->nConstraint; j++){ 2374 if( aUsage[j].omit ){ 2375 int iTerm = aConstraint[j].iTermOffset; 2376 disableTerm(pLevel, &wc.a[iTerm]); 2377 } 2378 } 2379 pLevel->op = OP_VNext; 2380 pLevel->p1 = iCur; 2381 pLevel->p2 = sqlite3VdbeCurrentAddr(v); 2382 }else 2383 #endif /* SQLITE_OMIT_VIRTUALTABLE */ 2384 2385 if( pLevel->flags & WHERE_ROWID_EQ ){ 2386 /* Case 1: We can directly reference a single row using an 2387 ** equality comparison against the ROWID field. Or 2388 ** we reference multiple rows using a "rowid IN (...)" 2389 ** construct. 2390 */ 2391 int r1; 2392 pTerm = findTerm(&wc, iCur, -1, notReady, WO_EQ|WO_IN, 0); 2393 assert( pTerm!=0 ); 2394 assert( pTerm->pExpr!=0 ); 2395 assert( pTerm->leftCursor==iCur ); 2396 assert( omitTable==0 ); 2397 r1 = codeEqualityTerm(pParse, pTerm, pLevel, 0); 2398 nxt = pLevel->nxt; 2399 sqlite3VdbeAddOp2(v, OP_MustBeInt, r1, nxt); 2400 sqlite3VdbeAddOp3(v, OP_NotExists, iCur, nxt, r1); 2401 VdbeComment((v, "pk")); 2402 pLevel->op = OP_Noop; 2403 }else if( pLevel->flags & WHERE_ROWID_RANGE ){ 2404 /* Case 2: We have an inequality comparison against the ROWID field. 2405 */ 2406 int testOp = OP_Noop; 2407 int start; 2408 WhereTerm *pStart, *pEnd; 2409 2410 assert( omitTable==0 ); 2411 pStart = findTerm(&wc, iCur, -1, notReady, WO_GT|WO_GE, 0); 2412 pEnd = findTerm(&wc, iCur, -1, notReady, WO_LT|WO_LE, 0); 2413 if( bRev ){ 2414 pTerm = pStart; 2415 pStart = pEnd; 2416 pEnd = pTerm; 2417 } 2418 if( pStart ){ 2419 Expr *pX; 2420 int r1, regFree1; 2421 pX = pStart->pExpr; 2422 assert( pX!=0 ); 2423 assert( pStart->leftCursor==iCur ); 2424 r1 = sqlite3ExprCodeTemp(pParse, pX->pRight, ®Free1); 2425 sqlite3VdbeAddOp3(v, OP_ForceInt, r1, brk, 2426 pX->op==TK_LE || pX->op==TK_GT); 2427 sqlite3VdbeAddOp3(v, bRev ? OP_MoveLt : OP_MoveGe, iCur, brk, r1); 2428 VdbeComment((v, "pk")); 2429 sqlite3ReleaseTempReg(pParse, regFree1); 2430 disableTerm(pLevel, pStart); 2431 }else{ 2432 sqlite3VdbeAddOp2(v, bRev ? OP_Last : OP_Rewind, iCur, brk); 2433 } 2434 if( pEnd ){ 2435 Expr *pX; 2436 pX = pEnd->pExpr; 2437 assert( pX!=0 ); 2438 assert( pEnd->leftCursor==iCur ); 2439 pLevel->iMem = ++pParse->nMem; 2440 sqlite3ExprCode(pParse, pX->pRight, pLevel->iMem); 2441 if( pX->op==TK_LT || pX->op==TK_GT ){ 2442 testOp = bRev ? OP_Le : OP_Ge; 2443 }else{ 2444 testOp = bRev ? OP_Lt : OP_Gt; 2445 } 2446 disableTerm(pLevel, pEnd); 2447 } 2448 start = sqlite3VdbeCurrentAddr(v); 2449 pLevel->op = bRev ? OP_Prev : OP_Next; 2450 pLevel->p1 = iCur; 2451 pLevel->p2 = start; 2452 if( testOp!=OP_Noop ){ 2453 int r1 = sqlite3GetTempReg(pParse); 2454 sqlite3VdbeAddOp2(v, OP_Rowid, iCur, r1); 2455 /* sqlite3VdbeAddOp2(v, OP_SCopy, pLevel->iMem, 0); */ 2456 sqlite3VdbeAddOp3(v, testOp, pLevel->iMem, brk, r1); 2457 sqlite3VdbeChangeP5(v, SQLITE_AFF_NUMERIC | SQLITE_JUMPIFNULL); 2458 sqlite3ReleaseTempReg(pParse, r1); 2459 } 2460 }else if( pLevel->flags & WHERE_COLUMN_RANGE ){ 2461 /* Case 3: The WHERE clause term that refers to the right-most 2462 ** column of the index is an inequality. For example, if 2463 ** the index is on (x,y,z) and the WHERE clause is of the 2464 ** form "x=5 AND y<10" then this case is used. Only the 2465 ** right-most column can be an inequality - the rest must 2466 ** use the "==" and "IN" operators. 2467 ** 2468 ** This case is also used when there are no WHERE clause 2469 ** constraints but an index is selected anyway, in order 2470 ** to force the output order to conform to an ORDER BY. 2471 */ 2472 int start; 2473 int nEq = pLevel->nEq; 2474 int topEq=0; /* True if top limit uses ==. False is strictly < */ 2475 int btmEq=0; /* True if btm limit uses ==. False if strictly > */ 2476 int topOp, btmOp; /* Operators for the top and bottom search bounds */ 2477 int testOp; 2478 int topLimit = (pLevel->flags & WHERE_TOP_LIMIT)!=0; 2479 int btmLimit = (pLevel->flags & WHERE_BTM_LIMIT)!=0; 2480 int isMinQuery = 0; /* If this is an optimized SELECT min(x) ... */ 2481 int regBase; /* Base register holding constraint values */ 2482 int r1; /* Temp register */ 2483 2484 /* Generate code to evaluate all constraint terms using == or IN 2485 ** and level the values of those terms on the stack. 2486 */ 2487 regBase = codeAllEqualityTerms(pParse, pLevel, &wc, notReady, 2); 2488 2489 /* Figure out what comparison operators to use for top and bottom 2490 ** search bounds. For an ascending index, the bottom bound is a > or >= 2491 ** operator and the top bound is a < or <= operator. For a descending 2492 ** index the operators are reversed. 2493 */ 2494 if( pIdx->aSortOrder[nEq]==SQLITE_SO_ASC ){ 2495 topOp = WO_LT|WO_LE; 2496 btmOp = WO_GT|WO_GE; 2497 }else{ 2498 topOp = WO_GT|WO_GE; 2499 btmOp = WO_LT|WO_LE; 2500 SWAP(int, topLimit, btmLimit); 2501 } 2502 2503 /* If this loop satisfies a sort order (pOrderBy) request that 2504 ** was passed to this function to implement a "SELECT min(x) ..." 2505 ** query, then the caller will only allow the loop to run for 2506 ** a single iteration. This means that the first row returned 2507 ** should not have a NULL value stored in 'x'. If column 'x' is 2508 ** the first one after the nEq equality constraints in the index, 2509 ** this requires some special handling. 2510 */ 2511 if( (wflags&WHERE_ORDERBY_MIN)!=0 2512 && (pLevel->flags&WHERE_ORDERBY) 2513 && (pIdx->nColumn>nEq) 2514 && (pOrderBy->a[0].pExpr->iColumn==pIdx->aiColumn[nEq]) 2515 ){ 2516 isMinQuery = 1; 2517 } 2518 2519 /* Generate the termination key. This is the key value that 2520 ** will end the search. There is no termination key if there 2521 ** are no equality terms and no "X<..." term. 2522 ** 2523 ** 2002-Dec-04: On a reverse-order scan, the so-called "termination" 2524 ** key computed here really ends up being the start key. 2525 */ 2526 nxt = pLevel->nxt; 2527 if( topLimit ){ 2528 Expr *pX; 2529 int k = pIdx->aiColumn[nEq]; 2530 pTerm = findTerm(&wc, iCur, k, notReady, topOp, pIdx); 2531 assert( pTerm!=0 ); 2532 pX = pTerm->pExpr; 2533 assert( (pTerm->flags & TERM_CODED)==0 ); 2534 sqlite3ExprCode(pParse, pX->pRight, regBase+nEq); 2535 sqlite3VdbeAddOp2(v, OP_IsNull, regBase+nEq, nxt); 2536 topEq = pTerm->eOperator & (WO_LE|WO_GE); 2537 disableTerm(pLevel, pTerm); 2538 testOp = OP_IdxGE; 2539 }else{ 2540 testOp = nEq>0 ? OP_IdxGE : OP_Noop; 2541 topEq = 1; 2542 } 2543 if( testOp!=OP_Noop || (isMinQuery&&bRev) ){ 2544 int nCol = nEq + topLimit; 2545 if( isMinQuery && bRev && !topLimit ){ 2546 sqlite3VdbeAddOp2(v, OP_Null, 0, regBase+nCol); 2547 nCol++; 2548 topEq = 0; 2549 } 2550 buildIndexProbe(pParse, nCol, pIdx, regBase, pLevel->iMem); 2551 if( bRev ){ 2552 int op = topEq ? OP_MoveLe : OP_MoveLt; 2553 sqlite3VdbeAddOp3(v, op, iIdxCur, nxt, pLevel->iMem); 2554 } 2555 }else if( bRev ){ 2556 sqlite3VdbeAddOp2(v, OP_Last, iIdxCur, brk); 2557 } 2558 2559 /* Generate the start key. This is the key that defines the lower 2560 ** bound on the search. There is no start key if there are no 2561 ** equality terms and if there is no "X>..." term. In 2562 ** that case, generate a "Rewind" instruction in place of the 2563 ** start key search. 2564 ** 2565 ** 2002-Dec-04: In the case of a reverse-order search, the so-called 2566 ** "start" key really ends up being used as the termination key. 2567 */ 2568 if( btmLimit ){ 2569 Expr *pX; 2570 int k = pIdx->aiColumn[nEq]; 2571 pTerm = findTerm(&wc, iCur, k, notReady, btmOp, pIdx); 2572 assert( pTerm!=0 ); 2573 pX = pTerm->pExpr; 2574 assert( (pTerm->flags & TERM_CODED)==0 ); 2575 sqlite3ExprCode(pParse, pX->pRight, regBase+nEq); 2576 sqlite3VdbeAddOp2(v, OP_IsNull, regBase+nEq, nxt); 2577 btmEq = pTerm->eOperator & (WO_LE|WO_GE); 2578 disableTerm(pLevel, pTerm); 2579 }else{ 2580 btmEq = 1; 2581 } 2582 if( nEq>0 || btmLimit || (isMinQuery&&!bRev) ){ 2583 int nCol = nEq + btmLimit; 2584 if( isMinQuery && !bRev && !btmLimit ){ 2585 sqlite3VdbeAddOp2(v, OP_Null, 0, regBase+nCol); 2586 nCol++; 2587 btmEq = 0; 2588 } 2589 if( bRev ){ 2590 r1 = pLevel->iMem; 2591 testOp = OP_IdxLT; 2592 }else{ 2593 r1 = sqlite3GetTempReg(pParse); 2594 } 2595 buildIndexProbe(pParse, nCol, pIdx, regBase, r1); 2596 if( !bRev ){ 2597 int op = btmEq ? OP_MoveGe : OP_MoveGt; 2598 sqlite3VdbeAddOp3(v, op, iIdxCur, nxt, r1); 2599 sqlite3ReleaseTempReg(pParse, r1); 2600 } 2601 }else if( bRev ){ 2602 testOp = OP_Noop; 2603 }else{ 2604 sqlite3VdbeAddOp2(v, OP_Rewind, iIdxCur, brk); 2605 } 2606 2607 /* Generate the the top of the loop. If there is a termination 2608 ** key we have to test for that key and abort at the top of the 2609 ** loop. 2610 */ 2611 start = sqlite3VdbeCurrentAddr(v); 2612 if( testOp!=OP_Noop ){ 2613 sqlite3VdbeAddOp3(v, testOp, iIdxCur, nxt, pLevel->iMem); 2614 if( (topEq && !bRev) || (!btmEq && bRev) ){ 2615 sqlite3VdbeChangeP5(v, 1); 2616 } 2617 } 2618 r1 = sqlite3GetTempReg(pParse); 2619 if( topLimit | btmLimit ){ 2620 sqlite3VdbeAddOp3(v, OP_Column, iIdxCur, nEq, r1); 2621 sqlite3VdbeAddOp2(v, OP_IsNull, r1, cont); 2622 } 2623 if( !omitTable ){ 2624 sqlite3VdbeAddOp2(v, OP_IdxRowid, iIdxCur, r1); 2625 sqlite3VdbeAddOp3(v, OP_MoveGe, iCur, 0, r1); /* Deferred seek */ 2626 } 2627 sqlite3ReleaseTempReg(pParse, r1); 2628 2629 /* Record the instruction used to terminate the loop. 2630 */ 2631 pLevel->op = bRev ? OP_Prev : OP_Next; 2632 pLevel->p1 = iIdxCur; 2633 pLevel->p2 = start; 2634 }else if( pLevel->flags & WHERE_COLUMN_EQ ){ 2635 /* Case 4: There is an index and all terms of the WHERE clause that 2636 ** refer to the index using the "==" or "IN" operators. 2637 */ 2638 int start; 2639 int nEq = pLevel->nEq; 2640 int isMinQuery = 0; /* If this is an optimized SELECT min(x) ... */ 2641 int regBase; /* Base register of array holding constraints */ 2642 int r1; 2643 2644 /* Generate code to evaluate all constraint terms using == or IN 2645 ** and leave the values of those terms on the stack. 2646 */ 2647 regBase = codeAllEqualityTerms(pParse, pLevel, &wc, notReady, 1); 2648 nxt = pLevel->nxt; 2649 2650 if( (wflags&WHERE_ORDERBY_MIN)!=0 2651 && (pLevel->flags&WHERE_ORDERBY) 2652 && (pIdx->nColumn>nEq) 2653 && (pOrderBy->a[0].pExpr->iColumn==pIdx->aiColumn[nEq]) 2654 ){ 2655 isMinQuery = 1; 2656 buildIndexProbe(pParse, nEq, pIdx, regBase, pLevel->iMem); 2657 sqlite3VdbeAddOp2(v, OP_Null, 0, regBase+nEq); 2658 r1 = ++pParse->nMem; 2659 buildIndexProbe(pParse, nEq+1, pIdx, regBase, r1); 2660 }else{ 2661 /* Generate a single key that will be used to both start and 2662 ** terminate the search 2663 */ 2664 r1 = pLevel->iMem; 2665 buildIndexProbe(pParse, nEq, pIdx, regBase, r1); 2666 } 2667 2668 /* Generate code (1) to move to the first matching element of the table. 2669 ** Then generate code (2) that jumps to "nxt" after the cursor is past 2670 ** the last matching element of the table. The code (1) is executed 2671 ** once to initialize the search, the code (2) is executed before each 2672 ** iteration of the scan to see if the scan has finished. */ 2673 if( bRev ){ 2674 /* Scan in reverse order */ 2675 int op; 2676 if( isMinQuery ){ 2677 op = OP_MoveLt; 2678 }else{ 2679 op = OP_MoveLe; 2680 } 2681 sqlite3VdbeAddOp3(v, op, iIdxCur, nxt, r1); 2682 start = sqlite3VdbeAddOp3(v, OP_IdxLT, iIdxCur, nxt, pLevel->iMem); 2683 pLevel->op = OP_Prev; 2684 }else{ 2685 /* Scan in the forward order */ 2686 int op; 2687 if( isMinQuery ){ 2688 op = OP_MoveGt; 2689 }else{ 2690 op = OP_MoveGe; 2691 } 2692 sqlite3VdbeAddOp3(v, op, iIdxCur, nxt, r1); 2693 start = sqlite3VdbeAddOp3(v, OP_IdxGE, iIdxCur, nxt, pLevel->iMem); 2694 sqlite3VdbeChangeP5(v, 1); 2695 pLevel->op = OP_Next; 2696 } 2697 if( !omitTable ){ 2698 r1 = sqlite3GetTempReg(pParse); 2699 sqlite3VdbeAddOp2(v, OP_IdxRowid, iIdxCur, r1); 2700 sqlite3VdbeAddOp3(v, OP_MoveGe, iCur, 0, r1); /* Deferred seek */ 2701 sqlite3ReleaseTempReg(pParse, r1); 2702 } 2703 pLevel->p1 = iIdxCur; 2704 pLevel->p2 = start; 2705 }else{ 2706 /* Case 5: There is no usable index. We must do a complete 2707 ** scan of the entire table. 2708 */ 2709 assert( omitTable==0 ); 2710 assert( bRev==0 ); 2711 pLevel->op = OP_Next; 2712 pLevel->p1 = iCur; 2713 pLevel->p2 = 1 + sqlite3VdbeAddOp2(v, OP_Rewind, iCur, brk); 2714 } 2715 notReady &= ~getMask(&maskSet, iCur); 2716 2717 /* Insert code to test every subexpression that can be completely 2718 ** computed using the current set of tables. 2719 */ 2720 for(pTerm=wc.a, j=wc.nTerm; j>0; j--, pTerm++){ 2721 Expr *pE; 2722 if( pTerm->flags & (TERM_VIRTUAL|TERM_CODED) ) continue; 2723 if( (pTerm->prereqAll & notReady)!=0 ) continue; 2724 pE = pTerm->pExpr; 2725 assert( pE!=0 ); 2726 if( pLevel->iLeftJoin && !ExprHasProperty(pE, EP_FromJoin) ){ 2727 continue; 2728 } 2729 sqlite3ExprIfFalse(pParse, pE, cont, SQLITE_JUMPIFNULL); 2730 pTerm->flags |= TERM_CODED; 2731 } 2732 2733 /* For a LEFT OUTER JOIN, generate code that will record the fact that 2734 ** at least one row of the right table has matched the left table. 2735 */ 2736 if( pLevel->iLeftJoin ){ 2737 pLevel->top = sqlite3VdbeCurrentAddr(v); 2738 sqlite3VdbeAddOp2(v, OP_Integer, 1, pLevel->iLeftJoin); 2739 VdbeComment((v, "record LEFT JOIN hit")); 2740 sqlite3ExprClearColumnCache(pParse, pLevel->iTabCur); 2741 sqlite3ExprClearColumnCache(pParse, pLevel->iIdxCur); 2742 for(pTerm=wc.a, j=0; j<wc.nTerm; j++, pTerm++){ 2743 if( pTerm->flags & (TERM_VIRTUAL|TERM_CODED) ) continue; 2744 if( (pTerm->prereqAll & notReady)!=0 ) continue; 2745 assert( pTerm->pExpr ); 2746 sqlite3ExprIfFalse(pParse, pTerm->pExpr, cont, SQLITE_JUMPIFNULL); 2747 pTerm->flags |= TERM_CODED; 2748 } 2749 } 2750 } 2751 2752 #ifdef SQLITE_TEST /* For testing and debugging use only */ 2753 /* Record in the query plan information about the current table 2754 ** and the index used to access it (if any). If the table itself 2755 ** is not used, its name is just '{}'. If no index is used 2756 ** the index is listed as "{}". If the primary key is used the 2757 ** index name is '*'. 2758 */ 2759 for(i=0; i<pTabList->nSrc; i++){ 2760 char *z; 2761 int n; 2762 pLevel = &pWInfo->a[i]; 2763 pTabItem = &pTabList->a[pLevel->iFrom]; 2764 z = pTabItem->zAlias; 2765 if( z==0 ) z = pTabItem->pTab->zName; 2766 n = strlen(z); 2767 if( n+nQPlan < sizeof(sqlite3_query_plan)-10 ){ 2768 if( pLevel->flags & WHERE_IDX_ONLY ){ 2769 memcpy(&sqlite3_query_plan[nQPlan], "{}", 2); 2770 nQPlan += 2; 2771 }else{ 2772 memcpy(&sqlite3_query_plan[nQPlan], z, n); 2773 nQPlan += n; 2774 } 2775 sqlite3_query_plan[nQPlan++] = ' '; 2776 } 2777 if( pLevel->flags & (WHERE_ROWID_EQ|WHERE_ROWID_RANGE) ){ 2778 memcpy(&sqlite3_query_plan[nQPlan], "* ", 2); 2779 nQPlan += 2; 2780 }else if( pLevel->pIdx==0 ){ 2781 memcpy(&sqlite3_query_plan[nQPlan], "{} ", 3); 2782 nQPlan += 3; 2783 }else{ 2784 n = strlen(pLevel->pIdx->zName); 2785 if( n+nQPlan < sizeof(sqlite3_query_plan)-2 ){ 2786 memcpy(&sqlite3_query_plan[nQPlan], pLevel->pIdx->zName, n); 2787 nQPlan += n; 2788 sqlite3_query_plan[nQPlan++] = ' '; 2789 } 2790 } 2791 } 2792 while( nQPlan>0 && sqlite3_query_plan[nQPlan-1]==' ' ){ 2793 sqlite3_query_plan[--nQPlan] = 0; 2794 } 2795 sqlite3_query_plan[nQPlan] = 0; 2796 nQPlan = 0; 2797 #endif /* SQLITE_TEST // Testing and debugging use only */ 2798 2799 /* Record the continuation address in the WhereInfo structure. Then 2800 ** clean up and return. 2801 */ 2802 pWInfo->iContinue = cont; 2803 whereClauseClear(&wc); 2804 return pWInfo; 2805 2806 /* Jump here if malloc fails */ 2807 whereBeginNoMem: 2808 whereClauseClear(&wc); 2809 whereInfoFree(pWInfo); 2810 return 0; 2811 } 2812 2813 /* 2814 ** Generate the end of the WHERE loop. See comments on 2815 ** sqlite3WhereBegin() for additional information. 2816 */ 2817 void sqlite3WhereEnd(WhereInfo *pWInfo){ 2818 Vdbe *v = pWInfo->pParse->pVdbe; 2819 int i; 2820 WhereLevel *pLevel; 2821 SrcList *pTabList = pWInfo->pTabList; 2822 2823 /* Generate loop termination code. 2824 */ 2825 sqlite3ExprClearColumnCache(pWInfo->pParse, -1); 2826 for(i=pTabList->nSrc-1; i>=0; i--){ 2827 pLevel = &pWInfo->a[i]; 2828 sqlite3VdbeResolveLabel(v, pLevel->cont); 2829 if( pLevel->op!=OP_Noop ){ 2830 sqlite3VdbeAddOp2(v, pLevel->op, pLevel->p1, pLevel->p2); 2831 } 2832 if( pLevel->nIn ){ 2833 struct InLoop *pIn; 2834 int j; 2835 sqlite3VdbeResolveLabel(v, pLevel->nxt); 2836 for(j=pLevel->nIn, pIn=&pLevel->aInLoop[j-1]; j>0; j--, pIn--){ 2837 sqlite3VdbeJumpHere(v, pIn->topAddr+1); 2838 sqlite3VdbeAddOp2(v, OP_Next, pIn->iCur, pIn->topAddr); 2839 sqlite3VdbeJumpHere(v, pIn->topAddr-1); 2840 } 2841 sqlite3_free(pLevel->aInLoop); 2842 } 2843 sqlite3VdbeResolveLabel(v, pLevel->brk); 2844 if( pLevel->iLeftJoin ){ 2845 int addr; 2846 addr = sqlite3VdbeAddOp1(v, OP_IfPos, pLevel->iLeftJoin); 2847 sqlite3VdbeAddOp1(v, OP_NullRow, pTabList->a[i].iCursor); 2848 if( pLevel->iIdxCur>=0 ){ 2849 sqlite3VdbeAddOp1(v, OP_NullRow, pLevel->iIdxCur); 2850 } 2851 sqlite3VdbeAddOp2(v, OP_Goto, 0, pLevel->top); 2852 sqlite3VdbeJumpHere(v, addr); 2853 } 2854 } 2855 2856 /* The "break" point is here, just past the end of the outer loop. 2857 ** Set it. 2858 */ 2859 sqlite3VdbeResolveLabel(v, pWInfo->iBreak); 2860 2861 /* Close all of the cursors that were opened by sqlite3WhereBegin. 2862 */ 2863 for(i=0, pLevel=pWInfo->a; i<pTabList->nSrc; i++, pLevel++){ 2864 struct SrcList_item *pTabItem = &pTabList->a[pLevel->iFrom]; 2865 Table *pTab = pTabItem->pTab; 2866 assert( pTab!=0 ); 2867 if( pTab->isEphem || pTab->pSelect ) continue; 2868 if( !pWInfo->okOnePass && (pLevel->flags & WHERE_IDX_ONLY)==0 ){ 2869 sqlite3VdbeAddOp1(v, OP_Close, pTabItem->iCursor); 2870 } 2871 if( pLevel->pIdx!=0 ){ 2872 sqlite3VdbeAddOp1(v, OP_Close, pLevel->iIdxCur); 2873 } 2874 2875 /* If this scan uses an index, make code substitutions to read data 2876 ** from the index in preference to the table. Sometimes, this means 2877 ** the table need never be read from. This is a performance boost, 2878 ** as the vdbe level waits until the table is read before actually 2879 ** seeking the table cursor to the record corresponding to the current 2880 ** position in the index. 2881 ** 2882 ** Calls to the code generator in between sqlite3WhereBegin and 2883 ** sqlite3WhereEnd will have created code that references the table 2884 ** directly. This loop scans all that code looking for opcodes 2885 ** that reference the table and converts them into opcodes that 2886 ** reference the index. 2887 */ 2888 if( pLevel->pIdx ){ 2889 int k, j, last; 2890 VdbeOp *pOp; 2891 Index *pIdx = pLevel->pIdx; 2892 int useIndexOnly = pLevel->flags & WHERE_IDX_ONLY; 2893 2894 assert( pIdx!=0 ); 2895 pOp = sqlite3VdbeGetOp(v, pWInfo->iTop); 2896 last = sqlite3VdbeCurrentAddr(v); 2897 for(k=pWInfo->iTop; k<last; k++, pOp++){ 2898 if( pOp->p1!=pLevel->iTabCur ) continue; 2899 if( pOp->opcode==OP_Column ){ 2900 for(j=0; j<pIdx->nColumn; j++){ 2901 if( pOp->p2==pIdx->aiColumn[j] ){ 2902 pOp->p2 = j; 2903 pOp->p1 = pLevel->iIdxCur; 2904 break; 2905 } 2906 } 2907 assert(!useIndexOnly || j<pIdx->nColumn); 2908 }else if( pOp->opcode==OP_Rowid ){ 2909 pOp->p1 = pLevel->iIdxCur; 2910 pOp->opcode = OP_IdxRowid; 2911 }else if( pOp->opcode==OP_NullRow && useIndexOnly ){ 2912 pOp->opcode = OP_Noop; 2913 } 2914 } 2915 } 2916 } 2917 2918 /* Final cleanup 2919 */ 2920 whereInfoFree(pWInfo); 2921 return; 2922 } 2923