Extra ".selecttrace" output following aggregate analysis. No changes to
[sqlite.git] / src / select.c
blob0f73c30f64fa2aac2b21d10675873dc897386953
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.
11 *************************************************************************
12 ** This file contains C code routines that are called by the parser
13 ** to handle SELECT statements in SQLite.
15 #include "sqliteInt.h"
18 ** Trace output macros
20 #if SELECTTRACE_ENABLED
21 /***/ int sqlite3SelectTrace = 0;
22 # define SELECTTRACE(K,P,S,X) \
23 if(sqlite3SelectTrace&(K)) \
24 sqlite3DebugPrintf("%*s%s.%p: ",(P)->nSelectIndent*2-2,"",\
25 (S)->zSelName,(S)),\
26 sqlite3DebugPrintf X
27 #else
28 # define SELECTTRACE(K,P,S,X)
29 #endif
33 ** An instance of the following object is used to record information about
34 ** how to process the DISTINCT keyword, to simplify passing that information
35 ** into the selectInnerLoop() routine.
37 typedef struct DistinctCtx DistinctCtx;
38 struct DistinctCtx {
39 u8 isTnct; /* True if the DISTINCT keyword is present */
40 u8 eTnctType; /* One of the WHERE_DISTINCT_* operators */
41 int tabTnct; /* Ephemeral table used for DISTINCT processing */
42 int addrTnct; /* Address of OP_OpenEphemeral opcode for tabTnct */
46 ** An instance of the following object is used to record information about
47 ** the ORDER BY (or GROUP BY) clause of query is being coded.
49 typedef struct SortCtx SortCtx;
50 struct SortCtx {
51 ExprList *pOrderBy; /* The ORDER BY (or GROUP BY clause) */
52 int nOBSat; /* Number of ORDER BY terms satisfied by indices */
53 int iECursor; /* Cursor number for the sorter */
54 int regReturn; /* Register holding block-output return address */
55 int labelBkOut; /* Start label for the block-output subroutine */
56 int addrSortIndex; /* Address of the OP_SorterOpen or OP_OpenEphemeral */
57 int labelDone; /* Jump here when done, ex: LIMIT reached */
58 u8 sortFlags; /* Zero or more SORTFLAG_* bits */
59 u8 bOrderedInnerLoop; /* ORDER BY correctly sorts the inner loop */
61 #define SORTFLAG_UseSorter 0x01 /* Use SorterOpen instead of OpenEphemeral */
64 ** Delete all the content of a Select structure. Deallocate the structure
65 ** itself only if bFree is true.
67 static void clearSelect(sqlite3 *db, Select *p, int bFree){
68 while( p ){
69 Select *pPrior = p->pPrior;
70 sqlite3ExprListDelete(db, p->pEList);
71 sqlite3SrcListDelete(db, p->pSrc);
72 sqlite3ExprDelete(db, p->pWhere);
73 sqlite3ExprListDelete(db, p->pGroupBy);
74 sqlite3ExprDelete(db, p->pHaving);
75 sqlite3ExprListDelete(db, p->pOrderBy);
76 sqlite3ExprDelete(db, p->pLimit);
77 sqlite3ExprDelete(db, p->pOffset);
78 if( OK_IF_ALWAYS_TRUE(p->pWith) ) sqlite3WithDelete(db, p->pWith);
79 if( bFree ) sqlite3DbFreeNN(db, p);
80 p = pPrior;
81 bFree = 1;
86 ** Initialize a SelectDest structure.
88 void sqlite3SelectDestInit(SelectDest *pDest, int eDest, int iParm){
89 pDest->eDest = (u8)eDest;
90 pDest->iSDParm = iParm;
91 pDest->zAffSdst = 0;
92 pDest->iSdst = 0;
93 pDest->nSdst = 0;
98 ** Allocate a new Select structure and return a pointer to that
99 ** structure.
101 Select *sqlite3SelectNew(
102 Parse *pParse, /* Parsing context */
103 ExprList *pEList, /* which columns to include in the result */
104 SrcList *pSrc, /* the FROM clause -- which tables to scan */
105 Expr *pWhere, /* the WHERE clause */
106 ExprList *pGroupBy, /* the GROUP BY clause */
107 Expr *pHaving, /* the HAVING clause */
108 ExprList *pOrderBy, /* the ORDER BY clause */
109 u32 selFlags, /* Flag parameters, such as SF_Distinct */
110 Expr *pLimit, /* LIMIT value. NULL means not used */
111 Expr *pOffset /* OFFSET value. NULL means no offset */
113 Select *pNew;
114 Select standin;
115 pNew = sqlite3DbMallocRawNN(pParse->db, sizeof(*pNew) );
116 if( pNew==0 ){
117 assert( pParse->db->mallocFailed );
118 pNew = &standin;
120 if( pEList==0 ){
121 pEList = sqlite3ExprListAppend(pParse, 0,
122 sqlite3Expr(pParse->db,TK_ASTERISK,0));
124 pNew->pEList = pEList;
125 pNew->op = TK_SELECT;
126 pNew->selFlags = selFlags;
127 pNew->iLimit = 0;
128 pNew->iOffset = 0;
129 #if SELECTTRACE_ENABLED
130 pNew->zSelName[0] = 0;
131 #endif
132 pNew->addrOpenEphm[0] = -1;
133 pNew->addrOpenEphm[1] = -1;
134 pNew->nSelectRow = 0;
135 if( pSrc==0 ) pSrc = sqlite3DbMallocZero(pParse->db, sizeof(*pSrc));
136 pNew->pSrc = pSrc;
137 pNew->pWhere = pWhere;
138 pNew->pGroupBy = pGroupBy;
139 pNew->pHaving = pHaving;
140 pNew->pOrderBy = pOrderBy;
141 pNew->pPrior = 0;
142 pNew->pNext = 0;
143 pNew->pLimit = pLimit;
144 pNew->pOffset = pOffset;
145 pNew->pWith = 0;
146 assert( pOffset==0 || pLimit!=0 || pParse->nErr>0
147 || pParse->db->mallocFailed!=0 );
148 if( pParse->db->mallocFailed ) {
149 clearSelect(pParse->db, pNew, pNew!=&standin);
150 pNew = 0;
151 }else{
152 assert( pNew->pSrc!=0 || pParse->nErr>0 );
154 assert( pNew!=&standin );
155 return pNew;
158 #if SELECTTRACE_ENABLED
160 ** Set the name of a Select object
162 void sqlite3SelectSetName(Select *p, const char *zName){
163 if( p && zName ){
164 sqlite3_snprintf(sizeof(p->zSelName), p->zSelName, "%s", zName);
167 #endif
171 ** Delete the given Select structure and all of its substructures.
173 void sqlite3SelectDelete(sqlite3 *db, Select *p){
174 if( OK_IF_ALWAYS_TRUE(p) ) clearSelect(db, p, 1);
178 ** Return a pointer to the right-most SELECT statement in a compound.
180 static Select *findRightmost(Select *p){
181 while( p->pNext ) p = p->pNext;
182 return p;
186 ** Given 1 to 3 identifiers preceding the JOIN keyword, determine the
187 ** type of join. Return an integer constant that expresses that type
188 ** in terms of the following bit values:
190 ** JT_INNER
191 ** JT_CROSS
192 ** JT_OUTER
193 ** JT_NATURAL
194 ** JT_LEFT
195 ** JT_RIGHT
197 ** A full outer join is the combination of JT_LEFT and JT_RIGHT.
199 ** If an illegal or unsupported join type is seen, then still return
200 ** a join type, but put an error in the pParse structure.
202 int sqlite3JoinType(Parse *pParse, Token *pA, Token *pB, Token *pC){
203 int jointype = 0;
204 Token *apAll[3];
205 Token *p;
206 /* 0123456789 123456789 123456789 123 */
207 static const char zKeyText[] = "naturaleftouterightfullinnercross";
208 static const struct {
209 u8 i; /* Beginning of keyword text in zKeyText[] */
210 u8 nChar; /* Length of the keyword in characters */
211 u8 code; /* Join type mask */
212 } aKeyword[] = {
213 /* natural */ { 0, 7, JT_NATURAL },
214 /* left */ { 6, 4, JT_LEFT|JT_OUTER },
215 /* outer */ { 10, 5, JT_OUTER },
216 /* right */ { 14, 5, JT_RIGHT|JT_OUTER },
217 /* full */ { 19, 4, JT_LEFT|JT_RIGHT|JT_OUTER },
218 /* inner */ { 23, 5, JT_INNER },
219 /* cross */ { 28, 5, JT_INNER|JT_CROSS },
221 int i, j;
222 apAll[0] = pA;
223 apAll[1] = pB;
224 apAll[2] = pC;
225 for(i=0; i<3 && apAll[i]; i++){
226 p = apAll[i];
227 for(j=0; j<ArraySize(aKeyword); j++){
228 if( p->n==aKeyword[j].nChar
229 && sqlite3StrNICmp((char*)p->z, &zKeyText[aKeyword[j].i], p->n)==0 ){
230 jointype |= aKeyword[j].code;
231 break;
234 testcase( j==0 || j==1 || j==2 || j==3 || j==4 || j==5 || j==6 );
235 if( j>=ArraySize(aKeyword) ){
236 jointype |= JT_ERROR;
237 break;
241 (jointype & (JT_INNER|JT_OUTER))==(JT_INNER|JT_OUTER) ||
242 (jointype & JT_ERROR)!=0
244 const char *zSp = " ";
245 assert( pB!=0 );
246 if( pC==0 ){ zSp++; }
247 sqlite3ErrorMsg(pParse, "unknown or unsupported join type: "
248 "%T %T%s%T", pA, pB, zSp, pC);
249 jointype = JT_INNER;
250 }else if( (jointype & JT_OUTER)!=0
251 && (jointype & (JT_LEFT|JT_RIGHT))!=JT_LEFT ){
252 sqlite3ErrorMsg(pParse,
253 "RIGHT and FULL OUTER JOINs are not currently supported");
254 jointype = JT_INNER;
256 return jointype;
260 ** Return the index of a column in a table. Return -1 if the column
261 ** is not contained in the table.
263 static int columnIndex(Table *pTab, const char *zCol){
264 int i;
265 for(i=0; i<pTab->nCol; i++){
266 if( sqlite3StrICmp(pTab->aCol[i].zName, zCol)==0 ) return i;
268 return -1;
272 ** Search the first N tables in pSrc, from left to right, looking for a
273 ** table that has a column named zCol.
275 ** When found, set *piTab and *piCol to the table index and column index
276 ** of the matching column and return TRUE.
278 ** If not found, return FALSE.
280 static int tableAndColumnIndex(
281 SrcList *pSrc, /* Array of tables to search */
282 int N, /* Number of tables in pSrc->a[] to search */
283 const char *zCol, /* Name of the column we are looking for */
284 int *piTab, /* Write index of pSrc->a[] here */
285 int *piCol /* Write index of pSrc->a[*piTab].pTab->aCol[] here */
287 int i; /* For looping over tables in pSrc */
288 int iCol; /* Index of column matching zCol */
290 assert( (piTab==0)==(piCol==0) ); /* Both or neither are NULL */
291 for(i=0; i<N; i++){
292 iCol = columnIndex(pSrc->a[i].pTab, zCol);
293 if( iCol>=0 ){
294 if( piTab ){
295 *piTab = i;
296 *piCol = iCol;
298 return 1;
301 return 0;
305 ** This function is used to add terms implied by JOIN syntax to the
306 ** WHERE clause expression of a SELECT statement. The new term, which
307 ** is ANDed with the existing WHERE clause, is of the form:
309 ** (tab1.col1 = tab2.col2)
311 ** where tab1 is the iSrc'th table in SrcList pSrc and tab2 is the
312 ** (iSrc+1)'th. Column col1 is column iColLeft of tab1, and col2 is
313 ** column iColRight of tab2.
315 static void addWhereTerm(
316 Parse *pParse, /* Parsing context */
317 SrcList *pSrc, /* List of tables in FROM clause */
318 int iLeft, /* Index of first table to join in pSrc */
319 int iColLeft, /* Index of column in first table */
320 int iRight, /* Index of second table in pSrc */
321 int iColRight, /* Index of column in second table */
322 int isOuterJoin, /* True if this is an OUTER join */
323 Expr **ppWhere /* IN/OUT: The WHERE clause to add to */
325 sqlite3 *db = pParse->db;
326 Expr *pE1;
327 Expr *pE2;
328 Expr *pEq;
330 assert( iLeft<iRight );
331 assert( pSrc->nSrc>iRight );
332 assert( pSrc->a[iLeft].pTab );
333 assert( pSrc->a[iRight].pTab );
335 pE1 = sqlite3CreateColumnExpr(db, pSrc, iLeft, iColLeft);
336 pE2 = sqlite3CreateColumnExpr(db, pSrc, iRight, iColRight);
338 pEq = sqlite3PExpr(pParse, TK_EQ, pE1, pE2);
339 if( pEq && isOuterJoin ){
340 ExprSetProperty(pEq, EP_FromJoin);
341 assert( !ExprHasProperty(pEq, EP_TokenOnly|EP_Reduced) );
342 ExprSetVVAProperty(pEq, EP_NoReduce);
343 pEq->iRightJoinTable = (i16)pE2->iTable;
345 *ppWhere = sqlite3ExprAnd(db, *ppWhere, pEq);
349 ** Set the EP_FromJoin property on all terms of the given expression.
350 ** And set the Expr.iRightJoinTable to iTable for every term in the
351 ** expression.
353 ** The EP_FromJoin property is used on terms of an expression to tell
354 ** the LEFT OUTER JOIN processing logic that this term is part of the
355 ** join restriction specified in the ON or USING clause and not a part
356 ** of the more general WHERE clause. These terms are moved over to the
357 ** WHERE clause during join processing but we need to remember that they
358 ** originated in the ON or USING clause.
360 ** The Expr.iRightJoinTable tells the WHERE clause processing that the
361 ** expression depends on table iRightJoinTable even if that table is not
362 ** explicitly mentioned in the expression. That information is needed
363 ** for cases like this:
365 ** SELECT * FROM t1 LEFT JOIN t2 ON t1.a=t2.b AND t1.x=5
367 ** The where clause needs to defer the handling of the t1.x=5
368 ** term until after the t2 loop of the join. In that way, a
369 ** NULL t2 row will be inserted whenever t1.x!=5. If we do not
370 ** defer the handling of t1.x=5, it will be processed immediately
371 ** after the t1 loop and rows with t1.x!=5 will never appear in
372 ** the output, which is incorrect.
374 static void setJoinExpr(Expr *p, int iTable){
375 while( p ){
376 ExprSetProperty(p, EP_FromJoin);
377 assert( !ExprHasProperty(p, EP_TokenOnly|EP_Reduced) );
378 ExprSetVVAProperty(p, EP_NoReduce);
379 p->iRightJoinTable = (i16)iTable;
380 if( p->op==TK_FUNCTION && p->x.pList ){
381 int i;
382 for(i=0; i<p->x.pList->nExpr; i++){
383 setJoinExpr(p->x.pList->a[i].pExpr, iTable);
386 setJoinExpr(p->pLeft, iTable);
387 p = p->pRight;
392 ** This routine processes the join information for a SELECT statement.
393 ** ON and USING clauses are converted into extra terms of the WHERE clause.
394 ** NATURAL joins also create extra WHERE clause terms.
396 ** The terms of a FROM clause are contained in the Select.pSrc structure.
397 ** The left most table is the first entry in Select.pSrc. The right-most
398 ** table is the last entry. The join operator is held in the entry to
399 ** the left. Thus entry 0 contains the join operator for the join between
400 ** entries 0 and 1. Any ON or USING clauses associated with the join are
401 ** also attached to the left entry.
403 ** This routine returns the number of errors encountered.
405 static int sqliteProcessJoin(Parse *pParse, Select *p){
406 SrcList *pSrc; /* All tables in the FROM clause */
407 int i, j; /* Loop counters */
408 struct SrcList_item *pLeft; /* Left table being joined */
409 struct SrcList_item *pRight; /* Right table being joined */
411 pSrc = p->pSrc;
412 pLeft = &pSrc->a[0];
413 pRight = &pLeft[1];
414 for(i=0; i<pSrc->nSrc-1; i++, pRight++, pLeft++){
415 Table *pRightTab = pRight->pTab;
416 int isOuter;
418 if( NEVER(pLeft->pTab==0 || pRightTab==0) ) continue;
419 isOuter = (pRight->fg.jointype & JT_OUTER)!=0;
421 /* When the NATURAL keyword is present, add WHERE clause terms for
422 ** every column that the two tables have in common.
424 if( pRight->fg.jointype & JT_NATURAL ){
425 if( pRight->pOn || pRight->pUsing ){
426 sqlite3ErrorMsg(pParse, "a NATURAL join may not have "
427 "an ON or USING clause", 0);
428 return 1;
430 for(j=0; j<pRightTab->nCol; j++){
431 char *zName; /* Name of column in the right table */
432 int iLeft; /* Matching left table */
433 int iLeftCol; /* Matching column in the left table */
435 zName = pRightTab->aCol[j].zName;
436 if( tableAndColumnIndex(pSrc, i+1, zName, &iLeft, &iLeftCol) ){
437 addWhereTerm(pParse, pSrc, iLeft, iLeftCol, i+1, j,
438 isOuter, &p->pWhere);
443 /* Disallow both ON and USING clauses in the same join
445 if( pRight->pOn && pRight->pUsing ){
446 sqlite3ErrorMsg(pParse, "cannot have both ON and USING "
447 "clauses in the same join");
448 return 1;
451 /* Add the ON clause to the end of the WHERE clause, connected by
452 ** an AND operator.
454 if( pRight->pOn ){
455 if( isOuter ) setJoinExpr(pRight->pOn, pRight->iCursor);
456 p->pWhere = sqlite3ExprAnd(pParse->db, p->pWhere, pRight->pOn);
457 pRight->pOn = 0;
460 /* Create extra terms on the WHERE clause for each column named
461 ** in the USING clause. Example: If the two tables to be joined are
462 ** A and B and the USING clause names X, Y, and Z, then add this
463 ** to the WHERE clause: A.X=B.X AND A.Y=B.Y AND A.Z=B.Z
464 ** Report an error if any column mentioned in the USING clause is
465 ** not contained in both tables to be joined.
467 if( pRight->pUsing ){
468 IdList *pList = pRight->pUsing;
469 for(j=0; j<pList->nId; j++){
470 char *zName; /* Name of the term in the USING clause */
471 int iLeft; /* Table on the left with matching column name */
472 int iLeftCol; /* Column number of matching column on the left */
473 int iRightCol; /* Column number of matching column on the right */
475 zName = pList->a[j].zName;
476 iRightCol = columnIndex(pRightTab, zName);
477 if( iRightCol<0
478 || !tableAndColumnIndex(pSrc, i+1, zName, &iLeft, &iLeftCol)
480 sqlite3ErrorMsg(pParse, "cannot join using column %s - column "
481 "not present in both tables", zName);
482 return 1;
484 addWhereTerm(pParse, pSrc, iLeft, iLeftCol, i+1, iRightCol,
485 isOuter, &p->pWhere);
489 return 0;
492 /* Forward reference */
493 static KeyInfo *keyInfoFromExprList(
494 Parse *pParse, /* Parsing context */
495 ExprList *pList, /* Form the KeyInfo object from this ExprList */
496 int iStart, /* Begin with this column of pList */
497 int nExtra /* Add this many extra columns to the end */
501 ** Generate code that will push the record in registers regData
502 ** through regData+nData-1 onto the sorter.
504 static void pushOntoSorter(
505 Parse *pParse, /* Parser context */
506 SortCtx *pSort, /* Information about the ORDER BY clause */
507 Select *pSelect, /* The whole SELECT statement */
508 int regData, /* First register holding data to be sorted */
509 int regOrigData, /* First register holding data before packing */
510 int nData, /* Number of elements in the data array */
511 int nPrefixReg /* No. of reg prior to regData available for use */
513 Vdbe *v = pParse->pVdbe; /* Stmt under construction */
514 int bSeq = ((pSort->sortFlags & SORTFLAG_UseSorter)==0);
515 int nExpr = pSort->pOrderBy->nExpr; /* No. of ORDER BY terms */
516 int nBase = nExpr + bSeq + nData; /* Fields in sorter record */
517 int regBase; /* Regs for sorter record */
518 int regRecord = ++pParse->nMem; /* Assembled sorter record */
519 int nOBSat = pSort->nOBSat; /* ORDER BY terms to skip */
520 int op; /* Opcode to add sorter record to sorter */
521 int iLimit; /* LIMIT counter */
523 assert( bSeq==0 || bSeq==1 );
524 assert( nData==1 || regData==regOrigData || regOrigData==0 );
525 if( nPrefixReg ){
526 assert( nPrefixReg==nExpr+bSeq );
527 regBase = regData - nExpr - bSeq;
528 }else{
529 regBase = pParse->nMem + 1;
530 pParse->nMem += nBase;
532 assert( pSelect->iOffset==0 || pSelect->iLimit!=0 );
533 iLimit = pSelect->iOffset ? pSelect->iOffset+1 : pSelect->iLimit;
534 pSort->labelDone = sqlite3VdbeMakeLabel(v);
535 sqlite3ExprCodeExprList(pParse, pSort->pOrderBy, regBase, regOrigData,
536 SQLITE_ECEL_DUP | (regOrigData? SQLITE_ECEL_REF : 0));
537 if( bSeq ){
538 sqlite3VdbeAddOp2(v, OP_Sequence, pSort->iECursor, regBase+nExpr);
540 if( nPrefixReg==0 && nData>0 ){
541 sqlite3ExprCodeMove(pParse, regData, regBase+nExpr+bSeq, nData);
543 sqlite3VdbeAddOp3(v, OP_MakeRecord, regBase+nOBSat, nBase-nOBSat, regRecord);
544 if( nOBSat>0 ){
545 int regPrevKey; /* The first nOBSat columns of the previous row */
546 int addrFirst; /* Address of the OP_IfNot opcode */
547 int addrJmp; /* Address of the OP_Jump opcode */
548 VdbeOp *pOp; /* Opcode that opens the sorter */
549 int nKey; /* Number of sorting key columns, including OP_Sequence */
550 KeyInfo *pKI; /* Original KeyInfo on the sorter table */
552 regPrevKey = pParse->nMem+1;
553 pParse->nMem += pSort->nOBSat;
554 nKey = nExpr - pSort->nOBSat + bSeq;
555 if( bSeq ){
556 addrFirst = sqlite3VdbeAddOp1(v, OP_IfNot, regBase+nExpr);
557 }else{
558 addrFirst = sqlite3VdbeAddOp1(v, OP_SequenceTest, pSort->iECursor);
560 VdbeCoverage(v);
561 sqlite3VdbeAddOp3(v, OP_Compare, regPrevKey, regBase, pSort->nOBSat);
562 pOp = sqlite3VdbeGetOp(v, pSort->addrSortIndex);
563 if( pParse->db->mallocFailed ) return;
564 pOp->p2 = nKey + nData;
565 pKI = pOp->p4.pKeyInfo;
566 memset(pKI->aSortOrder, 0, pKI->nKeyField); /* Makes OP_Jump testable */
567 sqlite3VdbeChangeP4(v, -1, (char*)pKI, P4_KEYINFO);
568 testcase( pKI->nAllField > pKI->nKeyField+2 );
569 pOp->p4.pKeyInfo = keyInfoFromExprList(pParse, pSort->pOrderBy, nOBSat,
570 pKI->nAllField-pKI->nKeyField-1);
571 addrJmp = sqlite3VdbeCurrentAddr(v);
572 sqlite3VdbeAddOp3(v, OP_Jump, addrJmp+1, 0, addrJmp+1); VdbeCoverage(v);
573 pSort->labelBkOut = sqlite3VdbeMakeLabel(v);
574 pSort->regReturn = ++pParse->nMem;
575 sqlite3VdbeAddOp2(v, OP_Gosub, pSort->regReturn, pSort->labelBkOut);
576 sqlite3VdbeAddOp1(v, OP_ResetSorter, pSort->iECursor);
577 if( iLimit ){
578 sqlite3VdbeAddOp2(v, OP_IfNot, iLimit, pSort->labelDone);
579 VdbeCoverage(v);
581 sqlite3VdbeJumpHere(v, addrFirst);
582 sqlite3ExprCodeMove(pParse, regBase, regPrevKey, pSort->nOBSat);
583 sqlite3VdbeJumpHere(v, addrJmp);
585 if( pSort->sortFlags & SORTFLAG_UseSorter ){
586 op = OP_SorterInsert;
587 }else{
588 op = OP_IdxInsert;
590 sqlite3VdbeAddOp4Int(v, op, pSort->iECursor, regRecord,
591 regBase+nOBSat, nBase-nOBSat);
592 if( iLimit ){
593 int addr;
594 int r1 = 0;
595 /* Fill the sorter until it contains LIMIT+OFFSET entries. (The iLimit
596 ** register is initialized with value of LIMIT+OFFSET.) After the sorter
597 ** fills up, delete the least entry in the sorter after each insert.
598 ** Thus we never hold more than the LIMIT+OFFSET rows in memory at once */
599 addr = sqlite3VdbeAddOp1(v, OP_IfNotZero, iLimit); VdbeCoverage(v);
600 sqlite3VdbeAddOp1(v, OP_Last, pSort->iECursor);
601 if( pSort->bOrderedInnerLoop ){
602 r1 = ++pParse->nMem;
603 sqlite3VdbeAddOp3(v, OP_Column, pSort->iECursor, nExpr, r1);
604 VdbeComment((v, "seq"));
606 sqlite3VdbeAddOp1(v, OP_Delete, pSort->iECursor);
607 if( pSort->bOrderedInnerLoop ){
608 /* If the inner loop is driven by an index such that values from
609 ** the same iteration of the inner loop are in sorted order, then
610 ** immediately jump to the next iteration of an inner loop if the
611 ** entry from the current iteration does not fit into the top
612 ** LIMIT+OFFSET entries of the sorter. */
613 int iBrk = sqlite3VdbeCurrentAddr(v) + 2;
614 sqlite3VdbeAddOp3(v, OP_Eq, regBase+nExpr, iBrk, r1);
615 sqlite3VdbeChangeP5(v, SQLITE_NULLEQ);
616 VdbeCoverage(v);
618 sqlite3VdbeJumpHere(v, addr);
623 ** Add code to implement the OFFSET
625 static void codeOffset(
626 Vdbe *v, /* Generate code into this VM */
627 int iOffset, /* Register holding the offset counter */
628 int iContinue /* Jump here to skip the current record */
630 if( iOffset>0 ){
631 sqlite3VdbeAddOp3(v, OP_IfPos, iOffset, iContinue, 1); VdbeCoverage(v);
632 VdbeComment((v, "OFFSET"));
637 ** Add code that will check to make sure the N registers starting at iMem
638 ** form a distinct entry. iTab is a sorting index that holds previously
639 ** seen combinations of the N values. A new entry is made in iTab
640 ** if the current N values are new.
642 ** A jump to addrRepeat is made and the N+1 values are popped from the
643 ** stack if the top N elements are not distinct.
645 static void codeDistinct(
646 Parse *pParse, /* Parsing and code generating context */
647 int iTab, /* A sorting index used to test for distinctness */
648 int addrRepeat, /* Jump to here if not distinct */
649 int N, /* Number of elements */
650 int iMem /* First element */
652 Vdbe *v;
653 int r1;
655 v = pParse->pVdbe;
656 r1 = sqlite3GetTempReg(pParse);
657 sqlite3VdbeAddOp4Int(v, OP_Found, iTab, addrRepeat, iMem, N); VdbeCoverage(v);
658 sqlite3VdbeAddOp3(v, OP_MakeRecord, iMem, N, r1);
659 sqlite3VdbeAddOp4Int(v, OP_IdxInsert, iTab, r1, iMem, N);
660 sqlite3VdbeChangeP5(v, OPFLAG_USESEEKRESULT);
661 sqlite3ReleaseTempReg(pParse, r1);
665 ** This routine generates the code for the inside of the inner loop
666 ** of a SELECT.
668 ** If srcTab is negative, then the p->pEList expressions
669 ** are evaluated in order to get the data for this row. If srcTab is
670 ** zero or more, then data is pulled from srcTab and p->pEList is used only
671 ** to get the number of columns and the collation sequence for each column.
673 static void selectInnerLoop(
674 Parse *pParse, /* The parser context */
675 Select *p, /* The complete select statement being coded */
676 int srcTab, /* Pull data from this table if non-negative */
677 SortCtx *pSort, /* If not NULL, info on how to process ORDER BY */
678 DistinctCtx *pDistinct, /* If not NULL, info on how to process DISTINCT */
679 SelectDest *pDest, /* How to dispose of the results */
680 int iContinue, /* Jump here to continue with next row */
681 int iBreak /* Jump here to break out of the inner loop */
683 Vdbe *v = pParse->pVdbe;
684 int i;
685 int hasDistinct; /* True if the DISTINCT keyword is present */
686 int eDest = pDest->eDest; /* How to dispose of results */
687 int iParm = pDest->iSDParm; /* First argument to disposal method */
688 int nResultCol; /* Number of result columns */
689 int nPrefixReg = 0; /* Number of extra registers before regResult */
691 /* Usually, regResult is the first cell in an array of memory cells
692 ** containing the current result row. In this case regOrig is set to the
693 ** same value. However, if the results are being sent to the sorter, the
694 ** values for any expressions that are also part of the sort-key are omitted
695 ** from this array. In this case regOrig is set to zero. */
696 int regResult; /* Start of memory holding current results */
697 int regOrig; /* Start of memory holding full result (or 0) */
699 assert( v );
700 assert( p->pEList!=0 );
701 hasDistinct = pDistinct ? pDistinct->eTnctType : WHERE_DISTINCT_NOOP;
702 if( pSort && pSort->pOrderBy==0 ) pSort = 0;
703 if( pSort==0 && !hasDistinct ){
704 assert( iContinue!=0 );
705 codeOffset(v, p->iOffset, iContinue);
708 /* Pull the requested columns.
710 nResultCol = p->pEList->nExpr;
712 if( pDest->iSdst==0 ){
713 if( pSort ){
714 nPrefixReg = pSort->pOrderBy->nExpr;
715 if( !(pSort->sortFlags & SORTFLAG_UseSorter) ) nPrefixReg++;
716 pParse->nMem += nPrefixReg;
718 pDest->iSdst = pParse->nMem+1;
719 pParse->nMem += nResultCol;
720 }else if( pDest->iSdst+nResultCol > pParse->nMem ){
721 /* This is an error condition that can result, for example, when a SELECT
722 ** on the right-hand side of an INSERT contains more result columns than
723 ** there are columns in the table on the left. The error will be caught
724 ** and reported later. But we need to make sure enough memory is allocated
725 ** to avoid other spurious errors in the meantime. */
726 pParse->nMem += nResultCol;
728 pDest->nSdst = nResultCol;
729 regOrig = regResult = pDest->iSdst;
730 if( srcTab>=0 ){
731 for(i=0; i<nResultCol; i++){
732 sqlite3VdbeAddOp3(v, OP_Column, srcTab, i, regResult+i);
733 VdbeComment((v, "%s", p->pEList->a[i].zName));
735 }else if( eDest!=SRT_Exists ){
736 /* If the destination is an EXISTS(...) expression, the actual
737 ** values returned by the SELECT are not required.
739 u8 ecelFlags;
740 if( eDest==SRT_Mem || eDest==SRT_Output || eDest==SRT_Coroutine ){
741 ecelFlags = SQLITE_ECEL_DUP;
742 }else{
743 ecelFlags = 0;
745 if( pSort && hasDistinct==0 && eDest!=SRT_EphemTab && eDest!=SRT_Table ){
746 /* For each expression in p->pEList that is a copy of an expression in
747 ** the ORDER BY clause (pSort->pOrderBy), set the associated
748 ** iOrderByCol value to one more than the index of the ORDER BY
749 ** expression within the sort-key that pushOntoSorter() will generate.
750 ** This allows the p->pEList field to be omitted from the sorted record,
751 ** saving space and CPU cycles. */
752 ecelFlags |= (SQLITE_ECEL_OMITREF|SQLITE_ECEL_REF);
753 for(i=pSort->nOBSat; i<pSort->pOrderBy->nExpr; i++){
754 int j;
755 if( (j = pSort->pOrderBy->a[i].u.x.iOrderByCol)>0 ){
756 p->pEList->a[j-1].u.x.iOrderByCol = i+1-pSort->nOBSat;
759 regOrig = 0;
760 assert( eDest==SRT_Set || eDest==SRT_Mem
761 || eDest==SRT_Coroutine || eDest==SRT_Output );
763 nResultCol = sqlite3ExprCodeExprList(pParse,p->pEList,regResult,
764 0,ecelFlags);
767 /* If the DISTINCT keyword was present on the SELECT statement
768 ** and this row has been seen before, then do not make this row
769 ** part of the result.
771 if( hasDistinct ){
772 switch( pDistinct->eTnctType ){
773 case WHERE_DISTINCT_ORDERED: {
774 VdbeOp *pOp; /* No longer required OpenEphemeral instr. */
775 int iJump; /* Jump destination */
776 int regPrev; /* Previous row content */
778 /* Allocate space for the previous row */
779 regPrev = pParse->nMem+1;
780 pParse->nMem += nResultCol;
782 /* Change the OP_OpenEphemeral coded earlier to an OP_Null
783 ** sets the MEM_Cleared bit on the first register of the
784 ** previous value. This will cause the OP_Ne below to always
785 ** fail on the first iteration of the loop even if the first
786 ** row is all NULLs.
788 sqlite3VdbeChangeToNoop(v, pDistinct->addrTnct);
789 pOp = sqlite3VdbeGetOp(v, pDistinct->addrTnct);
790 pOp->opcode = OP_Null;
791 pOp->p1 = 1;
792 pOp->p2 = regPrev;
794 iJump = sqlite3VdbeCurrentAddr(v) + nResultCol;
795 for(i=0; i<nResultCol; i++){
796 CollSeq *pColl = sqlite3ExprCollSeq(pParse, p->pEList->a[i].pExpr);
797 if( i<nResultCol-1 ){
798 sqlite3VdbeAddOp3(v, OP_Ne, regResult+i, iJump, regPrev+i);
799 VdbeCoverage(v);
800 }else{
801 sqlite3VdbeAddOp3(v, OP_Eq, regResult+i, iContinue, regPrev+i);
802 VdbeCoverage(v);
804 sqlite3VdbeChangeP4(v, -1, (const char *)pColl, P4_COLLSEQ);
805 sqlite3VdbeChangeP5(v, SQLITE_NULLEQ);
807 assert( sqlite3VdbeCurrentAddr(v)==iJump || pParse->db->mallocFailed );
808 sqlite3VdbeAddOp3(v, OP_Copy, regResult, regPrev, nResultCol-1);
809 break;
812 case WHERE_DISTINCT_UNIQUE: {
813 sqlite3VdbeChangeToNoop(v, pDistinct->addrTnct);
814 break;
817 default: {
818 assert( pDistinct->eTnctType==WHERE_DISTINCT_UNORDERED );
819 codeDistinct(pParse, pDistinct->tabTnct, iContinue, nResultCol,
820 regResult);
821 break;
824 if( pSort==0 ){
825 codeOffset(v, p->iOffset, iContinue);
829 switch( eDest ){
830 /* In this mode, write each query result to the key of the temporary
831 ** table iParm.
833 #ifndef SQLITE_OMIT_COMPOUND_SELECT
834 case SRT_Union: {
835 int r1;
836 r1 = sqlite3GetTempReg(pParse);
837 sqlite3VdbeAddOp3(v, OP_MakeRecord, regResult, nResultCol, r1);
838 sqlite3VdbeAddOp4Int(v, OP_IdxInsert, iParm, r1, regResult, nResultCol);
839 sqlite3ReleaseTempReg(pParse, r1);
840 break;
843 /* Construct a record from the query result, but instead of
844 ** saving that record, use it as a key to delete elements from
845 ** the temporary table iParm.
847 case SRT_Except: {
848 sqlite3VdbeAddOp3(v, OP_IdxDelete, iParm, regResult, nResultCol);
849 break;
851 #endif /* SQLITE_OMIT_COMPOUND_SELECT */
853 /* Store the result as data using a unique key.
855 case SRT_Fifo:
856 case SRT_DistFifo:
857 case SRT_Table:
858 case SRT_EphemTab: {
859 int r1 = sqlite3GetTempRange(pParse, nPrefixReg+1);
860 testcase( eDest==SRT_Table );
861 testcase( eDest==SRT_EphemTab );
862 testcase( eDest==SRT_Fifo );
863 testcase( eDest==SRT_DistFifo );
864 sqlite3VdbeAddOp3(v, OP_MakeRecord, regResult, nResultCol, r1+nPrefixReg);
865 #ifndef SQLITE_OMIT_CTE
866 if( eDest==SRT_DistFifo ){
867 /* If the destination is DistFifo, then cursor (iParm+1) is open
868 ** on an ephemeral index. If the current row is already present
869 ** in the index, do not write it to the output. If not, add the
870 ** current row to the index and proceed with writing it to the
871 ** output table as well. */
872 int addr = sqlite3VdbeCurrentAddr(v) + 4;
873 sqlite3VdbeAddOp4Int(v, OP_Found, iParm+1, addr, r1, 0);
874 VdbeCoverage(v);
875 sqlite3VdbeAddOp4Int(v, OP_IdxInsert, iParm+1, r1,regResult,nResultCol);
876 assert( pSort==0 );
878 #endif
879 if( pSort ){
880 pushOntoSorter(pParse, pSort, p, r1+nPrefixReg,regResult,1,nPrefixReg);
881 }else{
882 int r2 = sqlite3GetTempReg(pParse);
883 sqlite3VdbeAddOp2(v, OP_NewRowid, iParm, r2);
884 sqlite3VdbeAddOp3(v, OP_Insert, iParm, r1, r2);
885 sqlite3VdbeChangeP5(v, OPFLAG_APPEND);
886 sqlite3ReleaseTempReg(pParse, r2);
888 sqlite3ReleaseTempRange(pParse, r1, nPrefixReg+1);
889 break;
892 #ifndef SQLITE_OMIT_SUBQUERY
893 /* If we are creating a set for an "expr IN (SELECT ...)" construct,
894 ** then there should be a single item on the stack. Write this
895 ** item into the set table with bogus data.
897 case SRT_Set: {
898 if( pSort ){
899 /* At first glance you would think we could optimize out the
900 ** ORDER BY in this case since the order of entries in the set
901 ** does not matter. But there might be a LIMIT clause, in which
902 ** case the order does matter */
903 pushOntoSorter(
904 pParse, pSort, p, regResult, regOrig, nResultCol, nPrefixReg);
905 }else{
906 int r1 = sqlite3GetTempReg(pParse);
907 assert( sqlite3Strlen30(pDest->zAffSdst)==nResultCol );
908 sqlite3VdbeAddOp4(v, OP_MakeRecord, regResult, nResultCol,
909 r1, pDest->zAffSdst, nResultCol);
910 sqlite3ExprCacheAffinityChange(pParse, regResult, nResultCol);
911 sqlite3VdbeAddOp4Int(v, OP_IdxInsert, iParm, r1, regResult, nResultCol);
912 sqlite3ReleaseTempReg(pParse, r1);
914 break;
917 /* If any row exist in the result set, record that fact and abort.
919 case SRT_Exists: {
920 sqlite3VdbeAddOp2(v, OP_Integer, 1, iParm);
921 /* The LIMIT clause will terminate the loop for us */
922 break;
925 /* If this is a scalar select that is part of an expression, then
926 ** store the results in the appropriate memory cell or array of
927 ** memory cells and break out of the scan loop.
929 case SRT_Mem: {
930 if( pSort ){
931 assert( nResultCol<=pDest->nSdst );
932 pushOntoSorter(
933 pParse, pSort, p, regResult, regOrig, nResultCol, nPrefixReg);
934 }else{
935 assert( nResultCol==pDest->nSdst );
936 assert( regResult==iParm );
937 /* The LIMIT clause will jump out of the loop for us */
939 break;
941 #endif /* #ifndef SQLITE_OMIT_SUBQUERY */
943 case SRT_Coroutine: /* Send data to a co-routine */
944 case SRT_Output: { /* Return the results */
945 testcase( eDest==SRT_Coroutine );
946 testcase( eDest==SRT_Output );
947 if( pSort ){
948 pushOntoSorter(pParse, pSort, p, regResult, regOrig, nResultCol,
949 nPrefixReg);
950 }else if( eDest==SRT_Coroutine ){
951 sqlite3VdbeAddOp1(v, OP_Yield, pDest->iSDParm);
952 }else{
953 sqlite3VdbeAddOp2(v, OP_ResultRow, regResult, nResultCol);
954 sqlite3ExprCacheAffinityChange(pParse, regResult, nResultCol);
956 break;
959 #ifndef SQLITE_OMIT_CTE
960 /* Write the results into a priority queue that is order according to
961 ** pDest->pOrderBy (in pSO). pDest->iSDParm (in iParm) is the cursor for an
962 ** index with pSO->nExpr+2 columns. Build a key using pSO for the first
963 ** pSO->nExpr columns, then make sure all keys are unique by adding a
964 ** final OP_Sequence column. The last column is the record as a blob.
966 case SRT_DistQueue:
967 case SRT_Queue: {
968 int nKey;
969 int r1, r2, r3;
970 int addrTest = 0;
971 ExprList *pSO;
972 pSO = pDest->pOrderBy;
973 assert( pSO );
974 nKey = pSO->nExpr;
975 r1 = sqlite3GetTempReg(pParse);
976 r2 = sqlite3GetTempRange(pParse, nKey+2);
977 r3 = r2+nKey+1;
978 if( eDest==SRT_DistQueue ){
979 /* If the destination is DistQueue, then cursor (iParm+1) is open
980 ** on a second ephemeral index that holds all values every previously
981 ** added to the queue. */
982 addrTest = sqlite3VdbeAddOp4Int(v, OP_Found, iParm+1, 0,
983 regResult, nResultCol);
984 VdbeCoverage(v);
986 sqlite3VdbeAddOp3(v, OP_MakeRecord, regResult, nResultCol, r3);
987 if( eDest==SRT_DistQueue ){
988 sqlite3VdbeAddOp2(v, OP_IdxInsert, iParm+1, r3);
989 sqlite3VdbeChangeP5(v, OPFLAG_USESEEKRESULT);
991 for(i=0; i<nKey; i++){
992 sqlite3VdbeAddOp2(v, OP_SCopy,
993 regResult + pSO->a[i].u.x.iOrderByCol - 1,
994 r2+i);
996 sqlite3VdbeAddOp2(v, OP_Sequence, iParm, r2+nKey);
997 sqlite3VdbeAddOp3(v, OP_MakeRecord, r2, nKey+2, r1);
998 sqlite3VdbeAddOp4Int(v, OP_IdxInsert, iParm, r1, r2, nKey+2);
999 if( addrTest ) sqlite3VdbeJumpHere(v, addrTest);
1000 sqlite3ReleaseTempReg(pParse, r1);
1001 sqlite3ReleaseTempRange(pParse, r2, nKey+2);
1002 break;
1004 #endif /* SQLITE_OMIT_CTE */
1008 #if !defined(SQLITE_OMIT_TRIGGER)
1009 /* Discard the results. This is used for SELECT statements inside
1010 ** the body of a TRIGGER. The purpose of such selects is to call
1011 ** user-defined functions that have side effects. We do not care
1012 ** about the actual results of the select.
1014 default: {
1015 assert( eDest==SRT_Discard );
1016 break;
1018 #endif
1021 /* Jump to the end of the loop if the LIMIT is reached. Except, if
1022 ** there is a sorter, in which case the sorter has already limited
1023 ** the output for us.
1025 if( pSort==0 && p->iLimit ){
1026 sqlite3VdbeAddOp2(v, OP_DecrJumpZero, p->iLimit, iBreak); VdbeCoverage(v);
1031 ** Allocate a KeyInfo object sufficient for an index of N key columns and
1032 ** X extra columns.
1034 KeyInfo *sqlite3KeyInfoAlloc(sqlite3 *db, int N, int X){
1035 int nExtra = (N+X)*(sizeof(CollSeq*)+1) - sizeof(CollSeq*);
1036 KeyInfo *p = sqlite3DbMallocRawNN(db, sizeof(KeyInfo) + nExtra);
1037 if( p ){
1038 p->aSortOrder = (u8*)&p->aColl[N+X];
1039 p->nKeyField = (u16)N;
1040 p->nAllField = (u16)(N+X);
1041 p->enc = ENC(db);
1042 p->db = db;
1043 p->nRef = 1;
1044 memset(&p[1], 0, nExtra);
1045 }else{
1046 sqlite3OomFault(db);
1048 return p;
1052 ** Deallocate a KeyInfo object
1054 void sqlite3KeyInfoUnref(KeyInfo *p){
1055 if( p ){
1056 assert( p->nRef>0 );
1057 p->nRef--;
1058 if( p->nRef==0 ) sqlite3DbFreeNN(p->db, p);
1063 ** Make a new pointer to a KeyInfo object
1065 KeyInfo *sqlite3KeyInfoRef(KeyInfo *p){
1066 if( p ){
1067 assert( p->nRef>0 );
1068 p->nRef++;
1070 return p;
1073 #ifdef SQLITE_DEBUG
1075 ** Return TRUE if a KeyInfo object can be change. The KeyInfo object
1076 ** can only be changed if this is just a single reference to the object.
1078 ** This routine is used only inside of assert() statements.
1080 int sqlite3KeyInfoIsWriteable(KeyInfo *p){ return p->nRef==1; }
1081 #endif /* SQLITE_DEBUG */
1084 ** Given an expression list, generate a KeyInfo structure that records
1085 ** the collating sequence for each expression in that expression list.
1087 ** If the ExprList is an ORDER BY or GROUP BY clause then the resulting
1088 ** KeyInfo structure is appropriate for initializing a virtual index to
1089 ** implement that clause. If the ExprList is the result set of a SELECT
1090 ** then the KeyInfo structure is appropriate for initializing a virtual
1091 ** index to implement a DISTINCT test.
1093 ** Space to hold the KeyInfo structure is obtained from malloc. The calling
1094 ** function is responsible for seeing that this structure is eventually
1095 ** freed.
1097 static KeyInfo *keyInfoFromExprList(
1098 Parse *pParse, /* Parsing context */
1099 ExprList *pList, /* Form the KeyInfo object from this ExprList */
1100 int iStart, /* Begin with this column of pList */
1101 int nExtra /* Add this many extra columns to the end */
1103 int nExpr;
1104 KeyInfo *pInfo;
1105 struct ExprList_item *pItem;
1106 sqlite3 *db = pParse->db;
1107 int i;
1109 nExpr = pList->nExpr;
1110 pInfo = sqlite3KeyInfoAlloc(db, nExpr-iStart, nExtra+1);
1111 if( pInfo ){
1112 assert( sqlite3KeyInfoIsWriteable(pInfo) );
1113 for(i=iStart, pItem=pList->a+iStart; i<nExpr; i++, pItem++){
1114 pInfo->aColl[i-iStart] = sqlite3ExprNNCollSeq(pParse, pItem->pExpr);
1115 pInfo->aSortOrder[i-iStart] = pItem->sortOrder;
1118 return pInfo;
1122 ** Name of the connection operator, used for error messages.
1124 static const char *selectOpName(int id){
1125 char *z;
1126 switch( id ){
1127 case TK_ALL: z = "UNION ALL"; break;
1128 case TK_INTERSECT: z = "INTERSECT"; break;
1129 case TK_EXCEPT: z = "EXCEPT"; break;
1130 default: z = "UNION"; break;
1132 return z;
1135 #ifndef SQLITE_OMIT_EXPLAIN
1137 ** Unless an "EXPLAIN QUERY PLAN" command is being processed, this function
1138 ** is a no-op. Otherwise, it adds a single row of output to the EQP result,
1139 ** where the caption is of the form:
1141 ** "USE TEMP B-TREE FOR xxx"
1143 ** where xxx is one of "DISTINCT", "ORDER BY" or "GROUP BY". Exactly which
1144 ** is determined by the zUsage argument.
1146 static void explainTempTable(Parse *pParse, const char *zUsage){
1147 if( pParse->explain==2 ){
1148 Vdbe *v = pParse->pVdbe;
1149 char *zMsg = sqlite3MPrintf(pParse->db, "USE TEMP B-TREE FOR %s", zUsage);
1150 sqlite3VdbeAddOp4(v, OP_Explain, pParse->iSelectId, 0, 0, zMsg, P4_DYNAMIC);
1155 ** Assign expression b to lvalue a. A second, no-op, version of this macro
1156 ** is provided when SQLITE_OMIT_EXPLAIN is defined. This allows the code
1157 ** in sqlite3Select() to assign values to structure member variables that
1158 ** only exist if SQLITE_OMIT_EXPLAIN is not defined without polluting the
1159 ** code with #ifndef directives.
1161 # define explainSetInteger(a, b) a = b
1163 #else
1164 /* No-op versions of the explainXXX() functions and macros. */
1165 # define explainTempTable(y,z)
1166 # define explainSetInteger(y,z)
1167 #endif
1169 #if !defined(SQLITE_OMIT_EXPLAIN) && !defined(SQLITE_OMIT_COMPOUND_SELECT)
1171 ** Unless an "EXPLAIN QUERY PLAN" command is being processed, this function
1172 ** is a no-op. Otherwise, it adds a single row of output to the EQP result,
1173 ** where the caption is of one of the two forms:
1175 ** "COMPOSITE SUBQUERIES iSub1 and iSub2 (op)"
1176 ** "COMPOSITE SUBQUERIES iSub1 and iSub2 USING TEMP B-TREE (op)"
1178 ** where iSub1 and iSub2 are the integers passed as the corresponding
1179 ** function parameters, and op is the text representation of the parameter
1180 ** of the same name. The parameter "op" must be one of TK_UNION, TK_EXCEPT,
1181 ** TK_INTERSECT or TK_ALL. The first form is used if argument bUseTmp is
1182 ** false, or the second form if it is true.
1184 static void explainComposite(
1185 Parse *pParse, /* Parse context */
1186 int op, /* One of TK_UNION, TK_EXCEPT etc. */
1187 int iSub1, /* Subquery id 1 */
1188 int iSub2, /* Subquery id 2 */
1189 int bUseTmp /* True if a temp table was used */
1191 assert( op==TK_UNION || op==TK_EXCEPT || op==TK_INTERSECT || op==TK_ALL );
1192 if( pParse->explain==2 ){
1193 Vdbe *v = pParse->pVdbe;
1194 char *zMsg = sqlite3MPrintf(
1195 pParse->db, "COMPOUND SUBQUERIES %d AND %d %s(%s)", iSub1, iSub2,
1196 bUseTmp?"USING TEMP B-TREE ":"", selectOpName(op)
1198 sqlite3VdbeAddOp4(v, OP_Explain, pParse->iSelectId, 0, 0, zMsg, P4_DYNAMIC);
1201 #else
1202 /* No-op versions of the explainXXX() functions and macros. */
1203 # define explainComposite(v,w,x,y,z)
1204 #endif
1207 ** If the inner loop was generated using a non-null pOrderBy argument,
1208 ** then the results were placed in a sorter. After the loop is terminated
1209 ** we need to run the sorter and output the results. The following
1210 ** routine generates the code needed to do that.
1212 static void generateSortTail(
1213 Parse *pParse, /* Parsing context */
1214 Select *p, /* The SELECT statement */
1215 SortCtx *pSort, /* Information on the ORDER BY clause */
1216 int nColumn, /* Number of columns of data */
1217 SelectDest *pDest /* Write the sorted results here */
1219 Vdbe *v = pParse->pVdbe; /* The prepared statement */
1220 int addrBreak = pSort->labelDone; /* Jump here to exit loop */
1221 int addrContinue = sqlite3VdbeMakeLabel(v); /* Jump here for next cycle */
1222 int addr;
1223 int addrOnce = 0;
1224 int iTab;
1225 ExprList *pOrderBy = pSort->pOrderBy;
1226 int eDest = pDest->eDest;
1227 int iParm = pDest->iSDParm;
1228 int regRow;
1229 int regRowid;
1230 int iCol;
1231 int nKey;
1232 int iSortTab; /* Sorter cursor to read from */
1233 int nSortData; /* Trailing values to read from sorter */
1234 int i;
1235 int bSeq; /* True if sorter record includes seq. no. */
1236 struct ExprList_item *aOutEx = p->pEList->a;
1238 assert( addrBreak<0 );
1239 if( pSort->labelBkOut ){
1240 sqlite3VdbeAddOp2(v, OP_Gosub, pSort->regReturn, pSort->labelBkOut);
1241 sqlite3VdbeGoto(v, addrBreak);
1242 sqlite3VdbeResolveLabel(v, pSort->labelBkOut);
1244 iTab = pSort->iECursor;
1245 if( eDest==SRT_Output || eDest==SRT_Coroutine || eDest==SRT_Mem ){
1246 regRowid = 0;
1247 regRow = pDest->iSdst;
1248 nSortData = nColumn;
1249 }else{
1250 regRowid = sqlite3GetTempReg(pParse);
1251 regRow = sqlite3GetTempRange(pParse, nColumn);
1252 nSortData = nColumn;
1254 nKey = pOrderBy->nExpr - pSort->nOBSat;
1255 if( pSort->sortFlags & SORTFLAG_UseSorter ){
1256 int regSortOut = ++pParse->nMem;
1257 iSortTab = pParse->nTab++;
1258 if( pSort->labelBkOut ){
1259 addrOnce = sqlite3VdbeAddOp0(v, OP_Once); VdbeCoverage(v);
1261 sqlite3VdbeAddOp3(v, OP_OpenPseudo, iSortTab, regSortOut, nKey+1+nSortData);
1262 if( addrOnce ) sqlite3VdbeJumpHere(v, addrOnce);
1263 addr = 1 + sqlite3VdbeAddOp2(v, OP_SorterSort, iTab, addrBreak);
1264 VdbeCoverage(v);
1265 codeOffset(v, p->iOffset, addrContinue);
1266 sqlite3VdbeAddOp3(v, OP_SorterData, iTab, regSortOut, iSortTab);
1267 bSeq = 0;
1268 }else{
1269 addr = 1 + sqlite3VdbeAddOp2(v, OP_Sort, iTab, addrBreak); VdbeCoverage(v);
1270 codeOffset(v, p->iOffset, addrContinue);
1271 iSortTab = iTab;
1272 bSeq = 1;
1274 for(i=0, iCol=nKey+bSeq; i<nSortData; i++){
1275 int iRead;
1276 if( aOutEx[i].u.x.iOrderByCol ){
1277 iRead = aOutEx[i].u.x.iOrderByCol-1;
1278 }else{
1279 iRead = iCol++;
1281 sqlite3VdbeAddOp3(v, OP_Column, iSortTab, iRead, regRow+i);
1282 VdbeComment((v, "%s", aOutEx[i].zName ? aOutEx[i].zName : aOutEx[i].zSpan));
1284 switch( eDest ){
1285 case SRT_Table:
1286 case SRT_EphemTab: {
1287 sqlite3VdbeAddOp2(v, OP_NewRowid, iParm, regRowid);
1288 sqlite3VdbeAddOp3(v, OP_Insert, iParm, regRow, regRowid);
1289 sqlite3VdbeChangeP5(v, OPFLAG_APPEND);
1290 break;
1292 #ifndef SQLITE_OMIT_SUBQUERY
1293 case SRT_Set: {
1294 assert( nColumn==sqlite3Strlen30(pDest->zAffSdst) );
1295 sqlite3VdbeAddOp4(v, OP_MakeRecord, regRow, nColumn, regRowid,
1296 pDest->zAffSdst, nColumn);
1297 sqlite3ExprCacheAffinityChange(pParse, regRow, nColumn);
1298 sqlite3VdbeAddOp4Int(v, OP_IdxInsert, iParm, regRowid, regRow, nColumn);
1299 break;
1301 case SRT_Mem: {
1302 /* The LIMIT clause will terminate the loop for us */
1303 break;
1305 #endif
1306 default: {
1307 assert( eDest==SRT_Output || eDest==SRT_Coroutine );
1308 testcase( eDest==SRT_Output );
1309 testcase( eDest==SRT_Coroutine );
1310 if( eDest==SRT_Output ){
1311 sqlite3VdbeAddOp2(v, OP_ResultRow, pDest->iSdst, nColumn);
1312 sqlite3ExprCacheAffinityChange(pParse, pDest->iSdst, nColumn);
1313 }else{
1314 sqlite3VdbeAddOp1(v, OP_Yield, pDest->iSDParm);
1316 break;
1319 if( regRowid ){
1320 if( eDest==SRT_Set ){
1321 sqlite3ReleaseTempRange(pParse, regRow, nColumn);
1322 }else{
1323 sqlite3ReleaseTempReg(pParse, regRow);
1325 sqlite3ReleaseTempReg(pParse, regRowid);
1327 /* The bottom of the loop
1329 sqlite3VdbeResolveLabel(v, addrContinue);
1330 if( pSort->sortFlags & SORTFLAG_UseSorter ){
1331 sqlite3VdbeAddOp2(v, OP_SorterNext, iTab, addr); VdbeCoverage(v);
1332 }else{
1333 sqlite3VdbeAddOp2(v, OP_Next, iTab, addr); VdbeCoverage(v);
1335 if( pSort->regReturn ) sqlite3VdbeAddOp1(v, OP_Return, pSort->regReturn);
1336 sqlite3VdbeResolveLabel(v, addrBreak);
1340 ** Return a pointer to a string containing the 'declaration type' of the
1341 ** expression pExpr. The string may be treated as static by the caller.
1343 ** Also try to estimate the size of the returned value and return that
1344 ** result in *pEstWidth.
1346 ** The declaration type is the exact datatype definition extracted from the
1347 ** original CREATE TABLE statement if the expression is a column. The
1348 ** declaration type for a ROWID field is INTEGER. Exactly when an expression
1349 ** is considered a column can be complex in the presence of subqueries. The
1350 ** result-set expression in all of the following SELECT statements is
1351 ** considered a column by this function.
1353 ** SELECT col FROM tbl;
1354 ** SELECT (SELECT col FROM tbl;
1355 ** SELECT (SELECT col FROM tbl);
1356 ** SELECT abc FROM (SELECT col AS abc FROM tbl);
1358 ** The declaration type for any expression other than a column is NULL.
1360 ** This routine has either 3 or 6 parameters depending on whether or not
1361 ** the SQLITE_ENABLE_COLUMN_METADATA compile-time option is used.
1363 #ifdef SQLITE_ENABLE_COLUMN_METADATA
1364 # define columnType(A,B,C,D,E) columnTypeImpl(A,B,C,D,E)
1365 #else /* if !defined(SQLITE_ENABLE_COLUMN_METADATA) */
1366 # define columnType(A,B,C,D,E) columnTypeImpl(A,B)
1367 #endif
1368 static const char *columnTypeImpl(
1369 NameContext *pNC,
1370 #ifndef SQLITE_ENABLE_COLUMN_METADATA
1371 Expr *pExpr
1372 #else
1373 Expr *pExpr,
1374 const char **pzOrigDb,
1375 const char **pzOrigTab,
1376 const char **pzOrigCol
1377 #endif
1379 char const *zType = 0;
1380 int j;
1381 #ifdef SQLITE_ENABLE_COLUMN_METADATA
1382 char const *zOrigDb = 0;
1383 char const *zOrigTab = 0;
1384 char const *zOrigCol = 0;
1385 #endif
1387 assert( pExpr!=0 );
1388 assert( pNC->pSrcList!=0 );
1389 switch( pExpr->op ){
1390 case TK_AGG_COLUMN:
1391 case TK_COLUMN: {
1392 /* The expression is a column. Locate the table the column is being
1393 ** extracted from in NameContext.pSrcList. This table may be real
1394 ** database table or a subquery.
1396 Table *pTab = 0; /* Table structure column is extracted from */
1397 Select *pS = 0; /* Select the column is extracted from */
1398 int iCol = pExpr->iColumn; /* Index of column in pTab */
1399 testcase( pExpr->op==TK_AGG_COLUMN );
1400 testcase( pExpr->op==TK_COLUMN );
1401 while( pNC && !pTab ){
1402 SrcList *pTabList = pNC->pSrcList;
1403 for(j=0;j<pTabList->nSrc && pTabList->a[j].iCursor!=pExpr->iTable;j++);
1404 if( j<pTabList->nSrc ){
1405 pTab = pTabList->a[j].pTab;
1406 pS = pTabList->a[j].pSelect;
1407 }else{
1408 pNC = pNC->pNext;
1412 if( pTab==0 ){
1413 /* At one time, code such as "SELECT new.x" within a trigger would
1414 ** cause this condition to run. Since then, we have restructured how
1415 ** trigger code is generated and so this condition is no longer
1416 ** possible. However, it can still be true for statements like
1417 ** the following:
1419 ** CREATE TABLE t1(col INTEGER);
1420 ** SELECT (SELECT t1.col) FROM FROM t1;
1422 ** when columnType() is called on the expression "t1.col" in the
1423 ** sub-select. In this case, set the column type to NULL, even
1424 ** though it should really be "INTEGER".
1426 ** This is not a problem, as the column type of "t1.col" is never
1427 ** used. When columnType() is called on the expression
1428 ** "(SELECT t1.col)", the correct type is returned (see the TK_SELECT
1429 ** branch below. */
1430 break;
1433 assert( pTab && pExpr->pTab==pTab );
1434 if( pS ){
1435 /* The "table" is actually a sub-select or a view in the FROM clause
1436 ** of the SELECT statement. Return the declaration type and origin
1437 ** data for the result-set column of the sub-select.
1439 if( iCol>=0 && iCol<pS->pEList->nExpr ){
1440 /* If iCol is less than zero, then the expression requests the
1441 ** rowid of the sub-select or view. This expression is legal (see
1442 ** test case misc2.2.2) - it always evaluates to NULL.
1444 NameContext sNC;
1445 Expr *p = pS->pEList->a[iCol].pExpr;
1446 sNC.pSrcList = pS->pSrc;
1447 sNC.pNext = pNC;
1448 sNC.pParse = pNC->pParse;
1449 zType = columnType(&sNC, p,&zOrigDb,&zOrigTab,&zOrigCol);
1451 }else{
1452 /* A real table or a CTE table */
1453 assert( !pS );
1454 #ifdef SQLITE_ENABLE_COLUMN_METADATA
1455 if( iCol<0 ) iCol = pTab->iPKey;
1456 assert( iCol==XN_ROWID || (iCol>=0 && iCol<pTab->nCol) );
1457 if( iCol<0 ){
1458 zType = "INTEGER";
1459 zOrigCol = "rowid";
1460 }else{
1461 zOrigCol = pTab->aCol[iCol].zName;
1462 zType = sqlite3ColumnType(&pTab->aCol[iCol],0);
1464 zOrigTab = pTab->zName;
1465 if( pNC->pParse && pTab->pSchema ){
1466 int iDb = sqlite3SchemaToIndex(pNC->pParse->db, pTab->pSchema);
1467 zOrigDb = pNC->pParse->db->aDb[iDb].zDbSName;
1469 #else
1470 assert( iCol==XN_ROWID || (iCol>=0 && iCol<pTab->nCol) );
1471 if( iCol<0 ){
1472 zType = "INTEGER";
1473 }else{
1474 zType = sqlite3ColumnType(&pTab->aCol[iCol],0);
1476 #endif
1478 break;
1480 #ifndef SQLITE_OMIT_SUBQUERY
1481 case TK_SELECT: {
1482 /* The expression is a sub-select. Return the declaration type and
1483 ** origin info for the single column in the result set of the SELECT
1484 ** statement.
1486 NameContext sNC;
1487 Select *pS = pExpr->x.pSelect;
1488 Expr *p = pS->pEList->a[0].pExpr;
1489 assert( ExprHasProperty(pExpr, EP_xIsSelect) );
1490 sNC.pSrcList = pS->pSrc;
1491 sNC.pNext = pNC;
1492 sNC.pParse = pNC->pParse;
1493 zType = columnType(&sNC, p, &zOrigDb, &zOrigTab, &zOrigCol);
1494 break;
1496 #endif
1499 #ifdef SQLITE_ENABLE_COLUMN_METADATA
1500 if( pzOrigDb ){
1501 assert( pzOrigTab && pzOrigCol );
1502 *pzOrigDb = zOrigDb;
1503 *pzOrigTab = zOrigTab;
1504 *pzOrigCol = zOrigCol;
1506 #endif
1507 return zType;
1511 ** Generate code that will tell the VDBE the declaration types of columns
1512 ** in the result set.
1514 static void generateColumnTypes(
1515 Parse *pParse, /* Parser context */
1516 SrcList *pTabList, /* List of tables */
1517 ExprList *pEList /* Expressions defining the result set */
1519 #ifndef SQLITE_OMIT_DECLTYPE
1520 Vdbe *v = pParse->pVdbe;
1521 int i;
1522 NameContext sNC;
1523 sNC.pSrcList = pTabList;
1524 sNC.pParse = pParse;
1525 sNC.pNext = 0;
1526 for(i=0; i<pEList->nExpr; i++){
1527 Expr *p = pEList->a[i].pExpr;
1528 const char *zType;
1529 #ifdef SQLITE_ENABLE_COLUMN_METADATA
1530 const char *zOrigDb = 0;
1531 const char *zOrigTab = 0;
1532 const char *zOrigCol = 0;
1533 zType = columnType(&sNC, p, &zOrigDb, &zOrigTab, &zOrigCol);
1535 /* The vdbe must make its own copy of the column-type and other
1536 ** column specific strings, in case the schema is reset before this
1537 ** virtual machine is deleted.
1539 sqlite3VdbeSetColName(v, i, COLNAME_DATABASE, zOrigDb, SQLITE_TRANSIENT);
1540 sqlite3VdbeSetColName(v, i, COLNAME_TABLE, zOrigTab, SQLITE_TRANSIENT);
1541 sqlite3VdbeSetColName(v, i, COLNAME_COLUMN, zOrigCol, SQLITE_TRANSIENT);
1542 #else
1543 zType = columnType(&sNC, p, 0, 0, 0);
1544 #endif
1545 sqlite3VdbeSetColName(v, i, COLNAME_DECLTYPE, zType, SQLITE_TRANSIENT);
1547 #endif /* !defined(SQLITE_OMIT_DECLTYPE) */
1552 ** Compute the column names for a SELECT statement.
1554 ** The only guarantee that SQLite makes about column names is that if the
1555 ** column has an AS clause assigning it a name, that will be the name used.
1556 ** That is the only documented guarantee. However, countless applications
1557 ** developed over the years have made baseless assumptions about column names
1558 ** and will break if those assumptions changes. Hence, use extreme caution
1559 ** when modifying this routine to avoid breaking legacy.
1561 ** See Also: sqlite3ColumnsFromExprList()
1563 ** The PRAGMA short_column_names and PRAGMA full_column_names settings are
1564 ** deprecated. The default setting is short=ON, full=OFF. 99.9% of all
1565 ** applications should operate this way. Nevertheless, we need to support the
1566 ** other modes for legacy:
1568 ** short=OFF, full=OFF: Column name is the text of the expression has it
1569 ** originally appears in the SELECT statement. In
1570 ** other words, the zSpan of the result expression.
1572 ** short=ON, full=OFF: (This is the default setting). If the result
1573 ** refers directly to a table column, then the
1574 ** result column name is just the table column
1575 ** name: COLUMN. Otherwise use zSpan.
1577 ** full=ON, short=ANY: If the result refers directly to a table column,
1578 ** then the result column name with the table name
1579 ** prefix, ex: TABLE.COLUMN. Otherwise use zSpan.
1581 static void generateColumnNames(
1582 Parse *pParse, /* Parser context */
1583 Select *pSelect /* Generate column names for this SELECT statement */
1585 Vdbe *v = pParse->pVdbe;
1586 int i;
1587 Table *pTab;
1588 SrcList *pTabList;
1589 ExprList *pEList;
1590 sqlite3 *db = pParse->db;
1591 int fullName; /* TABLE.COLUMN if no AS clause and is a direct table ref */
1592 int srcName; /* COLUMN or TABLE.COLUMN if no AS clause and is direct */
1594 #ifndef SQLITE_OMIT_EXPLAIN
1595 /* If this is an EXPLAIN, skip this step */
1596 if( pParse->explain ){
1597 return;
1599 #endif
1601 if( pParse->colNamesSet || db->mallocFailed ) return;
1602 /* Column names are determined by the left-most term of a compound select */
1603 while( pSelect->pPrior ) pSelect = pSelect->pPrior;
1604 pTabList = pSelect->pSrc;
1605 pEList = pSelect->pEList;
1606 assert( v!=0 );
1607 assert( pTabList!=0 );
1608 pParse->colNamesSet = 1;
1609 fullName = (db->flags & SQLITE_FullColNames)!=0;
1610 srcName = (db->flags & SQLITE_ShortColNames)!=0 || fullName;
1611 sqlite3VdbeSetNumCols(v, pEList->nExpr);
1612 for(i=0; i<pEList->nExpr; i++){
1613 Expr *p = pEList->a[i].pExpr;
1615 assert( p!=0 );
1616 assert( p->op!=TK_AGG_COLUMN ); /* Agg processing has not run yet */
1617 assert( p->op!=TK_COLUMN || p->pTab!=0 ); /* Covering idx not yet coded */
1618 if( pEList->a[i].zName ){
1619 /* An AS clause always takes first priority */
1620 char *zName = pEList->a[i].zName;
1621 sqlite3VdbeSetColName(v, i, COLNAME_NAME, zName, SQLITE_TRANSIENT);
1622 }else if( srcName && p->op==TK_COLUMN ){
1623 char *zCol;
1624 int iCol = p->iColumn;
1625 pTab = p->pTab;
1626 assert( pTab!=0 );
1627 if( iCol<0 ) iCol = pTab->iPKey;
1628 assert( iCol==-1 || (iCol>=0 && iCol<pTab->nCol) );
1629 if( iCol<0 ){
1630 zCol = "rowid";
1631 }else{
1632 zCol = pTab->aCol[iCol].zName;
1634 if( fullName ){
1635 char *zName = 0;
1636 zName = sqlite3MPrintf(db, "%s.%s", pTab->zName, zCol);
1637 sqlite3VdbeSetColName(v, i, COLNAME_NAME, zName, SQLITE_DYNAMIC);
1638 }else{
1639 sqlite3VdbeSetColName(v, i, COLNAME_NAME, zCol, SQLITE_TRANSIENT);
1641 }else{
1642 const char *z = pEList->a[i].zSpan;
1643 z = z==0 ? sqlite3MPrintf(db, "column%d", i+1) : sqlite3DbStrDup(db, z);
1644 sqlite3VdbeSetColName(v, i, COLNAME_NAME, z, SQLITE_DYNAMIC);
1647 generateColumnTypes(pParse, pTabList, pEList);
1651 ** Given an expression list (which is really the list of expressions
1652 ** that form the result set of a SELECT statement) compute appropriate
1653 ** column names for a table that would hold the expression list.
1655 ** All column names will be unique.
1657 ** Only the column names are computed. Column.zType, Column.zColl,
1658 ** and other fields of Column are zeroed.
1660 ** Return SQLITE_OK on success. If a memory allocation error occurs,
1661 ** store NULL in *paCol and 0 in *pnCol and return SQLITE_NOMEM.
1663 ** The only guarantee that SQLite makes about column names is that if the
1664 ** column has an AS clause assigning it a name, that will be the name used.
1665 ** That is the only documented guarantee. However, countless applications
1666 ** developed over the years have made baseless assumptions about column names
1667 ** and will break if those assumptions changes. Hence, use extreme caution
1668 ** when modifying this routine to avoid breaking legacy.
1670 ** See Also: generateColumnNames()
1672 int sqlite3ColumnsFromExprList(
1673 Parse *pParse, /* Parsing context */
1674 ExprList *pEList, /* Expr list from which to derive column names */
1675 i16 *pnCol, /* Write the number of columns here */
1676 Column **paCol /* Write the new column list here */
1678 sqlite3 *db = pParse->db; /* Database connection */
1679 int i, j; /* Loop counters */
1680 u32 cnt; /* Index added to make the name unique */
1681 Column *aCol, *pCol; /* For looping over result columns */
1682 int nCol; /* Number of columns in the result set */
1683 char *zName; /* Column name */
1684 int nName; /* Size of name in zName[] */
1685 Hash ht; /* Hash table of column names */
1687 sqlite3HashInit(&ht);
1688 if( pEList ){
1689 nCol = pEList->nExpr;
1690 aCol = sqlite3DbMallocZero(db, sizeof(aCol[0])*nCol);
1691 testcase( aCol==0 );
1692 if( nCol>32767 ) nCol = 32767;
1693 }else{
1694 nCol = 0;
1695 aCol = 0;
1697 assert( nCol==(i16)nCol );
1698 *pnCol = nCol;
1699 *paCol = aCol;
1701 for(i=0, pCol=aCol; i<nCol && !db->mallocFailed; i++, pCol++){
1702 /* Get an appropriate name for the column
1704 if( (zName = pEList->a[i].zName)!=0 ){
1705 /* If the column contains an "AS <name>" phrase, use <name> as the name */
1706 }else{
1707 Expr *pColExpr = sqlite3ExprSkipCollate(pEList->a[i].pExpr);
1708 while( pColExpr->op==TK_DOT ){
1709 pColExpr = pColExpr->pRight;
1710 assert( pColExpr!=0 );
1712 if( (pColExpr->op==TK_COLUMN || pColExpr->op==TK_AGG_COLUMN)
1713 && pColExpr->pTab!=0
1715 /* For columns use the column name name */
1716 int iCol = pColExpr->iColumn;
1717 Table *pTab = pColExpr->pTab;
1718 if( iCol<0 ) iCol = pTab->iPKey;
1719 zName = iCol>=0 ? pTab->aCol[iCol].zName : "rowid";
1720 }else if( pColExpr->op==TK_ID ){
1721 assert( !ExprHasProperty(pColExpr, EP_IntValue) );
1722 zName = pColExpr->u.zToken;
1723 }else{
1724 /* Use the original text of the column expression as its name */
1725 zName = pEList->a[i].zSpan;
1728 if( zName ){
1729 zName = sqlite3DbStrDup(db, zName);
1730 }else{
1731 zName = sqlite3MPrintf(db,"column%d",i+1);
1734 /* Make sure the column name is unique. If the name is not unique,
1735 ** append an integer to the name so that it becomes unique.
1737 cnt = 0;
1738 while( zName && sqlite3HashFind(&ht, zName)!=0 ){
1739 nName = sqlite3Strlen30(zName);
1740 if( nName>0 ){
1741 for(j=nName-1; j>0 && sqlite3Isdigit(zName[j]); j--){}
1742 if( zName[j]==':' ) nName = j;
1744 zName = sqlite3MPrintf(db, "%.*z:%u", nName, zName, ++cnt);
1745 if( cnt>3 ) sqlite3_randomness(sizeof(cnt), &cnt);
1747 pCol->zName = zName;
1748 sqlite3ColumnPropertiesFromName(0, pCol);
1749 if( zName && sqlite3HashInsert(&ht, zName, pCol)==pCol ){
1750 sqlite3OomFault(db);
1753 sqlite3HashClear(&ht);
1754 if( db->mallocFailed ){
1755 for(j=0; j<i; j++){
1756 sqlite3DbFree(db, aCol[j].zName);
1758 sqlite3DbFree(db, aCol);
1759 *paCol = 0;
1760 *pnCol = 0;
1761 return SQLITE_NOMEM_BKPT;
1763 return SQLITE_OK;
1767 ** Add type and collation information to a column list based on
1768 ** a SELECT statement.
1770 ** The column list presumably came from selectColumnNamesFromExprList().
1771 ** The column list has only names, not types or collations. This
1772 ** routine goes through and adds the types and collations.
1774 ** This routine requires that all identifiers in the SELECT
1775 ** statement be resolved.
1777 void sqlite3SelectAddColumnTypeAndCollation(
1778 Parse *pParse, /* Parsing contexts */
1779 Table *pTab, /* Add column type information to this table */
1780 Select *pSelect /* SELECT used to determine types and collations */
1782 sqlite3 *db = pParse->db;
1783 NameContext sNC;
1784 Column *pCol;
1785 CollSeq *pColl;
1786 int i;
1787 Expr *p;
1788 struct ExprList_item *a;
1790 assert( pSelect!=0 );
1791 assert( (pSelect->selFlags & SF_Resolved)!=0 );
1792 assert( pTab->nCol==pSelect->pEList->nExpr || db->mallocFailed );
1793 if( db->mallocFailed ) return;
1794 memset(&sNC, 0, sizeof(sNC));
1795 sNC.pSrcList = pSelect->pSrc;
1796 a = pSelect->pEList->a;
1797 for(i=0, pCol=pTab->aCol; i<pTab->nCol; i++, pCol++){
1798 const char *zType;
1799 int n, m;
1800 p = a[i].pExpr;
1801 zType = columnType(&sNC, p, 0, 0, 0);
1802 /* pCol->szEst = ... // Column size est for SELECT tables never used */
1803 pCol->affinity = sqlite3ExprAffinity(p);
1804 if( zType ){
1805 m = sqlite3Strlen30(zType);
1806 n = sqlite3Strlen30(pCol->zName);
1807 pCol->zName = sqlite3DbReallocOrFree(db, pCol->zName, n+m+2);
1808 if( pCol->zName ){
1809 memcpy(&pCol->zName[n+1], zType, m+1);
1810 pCol->colFlags |= COLFLAG_HASTYPE;
1813 if( pCol->affinity==0 ) pCol->affinity = SQLITE_AFF_BLOB;
1814 pColl = sqlite3ExprCollSeq(pParse, p);
1815 if( pColl && pCol->zColl==0 ){
1816 pCol->zColl = sqlite3DbStrDup(db, pColl->zName);
1819 pTab->szTabRow = 1; /* Any non-zero value works */
1823 ** Given a SELECT statement, generate a Table structure that describes
1824 ** the result set of that SELECT.
1826 Table *sqlite3ResultSetOfSelect(Parse *pParse, Select *pSelect){
1827 Table *pTab;
1828 sqlite3 *db = pParse->db;
1829 int savedFlags;
1831 savedFlags = db->flags;
1832 db->flags &= ~SQLITE_FullColNames;
1833 db->flags |= SQLITE_ShortColNames;
1834 sqlite3SelectPrep(pParse, pSelect, 0);
1835 if( pParse->nErr ) return 0;
1836 while( pSelect->pPrior ) pSelect = pSelect->pPrior;
1837 db->flags = savedFlags;
1838 pTab = sqlite3DbMallocZero(db, sizeof(Table) );
1839 if( pTab==0 ){
1840 return 0;
1842 /* The sqlite3ResultSetOfSelect() is only used n contexts where lookaside
1843 ** is disabled */
1844 assert( db->lookaside.bDisable );
1845 pTab->nTabRef = 1;
1846 pTab->zName = 0;
1847 pTab->nRowLogEst = 200; assert( 200==sqlite3LogEst(1048576) );
1848 sqlite3ColumnsFromExprList(pParse, pSelect->pEList, &pTab->nCol, &pTab->aCol);
1849 sqlite3SelectAddColumnTypeAndCollation(pParse, pTab, pSelect);
1850 pTab->iPKey = -1;
1851 if( db->mallocFailed ){
1852 sqlite3DeleteTable(db, pTab);
1853 return 0;
1855 return pTab;
1859 ** Get a VDBE for the given parser context. Create a new one if necessary.
1860 ** If an error occurs, return NULL and leave a message in pParse.
1862 Vdbe *sqlite3GetVdbe(Parse *pParse){
1863 if( pParse->pVdbe ){
1864 return pParse->pVdbe;
1866 if( pParse->pToplevel==0
1867 && OptimizationEnabled(pParse->db,SQLITE_FactorOutConst)
1869 pParse->okConstFactor = 1;
1871 return sqlite3VdbeCreate(pParse);
1876 ** Compute the iLimit and iOffset fields of the SELECT based on the
1877 ** pLimit and pOffset expressions. pLimit and pOffset hold the expressions
1878 ** that appear in the original SQL statement after the LIMIT and OFFSET
1879 ** keywords. Or NULL if those keywords are omitted. iLimit and iOffset
1880 ** are the integer memory register numbers for counters used to compute
1881 ** the limit and offset. If there is no limit and/or offset, then
1882 ** iLimit and iOffset are negative.
1884 ** This routine changes the values of iLimit and iOffset only if
1885 ** a limit or offset is defined by pLimit and pOffset. iLimit and
1886 ** iOffset should have been preset to appropriate default values (zero)
1887 ** prior to calling this routine.
1889 ** The iOffset register (if it exists) is initialized to the value
1890 ** of the OFFSET. The iLimit register is initialized to LIMIT. Register
1891 ** iOffset+1 is initialized to LIMIT+OFFSET.
1893 ** Only if pLimit!=0 or pOffset!=0 do the limit registers get
1894 ** redefined. The UNION ALL operator uses this property to force
1895 ** the reuse of the same limit and offset registers across multiple
1896 ** SELECT statements.
1898 static void computeLimitRegisters(Parse *pParse, Select *p, int iBreak){
1899 Vdbe *v = 0;
1900 int iLimit = 0;
1901 int iOffset;
1902 int n;
1903 if( p->iLimit ) return;
1906 ** "LIMIT -1" always shows all rows. There is some
1907 ** controversy about what the correct behavior should be.
1908 ** The current implementation interprets "LIMIT 0" to mean
1909 ** no rows.
1911 sqlite3ExprCacheClear(pParse);
1912 assert( p->pOffset==0 || p->pLimit!=0 );
1913 if( p->pLimit ){
1914 p->iLimit = iLimit = ++pParse->nMem;
1915 v = sqlite3GetVdbe(pParse);
1916 assert( v!=0 );
1917 if( sqlite3ExprIsInteger(p->pLimit, &n) ){
1918 sqlite3VdbeAddOp2(v, OP_Integer, n, iLimit);
1919 VdbeComment((v, "LIMIT counter"));
1920 if( n==0 ){
1921 sqlite3VdbeGoto(v, iBreak);
1922 }else if( n>=0 && p->nSelectRow>sqlite3LogEst((u64)n) ){
1923 p->nSelectRow = sqlite3LogEst((u64)n);
1924 p->selFlags |= SF_FixedLimit;
1926 }else{
1927 sqlite3ExprCode(pParse, p->pLimit, iLimit);
1928 sqlite3VdbeAddOp1(v, OP_MustBeInt, iLimit); VdbeCoverage(v);
1929 VdbeComment((v, "LIMIT counter"));
1930 sqlite3VdbeAddOp2(v, OP_IfNot, iLimit, iBreak); VdbeCoverage(v);
1932 if( p->pOffset ){
1933 p->iOffset = iOffset = ++pParse->nMem;
1934 pParse->nMem++; /* Allocate an extra register for limit+offset */
1935 sqlite3ExprCode(pParse, p->pOffset, iOffset);
1936 sqlite3VdbeAddOp1(v, OP_MustBeInt, iOffset); VdbeCoverage(v);
1937 VdbeComment((v, "OFFSET counter"));
1938 sqlite3VdbeAddOp3(v, OP_OffsetLimit, iLimit, iOffset+1, iOffset);
1939 VdbeComment((v, "LIMIT+OFFSET"));
1944 #ifndef SQLITE_OMIT_COMPOUND_SELECT
1946 ** Return the appropriate collating sequence for the iCol-th column of
1947 ** the result set for the compound-select statement "p". Return NULL if
1948 ** the column has no default collating sequence.
1950 ** The collating sequence for the compound select is taken from the
1951 ** left-most term of the select that has a collating sequence.
1953 static CollSeq *multiSelectCollSeq(Parse *pParse, Select *p, int iCol){
1954 CollSeq *pRet;
1955 if( p->pPrior ){
1956 pRet = multiSelectCollSeq(pParse, p->pPrior, iCol);
1957 }else{
1958 pRet = 0;
1960 assert( iCol>=0 );
1961 /* iCol must be less than p->pEList->nExpr. Otherwise an error would
1962 ** have been thrown during name resolution and we would not have gotten
1963 ** this far */
1964 if( pRet==0 && ALWAYS(iCol<p->pEList->nExpr) ){
1965 pRet = sqlite3ExprCollSeq(pParse, p->pEList->a[iCol].pExpr);
1967 return pRet;
1971 ** The select statement passed as the second parameter is a compound SELECT
1972 ** with an ORDER BY clause. This function allocates and returns a KeyInfo
1973 ** structure suitable for implementing the ORDER BY.
1975 ** Space to hold the KeyInfo structure is obtained from malloc. The calling
1976 ** function is responsible for ensuring that this structure is eventually
1977 ** freed.
1979 static KeyInfo *multiSelectOrderByKeyInfo(Parse *pParse, Select *p, int nExtra){
1980 ExprList *pOrderBy = p->pOrderBy;
1981 int nOrderBy = p->pOrderBy->nExpr;
1982 sqlite3 *db = pParse->db;
1983 KeyInfo *pRet = sqlite3KeyInfoAlloc(db, nOrderBy+nExtra, 1);
1984 if( pRet ){
1985 int i;
1986 for(i=0; i<nOrderBy; i++){
1987 struct ExprList_item *pItem = &pOrderBy->a[i];
1988 Expr *pTerm = pItem->pExpr;
1989 CollSeq *pColl;
1991 if( pTerm->flags & EP_Collate ){
1992 pColl = sqlite3ExprCollSeq(pParse, pTerm);
1993 }else{
1994 pColl = multiSelectCollSeq(pParse, p, pItem->u.x.iOrderByCol-1);
1995 if( pColl==0 ) pColl = db->pDfltColl;
1996 pOrderBy->a[i].pExpr =
1997 sqlite3ExprAddCollateString(pParse, pTerm, pColl->zName);
1999 assert( sqlite3KeyInfoIsWriteable(pRet) );
2000 pRet->aColl[i] = pColl;
2001 pRet->aSortOrder[i] = pOrderBy->a[i].sortOrder;
2005 return pRet;
2008 #ifndef SQLITE_OMIT_CTE
2010 ** This routine generates VDBE code to compute the content of a WITH RECURSIVE
2011 ** query of the form:
2013 ** <recursive-table> AS (<setup-query> UNION [ALL] <recursive-query>)
2014 ** \___________/ \_______________/
2015 ** p->pPrior p
2018 ** There is exactly one reference to the recursive-table in the FROM clause
2019 ** of recursive-query, marked with the SrcList->a[].fg.isRecursive flag.
2021 ** The setup-query runs once to generate an initial set of rows that go
2022 ** into a Queue table. Rows are extracted from the Queue table one by
2023 ** one. Each row extracted from Queue is output to pDest. Then the single
2024 ** extracted row (now in the iCurrent table) becomes the content of the
2025 ** recursive-table for a recursive-query run. The output of the recursive-query
2026 ** is added back into the Queue table. Then another row is extracted from Queue
2027 ** and the iteration continues until the Queue table is empty.
2029 ** If the compound query operator is UNION then no duplicate rows are ever
2030 ** inserted into the Queue table. The iDistinct table keeps a copy of all rows
2031 ** that have ever been inserted into Queue and causes duplicates to be
2032 ** discarded. If the operator is UNION ALL, then duplicates are allowed.
2034 ** If the query has an ORDER BY, then entries in the Queue table are kept in
2035 ** ORDER BY order and the first entry is extracted for each cycle. Without
2036 ** an ORDER BY, the Queue table is just a FIFO.
2038 ** If a LIMIT clause is provided, then the iteration stops after LIMIT rows
2039 ** have been output to pDest. A LIMIT of zero means to output no rows and a
2040 ** negative LIMIT means to output all rows. If there is also an OFFSET clause
2041 ** with a positive value, then the first OFFSET outputs are discarded rather
2042 ** than being sent to pDest. The LIMIT count does not begin until after OFFSET
2043 ** rows have been skipped.
2045 static void generateWithRecursiveQuery(
2046 Parse *pParse, /* Parsing context */
2047 Select *p, /* The recursive SELECT to be coded */
2048 SelectDest *pDest /* What to do with query results */
2050 SrcList *pSrc = p->pSrc; /* The FROM clause of the recursive query */
2051 int nCol = p->pEList->nExpr; /* Number of columns in the recursive table */
2052 Vdbe *v = pParse->pVdbe; /* The prepared statement under construction */
2053 Select *pSetup = p->pPrior; /* The setup query */
2054 int addrTop; /* Top of the loop */
2055 int addrCont, addrBreak; /* CONTINUE and BREAK addresses */
2056 int iCurrent = 0; /* The Current table */
2057 int regCurrent; /* Register holding Current table */
2058 int iQueue; /* The Queue table */
2059 int iDistinct = 0; /* To ensure unique results if UNION */
2060 int eDest = SRT_Fifo; /* How to write to Queue */
2061 SelectDest destQueue; /* SelectDest targetting the Queue table */
2062 int i; /* Loop counter */
2063 int rc; /* Result code */
2064 ExprList *pOrderBy; /* The ORDER BY clause */
2065 Expr *pLimit, *pOffset; /* Saved LIMIT and OFFSET */
2066 int regLimit, regOffset; /* Registers used by LIMIT and OFFSET */
2068 /* Obtain authorization to do a recursive query */
2069 if( sqlite3AuthCheck(pParse, SQLITE_RECURSIVE, 0, 0, 0) ) return;
2071 /* Process the LIMIT and OFFSET clauses, if they exist */
2072 addrBreak = sqlite3VdbeMakeLabel(v);
2073 p->nSelectRow = 320; /* 4 billion rows */
2074 computeLimitRegisters(pParse, p, addrBreak);
2075 pLimit = p->pLimit;
2076 pOffset = p->pOffset;
2077 regLimit = p->iLimit;
2078 regOffset = p->iOffset;
2079 p->pLimit = p->pOffset = 0;
2080 p->iLimit = p->iOffset = 0;
2081 pOrderBy = p->pOrderBy;
2083 /* Locate the cursor number of the Current table */
2084 for(i=0; ALWAYS(i<pSrc->nSrc); i++){
2085 if( pSrc->a[i].fg.isRecursive ){
2086 iCurrent = pSrc->a[i].iCursor;
2087 break;
2091 /* Allocate cursors numbers for Queue and Distinct. The cursor number for
2092 ** the Distinct table must be exactly one greater than Queue in order
2093 ** for the SRT_DistFifo and SRT_DistQueue destinations to work. */
2094 iQueue = pParse->nTab++;
2095 if( p->op==TK_UNION ){
2096 eDest = pOrderBy ? SRT_DistQueue : SRT_DistFifo;
2097 iDistinct = pParse->nTab++;
2098 }else{
2099 eDest = pOrderBy ? SRT_Queue : SRT_Fifo;
2101 sqlite3SelectDestInit(&destQueue, eDest, iQueue);
2103 /* Allocate cursors for Current, Queue, and Distinct. */
2104 regCurrent = ++pParse->nMem;
2105 sqlite3VdbeAddOp3(v, OP_OpenPseudo, iCurrent, regCurrent, nCol);
2106 if( pOrderBy ){
2107 KeyInfo *pKeyInfo = multiSelectOrderByKeyInfo(pParse, p, 1);
2108 sqlite3VdbeAddOp4(v, OP_OpenEphemeral, iQueue, pOrderBy->nExpr+2, 0,
2109 (char*)pKeyInfo, P4_KEYINFO);
2110 destQueue.pOrderBy = pOrderBy;
2111 }else{
2112 sqlite3VdbeAddOp2(v, OP_OpenEphemeral, iQueue, nCol);
2114 VdbeComment((v, "Queue table"));
2115 if( iDistinct ){
2116 p->addrOpenEphm[0] = sqlite3VdbeAddOp2(v, OP_OpenEphemeral, iDistinct, 0);
2117 p->selFlags |= SF_UsesEphemeral;
2120 /* Detach the ORDER BY clause from the compound SELECT */
2121 p->pOrderBy = 0;
2123 /* Store the results of the setup-query in Queue. */
2124 pSetup->pNext = 0;
2125 rc = sqlite3Select(pParse, pSetup, &destQueue);
2126 pSetup->pNext = p;
2127 if( rc ) goto end_of_recursive_query;
2129 /* Find the next row in the Queue and output that row */
2130 addrTop = sqlite3VdbeAddOp2(v, OP_Rewind, iQueue, addrBreak); VdbeCoverage(v);
2132 /* Transfer the next row in Queue over to Current */
2133 sqlite3VdbeAddOp1(v, OP_NullRow, iCurrent); /* To reset column cache */
2134 if( pOrderBy ){
2135 sqlite3VdbeAddOp3(v, OP_Column, iQueue, pOrderBy->nExpr+1, regCurrent);
2136 }else{
2137 sqlite3VdbeAddOp2(v, OP_RowData, iQueue, regCurrent);
2139 sqlite3VdbeAddOp1(v, OP_Delete, iQueue);
2141 /* Output the single row in Current */
2142 addrCont = sqlite3VdbeMakeLabel(v);
2143 codeOffset(v, regOffset, addrCont);
2144 selectInnerLoop(pParse, p, iCurrent,
2145 0, 0, pDest, addrCont, addrBreak);
2146 if( regLimit ){
2147 sqlite3VdbeAddOp2(v, OP_DecrJumpZero, regLimit, addrBreak);
2148 VdbeCoverage(v);
2150 sqlite3VdbeResolveLabel(v, addrCont);
2152 /* Execute the recursive SELECT taking the single row in Current as
2153 ** the value for the recursive-table. Store the results in the Queue.
2155 if( p->selFlags & SF_Aggregate ){
2156 sqlite3ErrorMsg(pParse, "recursive aggregate queries not supported");
2157 }else{
2158 p->pPrior = 0;
2159 sqlite3Select(pParse, p, &destQueue);
2160 assert( p->pPrior==0 );
2161 p->pPrior = pSetup;
2164 /* Keep running the loop until the Queue is empty */
2165 sqlite3VdbeGoto(v, addrTop);
2166 sqlite3VdbeResolveLabel(v, addrBreak);
2168 end_of_recursive_query:
2169 sqlite3ExprListDelete(pParse->db, p->pOrderBy);
2170 p->pOrderBy = pOrderBy;
2171 p->pLimit = pLimit;
2172 p->pOffset = pOffset;
2173 return;
2175 #endif /* SQLITE_OMIT_CTE */
2177 /* Forward references */
2178 static int multiSelectOrderBy(
2179 Parse *pParse, /* Parsing context */
2180 Select *p, /* The right-most of SELECTs to be coded */
2181 SelectDest *pDest /* What to do with query results */
2185 ** Handle the special case of a compound-select that originates from a
2186 ** VALUES clause. By handling this as a special case, we avoid deep
2187 ** recursion, and thus do not need to enforce the SQLITE_LIMIT_COMPOUND_SELECT
2188 ** on a VALUES clause.
2190 ** Because the Select object originates from a VALUES clause:
2191 ** (1) It has no LIMIT or OFFSET
2192 ** (2) All terms are UNION ALL
2193 ** (3) There is no ORDER BY clause
2195 static int multiSelectValues(
2196 Parse *pParse, /* Parsing context */
2197 Select *p, /* The right-most of SELECTs to be coded */
2198 SelectDest *pDest /* What to do with query results */
2200 Select *pPrior;
2201 int nRow = 1;
2202 int rc = 0;
2203 assert( p->selFlags & SF_MultiValue );
2205 assert( p->selFlags & SF_Values );
2206 assert( p->op==TK_ALL || (p->op==TK_SELECT && p->pPrior==0) );
2207 assert( p->pLimit==0 );
2208 assert( p->pOffset==0 );
2209 assert( p->pNext==0 || p->pEList->nExpr==p->pNext->pEList->nExpr );
2210 if( p->pPrior==0 ) break;
2211 assert( p->pPrior->pNext==p );
2212 p = p->pPrior;
2213 nRow++;
2214 }while(1);
2215 while( p ){
2216 pPrior = p->pPrior;
2217 p->pPrior = 0;
2218 rc = sqlite3Select(pParse, p, pDest);
2219 p->pPrior = pPrior;
2220 if( rc ) break;
2221 p->nSelectRow = nRow;
2222 p = p->pNext;
2224 return rc;
2228 ** This routine is called to process a compound query form from
2229 ** two or more separate queries using UNION, UNION ALL, EXCEPT, or
2230 ** INTERSECT
2232 ** "p" points to the right-most of the two queries. the query on the
2233 ** left is p->pPrior. The left query could also be a compound query
2234 ** in which case this routine will be called recursively.
2236 ** The results of the total query are to be written into a destination
2237 ** of type eDest with parameter iParm.
2239 ** Example 1: Consider a three-way compound SQL statement.
2241 ** SELECT a FROM t1 UNION SELECT b FROM t2 UNION SELECT c FROM t3
2243 ** This statement is parsed up as follows:
2245 ** SELECT c FROM t3
2246 ** |
2247 ** `-----> SELECT b FROM t2
2248 ** |
2249 ** `------> SELECT a FROM t1
2251 ** The arrows in the diagram above represent the Select.pPrior pointer.
2252 ** So if this routine is called with p equal to the t3 query, then
2253 ** pPrior will be the t2 query. p->op will be TK_UNION in this case.
2255 ** Notice that because of the way SQLite parses compound SELECTs, the
2256 ** individual selects always group from left to right.
2258 static int multiSelect(
2259 Parse *pParse, /* Parsing context */
2260 Select *p, /* The right-most of SELECTs to be coded */
2261 SelectDest *pDest /* What to do with query results */
2263 int rc = SQLITE_OK; /* Success code from a subroutine */
2264 Select *pPrior; /* Another SELECT immediately to our left */
2265 Vdbe *v; /* Generate code to this VDBE */
2266 SelectDest dest; /* Alternative data destination */
2267 Select *pDelete = 0; /* Chain of simple selects to delete */
2268 sqlite3 *db; /* Database connection */
2269 #ifndef SQLITE_OMIT_EXPLAIN
2270 int iSub1 = 0; /* EQP id of left-hand query */
2271 int iSub2 = 0; /* EQP id of right-hand query */
2272 #endif
2274 /* Make sure there is no ORDER BY or LIMIT clause on prior SELECTs. Only
2275 ** the last (right-most) SELECT in the series may have an ORDER BY or LIMIT.
2277 assert( p && p->pPrior ); /* Calling function guarantees this much */
2278 assert( (p->selFlags & SF_Recursive)==0 || p->op==TK_ALL || p->op==TK_UNION );
2279 db = pParse->db;
2280 pPrior = p->pPrior;
2281 dest = *pDest;
2282 if( pPrior->pOrderBy || pPrior->pLimit ){
2283 sqlite3ErrorMsg(pParse,"%s clause should come after %s not before",
2284 pPrior->pOrderBy!=0 ? "ORDER BY" : "LIMIT", selectOpName(p->op));
2285 rc = 1;
2286 goto multi_select_end;
2289 v = sqlite3GetVdbe(pParse);
2290 assert( v!=0 ); /* The VDBE already created by calling function */
2292 /* Create the destination temporary table if necessary
2294 if( dest.eDest==SRT_EphemTab ){
2295 assert( p->pEList );
2296 sqlite3VdbeAddOp2(v, OP_OpenEphemeral, dest.iSDParm, p->pEList->nExpr);
2297 dest.eDest = SRT_Table;
2300 /* Special handling for a compound-select that originates as a VALUES clause.
2302 if( p->selFlags & SF_MultiValue ){
2303 rc = multiSelectValues(pParse, p, &dest);
2304 goto multi_select_end;
2307 /* Make sure all SELECTs in the statement have the same number of elements
2308 ** in their result sets.
2310 assert( p->pEList && pPrior->pEList );
2311 assert( p->pEList->nExpr==pPrior->pEList->nExpr );
2313 #ifndef SQLITE_OMIT_CTE
2314 if( p->selFlags & SF_Recursive ){
2315 generateWithRecursiveQuery(pParse, p, &dest);
2316 }else
2317 #endif
2319 /* Compound SELECTs that have an ORDER BY clause are handled separately.
2321 if( p->pOrderBy ){
2322 return multiSelectOrderBy(pParse, p, pDest);
2323 }else
2325 /* Generate code for the left and right SELECT statements.
2327 switch( p->op ){
2328 case TK_ALL: {
2329 int addr = 0;
2330 int nLimit;
2331 assert( !pPrior->pLimit );
2332 pPrior->iLimit = p->iLimit;
2333 pPrior->iOffset = p->iOffset;
2334 pPrior->pLimit = p->pLimit;
2335 pPrior->pOffset = p->pOffset;
2336 explainSetInteger(iSub1, pParse->iNextSelectId);
2337 rc = sqlite3Select(pParse, pPrior, &dest);
2338 p->pLimit = 0;
2339 p->pOffset = 0;
2340 if( rc ){
2341 goto multi_select_end;
2343 p->pPrior = 0;
2344 p->iLimit = pPrior->iLimit;
2345 p->iOffset = pPrior->iOffset;
2346 if( p->iLimit ){
2347 addr = sqlite3VdbeAddOp1(v, OP_IfNot, p->iLimit); VdbeCoverage(v);
2348 VdbeComment((v, "Jump ahead if LIMIT reached"));
2349 if( p->iOffset ){
2350 sqlite3VdbeAddOp3(v, OP_OffsetLimit,
2351 p->iLimit, p->iOffset+1, p->iOffset);
2354 explainSetInteger(iSub2, pParse->iNextSelectId);
2355 rc = sqlite3Select(pParse, p, &dest);
2356 testcase( rc!=SQLITE_OK );
2357 pDelete = p->pPrior;
2358 p->pPrior = pPrior;
2359 p->nSelectRow = sqlite3LogEstAdd(p->nSelectRow, pPrior->nSelectRow);
2360 if( pPrior->pLimit
2361 && sqlite3ExprIsInteger(pPrior->pLimit, &nLimit)
2362 && nLimit>0 && p->nSelectRow > sqlite3LogEst((u64)nLimit)
2364 p->nSelectRow = sqlite3LogEst((u64)nLimit);
2366 if( addr ){
2367 sqlite3VdbeJumpHere(v, addr);
2369 break;
2371 case TK_EXCEPT:
2372 case TK_UNION: {
2373 int unionTab; /* Cursor number of the temporary table holding result */
2374 u8 op = 0; /* One of the SRT_ operations to apply to self */
2375 int priorOp; /* The SRT_ operation to apply to prior selects */
2376 Expr *pLimit, *pOffset; /* Saved values of p->nLimit and p->nOffset */
2377 int addr;
2378 SelectDest uniondest;
2380 testcase( p->op==TK_EXCEPT );
2381 testcase( p->op==TK_UNION );
2382 priorOp = SRT_Union;
2383 if( dest.eDest==priorOp ){
2384 /* We can reuse a temporary table generated by a SELECT to our
2385 ** right.
2387 assert( p->pLimit==0 ); /* Not allowed on leftward elements */
2388 assert( p->pOffset==0 ); /* Not allowed on leftward elements */
2389 unionTab = dest.iSDParm;
2390 }else{
2391 /* We will need to create our own temporary table to hold the
2392 ** intermediate results.
2394 unionTab = pParse->nTab++;
2395 assert( p->pOrderBy==0 );
2396 addr = sqlite3VdbeAddOp2(v, OP_OpenEphemeral, unionTab, 0);
2397 assert( p->addrOpenEphm[0] == -1 );
2398 p->addrOpenEphm[0] = addr;
2399 findRightmost(p)->selFlags |= SF_UsesEphemeral;
2400 assert( p->pEList );
2403 /* Code the SELECT statements to our left
2405 assert( !pPrior->pOrderBy );
2406 sqlite3SelectDestInit(&uniondest, priorOp, unionTab);
2407 explainSetInteger(iSub1, pParse->iNextSelectId);
2408 rc = sqlite3Select(pParse, pPrior, &uniondest);
2409 if( rc ){
2410 goto multi_select_end;
2413 /* Code the current SELECT statement
2415 if( p->op==TK_EXCEPT ){
2416 op = SRT_Except;
2417 }else{
2418 assert( p->op==TK_UNION );
2419 op = SRT_Union;
2421 p->pPrior = 0;
2422 pLimit = p->pLimit;
2423 p->pLimit = 0;
2424 pOffset = p->pOffset;
2425 p->pOffset = 0;
2426 uniondest.eDest = op;
2427 explainSetInteger(iSub2, pParse->iNextSelectId);
2428 rc = sqlite3Select(pParse, p, &uniondest);
2429 testcase( rc!=SQLITE_OK );
2430 /* Query flattening in sqlite3Select() might refill p->pOrderBy.
2431 ** Be sure to delete p->pOrderBy, therefore, to avoid a memory leak. */
2432 sqlite3ExprListDelete(db, p->pOrderBy);
2433 pDelete = p->pPrior;
2434 p->pPrior = pPrior;
2435 p->pOrderBy = 0;
2436 if( p->op==TK_UNION ){
2437 p->nSelectRow = sqlite3LogEstAdd(p->nSelectRow, pPrior->nSelectRow);
2439 sqlite3ExprDelete(db, p->pLimit);
2440 p->pLimit = pLimit;
2441 p->pOffset = pOffset;
2442 p->iLimit = 0;
2443 p->iOffset = 0;
2445 /* Convert the data in the temporary table into whatever form
2446 ** it is that we currently need.
2448 assert( unionTab==dest.iSDParm || dest.eDest!=priorOp );
2449 if( dest.eDest!=priorOp ){
2450 int iCont, iBreak, iStart;
2451 assert( p->pEList );
2452 iBreak = sqlite3VdbeMakeLabel(v);
2453 iCont = sqlite3VdbeMakeLabel(v);
2454 computeLimitRegisters(pParse, p, iBreak);
2455 sqlite3VdbeAddOp2(v, OP_Rewind, unionTab, iBreak); VdbeCoverage(v);
2456 iStart = sqlite3VdbeCurrentAddr(v);
2457 selectInnerLoop(pParse, p, unionTab,
2458 0, 0, &dest, iCont, iBreak);
2459 sqlite3VdbeResolveLabel(v, iCont);
2460 sqlite3VdbeAddOp2(v, OP_Next, unionTab, iStart); VdbeCoverage(v);
2461 sqlite3VdbeResolveLabel(v, iBreak);
2462 sqlite3VdbeAddOp2(v, OP_Close, unionTab, 0);
2464 break;
2466 default: assert( p->op==TK_INTERSECT ); {
2467 int tab1, tab2;
2468 int iCont, iBreak, iStart;
2469 Expr *pLimit, *pOffset;
2470 int addr;
2471 SelectDest intersectdest;
2472 int r1;
2474 /* INTERSECT is different from the others since it requires
2475 ** two temporary tables. Hence it has its own case. Begin
2476 ** by allocating the tables we will need.
2478 tab1 = pParse->nTab++;
2479 tab2 = pParse->nTab++;
2480 assert( p->pOrderBy==0 );
2482 addr = sqlite3VdbeAddOp2(v, OP_OpenEphemeral, tab1, 0);
2483 assert( p->addrOpenEphm[0] == -1 );
2484 p->addrOpenEphm[0] = addr;
2485 findRightmost(p)->selFlags |= SF_UsesEphemeral;
2486 assert( p->pEList );
2488 /* Code the SELECTs to our left into temporary table "tab1".
2490 sqlite3SelectDestInit(&intersectdest, SRT_Union, tab1);
2491 explainSetInteger(iSub1, pParse->iNextSelectId);
2492 rc = sqlite3Select(pParse, pPrior, &intersectdest);
2493 if( rc ){
2494 goto multi_select_end;
2497 /* Code the current SELECT into temporary table "tab2"
2499 addr = sqlite3VdbeAddOp2(v, OP_OpenEphemeral, tab2, 0);
2500 assert( p->addrOpenEphm[1] == -1 );
2501 p->addrOpenEphm[1] = addr;
2502 p->pPrior = 0;
2503 pLimit = p->pLimit;
2504 p->pLimit = 0;
2505 pOffset = p->pOffset;
2506 p->pOffset = 0;
2507 intersectdest.iSDParm = tab2;
2508 explainSetInteger(iSub2, pParse->iNextSelectId);
2509 rc = sqlite3Select(pParse, p, &intersectdest);
2510 testcase( rc!=SQLITE_OK );
2511 pDelete = p->pPrior;
2512 p->pPrior = pPrior;
2513 if( p->nSelectRow>pPrior->nSelectRow ) p->nSelectRow = pPrior->nSelectRow;
2514 sqlite3ExprDelete(db, p->pLimit);
2515 p->pLimit = pLimit;
2516 p->pOffset = pOffset;
2518 /* Generate code to take the intersection of the two temporary
2519 ** tables.
2521 assert( p->pEList );
2522 iBreak = sqlite3VdbeMakeLabel(v);
2523 iCont = sqlite3VdbeMakeLabel(v);
2524 computeLimitRegisters(pParse, p, iBreak);
2525 sqlite3VdbeAddOp2(v, OP_Rewind, tab1, iBreak); VdbeCoverage(v);
2526 r1 = sqlite3GetTempReg(pParse);
2527 iStart = sqlite3VdbeAddOp2(v, OP_RowData, tab1, r1);
2528 sqlite3VdbeAddOp4Int(v, OP_NotFound, tab2, iCont, r1, 0); VdbeCoverage(v);
2529 sqlite3ReleaseTempReg(pParse, r1);
2530 selectInnerLoop(pParse, p, tab1,
2531 0, 0, &dest, iCont, iBreak);
2532 sqlite3VdbeResolveLabel(v, iCont);
2533 sqlite3VdbeAddOp2(v, OP_Next, tab1, iStart); VdbeCoverage(v);
2534 sqlite3VdbeResolveLabel(v, iBreak);
2535 sqlite3VdbeAddOp2(v, OP_Close, tab2, 0);
2536 sqlite3VdbeAddOp2(v, OP_Close, tab1, 0);
2537 break;
2541 explainComposite(pParse, p->op, iSub1, iSub2, p->op!=TK_ALL);
2543 /* Compute collating sequences used by
2544 ** temporary tables needed to implement the compound select.
2545 ** Attach the KeyInfo structure to all temporary tables.
2547 ** This section is run by the right-most SELECT statement only.
2548 ** SELECT statements to the left always skip this part. The right-most
2549 ** SELECT might also skip this part if it has no ORDER BY clause and
2550 ** no temp tables are required.
2552 if( p->selFlags & SF_UsesEphemeral ){
2553 int i; /* Loop counter */
2554 KeyInfo *pKeyInfo; /* Collating sequence for the result set */
2555 Select *pLoop; /* For looping through SELECT statements */
2556 CollSeq **apColl; /* For looping through pKeyInfo->aColl[] */
2557 int nCol; /* Number of columns in result set */
2559 assert( p->pNext==0 );
2560 nCol = p->pEList->nExpr;
2561 pKeyInfo = sqlite3KeyInfoAlloc(db, nCol, 1);
2562 if( !pKeyInfo ){
2563 rc = SQLITE_NOMEM_BKPT;
2564 goto multi_select_end;
2566 for(i=0, apColl=pKeyInfo->aColl; i<nCol; i++, apColl++){
2567 *apColl = multiSelectCollSeq(pParse, p, i);
2568 if( 0==*apColl ){
2569 *apColl = db->pDfltColl;
2573 for(pLoop=p; pLoop; pLoop=pLoop->pPrior){
2574 for(i=0; i<2; i++){
2575 int addr = pLoop->addrOpenEphm[i];
2576 if( addr<0 ){
2577 /* If [0] is unused then [1] is also unused. So we can
2578 ** always safely abort as soon as the first unused slot is found */
2579 assert( pLoop->addrOpenEphm[1]<0 );
2580 break;
2582 sqlite3VdbeChangeP2(v, addr, nCol);
2583 sqlite3VdbeChangeP4(v, addr, (char*)sqlite3KeyInfoRef(pKeyInfo),
2584 P4_KEYINFO);
2585 pLoop->addrOpenEphm[i] = -1;
2588 sqlite3KeyInfoUnref(pKeyInfo);
2591 multi_select_end:
2592 pDest->iSdst = dest.iSdst;
2593 pDest->nSdst = dest.nSdst;
2594 sqlite3SelectDelete(db, pDelete);
2595 return rc;
2597 #endif /* SQLITE_OMIT_COMPOUND_SELECT */
2600 ** Error message for when two or more terms of a compound select have different
2601 ** size result sets.
2603 void sqlite3SelectWrongNumTermsError(Parse *pParse, Select *p){
2604 if( p->selFlags & SF_Values ){
2605 sqlite3ErrorMsg(pParse, "all VALUES must have the same number of terms");
2606 }else{
2607 sqlite3ErrorMsg(pParse, "SELECTs to the left and right of %s"
2608 " do not have the same number of result columns", selectOpName(p->op));
2613 ** Code an output subroutine for a coroutine implementation of a
2614 ** SELECT statment.
2616 ** The data to be output is contained in pIn->iSdst. There are
2617 ** pIn->nSdst columns to be output. pDest is where the output should
2618 ** be sent.
2620 ** regReturn is the number of the register holding the subroutine
2621 ** return address.
2623 ** If regPrev>0 then it is the first register in a vector that
2624 ** records the previous output. mem[regPrev] is a flag that is false
2625 ** if there has been no previous output. If regPrev>0 then code is
2626 ** generated to suppress duplicates. pKeyInfo is used for comparing
2627 ** keys.
2629 ** If the LIMIT found in p->iLimit is reached, jump immediately to
2630 ** iBreak.
2632 static int generateOutputSubroutine(
2633 Parse *pParse, /* Parsing context */
2634 Select *p, /* The SELECT statement */
2635 SelectDest *pIn, /* Coroutine supplying data */
2636 SelectDest *pDest, /* Where to send the data */
2637 int regReturn, /* The return address register */
2638 int regPrev, /* Previous result register. No uniqueness if 0 */
2639 KeyInfo *pKeyInfo, /* For comparing with previous entry */
2640 int iBreak /* Jump here if we hit the LIMIT */
2642 Vdbe *v = pParse->pVdbe;
2643 int iContinue;
2644 int addr;
2646 addr = sqlite3VdbeCurrentAddr(v);
2647 iContinue = sqlite3VdbeMakeLabel(v);
2649 /* Suppress duplicates for UNION, EXCEPT, and INTERSECT
2651 if( regPrev ){
2652 int addr1, addr2;
2653 addr1 = sqlite3VdbeAddOp1(v, OP_IfNot, regPrev); VdbeCoverage(v);
2654 addr2 = sqlite3VdbeAddOp4(v, OP_Compare, pIn->iSdst, regPrev+1, pIn->nSdst,
2655 (char*)sqlite3KeyInfoRef(pKeyInfo), P4_KEYINFO);
2656 sqlite3VdbeAddOp3(v, OP_Jump, addr2+2, iContinue, addr2+2); VdbeCoverage(v);
2657 sqlite3VdbeJumpHere(v, addr1);
2658 sqlite3VdbeAddOp3(v, OP_Copy, pIn->iSdst, regPrev+1, pIn->nSdst-1);
2659 sqlite3VdbeAddOp2(v, OP_Integer, 1, regPrev);
2661 if( pParse->db->mallocFailed ) return 0;
2663 /* Suppress the first OFFSET entries if there is an OFFSET clause
2665 codeOffset(v, p->iOffset, iContinue);
2667 assert( pDest->eDest!=SRT_Exists );
2668 assert( pDest->eDest!=SRT_Table );
2669 switch( pDest->eDest ){
2670 /* Store the result as data using a unique key.
2672 case SRT_EphemTab: {
2673 int r1 = sqlite3GetTempReg(pParse);
2674 int r2 = sqlite3GetTempReg(pParse);
2675 sqlite3VdbeAddOp3(v, OP_MakeRecord, pIn->iSdst, pIn->nSdst, r1);
2676 sqlite3VdbeAddOp2(v, OP_NewRowid, pDest->iSDParm, r2);
2677 sqlite3VdbeAddOp3(v, OP_Insert, pDest->iSDParm, r1, r2);
2678 sqlite3VdbeChangeP5(v, OPFLAG_APPEND);
2679 sqlite3ReleaseTempReg(pParse, r2);
2680 sqlite3ReleaseTempReg(pParse, r1);
2681 break;
2684 #ifndef SQLITE_OMIT_SUBQUERY
2685 /* If we are creating a set for an "expr IN (SELECT ...)".
2687 case SRT_Set: {
2688 int r1;
2689 testcase( pIn->nSdst>1 );
2690 r1 = sqlite3GetTempReg(pParse);
2691 sqlite3VdbeAddOp4(v, OP_MakeRecord, pIn->iSdst, pIn->nSdst,
2692 r1, pDest->zAffSdst, pIn->nSdst);
2693 sqlite3ExprCacheAffinityChange(pParse, pIn->iSdst, pIn->nSdst);
2694 sqlite3VdbeAddOp4Int(v, OP_IdxInsert, pDest->iSDParm, r1,
2695 pIn->iSdst, pIn->nSdst);
2696 sqlite3ReleaseTempReg(pParse, r1);
2697 break;
2700 /* If this is a scalar select that is part of an expression, then
2701 ** store the results in the appropriate memory cell and break out
2702 ** of the scan loop.
2704 case SRT_Mem: {
2705 assert( pIn->nSdst==1 || pParse->nErr>0 ); testcase( pIn->nSdst!=1 );
2706 sqlite3ExprCodeMove(pParse, pIn->iSdst, pDest->iSDParm, 1);
2707 /* The LIMIT clause will jump out of the loop for us */
2708 break;
2710 #endif /* #ifndef SQLITE_OMIT_SUBQUERY */
2712 /* The results are stored in a sequence of registers
2713 ** starting at pDest->iSdst. Then the co-routine yields.
2715 case SRT_Coroutine: {
2716 if( pDest->iSdst==0 ){
2717 pDest->iSdst = sqlite3GetTempRange(pParse, pIn->nSdst);
2718 pDest->nSdst = pIn->nSdst;
2720 sqlite3ExprCodeMove(pParse, pIn->iSdst, pDest->iSdst, pIn->nSdst);
2721 sqlite3VdbeAddOp1(v, OP_Yield, pDest->iSDParm);
2722 break;
2725 /* If none of the above, then the result destination must be
2726 ** SRT_Output. This routine is never called with any other
2727 ** destination other than the ones handled above or SRT_Output.
2729 ** For SRT_Output, results are stored in a sequence of registers.
2730 ** Then the OP_ResultRow opcode is used to cause sqlite3_step() to
2731 ** return the next row of result.
2733 default: {
2734 assert( pDest->eDest==SRT_Output );
2735 sqlite3VdbeAddOp2(v, OP_ResultRow, pIn->iSdst, pIn->nSdst);
2736 sqlite3ExprCacheAffinityChange(pParse, pIn->iSdst, pIn->nSdst);
2737 break;
2741 /* Jump to the end of the loop if the LIMIT is reached.
2743 if( p->iLimit ){
2744 sqlite3VdbeAddOp2(v, OP_DecrJumpZero, p->iLimit, iBreak); VdbeCoverage(v);
2747 /* Generate the subroutine return
2749 sqlite3VdbeResolveLabel(v, iContinue);
2750 sqlite3VdbeAddOp1(v, OP_Return, regReturn);
2752 return addr;
2756 ** Alternative compound select code generator for cases when there
2757 ** is an ORDER BY clause.
2759 ** We assume a query of the following form:
2761 ** <selectA> <operator> <selectB> ORDER BY <orderbylist>
2763 ** <operator> is one of UNION ALL, UNION, EXCEPT, or INTERSECT. The idea
2764 ** is to code both <selectA> and <selectB> with the ORDER BY clause as
2765 ** co-routines. Then run the co-routines in parallel and merge the results
2766 ** into the output. In addition to the two coroutines (called selectA and
2767 ** selectB) there are 7 subroutines:
2769 ** outA: Move the output of the selectA coroutine into the output
2770 ** of the compound query.
2772 ** outB: Move the output of the selectB coroutine into the output
2773 ** of the compound query. (Only generated for UNION and
2774 ** UNION ALL. EXCEPT and INSERTSECT never output a row that
2775 ** appears only in B.)
2777 ** AltB: Called when there is data from both coroutines and A<B.
2779 ** AeqB: Called when there is data from both coroutines and A==B.
2781 ** AgtB: Called when there is data from both coroutines and A>B.
2783 ** EofA: Called when data is exhausted from selectA.
2785 ** EofB: Called when data is exhausted from selectB.
2787 ** The implementation of the latter five subroutines depend on which
2788 ** <operator> is used:
2791 ** UNION ALL UNION EXCEPT INTERSECT
2792 ** ------------- ----------------- -------------- -----------------
2793 ** AltB: outA, nextA outA, nextA outA, nextA nextA
2795 ** AeqB: outA, nextA nextA nextA outA, nextA
2797 ** AgtB: outB, nextB outB, nextB nextB nextB
2799 ** EofA: outB, nextB outB, nextB halt halt
2801 ** EofB: outA, nextA outA, nextA outA, nextA halt
2803 ** In the AltB, AeqB, and AgtB subroutines, an EOF on A following nextA
2804 ** causes an immediate jump to EofA and an EOF on B following nextB causes
2805 ** an immediate jump to EofB. Within EofA and EofB, and EOF on entry or
2806 ** following nextX causes a jump to the end of the select processing.
2808 ** Duplicate removal in the UNION, EXCEPT, and INTERSECT cases is handled
2809 ** within the output subroutine. The regPrev register set holds the previously
2810 ** output value. A comparison is made against this value and the output
2811 ** is skipped if the next results would be the same as the previous.
2813 ** The implementation plan is to implement the two coroutines and seven
2814 ** subroutines first, then put the control logic at the bottom. Like this:
2816 ** goto Init
2817 ** coA: coroutine for left query (A)
2818 ** coB: coroutine for right query (B)
2819 ** outA: output one row of A
2820 ** outB: output one row of B (UNION and UNION ALL only)
2821 ** EofA: ...
2822 ** EofB: ...
2823 ** AltB: ...
2824 ** AeqB: ...
2825 ** AgtB: ...
2826 ** Init: initialize coroutine registers
2827 ** yield coA
2828 ** if eof(A) goto EofA
2829 ** yield coB
2830 ** if eof(B) goto EofB
2831 ** Cmpr: Compare A, B
2832 ** Jump AltB, AeqB, AgtB
2833 ** End: ...
2835 ** We call AltB, AeqB, AgtB, EofA, and EofB "subroutines" but they are not
2836 ** actually called using Gosub and they do not Return. EofA and EofB loop
2837 ** until all data is exhausted then jump to the "end" labe. AltB, AeqB,
2838 ** and AgtB jump to either L2 or to one of EofA or EofB.
2840 #ifndef SQLITE_OMIT_COMPOUND_SELECT
2841 static int multiSelectOrderBy(
2842 Parse *pParse, /* Parsing context */
2843 Select *p, /* The right-most of SELECTs to be coded */
2844 SelectDest *pDest /* What to do with query results */
2846 int i, j; /* Loop counters */
2847 Select *pPrior; /* Another SELECT immediately to our left */
2848 Vdbe *v; /* Generate code to this VDBE */
2849 SelectDest destA; /* Destination for coroutine A */
2850 SelectDest destB; /* Destination for coroutine B */
2851 int regAddrA; /* Address register for select-A coroutine */
2852 int regAddrB; /* Address register for select-B coroutine */
2853 int addrSelectA; /* Address of the select-A coroutine */
2854 int addrSelectB; /* Address of the select-B coroutine */
2855 int regOutA; /* Address register for the output-A subroutine */
2856 int regOutB; /* Address register for the output-B subroutine */
2857 int addrOutA; /* Address of the output-A subroutine */
2858 int addrOutB = 0; /* Address of the output-B subroutine */
2859 int addrEofA; /* Address of the select-A-exhausted subroutine */
2860 int addrEofA_noB; /* Alternate addrEofA if B is uninitialized */
2861 int addrEofB; /* Address of the select-B-exhausted subroutine */
2862 int addrAltB; /* Address of the A<B subroutine */
2863 int addrAeqB; /* Address of the A==B subroutine */
2864 int addrAgtB; /* Address of the A>B subroutine */
2865 int regLimitA; /* Limit register for select-A */
2866 int regLimitB; /* Limit register for select-A */
2867 int regPrev; /* A range of registers to hold previous output */
2868 int savedLimit; /* Saved value of p->iLimit */
2869 int savedOffset; /* Saved value of p->iOffset */
2870 int labelCmpr; /* Label for the start of the merge algorithm */
2871 int labelEnd; /* Label for the end of the overall SELECT stmt */
2872 int addr1; /* Jump instructions that get retargetted */
2873 int op; /* One of TK_ALL, TK_UNION, TK_EXCEPT, TK_INTERSECT */
2874 KeyInfo *pKeyDup = 0; /* Comparison information for duplicate removal */
2875 KeyInfo *pKeyMerge; /* Comparison information for merging rows */
2876 sqlite3 *db; /* Database connection */
2877 ExprList *pOrderBy; /* The ORDER BY clause */
2878 int nOrderBy; /* Number of terms in the ORDER BY clause */
2879 int *aPermute; /* Mapping from ORDER BY terms to result set columns */
2880 #ifndef SQLITE_OMIT_EXPLAIN
2881 int iSub1; /* EQP id of left-hand query */
2882 int iSub2; /* EQP id of right-hand query */
2883 #endif
2885 assert( p->pOrderBy!=0 );
2886 assert( pKeyDup==0 ); /* "Managed" code needs this. Ticket #3382. */
2887 db = pParse->db;
2888 v = pParse->pVdbe;
2889 assert( v!=0 ); /* Already thrown the error if VDBE alloc failed */
2890 labelEnd = sqlite3VdbeMakeLabel(v);
2891 labelCmpr = sqlite3VdbeMakeLabel(v);
2894 /* Patch up the ORDER BY clause
2896 op = p->op;
2897 pPrior = p->pPrior;
2898 assert( pPrior->pOrderBy==0 );
2899 pOrderBy = p->pOrderBy;
2900 assert( pOrderBy );
2901 nOrderBy = pOrderBy->nExpr;
2903 /* For operators other than UNION ALL we have to make sure that
2904 ** the ORDER BY clause covers every term of the result set. Add
2905 ** terms to the ORDER BY clause as necessary.
2907 if( op!=TK_ALL ){
2908 for(i=1; db->mallocFailed==0 && i<=p->pEList->nExpr; i++){
2909 struct ExprList_item *pItem;
2910 for(j=0, pItem=pOrderBy->a; j<nOrderBy; j++, pItem++){
2911 assert( pItem->u.x.iOrderByCol>0 );
2912 if( pItem->u.x.iOrderByCol==i ) break;
2914 if( j==nOrderBy ){
2915 Expr *pNew = sqlite3Expr(db, TK_INTEGER, 0);
2916 if( pNew==0 ) return SQLITE_NOMEM_BKPT;
2917 pNew->flags |= EP_IntValue;
2918 pNew->u.iValue = i;
2919 p->pOrderBy = pOrderBy = sqlite3ExprListAppend(pParse, pOrderBy, pNew);
2920 if( pOrderBy ) pOrderBy->a[nOrderBy++].u.x.iOrderByCol = (u16)i;
2925 /* Compute the comparison permutation and keyinfo that is used with
2926 ** the permutation used to determine if the next
2927 ** row of results comes from selectA or selectB. Also add explicit
2928 ** collations to the ORDER BY clause terms so that when the subqueries
2929 ** to the right and the left are evaluated, they use the correct
2930 ** collation.
2932 aPermute = sqlite3DbMallocRawNN(db, sizeof(int)*(nOrderBy + 1));
2933 if( aPermute ){
2934 struct ExprList_item *pItem;
2935 aPermute[0] = nOrderBy;
2936 for(i=1, pItem=pOrderBy->a; i<=nOrderBy; i++, pItem++){
2937 assert( pItem->u.x.iOrderByCol>0 );
2938 assert( pItem->u.x.iOrderByCol<=p->pEList->nExpr );
2939 aPermute[i] = pItem->u.x.iOrderByCol - 1;
2941 pKeyMerge = multiSelectOrderByKeyInfo(pParse, p, 1);
2942 }else{
2943 pKeyMerge = 0;
2946 /* Reattach the ORDER BY clause to the query.
2948 p->pOrderBy = pOrderBy;
2949 pPrior->pOrderBy = sqlite3ExprListDup(pParse->db, pOrderBy, 0);
2951 /* Allocate a range of temporary registers and the KeyInfo needed
2952 ** for the logic that removes duplicate result rows when the
2953 ** operator is UNION, EXCEPT, or INTERSECT (but not UNION ALL).
2955 if( op==TK_ALL ){
2956 regPrev = 0;
2957 }else{
2958 int nExpr = p->pEList->nExpr;
2959 assert( nOrderBy>=nExpr || db->mallocFailed );
2960 regPrev = pParse->nMem+1;
2961 pParse->nMem += nExpr+1;
2962 sqlite3VdbeAddOp2(v, OP_Integer, 0, regPrev);
2963 pKeyDup = sqlite3KeyInfoAlloc(db, nExpr, 1);
2964 if( pKeyDup ){
2965 assert( sqlite3KeyInfoIsWriteable(pKeyDup) );
2966 for(i=0; i<nExpr; i++){
2967 pKeyDup->aColl[i] = multiSelectCollSeq(pParse, p, i);
2968 pKeyDup->aSortOrder[i] = 0;
2973 /* Separate the left and the right query from one another
2975 p->pPrior = 0;
2976 pPrior->pNext = 0;
2977 sqlite3ResolveOrderGroupBy(pParse, p, p->pOrderBy, "ORDER");
2978 if( pPrior->pPrior==0 ){
2979 sqlite3ResolveOrderGroupBy(pParse, pPrior, pPrior->pOrderBy, "ORDER");
2982 /* Compute the limit registers */
2983 computeLimitRegisters(pParse, p, labelEnd);
2984 if( p->iLimit && op==TK_ALL ){
2985 regLimitA = ++pParse->nMem;
2986 regLimitB = ++pParse->nMem;
2987 sqlite3VdbeAddOp2(v, OP_Copy, p->iOffset ? p->iOffset+1 : p->iLimit,
2988 regLimitA);
2989 sqlite3VdbeAddOp2(v, OP_Copy, regLimitA, regLimitB);
2990 }else{
2991 regLimitA = regLimitB = 0;
2993 sqlite3ExprDelete(db, p->pLimit);
2994 p->pLimit = 0;
2995 sqlite3ExprDelete(db, p->pOffset);
2996 p->pOffset = 0;
2998 regAddrA = ++pParse->nMem;
2999 regAddrB = ++pParse->nMem;
3000 regOutA = ++pParse->nMem;
3001 regOutB = ++pParse->nMem;
3002 sqlite3SelectDestInit(&destA, SRT_Coroutine, regAddrA);
3003 sqlite3SelectDestInit(&destB, SRT_Coroutine, regAddrB);
3005 /* Generate a coroutine to evaluate the SELECT statement to the
3006 ** left of the compound operator - the "A" select.
3008 addrSelectA = sqlite3VdbeCurrentAddr(v) + 1;
3009 addr1 = sqlite3VdbeAddOp3(v, OP_InitCoroutine, regAddrA, 0, addrSelectA);
3010 VdbeComment((v, "left SELECT"));
3011 pPrior->iLimit = regLimitA;
3012 explainSetInteger(iSub1, pParse->iNextSelectId);
3013 sqlite3Select(pParse, pPrior, &destA);
3014 sqlite3VdbeEndCoroutine(v, regAddrA);
3015 sqlite3VdbeJumpHere(v, addr1);
3017 /* Generate a coroutine to evaluate the SELECT statement on
3018 ** the right - the "B" select
3020 addrSelectB = sqlite3VdbeCurrentAddr(v) + 1;
3021 addr1 = sqlite3VdbeAddOp3(v, OP_InitCoroutine, regAddrB, 0, addrSelectB);
3022 VdbeComment((v, "right SELECT"));
3023 savedLimit = p->iLimit;
3024 savedOffset = p->iOffset;
3025 p->iLimit = regLimitB;
3026 p->iOffset = 0;
3027 explainSetInteger(iSub2, pParse->iNextSelectId);
3028 sqlite3Select(pParse, p, &destB);
3029 p->iLimit = savedLimit;
3030 p->iOffset = savedOffset;
3031 sqlite3VdbeEndCoroutine(v, regAddrB);
3033 /* Generate a subroutine that outputs the current row of the A
3034 ** select as the next output row of the compound select.
3036 VdbeNoopComment((v, "Output routine for A"));
3037 addrOutA = generateOutputSubroutine(pParse,
3038 p, &destA, pDest, regOutA,
3039 regPrev, pKeyDup, labelEnd);
3041 /* Generate a subroutine that outputs the current row of the B
3042 ** select as the next output row of the compound select.
3044 if( op==TK_ALL || op==TK_UNION ){
3045 VdbeNoopComment((v, "Output routine for B"));
3046 addrOutB = generateOutputSubroutine(pParse,
3047 p, &destB, pDest, regOutB,
3048 regPrev, pKeyDup, labelEnd);
3050 sqlite3KeyInfoUnref(pKeyDup);
3052 /* Generate a subroutine to run when the results from select A
3053 ** are exhausted and only data in select B remains.
3055 if( op==TK_EXCEPT || op==TK_INTERSECT ){
3056 addrEofA_noB = addrEofA = labelEnd;
3057 }else{
3058 VdbeNoopComment((v, "eof-A subroutine"));
3059 addrEofA = sqlite3VdbeAddOp2(v, OP_Gosub, regOutB, addrOutB);
3060 addrEofA_noB = sqlite3VdbeAddOp2(v, OP_Yield, regAddrB, labelEnd);
3061 VdbeCoverage(v);
3062 sqlite3VdbeGoto(v, addrEofA);
3063 p->nSelectRow = sqlite3LogEstAdd(p->nSelectRow, pPrior->nSelectRow);
3066 /* Generate a subroutine to run when the results from select B
3067 ** are exhausted and only data in select A remains.
3069 if( op==TK_INTERSECT ){
3070 addrEofB = addrEofA;
3071 if( p->nSelectRow > pPrior->nSelectRow ) p->nSelectRow = pPrior->nSelectRow;
3072 }else{
3073 VdbeNoopComment((v, "eof-B subroutine"));
3074 addrEofB = sqlite3VdbeAddOp2(v, OP_Gosub, regOutA, addrOutA);
3075 sqlite3VdbeAddOp2(v, OP_Yield, regAddrA, labelEnd); VdbeCoverage(v);
3076 sqlite3VdbeGoto(v, addrEofB);
3079 /* Generate code to handle the case of A<B
3081 VdbeNoopComment((v, "A-lt-B subroutine"));
3082 addrAltB = sqlite3VdbeAddOp2(v, OP_Gosub, regOutA, addrOutA);
3083 sqlite3VdbeAddOp2(v, OP_Yield, regAddrA, addrEofA); VdbeCoverage(v);
3084 sqlite3VdbeGoto(v, labelCmpr);
3086 /* Generate code to handle the case of A==B
3088 if( op==TK_ALL ){
3089 addrAeqB = addrAltB;
3090 }else if( op==TK_INTERSECT ){
3091 addrAeqB = addrAltB;
3092 addrAltB++;
3093 }else{
3094 VdbeNoopComment((v, "A-eq-B subroutine"));
3095 addrAeqB =
3096 sqlite3VdbeAddOp2(v, OP_Yield, regAddrA, addrEofA); VdbeCoverage(v);
3097 sqlite3VdbeGoto(v, labelCmpr);
3100 /* Generate code to handle the case of A>B
3102 VdbeNoopComment((v, "A-gt-B subroutine"));
3103 addrAgtB = sqlite3VdbeCurrentAddr(v);
3104 if( op==TK_ALL || op==TK_UNION ){
3105 sqlite3VdbeAddOp2(v, OP_Gosub, regOutB, addrOutB);
3107 sqlite3VdbeAddOp2(v, OP_Yield, regAddrB, addrEofB); VdbeCoverage(v);
3108 sqlite3VdbeGoto(v, labelCmpr);
3110 /* This code runs once to initialize everything.
3112 sqlite3VdbeJumpHere(v, addr1);
3113 sqlite3VdbeAddOp2(v, OP_Yield, regAddrA, addrEofA_noB); VdbeCoverage(v);
3114 sqlite3VdbeAddOp2(v, OP_Yield, regAddrB, addrEofB); VdbeCoverage(v);
3116 /* Implement the main merge loop
3118 sqlite3VdbeResolveLabel(v, labelCmpr);
3119 sqlite3VdbeAddOp4(v, OP_Permutation, 0, 0, 0, (char*)aPermute, P4_INTARRAY);
3120 sqlite3VdbeAddOp4(v, OP_Compare, destA.iSdst, destB.iSdst, nOrderBy,
3121 (char*)pKeyMerge, P4_KEYINFO);
3122 sqlite3VdbeChangeP5(v, OPFLAG_PERMUTE);
3123 sqlite3VdbeAddOp3(v, OP_Jump, addrAltB, addrAeqB, addrAgtB); VdbeCoverage(v);
3125 /* Jump to the this point in order to terminate the query.
3127 sqlite3VdbeResolveLabel(v, labelEnd);
3129 /* Reassembly the compound query so that it will be freed correctly
3130 ** by the calling function */
3131 if( p->pPrior ){
3132 sqlite3SelectDelete(db, p->pPrior);
3134 p->pPrior = pPrior;
3135 pPrior->pNext = p;
3137 /*** TBD: Insert subroutine calls to close cursors on incomplete
3138 **** subqueries ****/
3139 explainComposite(pParse, p->op, iSub1, iSub2, 0);
3140 return pParse->nErr!=0;
3142 #endif
3144 #if !defined(SQLITE_OMIT_SUBQUERY) || !defined(SQLITE_OMIT_VIEW)
3146 /* An instance of the SubstContext object describes an substitution edit
3147 ** to be performed on a parse tree.
3149 ** All references to columns in table iTable are to be replaced by corresponding
3150 ** expressions in pEList.
3152 typedef struct SubstContext {
3153 Parse *pParse; /* The parsing context */
3154 int iTable; /* Replace references to this table */
3155 int iNewTable; /* New table number */
3156 int isLeftJoin; /* Add TK_IF_NULL_ROW opcodes on each replacement */
3157 ExprList *pEList; /* Replacement expressions */
3158 } SubstContext;
3160 /* Forward Declarations */
3161 static void substExprList(SubstContext*, ExprList*);
3162 static void substSelect(SubstContext*, Select*, int);
3165 ** Scan through the expression pExpr. Replace every reference to
3166 ** a column in table number iTable with a copy of the iColumn-th
3167 ** entry in pEList. (But leave references to the ROWID column
3168 ** unchanged.)
3170 ** This routine is part of the flattening procedure. A subquery
3171 ** whose result set is defined by pEList appears as entry in the
3172 ** FROM clause of a SELECT such that the VDBE cursor assigned to that
3173 ** FORM clause entry is iTable. This routine makes the necessary
3174 ** changes to pExpr so that it refers directly to the source table
3175 ** of the subquery rather the result set of the subquery.
3177 static Expr *substExpr(
3178 SubstContext *pSubst, /* Description of the substitution */
3179 Expr *pExpr /* Expr in which substitution occurs */
3181 if( pExpr==0 ) return 0;
3182 if( ExprHasProperty(pExpr, EP_FromJoin)
3183 && pExpr->iRightJoinTable==pSubst->iTable
3185 pExpr->iRightJoinTable = pSubst->iNewTable;
3187 if( pExpr->op==TK_COLUMN && pExpr->iTable==pSubst->iTable ){
3188 if( pExpr->iColumn<0 ){
3189 pExpr->op = TK_NULL;
3190 }else{
3191 Expr *pNew;
3192 Expr *pCopy = pSubst->pEList->a[pExpr->iColumn].pExpr;
3193 Expr ifNullRow;
3194 assert( pSubst->pEList!=0 && pExpr->iColumn<pSubst->pEList->nExpr );
3195 assert( pExpr->pLeft==0 && pExpr->pRight==0 );
3196 if( sqlite3ExprIsVector(pCopy) ){
3197 sqlite3VectorErrorMsg(pSubst->pParse, pCopy);
3198 }else{
3199 sqlite3 *db = pSubst->pParse->db;
3200 if( pSubst->isLeftJoin && pCopy->op!=TK_COLUMN ){
3201 memset(&ifNullRow, 0, sizeof(ifNullRow));
3202 ifNullRow.op = TK_IF_NULL_ROW;
3203 ifNullRow.pLeft = pCopy;
3204 ifNullRow.iTable = pSubst->iNewTable;
3205 pCopy = &ifNullRow;
3207 pNew = sqlite3ExprDup(db, pCopy, 0);
3208 if( pNew && pSubst->isLeftJoin ){
3209 ExprSetProperty(pNew, EP_CanBeNull);
3211 if( pNew && ExprHasProperty(pExpr,EP_FromJoin) ){
3212 pNew->iRightJoinTable = pExpr->iRightJoinTable;
3213 ExprSetProperty(pNew, EP_FromJoin);
3215 sqlite3ExprDelete(db, pExpr);
3216 pExpr = pNew;
3219 }else{
3220 if( pExpr->op==TK_IF_NULL_ROW && pExpr->iTable==pSubst->iTable ){
3221 pExpr->iTable = pSubst->iNewTable;
3223 pExpr->pLeft = substExpr(pSubst, pExpr->pLeft);
3224 pExpr->pRight = substExpr(pSubst, pExpr->pRight);
3225 if( ExprHasProperty(pExpr, EP_xIsSelect) ){
3226 substSelect(pSubst, pExpr->x.pSelect, 1);
3227 }else{
3228 substExprList(pSubst, pExpr->x.pList);
3231 return pExpr;
3233 static void substExprList(
3234 SubstContext *pSubst, /* Description of the substitution */
3235 ExprList *pList /* List to scan and in which to make substitutes */
3237 int i;
3238 if( pList==0 ) return;
3239 for(i=0; i<pList->nExpr; i++){
3240 pList->a[i].pExpr = substExpr(pSubst, pList->a[i].pExpr);
3243 static void substSelect(
3244 SubstContext *pSubst, /* Description of the substitution */
3245 Select *p, /* SELECT statement in which to make substitutions */
3246 int doPrior /* Do substitutes on p->pPrior too */
3248 SrcList *pSrc;
3249 struct SrcList_item *pItem;
3250 int i;
3251 if( !p ) return;
3253 substExprList(pSubst, p->pEList);
3254 substExprList(pSubst, p->pGroupBy);
3255 substExprList(pSubst, p->pOrderBy);
3256 p->pHaving = substExpr(pSubst, p->pHaving);
3257 p->pWhere = substExpr(pSubst, p->pWhere);
3258 pSrc = p->pSrc;
3259 assert( pSrc!=0 );
3260 for(i=pSrc->nSrc, pItem=pSrc->a; i>0; i--, pItem++){
3261 substSelect(pSubst, pItem->pSelect, 1);
3262 if( pItem->fg.isTabFunc ){
3263 substExprList(pSubst, pItem->u1.pFuncArg);
3266 }while( doPrior && (p = p->pPrior)!=0 );
3268 #endif /* !defined(SQLITE_OMIT_SUBQUERY) || !defined(SQLITE_OMIT_VIEW) */
3270 #if !defined(SQLITE_OMIT_SUBQUERY) || !defined(SQLITE_OMIT_VIEW)
3272 ** This routine attempts to flatten subqueries as a performance optimization.
3273 ** This routine returns 1 if it makes changes and 0 if no flattening occurs.
3275 ** To understand the concept of flattening, consider the following
3276 ** query:
3278 ** SELECT a FROM (SELECT x+y AS a FROM t1 WHERE z<100) WHERE a>5
3280 ** The default way of implementing this query is to execute the
3281 ** subquery first and store the results in a temporary table, then
3282 ** run the outer query on that temporary table. This requires two
3283 ** passes over the data. Furthermore, because the temporary table
3284 ** has no indices, the WHERE clause on the outer query cannot be
3285 ** optimized.
3287 ** This routine attempts to rewrite queries such as the above into
3288 ** a single flat select, like this:
3290 ** SELECT x+y AS a FROM t1 WHERE z<100 AND a>5
3292 ** The code generated for this simplification gives the same result
3293 ** but only has to scan the data once. And because indices might
3294 ** exist on the table t1, a complete scan of the data might be
3295 ** avoided.
3297 ** Flattening is subject to the following constraints:
3299 ** (**) We no longer attempt to flatten aggregate subqueries. Was:
3300 ** The subquery and the outer query cannot both be aggregates.
3302 ** (**) We no longer attempt to flatten aggregate subqueries. Was:
3303 ** (2) If the subquery is an aggregate then
3304 ** (2a) the outer query must not be a join and
3305 ** (2b) the outer query must not use subqueries
3306 ** other than the one FROM-clause subquery that is a candidate
3307 ** for flattening. (This is due to ticket [2f7170d73bf9abf80]
3308 ** from 2015-02-09.)
3310 ** (3) If the subquery is the right operand of a LEFT JOIN then
3311 ** (3a) the subquery may not be a join and
3312 ** (3b) the FROM clause of the subquery may not contain a virtual
3313 ** table and
3314 ** (3c) the outer query may not be an aggregate.
3316 ** (4) The subquery can not be DISTINCT.
3318 ** (**) At one point restrictions (4) and (5) defined a subset of DISTINCT
3319 ** sub-queries that were excluded from this optimization. Restriction
3320 ** (4) has since been expanded to exclude all DISTINCT subqueries.
3322 ** (**) We no longer attempt to flatten aggregate subqueries. Was:
3323 ** If the subquery is aggregate, the outer query may not be DISTINCT.
3325 ** (7) The subquery must have a FROM clause. TODO: For subqueries without
3326 ** A FROM clause, consider adding a FROM clause with the special
3327 ** table sqlite_once that consists of a single row containing a
3328 ** single NULL.
3330 ** (8) If the subquery uses LIMIT then the outer query may not be a join.
3332 ** (9) If the subquery uses LIMIT then the outer query may not be aggregate.
3334 ** (**) Restriction (10) was removed from the code on 2005-02-05 but we
3335 ** accidently carried the comment forward until 2014-09-15. Original
3336 ** constraint: "If the subquery is aggregate then the outer query
3337 ** may not use LIMIT."
3339 ** (11) The subquery and the outer query may not both have ORDER BY clauses.
3341 ** (**) Not implemented. Subsumed into restriction (3). Was previously
3342 ** a separate restriction deriving from ticket #350.
3344 ** (13) The subquery and outer query may not both use LIMIT.
3346 ** (14) The subquery may not use OFFSET.
3348 ** (15) If the outer query is part of a compound select, then the
3349 ** subquery may not use LIMIT.
3350 ** (See ticket #2339 and ticket [02a8e81d44]).
3352 ** (16) If the outer query is aggregate, then the subquery may not
3353 ** use ORDER BY. (Ticket #2942) This used to not matter
3354 ** until we introduced the group_concat() function.
3356 ** (17) If the subquery is a compound select, then
3357 ** (17a) all compound operators must be a UNION ALL, and
3358 ** (17b) no terms within the subquery compound may be aggregate
3359 ** or DISTINCT, and
3360 ** (17c) every term within the subquery compound must have a FROM clause
3361 ** (17d) the outer query may not be
3362 ** (17d1) aggregate, or
3363 ** (17d2) DISTINCT, or
3364 ** (17d3) a join.
3366 ** The parent and sub-query may contain WHERE clauses. Subject to
3367 ** rules (11), (13) and (14), they may also contain ORDER BY,
3368 ** LIMIT and OFFSET clauses. The subquery cannot use any compound
3369 ** operator other than UNION ALL because all the other compound
3370 ** operators have an implied DISTINCT which is disallowed by
3371 ** restriction (4).
3373 ** Also, each component of the sub-query must return the same number
3374 ** of result columns. This is actually a requirement for any compound
3375 ** SELECT statement, but all the code here does is make sure that no
3376 ** such (illegal) sub-query is flattened. The caller will detect the
3377 ** syntax error and return a detailed message.
3379 ** (18) If the sub-query is a compound select, then all terms of the
3380 ** ORDER BY clause of the parent must be simple references to
3381 ** columns of the sub-query.
3383 ** (19) If the subquery uses LIMIT then the outer query may not
3384 ** have a WHERE clause.
3386 ** (**) Subsumed into (17d3). Was: If the sub-query is a compound select,
3387 ** then it must not use an ORDER BY clause - Ticket #3773. Because
3388 ** of (17d3), then only way to have a compound subquery is if it is
3389 ** the only term in the FROM clause of the outer query. But if the
3390 ** only term in the FROM clause has an ORDER BY, then it will be
3391 ** implemented as a co-routine and the flattener will never be called.
3393 ** (21) If the subquery uses LIMIT then the outer query may not be
3394 ** DISTINCT. (See ticket [752e1646fc]).
3396 ** (22) The subquery may not be a recursive CTE.
3398 ** (**) Subsumed into restriction (17d3). Was: If the outer query is
3399 ** a recursive CTE, then the sub-query may not be a compound query.
3400 ** This restriction is because transforming the
3401 ** parent to a compound query confuses the code that handles
3402 ** recursive queries in multiSelect().
3404 ** (**) We no longer attempt to flatten aggregate subqueries. Was:
3405 ** The subquery may not be an aggregate that uses the built-in min() or
3406 ** or max() functions. (Without this restriction, a query like:
3407 ** "SELECT x FROM (SELECT max(y), x FROM t1)" would not necessarily
3408 ** return the value X for which Y was maximal.)
3411 ** In this routine, the "p" parameter is a pointer to the outer query.
3412 ** The subquery is p->pSrc->a[iFrom]. isAgg is true if the outer query
3413 ** uses aggregates.
3415 ** If flattening is not attempted, this routine is a no-op and returns 0.
3416 ** If flattening is attempted this routine returns 1.
3418 ** All of the expression analysis must occur on both the outer query and
3419 ** the subquery before this routine runs.
3421 static int flattenSubquery(
3422 Parse *pParse, /* Parsing context */
3423 Select *p, /* The parent or outer SELECT statement */
3424 int iFrom, /* Index in p->pSrc->a[] of the inner subquery */
3425 int isAgg /* True if outer SELECT uses aggregate functions */
3427 const char *zSavedAuthContext = pParse->zAuthContext;
3428 Select *pParent; /* Current UNION ALL term of the other query */
3429 Select *pSub; /* The inner query or "subquery" */
3430 Select *pSub1; /* Pointer to the rightmost select in sub-query */
3431 SrcList *pSrc; /* The FROM clause of the outer query */
3432 SrcList *pSubSrc; /* The FROM clause of the subquery */
3433 int iParent; /* VDBE cursor number of the pSub result set temp table */
3434 int iNewParent = -1;/* Replacement table for iParent */
3435 int isLeftJoin = 0; /* True if pSub is the right side of a LEFT JOIN */
3436 int i; /* Loop counter */
3437 Expr *pWhere; /* The WHERE clause */
3438 struct SrcList_item *pSubitem; /* The subquery */
3439 sqlite3 *db = pParse->db;
3441 /* Check to see if flattening is permitted. Return 0 if not.
3443 assert( p!=0 );
3444 assert( p->pPrior==0 );
3445 if( OptimizationDisabled(db, SQLITE_QueryFlattener) ) return 0;
3446 pSrc = p->pSrc;
3447 assert( pSrc && iFrom>=0 && iFrom<pSrc->nSrc );
3448 pSubitem = &pSrc->a[iFrom];
3449 iParent = pSubitem->iCursor;
3450 pSub = pSubitem->pSelect;
3451 assert( pSub!=0 );
3453 pSubSrc = pSub->pSrc;
3454 assert( pSubSrc );
3455 /* Prior to version 3.1.2, when LIMIT and OFFSET had to be simple constants,
3456 ** not arbitrary expressions, we allowed some combining of LIMIT and OFFSET
3457 ** because they could be computed at compile-time. But when LIMIT and OFFSET
3458 ** became arbitrary expressions, we were forced to add restrictions (13)
3459 ** and (14). */
3460 if( pSub->pLimit && p->pLimit ) return 0; /* Restriction (13) */
3461 if( pSub->pOffset ) return 0; /* Restriction (14) */
3462 if( (p->selFlags & SF_Compound)!=0 && pSub->pLimit ){
3463 return 0; /* Restriction (15) */
3465 if( pSubSrc->nSrc==0 ) return 0; /* Restriction (7) */
3466 if( pSub->selFlags & SF_Distinct ) return 0; /* Restriction (4) */
3467 if( pSub->pLimit && (pSrc->nSrc>1 || isAgg) ){
3468 return 0; /* Restrictions (8)(9) */
3470 if( p->pOrderBy && pSub->pOrderBy ){
3471 return 0; /* Restriction (11) */
3473 if( isAgg && pSub->pOrderBy ) return 0; /* Restriction (16) */
3474 if( pSub->pLimit && p->pWhere ) return 0; /* Restriction (19) */
3475 if( pSub->pLimit && (p->selFlags & SF_Distinct)!=0 ){
3476 return 0; /* Restriction (21) */
3478 if( pSub->selFlags & (SF_Recursive) ){
3479 return 0; /* Restrictions (22) */
3483 ** If the subquery is the right operand of a LEFT JOIN, then the
3484 ** subquery may not be a join itself (3a). Example of why this is not
3485 ** allowed:
3487 ** t1 LEFT OUTER JOIN (t2 JOIN t3)
3489 ** If we flatten the above, we would get
3491 ** (t1 LEFT OUTER JOIN t2) JOIN t3
3493 ** which is not at all the same thing.
3495 ** If the subquery is the right operand of a LEFT JOIN, then the outer
3496 ** query cannot be an aggregate. (3c) This is an artifact of the way
3497 ** aggregates are processed - there is no mechanism to determine if
3498 ** the LEFT JOIN table should be all-NULL.
3500 ** See also tickets #306, #350, and #3300.
3502 if( (pSubitem->fg.jointype & JT_OUTER)!=0 ){
3503 isLeftJoin = 1;
3504 if( pSubSrc->nSrc>1 || isAgg || IsVirtual(pSubSrc->a[0].pTab) ){
3505 /* (3a) (3c) (3b) */
3506 return 0;
3509 #ifdef SQLITE_EXTRA_IFNULLROW
3510 else if( iFrom>0 && !isAgg ){
3511 /* Setting isLeftJoin to -1 causes OP_IfNullRow opcodes to be generated for
3512 ** every reference to any result column from subquery in a join, even
3513 ** though they are not necessary. This will stress-test the OP_IfNullRow
3514 ** opcode. */
3515 isLeftJoin = -1;
3517 #endif
3519 /* Restriction (17): If the sub-query is a compound SELECT, then it must
3520 ** use only the UNION ALL operator. And none of the simple select queries
3521 ** that make up the compound SELECT are allowed to be aggregate or distinct
3522 ** queries.
3524 if( pSub->pPrior ){
3525 if( isAgg || (p->selFlags & SF_Distinct)!=0 || pSrc->nSrc!=1 ){
3526 return 0; /* (17d1), (17d2), or (17d3) */
3528 for(pSub1=pSub; pSub1; pSub1=pSub1->pPrior){
3529 testcase( (pSub1->selFlags & (SF_Distinct|SF_Aggregate))==SF_Distinct );
3530 testcase( (pSub1->selFlags & (SF_Distinct|SF_Aggregate))==SF_Aggregate );
3531 assert( pSub->pSrc!=0 );
3532 assert( pSub->pEList->nExpr==pSub1->pEList->nExpr );
3533 if( (pSub1->selFlags & (SF_Distinct|SF_Aggregate))!=0 /* (17b) */
3534 || (pSub1->pPrior && pSub1->op!=TK_ALL) /* (17a) */
3535 || pSub1->pSrc->nSrc<1 /* (17c) */
3537 return 0;
3539 testcase( pSub1->pSrc->nSrc>1 );
3542 /* Restriction (18). */
3543 if( p->pOrderBy ){
3544 int ii;
3545 for(ii=0; ii<p->pOrderBy->nExpr; ii++){
3546 if( p->pOrderBy->a[ii].u.x.iOrderByCol==0 ) return 0;
3551 /* Ex-restriction (23):
3552 ** The only way that the recursive part of a CTE can contain a compound
3553 ** subquery is for the subquery to be one term of a join. But if the
3554 ** subquery is a join, then the flattening has already been stopped by
3555 ** restriction (17d3)
3557 assert( (p->selFlags & SF_Recursive)==0 || pSub->pPrior==0 );
3559 /* Ex-restriction (20):
3560 ** A compound subquery must be the only term in the FROM clause of the
3561 ** outer query by restriction (17d3). But if that term also has an
3562 ** ORDER BY clause, then the subquery will be implemented by co-routine
3563 ** and so the flattener will never be invoked. Hence, it is not possible
3564 ** for the subquery to be a compound and have an ORDER BY clause.
3566 assert( pSub->pPrior==0 || pSub->pOrderBy==0 );
3568 /***** If we reach this point, flattening is permitted. *****/
3569 SELECTTRACE(1,pParse,p,("flatten %s.%p from term %d\n",
3570 pSub->zSelName, pSub, iFrom));
3572 /* Authorize the subquery */
3573 pParse->zAuthContext = pSubitem->zName;
3574 TESTONLY(i =) sqlite3AuthCheck(pParse, SQLITE_SELECT, 0, 0, 0);
3575 testcase( i==SQLITE_DENY );
3576 pParse->zAuthContext = zSavedAuthContext;
3578 /* If the sub-query is a compound SELECT statement, then (by restrictions
3579 ** 17 and 18 above) it must be a UNION ALL and the parent query must
3580 ** be of the form:
3582 ** SELECT <expr-list> FROM (<sub-query>) <where-clause>
3584 ** followed by any ORDER BY, LIMIT and/or OFFSET clauses. This block
3585 ** creates N-1 copies of the parent query without any ORDER BY, LIMIT or
3586 ** OFFSET clauses and joins them to the left-hand-side of the original
3587 ** using UNION ALL operators. In this case N is the number of simple
3588 ** select statements in the compound sub-query.
3590 ** Example:
3592 ** SELECT a+1 FROM (
3593 ** SELECT x FROM tab
3594 ** UNION ALL
3595 ** SELECT y FROM tab
3596 ** UNION ALL
3597 ** SELECT abs(z*2) FROM tab2
3598 ** ) WHERE a!=5 ORDER BY 1
3600 ** Transformed into:
3602 ** SELECT x+1 FROM tab WHERE x+1!=5
3603 ** UNION ALL
3604 ** SELECT y+1 FROM tab WHERE y+1!=5
3605 ** UNION ALL
3606 ** SELECT abs(z*2)+1 FROM tab2 WHERE abs(z*2)+1!=5
3607 ** ORDER BY 1
3609 ** We call this the "compound-subquery flattening".
3611 for(pSub=pSub->pPrior; pSub; pSub=pSub->pPrior){
3612 Select *pNew;
3613 ExprList *pOrderBy = p->pOrderBy;
3614 Expr *pLimit = p->pLimit;
3615 Expr *pOffset = p->pOffset;
3616 Select *pPrior = p->pPrior;
3617 p->pOrderBy = 0;
3618 p->pSrc = 0;
3619 p->pPrior = 0;
3620 p->pLimit = 0;
3621 p->pOffset = 0;
3622 pNew = sqlite3SelectDup(db, p, 0);
3623 sqlite3SelectSetName(pNew, pSub->zSelName);
3624 p->pOffset = pOffset;
3625 p->pLimit = pLimit;
3626 p->pOrderBy = pOrderBy;
3627 p->pSrc = pSrc;
3628 p->op = TK_ALL;
3629 if( pNew==0 ){
3630 p->pPrior = pPrior;
3631 }else{
3632 pNew->pPrior = pPrior;
3633 if( pPrior ) pPrior->pNext = pNew;
3634 pNew->pNext = p;
3635 p->pPrior = pNew;
3636 SELECTTRACE(2,pParse,p,
3637 ("compound-subquery flattener creates %s.%p as peer\n",
3638 pNew->zSelName, pNew));
3640 if( db->mallocFailed ) return 1;
3643 /* Begin flattening the iFrom-th entry of the FROM clause
3644 ** in the outer query.
3646 pSub = pSub1 = pSubitem->pSelect;
3648 /* Delete the transient table structure associated with the
3649 ** subquery
3651 sqlite3DbFree(db, pSubitem->zDatabase);
3652 sqlite3DbFree(db, pSubitem->zName);
3653 sqlite3DbFree(db, pSubitem->zAlias);
3654 pSubitem->zDatabase = 0;
3655 pSubitem->zName = 0;
3656 pSubitem->zAlias = 0;
3657 pSubitem->pSelect = 0;
3659 /* Defer deleting the Table object associated with the
3660 ** subquery until code generation is
3661 ** complete, since there may still exist Expr.pTab entries that
3662 ** refer to the subquery even after flattening. Ticket #3346.
3664 ** pSubitem->pTab is always non-NULL by test restrictions and tests above.
3666 if( ALWAYS(pSubitem->pTab!=0) ){
3667 Table *pTabToDel = pSubitem->pTab;
3668 if( pTabToDel->nTabRef==1 ){
3669 Parse *pToplevel = sqlite3ParseToplevel(pParse);
3670 pTabToDel->pNextZombie = pToplevel->pZombieTab;
3671 pToplevel->pZombieTab = pTabToDel;
3672 }else{
3673 pTabToDel->nTabRef--;
3675 pSubitem->pTab = 0;
3678 /* The following loop runs once for each term in a compound-subquery
3679 ** flattening (as described above). If we are doing a different kind
3680 ** of flattening - a flattening other than a compound-subquery flattening -
3681 ** then this loop only runs once.
3683 ** This loop moves all of the FROM elements of the subquery into the
3684 ** the FROM clause of the outer query. Before doing this, remember
3685 ** the cursor number for the original outer query FROM element in
3686 ** iParent. The iParent cursor will never be used. Subsequent code
3687 ** will scan expressions looking for iParent references and replace
3688 ** those references with expressions that resolve to the subquery FROM
3689 ** elements we are now copying in.
3691 for(pParent=p; pParent; pParent=pParent->pPrior, pSub=pSub->pPrior){
3692 int nSubSrc;
3693 u8 jointype = 0;
3694 pSubSrc = pSub->pSrc; /* FROM clause of subquery */
3695 nSubSrc = pSubSrc->nSrc; /* Number of terms in subquery FROM clause */
3696 pSrc = pParent->pSrc; /* FROM clause of the outer query */
3698 if( pSrc ){
3699 assert( pParent==p ); /* First time through the loop */
3700 jointype = pSubitem->fg.jointype;
3701 }else{
3702 assert( pParent!=p ); /* 2nd and subsequent times through the loop */
3703 pSrc = pParent->pSrc = sqlite3SrcListAppend(db, 0, 0, 0);
3704 if( pSrc==0 ){
3705 assert( db->mallocFailed );
3706 break;
3710 /* The subquery uses a single slot of the FROM clause of the outer
3711 ** query. If the subquery has more than one element in its FROM clause,
3712 ** then expand the outer query to make space for it to hold all elements
3713 ** of the subquery.
3715 ** Example:
3717 ** SELECT * FROM tabA, (SELECT * FROM sub1, sub2), tabB;
3719 ** The outer query has 3 slots in its FROM clause. One slot of the
3720 ** outer query (the middle slot) is used by the subquery. The next
3721 ** block of code will expand the outer query FROM clause to 4 slots.
3722 ** The middle slot is expanded to two slots in order to make space
3723 ** for the two elements in the FROM clause of the subquery.
3725 if( nSubSrc>1 ){
3726 pParent->pSrc = pSrc = sqlite3SrcListEnlarge(db, pSrc, nSubSrc-1,iFrom+1);
3727 if( db->mallocFailed ){
3728 break;
3732 /* Transfer the FROM clause terms from the subquery into the
3733 ** outer query.
3735 for(i=0; i<nSubSrc; i++){
3736 sqlite3IdListDelete(db, pSrc->a[i+iFrom].pUsing);
3737 assert( pSrc->a[i+iFrom].fg.isTabFunc==0 );
3738 pSrc->a[i+iFrom] = pSubSrc->a[i];
3739 iNewParent = pSubSrc->a[i].iCursor;
3740 memset(&pSubSrc->a[i], 0, sizeof(pSubSrc->a[i]));
3742 pSrc->a[iFrom].fg.jointype = jointype;
3744 /* Now begin substituting subquery result set expressions for
3745 ** references to the iParent in the outer query.
3747 ** Example:
3749 ** SELECT a+5, b*10 FROM (SELECT x*3 AS a, y+10 AS b FROM t1) WHERE a>b;
3750 ** \ \_____________ subquery __________/ /
3751 ** \_____________________ outer query ______________________________/
3753 ** We look at every expression in the outer query and every place we see
3754 ** "a" we substitute "x*3" and every place we see "b" we substitute "y+10".
3756 if( pSub->pOrderBy ){
3757 /* At this point, any non-zero iOrderByCol values indicate that the
3758 ** ORDER BY column expression is identical to the iOrderByCol'th
3759 ** expression returned by SELECT statement pSub. Since these values
3760 ** do not necessarily correspond to columns in SELECT statement pParent,
3761 ** zero them before transfering the ORDER BY clause.
3763 ** Not doing this may cause an error if a subsequent call to this
3764 ** function attempts to flatten a compound sub-query into pParent
3765 ** (the only way this can happen is if the compound sub-query is
3766 ** currently part of pSub->pSrc). See ticket [d11a6e908f]. */
3767 ExprList *pOrderBy = pSub->pOrderBy;
3768 for(i=0; i<pOrderBy->nExpr; i++){
3769 pOrderBy->a[i].u.x.iOrderByCol = 0;
3771 assert( pParent->pOrderBy==0 );
3772 assert( pSub->pPrior==0 );
3773 pParent->pOrderBy = pOrderBy;
3774 pSub->pOrderBy = 0;
3776 pWhere = sqlite3ExprDup(db, pSub->pWhere, 0);
3777 if( isLeftJoin>0 ){
3778 setJoinExpr(pWhere, iNewParent);
3780 pParent->pWhere = sqlite3ExprAnd(db, pWhere, pParent->pWhere);
3781 if( db->mallocFailed==0 ){
3782 SubstContext x;
3783 x.pParse = pParse;
3784 x.iTable = iParent;
3785 x.iNewTable = iNewParent;
3786 x.isLeftJoin = isLeftJoin;
3787 x.pEList = pSub->pEList;
3788 substSelect(&x, pParent, 0);
3791 /* The flattened query is distinct if either the inner or the
3792 ** outer query is distinct.
3794 pParent->selFlags |= pSub->selFlags & SF_Distinct;
3797 ** SELECT ... FROM (SELECT ... LIMIT a OFFSET b) LIMIT x OFFSET y;
3799 ** One is tempted to try to add a and b to combine the limits. But this
3800 ** does not work if either limit is negative.
3802 if( pSub->pLimit ){
3803 pParent->pLimit = pSub->pLimit;
3804 pSub->pLimit = 0;
3808 /* Finially, delete what is left of the subquery and return
3809 ** success.
3811 sqlite3SelectDelete(db, pSub1);
3813 #if SELECTTRACE_ENABLED
3814 if( sqlite3SelectTrace & 0x100 ){
3815 SELECTTRACE(0x100,pParse,p,("After flattening:\n"));
3816 sqlite3TreeViewSelect(0, p, 0);
3818 #endif
3820 return 1;
3822 #endif /* !defined(SQLITE_OMIT_SUBQUERY) || !defined(SQLITE_OMIT_VIEW) */
3826 #if !defined(SQLITE_OMIT_SUBQUERY) || !defined(SQLITE_OMIT_VIEW)
3828 ** Make copies of relevant WHERE clause terms of the outer query into
3829 ** the WHERE clause of subquery. Example:
3831 ** SELECT * FROM (SELECT a AS x, c-d AS y FROM t1) WHERE x=5 AND y=10;
3833 ** Transformed into:
3835 ** SELECT * FROM (SELECT a AS x, c-d AS y FROM t1 WHERE a=5 AND c-d=10)
3836 ** WHERE x=5 AND y=10;
3838 ** The hope is that the terms added to the inner query will make it more
3839 ** efficient.
3841 ** Do not attempt this optimization if:
3843 ** (1) (** This restriction was removed on 2017-09-29. We used to
3844 ** disallow this optimization for aggregate subqueries, but now
3845 ** it is allowed by putting the extra terms on the HAVING clause.
3846 ** The added HAVING clause is pointless if the subquery lacks
3847 ** a GROUP BY clause. But such a HAVING clause is also harmless
3848 ** so there does not appear to be any reason to add extra logic
3849 ** to suppress it. **)
3851 ** (2) The inner query is the recursive part of a common table expression.
3853 ** (3) The inner query has a LIMIT clause (since the changes to the WHERE
3854 ** close would change the meaning of the LIMIT).
3856 ** (4) The inner query is the right operand of a LEFT JOIN. (The caller
3857 ** enforces this restriction since this routine does not have enough
3858 ** information to know.)
3860 ** (5) The WHERE clause expression originates in the ON or USING clause
3861 ** of a LEFT JOIN.
3863 ** Return 0 if no changes are made and non-zero if one or more WHERE clause
3864 ** terms are duplicated into the subquery.
3866 static int pushDownWhereTerms(
3867 Parse *pParse, /* Parse context (for malloc() and error reporting) */
3868 Select *pSubq, /* The subquery whose WHERE clause is to be augmented */
3869 Expr *pWhere, /* The WHERE clause of the outer query */
3870 int iCursor /* Cursor number of the subquery */
3872 Expr *pNew;
3873 int nChng = 0;
3874 if( pWhere==0 ) return 0;
3875 if( pSubq->selFlags & SF_Recursive ) return 0; /* restriction (2) */
3877 #ifdef SQLITE_DEBUG
3878 /* Only the first term of a compound can have a WITH clause. But make
3879 ** sure no other terms are marked SF_Recursive in case something changes
3880 ** in the future.
3883 Select *pX;
3884 for(pX=pSubq; pX; pX=pX->pPrior){
3885 assert( (pX->selFlags & (SF_Recursive))==0 );
3888 #endif
3890 if( pSubq->pLimit!=0 ){
3891 return 0; /* restriction (3) */
3893 while( pWhere->op==TK_AND ){
3894 nChng += pushDownWhereTerms(pParse, pSubq, pWhere->pRight, iCursor);
3895 pWhere = pWhere->pLeft;
3897 if( ExprHasProperty(pWhere,EP_FromJoin) ) return 0; /* restriction (5) */
3898 if( sqlite3ExprIsTableConstant(pWhere, iCursor) ){
3899 nChng++;
3900 while( pSubq ){
3901 SubstContext x;
3902 pNew = sqlite3ExprDup(pParse->db, pWhere, 0);
3903 x.pParse = pParse;
3904 x.iTable = iCursor;
3905 x.iNewTable = iCursor;
3906 x.isLeftJoin = 0;
3907 x.pEList = pSubq->pEList;
3908 pNew = substExpr(&x, pNew);
3909 if( pSubq->selFlags & SF_Aggregate ){
3910 pSubq->pHaving = sqlite3ExprAnd(pParse->db, pSubq->pHaving, pNew);
3911 }else{
3912 pSubq->pWhere = sqlite3ExprAnd(pParse->db, pSubq->pWhere, pNew);
3914 pSubq = pSubq->pPrior;
3917 return nChng;
3919 #endif /* !defined(SQLITE_OMIT_SUBQUERY) || !defined(SQLITE_OMIT_VIEW) */
3922 ** Based on the contents of the AggInfo structure indicated by the first
3923 ** argument, this function checks if the following are true:
3925 ** * the query contains just a single aggregate function,
3926 ** * the aggregate function is either min() or max(), and
3927 ** * the argument to the aggregate function is a column value.
3929 ** If all of the above are true, then WHERE_ORDERBY_MIN or WHERE_ORDERBY_MAX
3930 ** is returned as appropriate. Also, *ppMinMax is set to point to the
3931 ** list of arguments passed to the aggregate before returning.
3933 ** Or, if the conditions above are not met, *ppMinMax is set to 0 and
3934 ** WHERE_ORDERBY_NORMAL is returned.
3936 static u8 minMaxQuery(AggInfo *pAggInfo, ExprList **ppMinMax){
3937 int eRet = WHERE_ORDERBY_NORMAL; /* Return value */
3939 *ppMinMax = 0;
3940 if( pAggInfo->nFunc==1 ){
3941 Expr *pExpr = pAggInfo->aFunc[0].pExpr; /* Aggregate function */
3942 ExprList *pEList = pExpr->x.pList; /* Arguments to agg function */
3944 assert( pExpr->op==TK_AGG_FUNCTION );
3945 if( pEList && pEList->nExpr==1 && pEList->a[0].pExpr->op==TK_AGG_COLUMN ){
3946 const char *zFunc = pExpr->u.zToken;
3947 if( sqlite3StrICmp(zFunc, "min")==0 ){
3948 eRet = WHERE_ORDERBY_MIN;
3949 *ppMinMax = pEList;
3950 }else if( sqlite3StrICmp(zFunc, "max")==0 ){
3951 eRet = WHERE_ORDERBY_MAX;
3952 *ppMinMax = pEList;
3957 assert( *ppMinMax==0 || (*ppMinMax)->nExpr==1 );
3958 return eRet;
3962 ** The select statement passed as the first argument is an aggregate query.
3963 ** The second argument is the associated aggregate-info object. This
3964 ** function tests if the SELECT is of the form:
3966 ** SELECT count(*) FROM <tbl>
3968 ** where table is a database table, not a sub-select or view. If the query
3969 ** does match this pattern, then a pointer to the Table object representing
3970 ** <tbl> is returned. Otherwise, 0 is returned.
3972 static Table *isSimpleCount(Select *p, AggInfo *pAggInfo){
3973 Table *pTab;
3974 Expr *pExpr;
3976 assert( !p->pGroupBy );
3978 if( p->pWhere || p->pEList->nExpr!=1
3979 || p->pSrc->nSrc!=1 || p->pSrc->a[0].pSelect
3981 return 0;
3983 pTab = p->pSrc->a[0].pTab;
3984 pExpr = p->pEList->a[0].pExpr;
3985 assert( pTab && !pTab->pSelect && pExpr );
3987 if( IsVirtual(pTab) ) return 0;
3988 if( pExpr->op!=TK_AGG_FUNCTION ) return 0;
3989 if( NEVER(pAggInfo->nFunc==0) ) return 0;
3990 if( (pAggInfo->aFunc[0].pFunc->funcFlags&SQLITE_FUNC_COUNT)==0 ) return 0;
3991 if( pExpr->flags&EP_Distinct ) return 0;
3993 return pTab;
3997 ** If the source-list item passed as an argument was augmented with an
3998 ** INDEXED BY clause, then try to locate the specified index. If there
3999 ** was such a clause and the named index cannot be found, return
4000 ** SQLITE_ERROR and leave an error in pParse. Otherwise, populate
4001 ** pFrom->pIndex and return SQLITE_OK.
4003 int sqlite3IndexedByLookup(Parse *pParse, struct SrcList_item *pFrom){
4004 if( pFrom->pTab && pFrom->fg.isIndexedBy ){
4005 Table *pTab = pFrom->pTab;
4006 char *zIndexedBy = pFrom->u1.zIndexedBy;
4007 Index *pIdx;
4008 for(pIdx=pTab->pIndex;
4009 pIdx && sqlite3StrICmp(pIdx->zName, zIndexedBy);
4010 pIdx=pIdx->pNext
4012 if( !pIdx ){
4013 sqlite3ErrorMsg(pParse, "no such index: %s", zIndexedBy, 0);
4014 pParse->checkSchema = 1;
4015 return SQLITE_ERROR;
4017 pFrom->pIBIndex = pIdx;
4019 return SQLITE_OK;
4022 ** Detect compound SELECT statements that use an ORDER BY clause with
4023 ** an alternative collating sequence.
4025 ** SELECT ... FROM t1 EXCEPT SELECT ... FROM t2 ORDER BY .. COLLATE ...
4027 ** These are rewritten as a subquery:
4029 ** SELECT * FROM (SELECT ... FROM t1 EXCEPT SELECT ... FROM t2)
4030 ** ORDER BY ... COLLATE ...
4032 ** This transformation is necessary because the multiSelectOrderBy() routine
4033 ** above that generates the code for a compound SELECT with an ORDER BY clause
4034 ** uses a merge algorithm that requires the same collating sequence on the
4035 ** result columns as on the ORDER BY clause. See ticket
4036 ** http://www.sqlite.org/src/info/6709574d2a
4038 ** This transformation is only needed for EXCEPT, INTERSECT, and UNION.
4039 ** The UNION ALL operator works fine with multiSelectOrderBy() even when
4040 ** there are COLLATE terms in the ORDER BY.
4042 static int convertCompoundSelectToSubquery(Walker *pWalker, Select *p){
4043 int i;
4044 Select *pNew;
4045 Select *pX;
4046 sqlite3 *db;
4047 struct ExprList_item *a;
4048 SrcList *pNewSrc;
4049 Parse *pParse;
4050 Token dummy;
4052 if( p->pPrior==0 ) return WRC_Continue;
4053 if( p->pOrderBy==0 ) return WRC_Continue;
4054 for(pX=p; pX && (pX->op==TK_ALL || pX->op==TK_SELECT); pX=pX->pPrior){}
4055 if( pX==0 ) return WRC_Continue;
4056 a = p->pOrderBy->a;
4057 for(i=p->pOrderBy->nExpr-1; i>=0; i--){
4058 if( a[i].pExpr->flags & EP_Collate ) break;
4060 if( i<0 ) return WRC_Continue;
4062 /* If we reach this point, that means the transformation is required. */
4064 pParse = pWalker->pParse;
4065 db = pParse->db;
4066 pNew = sqlite3DbMallocZero(db, sizeof(*pNew) );
4067 if( pNew==0 ) return WRC_Abort;
4068 memset(&dummy, 0, sizeof(dummy));
4069 pNewSrc = sqlite3SrcListAppendFromTerm(pParse,0,0,0,&dummy,pNew,0,0);
4070 if( pNewSrc==0 ) return WRC_Abort;
4071 *pNew = *p;
4072 p->pSrc = pNewSrc;
4073 p->pEList = sqlite3ExprListAppend(pParse, 0, sqlite3Expr(db, TK_ASTERISK, 0));
4074 p->op = TK_SELECT;
4075 p->pWhere = 0;
4076 pNew->pGroupBy = 0;
4077 pNew->pHaving = 0;
4078 pNew->pOrderBy = 0;
4079 p->pPrior = 0;
4080 p->pNext = 0;
4081 p->pWith = 0;
4082 p->selFlags &= ~SF_Compound;
4083 assert( (p->selFlags & SF_Converted)==0 );
4084 p->selFlags |= SF_Converted;
4085 assert( pNew->pPrior!=0 );
4086 pNew->pPrior->pNext = pNew;
4087 pNew->pLimit = 0;
4088 pNew->pOffset = 0;
4089 return WRC_Continue;
4093 ** Check to see if the FROM clause term pFrom has table-valued function
4094 ** arguments. If it does, leave an error message in pParse and return
4095 ** non-zero, since pFrom is not allowed to be a table-valued function.
4097 static int cannotBeFunction(Parse *pParse, struct SrcList_item *pFrom){
4098 if( pFrom->fg.isTabFunc ){
4099 sqlite3ErrorMsg(pParse, "'%s' is not a function", pFrom->zName);
4100 return 1;
4102 return 0;
4105 #ifndef SQLITE_OMIT_CTE
4107 ** Argument pWith (which may be NULL) points to a linked list of nested
4108 ** WITH contexts, from inner to outermost. If the table identified by
4109 ** FROM clause element pItem is really a common-table-expression (CTE)
4110 ** then return a pointer to the CTE definition for that table. Otherwise
4111 ** return NULL.
4113 ** If a non-NULL value is returned, set *ppContext to point to the With
4114 ** object that the returned CTE belongs to.
4116 static struct Cte *searchWith(
4117 With *pWith, /* Current innermost WITH clause */
4118 struct SrcList_item *pItem, /* FROM clause element to resolve */
4119 With **ppContext /* OUT: WITH clause return value belongs to */
4121 const char *zName;
4122 if( pItem->zDatabase==0 && (zName = pItem->zName)!=0 ){
4123 With *p;
4124 for(p=pWith; p; p=p->pOuter){
4125 int i;
4126 for(i=0; i<p->nCte; i++){
4127 if( sqlite3StrICmp(zName, p->a[i].zName)==0 ){
4128 *ppContext = p;
4129 return &p->a[i];
4134 return 0;
4137 /* The code generator maintains a stack of active WITH clauses
4138 ** with the inner-most WITH clause being at the top of the stack.
4140 ** This routine pushes the WITH clause passed as the second argument
4141 ** onto the top of the stack. If argument bFree is true, then this
4142 ** WITH clause will never be popped from the stack. In this case it
4143 ** should be freed along with the Parse object. In other cases, when
4144 ** bFree==0, the With object will be freed along with the SELECT
4145 ** statement with which it is associated.
4147 void sqlite3WithPush(Parse *pParse, With *pWith, u8 bFree){
4148 assert( bFree==0 || (pParse->pWith==0 && pParse->pWithToFree==0) );
4149 if( pWith ){
4150 assert( pParse->pWith!=pWith );
4151 pWith->pOuter = pParse->pWith;
4152 pParse->pWith = pWith;
4153 if( bFree ) pParse->pWithToFree = pWith;
4158 ** This function checks if argument pFrom refers to a CTE declared by
4159 ** a WITH clause on the stack currently maintained by the parser. And,
4160 ** if currently processing a CTE expression, if it is a recursive
4161 ** reference to the current CTE.
4163 ** If pFrom falls into either of the two categories above, pFrom->pTab
4164 ** and other fields are populated accordingly. The caller should check
4165 ** (pFrom->pTab!=0) to determine whether or not a successful match
4166 ** was found.
4168 ** Whether or not a match is found, SQLITE_OK is returned if no error
4169 ** occurs. If an error does occur, an error message is stored in the
4170 ** parser and some error code other than SQLITE_OK returned.
4172 static int withExpand(
4173 Walker *pWalker,
4174 struct SrcList_item *pFrom
4176 Parse *pParse = pWalker->pParse;
4177 sqlite3 *db = pParse->db;
4178 struct Cte *pCte; /* Matched CTE (or NULL if no match) */
4179 With *pWith; /* WITH clause that pCte belongs to */
4181 assert( pFrom->pTab==0 );
4183 pCte = searchWith(pParse->pWith, pFrom, &pWith);
4184 if( pCte ){
4185 Table *pTab;
4186 ExprList *pEList;
4187 Select *pSel;
4188 Select *pLeft; /* Left-most SELECT statement */
4189 int bMayRecursive; /* True if compound joined by UNION [ALL] */
4190 With *pSavedWith; /* Initial value of pParse->pWith */
4192 /* If pCte->zCteErr is non-NULL at this point, then this is an illegal
4193 ** recursive reference to CTE pCte. Leave an error in pParse and return
4194 ** early. If pCte->zCteErr is NULL, then this is not a recursive reference.
4195 ** In this case, proceed. */
4196 if( pCte->zCteErr ){
4197 sqlite3ErrorMsg(pParse, pCte->zCteErr, pCte->zName);
4198 return SQLITE_ERROR;
4200 if( cannotBeFunction(pParse, pFrom) ) return SQLITE_ERROR;
4202 assert( pFrom->pTab==0 );
4203 pFrom->pTab = pTab = sqlite3DbMallocZero(db, sizeof(Table));
4204 if( pTab==0 ) return WRC_Abort;
4205 pTab->nTabRef = 1;
4206 pTab->zName = sqlite3DbStrDup(db, pCte->zName);
4207 pTab->iPKey = -1;
4208 pTab->nRowLogEst = 200; assert( 200==sqlite3LogEst(1048576) );
4209 pTab->tabFlags |= TF_Ephemeral | TF_NoVisibleRowid;
4210 pFrom->pSelect = sqlite3SelectDup(db, pCte->pSelect, 0);
4211 if( db->mallocFailed ) return SQLITE_NOMEM_BKPT;
4212 assert( pFrom->pSelect );
4214 /* Check if this is a recursive CTE. */
4215 pSel = pFrom->pSelect;
4216 bMayRecursive = ( pSel->op==TK_ALL || pSel->op==TK_UNION );
4217 if( bMayRecursive ){
4218 int i;
4219 SrcList *pSrc = pFrom->pSelect->pSrc;
4220 for(i=0; i<pSrc->nSrc; i++){
4221 struct SrcList_item *pItem = &pSrc->a[i];
4222 if( pItem->zDatabase==0
4223 && pItem->zName!=0
4224 && 0==sqlite3StrICmp(pItem->zName, pCte->zName)
4226 pItem->pTab = pTab;
4227 pItem->fg.isRecursive = 1;
4228 pTab->nTabRef++;
4229 pSel->selFlags |= SF_Recursive;
4234 /* Only one recursive reference is permitted. */
4235 if( pTab->nTabRef>2 ){
4236 sqlite3ErrorMsg(
4237 pParse, "multiple references to recursive table: %s", pCte->zName
4239 return SQLITE_ERROR;
4241 assert( pTab->nTabRef==1 ||
4242 ((pSel->selFlags&SF_Recursive) && pTab->nTabRef==2 ));
4244 pCte->zCteErr = "circular reference: %s";
4245 pSavedWith = pParse->pWith;
4246 pParse->pWith = pWith;
4247 if( bMayRecursive ){
4248 Select *pPrior = pSel->pPrior;
4249 assert( pPrior->pWith==0 );
4250 pPrior->pWith = pSel->pWith;
4251 sqlite3WalkSelect(pWalker, pPrior);
4252 pPrior->pWith = 0;
4253 }else{
4254 sqlite3WalkSelect(pWalker, pSel);
4256 pParse->pWith = pWith;
4258 for(pLeft=pSel; pLeft->pPrior; pLeft=pLeft->pPrior);
4259 pEList = pLeft->pEList;
4260 if( pCte->pCols ){
4261 if( pEList && pEList->nExpr!=pCte->pCols->nExpr ){
4262 sqlite3ErrorMsg(pParse, "table %s has %d values for %d columns",
4263 pCte->zName, pEList->nExpr, pCte->pCols->nExpr
4265 pParse->pWith = pSavedWith;
4266 return SQLITE_ERROR;
4268 pEList = pCte->pCols;
4271 sqlite3ColumnsFromExprList(pParse, pEList, &pTab->nCol, &pTab->aCol);
4272 if( bMayRecursive ){
4273 if( pSel->selFlags & SF_Recursive ){
4274 pCte->zCteErr = "multiple recursive references: %s";
4275 }else{
4276 pCte->zCteErr = "recursive reference in a subquery: %s";
4278 sqlite3WalkSelect(pWalker, pSel);
4280 pCte->zCteErr = 0;
4281 pParse->pWith = pSavedWith;
4284 return SQLITE_OK;
4286 #endif
4288 #ifndef SQLITE_OMIT_CTE
4290 ** If the SELECT passed as the second argument has an associated WITH
4291 ** clause, pop it from the stack stored as part of the Parse object.
4293 ** This function is used as the xSelectCallback2() callback by
4294 ** sqlite3SelectExpand() when walking a SELECT tree to resolve table
4295 ** names and other FROM clause elements.
4297 static void selectPopWith(Walker *pWalker, Select *p){
4298 Parse *pParse = pWalker->pParse;
4299 if( OK_IF_ALWAYS_TRUE(pParse->pWith) && p->pPrior==0 ){
4300 With *pWith = findRightmost(p)->pWith;
4301 if( pWith!=0 ){
4302 assert( pParse->pWith==pWith );
4303 pParse->pWith = pWith->pOuter;
4307 #else
4308 #define selectPopWith 0
4309 #endif
4312 ** This routine is a Walker callback for "expanding" a SELECT statement.
4313 ** "Expanding" means to do the following:
4315 ** (1) Make sure VDBE cursor numbers have been assigned to every
4316 ** element of the FROM clause.
4318 ** (2) Fill in the pTabList->a[].pTab fields in the SrcList that
4319 ** defines FROM clause. When views appear in the FROM clause,
4320 ** fill pTabList->a[].pSelect with a copy of the SELECT statement
4321 ** that implements the view. A copy is made of the view's SELECT
4322 ** statement so that we can freely modify or delete that statement
4323 ** without worrying about messing up the persistent representation
4324 ** of the view.
4326 ** (3) Add terms to the WHERE clause to accommodate the NATURAL keyword
4327 ** on joins and the ON and USING clause of joins.
4329 ** (4) Scan the list of columns in the result set (pEList) looking
4330 ** for instances of the "*" operator or the TABLE.* operator.
4331 ** If found, expand each "*" to be every column in every table
4332 ** and TABLE.* to be every column in TABLE.
4335 static int selectExpander(Walker *pWalker, Select *p){
4336 Parse *pParse = pWalker->pParse;
4337 int i, j, k;
4338 SrcList *pTabList;
4339 ExprList *pEList;
4340 struct SrcList_item *pFrom;
4341 sqlite3 *db = pParse->db;
4342 Expr *pE, *pRight, *pExpr;
4343 u16 selFlags = p->selFlags;
4345 p->selFlags |= SF_Expanded;
4346 if( db->mallocFailed ){
4347 return WRC_Abort;
4349 if( NEVER(p->pSrc==0) || (selFlags & SF_Expanded)!=0 ){
4350 return WRC_Prune;
4352 pTabList = p->pSrc;
4353 pEList = p->pEList;
4354 if( OK_IF_ALWAYS_TRUE(p->pWith) ){
4355 sqlite3WithPush(pParse, p->pWith, 0);
4358 /* Make sure cursor numbers have been assigned to all entries in
4359 ** the FROM clause of the SELECT statement.
4361 sqlite3SrcListAssignCursors(pParse, pTabList);
4363 /* Look up every table named in the FROM clause of the select. If
4364 ** an entry of the FROM clause is a subquery instead of a table or view,
4365 ** then create a transient table structure to describe the subquery.
4367 for(i=0, pFrom=pTabList->a; i<pTabList->nSrc; i++, pFrom++){
4368 Table *pTab;
4369 assert( pFrom->fg.isRecursive==0 || pFrom->pTab!=0 );
4370 if( pFrom->fg.isRecursive ) continue;
4371 assert( pFrom->pTab==0 );
4372 #ifndef SQLITE_OMIT_CTE
4373 if( withExpand(pWalker, pFrom) ) return WRC_Abort;
4374 if( pFrom->pTab ) {} else
4375 #endif
4376 if( pFrom->zName==0 ){
4377 #ifndef SQLITE_OMIT_SUBQUERY
4378 Select *pSel = pFrom->pSelect;
4379 /* A sub-query in the FROM clause of a SELECT */
4380 assert( pSel!=0 );
4381 assert( pFrom->pTab==0 );
4382 if( sqlite3WalkSelect(pWalker, pSel) ) return WRC_Abort;
4383 pFrom->pTab = pTab = sqlite3DbMallocZero(db, sizeof(Table));
4384 if( pTab==0 ) return WRC_Abort;
4385 pTab->nTabRef = 1;
4386 if( pFrom->zAlias ){
4387 pTab->zName = sqlite3DbStrDup(db, pFrom->zAlias);
4388 }else{
4389 pTab->zName = sqlite3MPrintf(db, "subquery_%p", (void*)pTab);
4391 while( pSel->pPrior ){ pSel = pSel->pPrior; }
4392 sqlite3ColumnsFromExprList(pParse, pSel->pEList,&pTab->nCol,&pTab->aCol);
4393 pTab->iPKey = -1;
4394 pTab->nRowLogEst = 200; assert( 200==sqlite3LogEst(1048576) );
4395 pTab->tabFlags |= TF_Ephemeral;
4396 #endif
4397 }else{
4398 /* An ordinary table or view name in the FROM clause */
4399 assert( pFrom->pTab==0 );
4400 pFrom->pTab = pTab = sqlite3LocateTableItem(pParse, 0, pFrom);
4401 if( pTab==0 ) return WRC_Abort;
4402 if( pTab->nTabRef>=0xffff ){
4403 sqlite3ErrorMsg(pParse, "too many references to \"%s\": max 65535",
4404 pTab->zName);
4405 pFrom->pTab = 0;
4406 return WRC_Abort;
4408 pTab->nTabRef++;
4409 if( !IsVirtual(pTab) && cannotBeFunction(pParse, pFrom) ){
4410 return WRC_Abort;
4412 #if !defined(SQLITE_OMIT_VIEW) || !defined (SQLITE_OMIT_VIRTUALTABLE)
4413 if( IsVirtual(pTab) || pTab->pSelect ){
4414 i16 nCol;
4415 if( sqlite3ViewGetColumnNames(pParse, pTab) ) return WRC_Abort;
4416 assert( pFrom->pSelect==0 );
4417 pFrom->pSelect = sqlite3SelectDup(db, pTab->pSelect, 0);
4418 sqlite3SelectSetName(pFrom->pSelect, pTab->zName);
4419 nCol = pTab->nCol;
4420 pTab->nCol = -1;
4421 sqlite3WalkSelect(pWalker, pFrom->pSelect);
4422 pTab->nCol = nCol;
4424 #endif
4427 /* Locate the index named by the INDEXED BY clause, if any. */
4428 if( sqlite3IndexedByLookup(pParse, pFrom) ){
4429 return WRC_Abort;
4433 /* Process NATURAL keywords, and ON and USING clauses of joins.
4435 if( db->mallocFailed || sqliteProcessJoin(pParse, p) ){
4436 return WRC_Abort;
4439 /* For every "*" that occurs in the column list, insert the names of
4440 ** all columns in all tables. And for every TABLE.* insert the names
4441 ** of all columns in TABLE. The parser inserted a special expression
4442 ** with the TK_ASTERISK operator for each "*" that it found in the column
4443 ** list. The following code just has to locate the TK_ASTERISK
4444 ** expressions and expand each one to the list of all columns in
4445 ** all tables.
4447 ** The first loop just checks to see if there are any "*" operators
4448 ** that need expanding.
4450 for(k=0; k<pEList->nExpr; k++){
4451 pE = pEList->a[k].pExpr;
4452 if( pE->op==TK_ASTERISK ) break;
4453 assert( pE->op!=TK_DOT || pE->pRight!=0 );
4454 assert( pE->op!=TK_DOT || (pE->pLeft!=0 && pE->pLeft->op==TK_ID) );
4455 if( pE->op==TK_DOT && pE->pRight->op==TK_ASTERISK ) break;
4457 if( k<pEList->nExpr ){
4459 ** If we get here it means the result set contains one or more "*"
4460 ** operators that need to be expanded. Loop through each expression
4461 ** in the result set and expand them one by one.
4463 struct ExprList_item *a = pEList->a;
4464 ExprList *pNew = 0;
4465 int flags = pParse->db->flags;
4466 int longNames = (flags & SQLITE_FullColNames)!=0
4467 && (flags & SQLITE_ShortColNames)==0;
4469 for(k=0; k<pEList->nExpr; k++){
4470 pE = a[k].pExpr;
4471 pRight = pE->pRight;
4472 assert( pE->op!=TK_DOT || pRight!=0 );
4473 if( pE->op!=TK_ASTERISK
4474 && (pE->op!=TK_DOT || pRight->op!=TK_ASTERISK)
4476 /* This particular expression does not need to be expanded.
4478 pNew = sqlite3ExprListAppend(pParse, pNew, a[k].pExpr);
4479 if( pNew ){
4480 pNew->a[pNew->nExpr-1].zName = a[k].zName;
4481 pNew->a[pNew->nExpr-1].zSpan = a[k].zSpan;
4482 a[k].zName = 0;
4483 a[k].zSpan = 0;
4485 a[k].pExpr = 0;
4486 }else{
4487 /* This expression is a "*" or a "TABLE.*" and needs to be
4488 ** expanded. */
4489 int tableSeen = 0; /* Set to 1 when TABLE matches */
4490 char *zTName = 0; /* text of name of TABLE */
4491 if( pE->op==TK_DOT ){
4492 assert( pE->pLeft!=0 );
4493 assert( !ExprHasProperty(pE->pLeft, EP_IntValue) );
4494 zTName = pE->pLeft->u.zToken;
4496 for(i=0, pFrom=pTabList->a; i<pTabList->nSrc; i++, pFrom++){
4497 Table *pTab = pFrom->pTab;
4498 Select *pSub = pFrom->pSelect;
4499 char *zTabName = pFrom->zAlias;
4500 const char *zSchemaName = 0;
4501 int iDb;
4502 if( zTabName==0 ){
4503 zTabName = pTab->zName;
4505 if( db->mallocFailed ) break;
4506 if( pSub==0 || (pSub->selFlags & SF_NestedFrom)==0 ){
4507 pSub = 0;
4508 if( zTName && sqlite3StrICmp(zTName, zTabName)!=0 ){
4509 continue;
4511 iDb = sqlite3SchemaToIndex(db, pTab->pSchema);
4512 zSchemaName = iDb>=0 ? db->aDb[iDb].zDbSName : "*";
4514 for(j=0; j<pTab->nCol; j++){
4515 char *zName = pTab->aCol[j].zName;
4516 char *zColname; /* The computed column name */
4517 char *zToFree; /* Malloced string that needs to be freed */
4518 Token sColname; /* Computed column name as a token */
4520 assert( zName );
4521 if( zTName && pSub
4522 && sqlite3MatchSpanName(pSub->pEList->a[j].zSpan, 0, zTName, 0)==0
4524 continue;
4527 /* If a column is marked as 'hidden', omit it from the expanded
4528 ** result-set list unless the SELECT has the SF_IncludeHidden
4529 ** bit set.
4531 if( (p->selFlags & SF_IncludeHidden)==0
4532 && IsHiddenColumn(&pTab->aCol[j])
4534 continue;
4536 tableSeen = 1;
4538 if( i>0 && zTName==0 ){
4539 if( (pFrom->fg.jointype & JT_NATURAL)!=0
4540 && tableAndColumnIndex(pTabList, i, zName, 0, 0)
4542 /* In a NATURAL join, omit the join columns from the
4543 ** table to the right of the join */
4544 continue;
4546 if( sqlite3IdListIndex(pFrom->pUsing, zName)>=0 ){
4547 /* In a join with a USING clause, omit columns in the
4548 ** using clause from the table on the right. */
4549 continue;
4552 pRight = sqlite3Expr(db, TK_ID, zName);
4553 zColname = zName;
4554 zToFree = 0;
4555 if( longNames || pTabList->nSrc>1 ){
4556 Expr *pLeft;
4557 pLeft = sqlite3Expr(db, TK_ID, zTabName);
4558 pExpr = sqlite3PExpr(pParse, TK_DOT, pLeft, pRight);
4559 if( zSchemaName ){
4560 pLeft = sqlite3Expr(db, TK_ID, zSchemaName);
4561 pExpr = sqlite3PExpr(pParse, TK_DOT, pLeft, pExpr);
4563 if( longNames ){
4564 zColname = sqlite3MPrintf(db, "%s.%s", zTabName, zName);
4565 zToFree = zColname;
4567 }else{
4568 pExpr = pRight;
4570 pNew = sqlite3ExprListAppend(pParse, pNew, pExpr);
4571 sqlite3TokenInit(&sColname, zColname);
4572 sqlite3ExprListSetName(pParse, pNew, &sColname, 0);
4573 if( pNew && (p->selFlags & SF_NestedFrom)!=0 ){
4574 struct ExprList_item *pX = &pNew->a[pNew->nExpr-1];
4575 if( pSub ){
4576 pX->zSpan = sqlite3DbStrDup(db, pSub->pEList->a[j].zSpan);
4577 testcase( pX->zSpan==0 );
4578 }else{
4579 pX->zSpan = sqlite3MPrintf(db, "%s.%s.%s",
4580 zSchemaName, zTabName, zColname);
4581 testcase( pX->zSpan==0 );
4583 pX->bSpanIsTab = 1;
4585 sqlite3DbFree(db, zToFree);
4588 if( !tableSeen ){
4589 if( zTName ){
4590 sqlite3ErrorMsg(pParse, "no such table: %s", zTName);
4591 }else{
4592 sqlite3ErrorMsg(pParse, "no tables specified");
4597 sqlite3ExprListDelete(db, pEList);
4598 p->pEList = pNew;
4600 if( p->pEList && p->pEList->nExpr>db->aLimit[SQLITE_LIMIT_COLUMN] ){
4601 sqlite3ErrorMsg(pParse, "too many columns in result set");
4602 return WRC_Abort;
4604 return WRC_Continue;
4608 ** No-op routine for the parse-tree walker.
4610 ** When this routine is the Walker.xExprCallback then expression trees
4611 ** are walked without any actions being taken at each node. Presumably,
4612 ** when this routine is used for Walker.xExprCallback then
4613 ** Walker.xSelectCallback is set to do something useful for every
4614 ** subquery in the parser tree.
4616 int sqlite3ExprWalkNoop(Walker *NotUsed, Expr *NotUsed2){
4617 UNUSED_PARAMETER2(NotUsed, NotUsed2);
4618 return WRC_Continue;
4622 ** No-op routine for the parse-tree walker for SELECT statements.
4623 ** subquery in the parser tree.
4625 int sqlite3SelectWalkNoop(Walker *NotUsed, Select *NotUsed2){
4626 UNUSED_PARAMETER2(NotUsed, NotUsed2);
4627 return WRC_Continue;
4630 #if SQLITE_DEBUG
4632 ** Always assert. This xSelectCallback2 implementation proves that the
4633 ** xSelectCallback2 is never invoked.
4635 void sqlite3SelectWalkAssert2(Walker *NotUsed, Select *NotUsed2){
4636 UNUSED_PARAMETER2(NotUsed, NotUsed2);
4637 assert( 0 );
4639 #endif
4641 ** This routine "expands" a SELECT statement and all of its subqueries.
4642 ** For additional information on what it means to "expand" a SELECT
4643 ** statement, see the comment on the selectExpand worker callback above.
4645 ** Expanding a SELECT statement is the first step in processing a
4646 ** SELECT statement. The SELECT statement must be expanded before
4647 ** name resolution is performed.
4649 ** If anything goes wrong, an error message is written into pParse.
4650 ** The calling function can detect the problem by looking at pParse->nErr
4651 ** and/or pParse->db->mallocFailed.
4653 static void sqlite3SelectExpand(Parse *pParse, Select *pSelect){
4654 Walker w;
4655 w.xExprCallback = sqlite3ExprWalkNoop;
4656 w.pParse = pParse;
4657 if( OK_IF_ALWAYS_TRUE(pParse->hasCompound) ){
4658 w.xSelectCallback = convertCompoundSelectToSubquery;
4659 w.xSelectCallback2 = 0;
4660 sqlite3WalkSelect(&w, pSelect);
4662 w.xSelectCallback = selectExpander;
4663 w.xSelectCallback2 = selectPopWith;
4664 sqlite3WalkSelect(&w, pSelect);
4668 #ifndef SQLITE_OMIT_SUBQUERY
4670 ** This is a Walker.xSelectCallback callback for the sqlite3SelectTypeInfo()
4671 ** interface.
4673 ** For each FROM-clause subquery, add Column.zType and Column.zColl
4674 ** information to the Table structure that represents the result set
4675 ** of that subquery.
4677 ** The Table structure that represents the result set was constructed
4678 ** by selectExpander() but the type and collation information was omitted
4679 ** at that point because identifiers had not yet been resolved. This
4680 ** routine is called after identifier resolution.
4682 static void selectAddSubqueryTypeInfo(Walker *pWalker, Select *p){
4683 Parse *pParse;
4684 int i;
4685 SrcList *pTabList;
4686 struct SrcList_item *pFrom;
4688 assert( p->selFlags & SF_Resolved );
4689 assert( (p->selFlags & SF_HasTypeInfo)==0 );
4690 p->selFlags |= SF_HasTypeInfo;
4691 pParse = pWalker->pParse;
4692 pTabList = p->pSrc;
4693 for(i=0, pFrom=pTabList->a; i<pTabList->nSrc; i++, pFrom++){
4694 Table *pTab = pFrom->pTab;
4695 assert( pTab!=0 );
4696 if( (pTab->tabFlags & TF_Ephemeral)!=0 ){
4697 /* A sub-query in the FROM clause of a SELECT */
4698 Select *pSel = pFrom->pSelect;
4699 if( pSel ){
4700 while( pSel->pPrior ) pSel = pSel->pPrior;
4701 sqlite3SelectAddColumnTypeAndCollation(pParse, pTab, pSel);
4706 #endif
4710 ** This routine adds datatype and collating sequence information to
4711 ** the Table structures of all FROM-clause subqueries in a
4712 ** SELECT statement.
4714 ** Use this routine after name resolution.
4716 static void sqlite3SelectAddTypeInfo(Parse *pParse, Select *pSelect){
4717 #ifndef SQLITE_OMIT_SUBQUERY
4718 Walker w;
4719 w.xSelectCallback = sqlite3SelectWalkNoop;
4720 w.xSelectCallback2 = selectAddSubqueryTypeInfo;
4721 w.xExprCallback = sqlite3ExprWalkNoop;
4722 w.pParse = pParse;
4723 sqlite3WalkSelect(&w, pSelect);
4724 #endif
4729 ** This routine sets up a SELECT statement for processing. The
4730 ** following is accomplished:
4732 ** * VDBE Cursor numbers are assigned to all FROM-clause terms.
4733 ** * Ephemeral Table objects are created for all FROM-clause subqueries.
4734 ** * ON and USING clauses are shifted into WHERE statements
4735 ** * Wildcards "*" and "TABLE.*" in result sets are expanded.
4736 ** * Identifiers in expression are matched to tables.
4738 ** This routine acts recursively on all subqueries within the SELECT.
4740 void sqlite3SelectPrep(
4741 Parse *pParse, /* The parser context */
4742 Select *p, /* The SELECT statement being coded. */
4743 NameContext *pOuterNC /* Name context for container */
4745 assert( p!=0 || pParse->db->mallocFailed );
4746 if( pParse->db->mallocFailed ) return;
4747 if( p->selFlags & SF_HasTypeInfo ) return;
4748 sqlite3SelectExpand(pParse, p);
4749 if( pParse->nErr || pParse->db->mallocFailed ) return;
4750 sqlite3ResolveSelectNames(pParse, p, pOuterNC);
4751 if( pParse->nErr || pParse->db->mallocFailed ) return;
4752 sqlite3SelectAddTypeInfo(pParse, p);
4756 ** Reset the aggregate accumulator.
4758 ** The aggregate accumulator is a set of memory cells that hold
4759 ** intermediate results while calculating an aggregate. This
4760 ** routine generates code that stores NULLs in all of those memory
4761 ** cells.
4763 static void resetAccumulator(Parse *pParse, AggInfo *pAggInfo){
4764 Vdbe *v = pParse->pVdbe;
4765 int i;
4766 struct AggInfo_func *pFunc;
4767 int nReg = pAggInfo->nFunc + pAggInfo->nColumn;
4768 if( nReg==0 ) return;
4769 #ifdef SQLITE_DEBUG
4770 /* Verify that all AggInfo registers are within the range specified by
4771 ** AggInfo.mnReg..AggInfo.mxReg */
4772 assert( nReg==pAggInfo->mxReg-pAggInfo->mnReg+1 );
4773 for(i=0; i<pAggInfo->nColumn; i++){
4774 assert( pAggInfo->aCol[i].iMem>=pAggInfo->mnReg
4775 && pAggInfo->aCol[i].iMem<=pAggInfo->mxReg );
4777 for(i=0; i<pAggInfo->nFunc; i++){
4778 assert( pAggInfo->aFunc[i].iMem>=pAggInfo->mnReg
4779 && pAggInfo->aFunc[i].iMem<=pAggInfo->mxReg );
4781 #endif
4782 sqlite3VdbeAddOp3(v, OP_Null, 0, pAggInfo->mnReg, pAggInfo->mxReg);
4783 for(pFunc=pAggInfo->aFunc, i=0; i<pAggInfo->nFunc; i++, pFunc++){
4784 if( pFunc->iDistinct>=0 ){
4785 Expr *pE = pFunc->pExpr;
4786 assert( !ExprHasProperty(pE, EP_xIsSelect) );
4787 if( pE->x.pList==0 || pE->x.pList->nExpr!=1 ){
4788 sqlite3ErrorMsg(pParse, "DISTINCT aggregates must have exactly one "
4789 "argument");
4790 pFunc->iDistinct = -1;
4791 }else{
4792 KeyInfo *pKeyInfo = keyInfoFromExprList(pParse, pE->x.pList, 0, 0);
4793 sqlite3VdbeAddOp4(v, OP_OpenEphemeral, pFunc->iDistinct, 0, 0,
4794 (char*)pKeyInfo, P4_KEYINFO);
4801 ** Invoke the OP_AggFinalize opcode for every aggregate function
4802 ** in the AggInfo structure.
4804 static void finalizeAggFunctions(Parse *pParse, AggInfo *pAggInfo){
4805 Vdbe *v = pParse->pVdbe;
4806 int i;
4807 struct AggInfo_func *pF;
4808 for(i=0, pF=pAggInfo->aFunc; i<pAggInfo->nFunc; i++, pF++){
4809 ExprList *pList = pF->pExpr->x.pList;
4810 assert( !ExprHasProperty(pF->pExpr, EP_xIsSelect) );
4811 sqlite3VdbeAddOp2(v, OP_AggFinal, pF->iMem, pList ? pList->nExpr : 0);
4812 sqlite3VdbeAppendP4(v, pF->pFunc, P4_FUNCDEF);
4817 ** Update the accumulator memory cells for an aggregate based on
4818 ** the current cursor position.
4820 static void updateAccumulator(Parse *pParse, AggInfo *pAggInfo){
4821 Vdbe *v = pParse->pVdbe;
4822 int i;
4823 int regHit = 0;
4824 int addrHitTest = 0;
4825 struct AggInfo_func *pF;
4826 struct AggInfo_col *pC;
4828 pAggInfo->directMode = 1;
4829 for(i=0, pF=pAggInfo->aFunc; i<pAggInfo->nFunc; i++, pF++){
4830 int nArg;
4831 int addrNext = 0;
4832 int regAgg;
4833 ExprList *pList = pF->pExpr->x.pList;
4834 assert( !ExprHasProperty(pF->pExpr, EP_xIsSelect) );
4835 if( pList ){
4836 nArg = pList->nExpr;
4837 regAgg = sqlite3GetTempRange(pParse, nArg);
4838 sqlite3ExprCodeExprList(pParse, pList, regAgg, 0, SQLITE_ECEL_DUP);
4839 }else{
4840 nArg = 0;
4841 regAgg = 0;
4843 if( pF->iDistinct>=0 ){
4844 addrNext = sqlite3VdbeMakeLabel(v);
4845 testcase( nArg==0 ); /* Error condition */
4846 testcase( nArg>1 ); /* Also an error */
4847 codeDistinct(pParse, pF->iDistinct, addrNext, 1, regAgg);
4849 if( pF->pFunc->funcFlags & SQLITE_FUNC_NEEDCOLL ){
4850 CollSeq *pColl = 0;
4851 struct ExprList_item *pItem;
4852 int j;
4853 assert( pList!=0 ); /* pList!=0 if pF->pFunc has NEEDCOLL */
4854 for(j=0, pItem=pList->a; !pColl && j<nArg; j++, pItem++){
4855 pColl = sqlite3ExprCollSeq(pParse, pItem->pExpr);
4857 if( !pColl ){
4858 pColl = pParse->db->pDfltColl;
4860 if( regHit==0 && pAggInfo->nAccumulator ) regHit = ++pParse->nMem;
4861 sqlite3VdbeAddOp4(v, OP_CollSeq, regHit, 0, 0, (char *)pColl, P4_COLLSEQ);
4863 sqlite3VdbeAddOp3(v, OP_AggStep0, 0, regAgg, pF->iMem);
4864 sqlite3VdbeAppendP4(v, pF->pFunc, P4_FUNCDEF);
4865 sqlite3VdbeChangeP5(v, (u8)nArg);
4866 sqlite3ExprCacheAffinityChange(pParse, regAgg, nArg);
4867 sqlite3ReleaseTempRange(pParse, regAgg, nArg);
4868 if( addrNext ){
4869 sqlite3VdbeResolveLabel(v, addrNext);
4870 sqlite3ExprCacheClear(pParse);
4874 /* Before populating the accumulator registers, clear the column cache.
4875 ** Otherwise, if any of the required column values are already present
4876 ** in registers, sqlite3ExprCode() may use OP_SCopy to copy the value
4877 ** to pC->iMem. But by the time the value is used, the original register
4878 ** may have been used, invalidating the underlying buffer holding the
4879 ** text or blob value. See ticket [883034dcb5].
4881 ** Another solution would be to change the OP_SCopy used to copy cached
4882 ** values to an OP_Copy.
4884 if( regHit ){
4885 addrHitTest = sqlite3VdbeAddOp1(v, OP_If, regHit); VdbeCoverage(v);
4887 sqlite3ExprCacheClear(pParse);
4888 for(i=0, pC=pAggInfo->aCol; i<pAggInfo->nAccumulator; i++, pC++){
4889 sqlite3ExprCode(pParse, pC->pExpr, pC->iMem);
4891 pAggInfo->directMode = 0;
4892 sqlite3ExprCacheClear(pParse);
4893 if( addrHitTest ){
4894 sqlite3VdbeJumpHere(v, addrHitTest);
4899 ** Add a single OP_Explain instruction to the VDBE to explain a simple
4900 ** count(*) query ("SELECT count(*) FROM pTab").
4902 #ifndef SQLITE_OMIT_EXPLAIN
4903 static void explainSimpleCount(
4904 Parse *pParse, /* Parse context */
4905 Table *pTab, /* Table being queried */
4906 Index *pIdx /* Index used to optimize scan, or NULL */
4908 if( pParse->explain==2 ){
4909 int bCover = (pIdx!=0 && (HasRowid(pTab) || !IsPrimaryKeyIndex(pIdx)));
4910 char *zEqp = sqlite3MPrintf(pParse->db, "SCAN TABLE %s%s%s",
4911 pTab->zName,
4912 bCover ? " USING COVERING INDEX " : "",
4913 bCover ? pIdx->zName : ""
4915 sqlite3VdbeAddOp4(
4916 pParse->pVdbe, OP_Explain, pParse->iSelectId, 0, 0, zEqp, P4_DYNAMIC
4920 #else
4921 # define explainSimpleCount(a,b,c)
4922 #endif
4925 ** Context object for havingToWhereExprCb().
4927 struct HavingToWhereCtx {
4928 Expr **ppWhere;
4929 ExprList *pGroupBy;
4933 ** sqlite3WalkExpr() callback used by havingToWhere().
4935 ** If the node passed to the callback is a TK_AND node, return
4936 ** WRC_Continue to tell sqlite3WalkExpr() to iterate through child nodes.
4938 ** Otherwise, return WRC_Prune. In this case, also check if the
4939 ** sub-expression matches the criteria for being moved to the WHERE
4940 ** clause. If so, add it to the WHERE clause and replace the sub-expression
4941 ** within the HAVING expression with a constant "1".
4943 static int havingToWhereExprCb(Walker *pWalker, Expr *pExpr){
4944 if( pExpr->op!=TK_AND ){
4945 struct HavingToWhereCtx *p = pWalker->u.pHavingCtx;
4946 if( sqlite3ExprIsConstantOrGroupBy(pWalker->pParse, pExpr, p->pGroupBy) ){
4947 sqlite3 *db = pWalker->pParse->db;
4948 Expr *pNew = sqlite3ExprAlloc(db, TK_INTEGER, &sqlite3IntTokens[1], 0);
4949 if( pNew ){
4950 Expr *pWhere = *(p->ppWhere);
4951 SWAP(Expr, *pNew, *pExpr);
4952 pNew = sqlite3ExprAnd(db, pWhere, pNew);
4953 *(p->ppWhere) = pNew;
4956 return WRC_Prune;
4958 return WRC_Continue;
4962 ** Transfer eligible terms from the HAVING clause of a query, which is
4963 ** processed after grouping, to the WHERE clause, which is processed before
4964 ** grouping. For example, the query:
4966 ** SELECT * FROM <tables> WHERE a=? GROUP BY b HAVING b=? AND c=?
4968 ** can be rewritten as:
4970 ** SELECT * FROM <tables> WHERE a=? AND b=? GROUP BY b HAVING c=?
4972 ** A term of the HAVING expression is eligible for transfer if it consists
4973 ** entirely of constants and expressions that are also GROUP BY terms that
4974 ** use the "BINARY" collation sequence.
4976 static void havingToWhere(
4977 Parse *pParse,
4978 ExprList *pGroupBy,
4979 Expr *pHaving,
4980 Expr **ppWhere
4982 struct HavingToWhereCtx sCtx;
4983 Walker sWalker;
4985 sCtx.ppWhere = ppWhere;
4986 sCtx.pGroupBy = pGroupBy;
4988 memset(&sWalker, 0, sizeof(sWalker));
4989 sWalker.pParse = pParse;
4990 sWalker.xExprCallback = havingToWhereExprCb;
4991 sWalker.u.pHavingCtx = &sCtx;
4992 sqlite3WalkExpr(&sWalker, pHaving);
4996 ** Check to see if the pThis entry of pTabList is a self-join of a prior view.
4997 ** If it is, then return the SrcList_item for the prior view. If it is not,
4998 ** then return 0.
5000 static struct SrcList_item *isSelfJoinView(
5001 SrcList *pTabList, /* Search for self-joins in this FROM clause */
5002 struct SrcList_item *pThis /* Search for prior reference to this subquery */
5004 struct SrcList_item *pItem;
5005 for(pItem = pTabList->a; pItem<pThis; pItem++){
5006 if( pItem->pSelect==0 ) continue;
5007 if( pItem->fg.viaCoroutine ) continue;
5008 if( pItem->zName==0 ) continue;
5009 if( sqlite3_stricmp(pItem->zDatabase, pThis->zDatabase)!=0 ) continue;
5010 if( sqlite3_stricmp(pItem->zName, pThis->zName)!=0 ) continue;
5011 if( sqlite3ExprCompare(0,
5012 pThis->pSelect->pWhere, pItem->pSelect->pWhere, -1)
5014 /* The view was modified by some other optimization such as
5015 ** pushDownWhereTerms() */
5016 continue;
5018 return pItem;
5020 return 0;
5023 #ifdef SQLITE_COUNTOFVIEW_OPTIMIZATION
5025 ** Attempt to transform a query of the form
5027 ** SELECT count(*) FROM (SELECT x FROM t1 UNION ALL SELECT y FROM t2)
5029 ** Into this:
5031 ** SELECT (SELECT count(*) FROM t1)+(SELECT count(*) FROM t2)
5033 ** The transformation only works if all of the following are true:
5035 ** * The subquery is a UNION ALL of two or more terms
5036 ** * There is no WHERE or GROUP BY or HAVING clauses on the subqueries
5037 ** * The outer query is a simple count(*)
5039 ** Return TRUE if the optimization is undertaken.
5041 static int countOfViewOptimization(Parse *pParse, Select *p){
5042 Select *pSub, *pPrior;
5043 Expr *pExpr;
5044 Expr *pCount;
5045 sqlite3 *db;
5046 if( (p->selFlags & SF_Aggregate)==0 ) return 0; /* This is an aggregate */
5047 if( p->pEList->nExpr!=1 ) return 0; /* Single result column */
5048 pExpr = p->pEList->a[0].pExpr;
5049 if( pExpr->op!=TK_AGG_FUNCTION ) return 0; /* Result is an aggregate */
5050 if( sqlite3_stricmp(pExpr->u.zToken,"count") ) return 0; /* Is count() */
5051 if( pExpr->x.pList!=0 ) return 0; /* Must be count(*) */
5052 if( p->pSrc->nSrc!=1 ) return 0; /* One table in FROM */
5053 pSub = p->pSrc->a[0].pSelect;
5054 if( pSub==0 ) return 0; /* The FROM is a subquery */
5055 if( pSub->pPrior==0 ) return 0; /* Must be a compound ry */
5057 if( pSub->op!=TK_ALL && pSub->pPrior ) return 0; /* Must be UNION ALL */
5058 if( pSub->pWhere ) return 0; /* No WHERE clause */
5059 if( pSub->selFlags & SF_Aggregate ) return 0; /* Not an aggregate */
5060 pSub = pSub->pPrior; /* Repeat over compound */
5061 }while( pSub );
5063 /* If we reach this point then it is OK to perform the transformation */
5065 db = pParse->db;
5066 pCount = pExpr;
5067 pExpr = 0;
5068 pSub = p->pSrc->a[0].pSelect;
5069 p->pSrc->a[0].pSelect = 0;
5070 sqlite3SrcListDelete(db, p->pSrc);
5071 p->pSrc = sqlite3DbMallocZero(pParse->db, sizeof(*p->pSrc));
5072 while( pSub ){
5073 Expr *pTerm;
5074 pPrior = pSub->pPrior;
5075 pSub->pPrior = 0;
5076 pSub->pNext = 0;
5077 pSub->selFlags |= SF_Aggregate;
5078 pSub->selFlags &= ~SF_Compound;
5079 pSub->nSelectRow = 0;
5080 sqlite3ExprListDelete(db, pSub->pEList);
5081 pTerm = pPrior ? sqlite3ExprDup(db, pCount, 0) : pCount;
5082 pSub->pEList = sqlite3ExprListAppend(pParse, 0, pTerm);
5083 pTerm = sqlite3PExpr(pParse, TK_SELECT, 0, 0);
5084 sqlite3PExprAddSelect(pParse, pTerm, pSub);
5085 if( pExpr==0 ){
5086 pExpr = pTerm;
5087 }else{
5088 pExpr = sqlite3PExpr(pParse, TK_PLUS, pTerm, pExpr);
5090 pSub = pPrior;
5092 p->pEList->a[0].pExpr = pExpr;
5093 p->selFlags &= ~SF_Aggregate;
5095 #if SELECTTRACE_ENABLED
5096 if( sqlite3SelectTrace & 0x400 ){
5097 SELECTTRACE(0x400,pParse,p,("After count-of-view optimization:\n"));
5098 sqlite3TreeViewSelect(0, p, 0);
5100 #endif
5101 return 1;
5103 #endif /* SQLITE_COUNTOFVIEW_OPTIMIZATION */
5106 ** Generate code for the SELECT statement given in the p argument.
5108 ** The results are returned according to the SelectDest structure.
5109 ** See comments in sqliteInt.h for further information.
5111 ** This routine returns the number of errors. If any errors are
5112 ** encountered, then an appropriate error message is left in
5113 ** pParse->zErrMsg.
5115 ** This routine does NOT free the Select structure passed in. The
5116 ** calling function needs to do that.
5118 int sqlite3Select(
5119 Parse *pParse, /* The parser context */
5120 Select *p, /* The SELECT statement being coded. */
5121 SelectDest *pDest /* What to do with the query results */
5123 int i, j; /* Loop counters */
5124 WhereInfo *pWInfo; /* Return from sqlite3WhereBegin() */
5125 Vdbe *v; /* The virtual machine under construction */
5126 int isAgg; /* True for select lists like "count(*)" */
5127 ExprList *pEList = 0; /* List of columns to extract. */
5128 SrcList *pTabList; /* List of tables to select from */
5129 Expr *pWhere; /* The WHERE clause. May be NULL */
5130 ExprList *pGroupBy; /* The GROUP BY clause. May be NULL */
5131 Expr *pHaving; /* The HAVING clause. May be NULL */
5132 int rc = 1; /* Value to return from this function */
5133 DistinctCtx sDistinct; /* Info on how to code the DISTINCT keyword */
5134 SortCtx sSort; /* Info on how to code the ORDER BY clause */
5135 AggInfo sAggInfo; /* Information used by aggregate queries */
5136 int iEnd; /* Address of the end of the query */
5137 sqlite3 *db; /* The database connection */
5139 #ifndef SQLITE_OMIT_EXPLAIN
5140 int iRestoreSelectId = pParse->iSelectId;
5141 pParse->iSelectId = pParse->iNextSelectId++;
5142 #endif
5144 db = pParse->db;
5145 if( p==0 || db->mallocFailed || pParse->nErr ){
5146 return 1;
5148 if( sqlite3AuthCheck(pParse, SQLITE_SELECT, 0, 0, 0) ) return 1;
5149 memset(&sAggInfo, 0, sizeof(sAggInfo));
5150 #if SELECTTRACE_ENABLED
5151 pParse->nSelectIndent++;
5152 SELECTTRACE(1,pParse,p, ("begin processing:\n"));
5153 if( sqlite3SelectTrace & 0x100 ){
5154 sqlite3TreeViewSelect(0, p, 0);
5156 #endif
5158 assert( p->pOrderBy==0 || pDest->eDest!=SRT_DistFifo );
5159 assert( p->pOrderBy==0 || pDest->eDest!=SRT_Fifo );
5160 assert( p->pOrderBy==0 || pDest->eDest!=SRT_DistQueue );
5161 assert( p->pOrderBy==0 || pDest->eDest!=SRT_Queue );
5162 if( IgnorableOrderby(pDest) ){
5163 assert(pDest->eDest==SRT_Exists || pDest->eDest==SRT_Union ||
5164 pDest->eDest==SRT_Except || pDest->eDest==SRT_Discard ||
5165 pDest->eDest==SRT_Queue || pDest->eDest==SRT_DistFifo ||
5166 pDest->eDest==SRT_DistQueue || pDest->eDest==SRT_Fifo);
5167 /* If ORDER BY makes no difference in the output then neither does
5168 ** DISTINCT so it can be removed too. */
5169 sqlite3ExprListDelete(db, p->pOrderBy);
5170 p->pOrderBy = 0;
5171 p->selFlags &= ~SF_Distinct;
5173 sqlite3SelectPrep(pParse, p, 0);
5174 memset(&sSort, 0, sizeof(sSort));
5175 sSort.pOrderBy = p->pOrderBy;
5176 pTabList = p->pSrc;
5177 if( pParse->nErr || db->mallocFailed ){
5178 goto select_end;
5180 assert( p->pEList!=0 );
5181 isAgg = (p->selFlags & SF_Aggregate)!=0;
5182 #if SELECTTRACE_ENABLED
5183 if( sqlite3SelectTrace & 0x100 ){
5184 SELECTTRACE(0x100,pParse,p, ("after name resolution:\n"));
5185 sqlite3TreeViewSelect(0, p, 0);
5187 #endif
5189 /* Get a pointer the VDBE under construction, allocating a new VDBE if one
5190 ** does not already exist */
5191 v = sqlite3GetVdbe(pParse);
5192 if( v==0 ) goto select_end;
5193 if( pDest->eDest==SRT_Output ){
5194 generateColumnNames(pParse, p);
5197 /* Try to flatten subqueries in the FROM clause up into the main query
5199 #if !defined(SQLITE_OMIT_SUBQUERY) || !defined(SQLITE_OMIT_VIEW)
5200 for(i=0; !p->pPrior && i<pTabList->nSrc; i++){
5201 struct SrcList_item *pItem = &pTabList->a[i];
5202 Select *pSub = pItem->pSelect;
5203 Table *pTab = pItem->pTab;
5204 if( pSub==0 ) continue;
5206 /* Catch mismatch in the declared columns of a view and the number of
5207 ** columns in the SELECT on the RHS */
5208 if( pTab->nCol!=pSub->pEList->nExpr ){
5209 sqlite3ErrorMsg(pParse, "expected %d columns for '%s' but got %d",
5210 pTab->nCol, pTab->zName, pSub->pEList->nExpr);
5211 goto select_end;
5214 /* Do not try to flatten an aggregate subquery.
5216 ** Flattening an aggregate subquery is only possible if the outer query
5217 ** is not a join. But if the outer query is not a join, then the subquery
5218 ** will be implemented as a co-routine and there is no advantage to
5219 ** flattening in that case.
5221 if( (pSub->selFlags & SF_Aggregate)!=0 ) continue;
5222 assert( pSub->pGroupBy==0 );
5224 /* If the subquery contains an ORDER BY clause and if
5225 ** it will be implemented as a co-routine, then do not flatten. This
5226 ** restriction allows SQL constructs like this:
5228 ** SELECT expensive_function(x)
5229 ** FROM (SELECT x FROM tab ORDER BY y LIMIT 10);
5231 ** The expensive_function() is only computed on the 10 rows that
5232 ** are output, rather than every row of the table.
5234 if( pSub->pOrderBy!=0
5235 && i==0
5236 && (pTabList->nSrc==1
5237 || (pTabList->a[1].fg.jointype&(JT_LEFT|JT_CROSS))!=0)
5239 continue;
5242 if( flattenSubquery(pParse, p, i, isAgg) ){
5243 /* This subquery can be absorbed into its parent. */
5244 i = -1;
5246 pTabList = p->pSrc;
5247 if( db->mallocFailed ) goto select_end;
5248 if( !IgnorableOrderby(pDest) ){
5249 sSort.pOrderBy = p->pOrderBy;
5252 #endif
5254 #ifndef SQLITE_OMIT_COMPOUND_SELECT
5255 /* Handle compound SELECT statements using the separate multiSelect()
5256 ** procedure.
5258 if( p->pPrior ){
5259 rc = multiSelect(pParse, p, pDest);
5260 explainSetInteger(pParse->iSelectId, iRestoreSelectId);
5261 #if SELECTTRACE_ENABLED
5262 SELECTTRACE(1,pParse,p,("end compound-select processing\n"));
5263 pParse->nSelectIndent--;
5264 #endif
5265 return rc;
5267 #endif
5269 /* For each term in the FROM clause, do two things:
5270 ** (1) Authorized unreferenced tables
5271 ** (2) Generate code for all sub-queries
5273 for(i=0; i<pTabList->nSrc; i++){
5274 struct SrcList_item *pItem = &pTabList->a[i];
5275 SelectDest dest;
5276 Select *pSub;
5277 #if !defined(SQLITE_OMIT_SUBQUERY) || !defined(SQLITE_OMIT_VIEW)
5278 const char *zSavedAuthContext;
5279 #endif
5281 /* Issue SQLITE_READ authorizations with a fake column name for any
5282 ** tables that are referenced but from which no values are extracted.
5283 ** Examples of where these kinds of null SQLITE_READ authorizations
5284 ** would occur:
5286 ** SELECT count(*) FROM t1; -- SQLITE_READ t1.""
5287 ** SELECT t1.* FROM t1, t2; -- SQLITE_READ t2.""
5289 ** The fake column name is an empty string. It is possible for a table to
5290 ** have a column named by the empty string, in which case there is no way to
5291 ** distinguish between an unreferenced table and an actual reference to the
5292 ** "" column. The original design was for the fake column name to be a NULL,
5293 ** which would be unambiguous. But legacy authorization callbacks might
5294 ** assume the column name is non-NULL and segfault. The use of an empty
5295 ** string for the fake column name seems safer.
5297 if( pItem->colUsed==0 ){
5298 sqlite3AuthCheck(pParse, SQLITE_READ, pItem->zName, "", pItem->zDatabase);
5301 #if !defined(SQLITE_OMIT_SUBQUERY) || !defined(SQLITE_OMIT_VIEW)
5302 /* Generate code for all sub-queries in the FROM clause
5304 pSub = pItem->pSelect;
5305 if( pSub==0 ) continue;
5307 /* Sometimes the code for a subquery will be generated more than
5308 ** once, if the subquery is part of the WHERE clause in a LEFT JOIN,
5309 ** for example. In that case, do not regenerate the code to manifest
5310 ** a view or the co-routine to implement a view. The first instance
5311 ** is sufficient, though the subroutine to manifest the view does need
5312 ** to be invoked again. */
5313 if( pItem->addrFillSub ){
5314 if( pItem->fg.viaCoroutine==0 ){
5315 /* The subroutine that manifests the view might be a one-time routine,
5316 ** or it might need to be rerun on each iteration because it
5317 ** encodes a correlated subquery. */
5318 testcase( sqlite3VdbeGetOp(v, pItem->addrFillSub)->opcode==OP_Once );
5319 sqlite3VdbeAddOp2(v, OP_Gosub, pItem->regReturn, pItem->addrFillSub);
5321 continue;
5324 /* Increment Parse.nHeight by the height of the largest expression
5325 ** tree referred to by this, the parent select. The child select
5326 ** may contain expression trees of at most
5327 ** (SQLITE_MAX_EXPR_DEPTH-Parse.nHeight) height. This is a bit
5328 ** more conservative than necessary, but much easier than enforcing
5329 ** an exact limit.
5331 pParse->nHeight += sqlite3SelectExprHeight(p);
5333 /* Make copies of constant WHERE-clause terms in the outer query down
5334 ** inside the subquery. This can help the subquery to run more efficiently.
5336 if( (pItem->fg.jointype & JT_OUTER)==0
5337 && pushDownWhereTerms(pParse, pSub, p->pWhere, pItem->iCursor)
5339 #if SELECTTRACE_ENABLED
5340 if( sqlite3SelectTrace & 0x100 ){
5341 SELECTTRACE(0x100,pParse,p,("After WHERE-clause push-down:\n"));
5342 sqlite3TreeViewSelect(0, p, 0);
5344 #endif
5347 zSavedAuthContext = pParse->zAuthContext;
5348 pParse->zAuthContext = pItem->zName;
5350 /* Generate code to implement the subquery
5352 ** The subquery is implemented as a co-routine if the subquery is
5353 ** guaranteed to be the outer loop (so that it does not need to be
5354 ** computed more than once)
5356 ** TODO: Are there other reasons beside (1) to use a co-routine
5357 ** implementation?
5359 if( i==0
5360 && (pTabList->nSrc==1
5361 || (pTabList->a[1].fg.jointype&(JT_LEFT|JT_CROSS))!=0) /* (1) */
5363 /* Implement a co-routine that will return a single row of the result
5364 ** set on each invocation.
5366 int addrTop = sqlite3VdbeCurrentAddr(v)+1;
5368 pItem->regReturn = ++pParse->nMem;
5369 sqlite3VdbeAddOp3(v, OP_InitCoroutine, pItem->regReturn, 0, addrTop);
5370 VdbeComment((v, "%s", pItem->pTab->zName));
5371 pItem->addrFillSub = addrTop;
5372 sqlite3SelectDestInit(&dest, SRT_Coroutine, pItem->regReturn);
5373 explainSetInteger(pItem->iSelectId, (u8)pParse->iNextSelectId);
5374 sqlite3Select(pParse, pSub, &dest);
5375 pItem->pTab->nRowLogEst = pSub->nSelectRow;
5376 pItem->fg.viaCoroutine = 1;
5377 pItem->regResult = dest.iSdst;
5378 sqlite3VdbeEndCoroutine(v, pItem->regReturn);
5379 sqlite3VdbeJumpHere(v, addrTop-1);
5380 sqlite3ClearTempRegCache(pParse);
5381 }else{
5382 /* Generate a subroutine that will fill an ephemeral table with
5383 ** the content of this subquery. pItem->addrFillSub will point
5384 ** to the address of the generated subroutine. pItem->regReturn
5385 ** is a register allocated to hold the subroutine return address
5387 int topAddr;
5388 int onceAddr = 0;
5389 int retAddr;
5390 struct SrcList_item *pPrior;
5392 assert( pItem->addrFillSub==0 );
5393 pItem->regReturn = ++pParse->nMem;
5394 topAddr = sqlite3VdbeAddOp2(v, OP_Integer, 0, pItem->regReturn);
5395 pItem->addrFillSub = topAddr+1;
5396 if( pItem->fg.isCorrelated==0 ){
5397 /* If the subquery is not correlated and if we are not inside of
5398 ** a trigger, then we only need to compute the value of the subquery
5399 ** once. */
5400 onceAddr = sqlite3VdbeAddOp0(v, OP_Once); VdbeCoverage(v);
5401 VdbeComment((v, "materialize \"%s\"", pItem->pTab->zName));
5402 }else{
5403 VdbeNoopComment((v, "materialize \"%s\"", pItem->pTab->zName));
5405 pPrior = isSelfJoinView(pTabList, pItem);
5406 if( pPrior ){
5407 sqlite3VdbeAddOp2(v, OP_OpenDup, pItem->iCursor, pPrior->iCursor);
5408 explainSetInteger(pItem->iSelectId, pPrior->iSelectId);
5409 assert( pPrior->pSelect!=0 );
5410 pSub->nSelectRow = pPrior->pSelect->nSelectRow;
5411 }else{
5412 sqlite3SelectDestInit(&dest, SRT_EphemTab, pItem->iCursor);
5413 explainSetInteger(pItem->iSelectId, (u8)pParse->iNextSelectId);
5414 sqlite3Select(pParse, pSub, &dest);
5416 pItem->pTab->nRowLogEst = pSub->nSelectRow;
5417 if( onceAddr ) sqlite3VdbeJumpHere(v, onceAddr);
5418 retAddr = sqlite3VdbeAddOp1(v, OP_Return, pItem->regReturn);
5419 VdbeComment((v, "end %s", pItem->pTab->zName));
5420 sqlite3VdbeChangeP1(v, topAddr, retAddr);
5421 sqlite3ClearTempRegCache(pParse);
5423 if( db->mallocFailed ) goto select_end;
5424 pParse->nHeight -= sqlite3SelectExprHeight(p);
5425 pParse->zAuthContext = zSavedAuthContext;
5426 #endif
5429 /* Various elements of the SELECT copied into local variables for
5430 ** convenience */
5431 pEList = p->pEList;
5432 pWhere = p->pWhere;
5433 pGroupBy = p->pGroupBy;
5434 pHaving = p->pHaving;
5435 sDistinct.isTnct = (p->selFlags & SF_Distinct)!=0;
5437 #if SELECTTRACE_ENABLED
5438 if( sqlite3SelectTrace & 0x400 ){
5439 SELECTTRACE(0x400,pParse,p,("After all FROM-clause analysis:\n"));
5440 sqlite3TreeViewSelect(0, p, 0);
5442 #endif
5444 #ifdef SQLITE_COUNTOFVIEW_OPTIMIZATION
5445 if( OptimizationEnabled(db, SQLITE_QueryFlattener|SQLITE_CountOfView)
5446 && countOfViewOptimization(pParse, p)
5448 if( db->mallocFailed ) goto select_end;
5449 pEList = p->pEList;
5450 pTabList = p->pSrc;
5452 #endif
5454 /* If the query is DISTINCT with an ORDER BY but is not an aggregate, and
5455 ** if the select-list is the same as the ORDER BY list, then this query
5456 ** can be rewritten as a GROUP BY. In other words, this:
5458 ** SELECT DISTINCT xyz FROM ... ORDER BY xyz
5460 ** is transformed to:
5462 ** SELECT xyz FROM ... GROUP BY xyz ORDER BY xyz
5464 ** The second form is preferred as a single index (or temp-table) may be
5465 ** used for both the ORDER BY and DISTINCT processing. As originally
5466 ** written the query must use a temp-table for at least one of the ORDER
5467 ** BY and DISTINCT, and an index or separate temp-table for the other.
5469 if( (p->selFlags & (SF_Distinct|SF_Aggregate))==SF_Distinct
5470 && sqlite3ExprListCompare(sSort.pOrderBy, pEList, -1)==0
5472 p->selFlags &= ~SF_Distinct;
5473 pGroupBy = p->pGroupBy = sqlite3ExprListDup(db, pEList, 0);
5474 /* Notice that even thought SF_Distinct has been cleared from p->selFlags,
5475 ** the sDistinct.isTnct is still set. Hence, isTnct represents the
5476 ** original setting of the SF_Distinct flag, not the current setting */
5477 assert( sDistinct.isTnct );
5479 #if SELECTTRACE_ENABLED
5480 if( sqlite3SelectTrace & 0x400 ){
5481 SELECTTRACE(0x400,pParse,p,("Transform DISTINCT into GROUP BY:\n"));
5482 sqlite3TreeViewSelect(0, p, 0);
5484 #endif
5487 /* If there is an ORDER BY clause, then create an ephemeral index to
5488 ** do the sorting. But this sorting ephemeral index might end up
5489 ** being unused if the data can be extracted in pre-sorted order.
5490 ** If that is the case, then the OP_OpenEphemeral instruction will be
5491 ** changed to an OP_Noop once we figure out that the sorting index is
5492 ** not needed. The sSort.addrSortIndex variable is used to facilitate
5493 ** that change.
5495 if( sSort.pOrderBy ){
5496 KeyInfo *pKeyInfo;
5497 pKeyInfo = keyInfoFromExprList(pParse, sSort.pOrderBy, 0, pEList->nExpr);
5498 sSort.iECursor = pParse->nTab++;
5499 sSort.addrSortIndex =
5500 sqlite3VdbeAddOp4(v, OP_OpenEphemeral,
5501 sSort.iECursor, sSort.pOrderBy->nExpr+1+pEList->nExpr, 0,
5502 (char*)pKeyInfo, P4_KEYINFO
5504 }else{
5505 sSort.addrSortIndex = -1;
5508 /* If the output is destined for a temporary table, open that table.
5510 if( pDest->eDest==SRT_EphemTab ){
5511 sqlite3VdbeAddOp2(v, OP_OpenEphemeral, pDest->iSDParm, pEList->nExpr);
5514 /* Set the limiter.
5516 iEnd = sqlite3VdbeMakeLabel(v);
5517 if( (p->selFlags & SF_FixedLimit)==0 ){
5518 p->nSelectRow = 320; /* 4 billion rows */
5520 computeLimitRegisters(pParse, p, iEnd);
5521 if( p->iLimit==0 && sSort.addrSortIndex>=0 ){
5522 sqlite3VdbeChangeOpcode(v, sSort.addrSortIndex, OP_SorterOpen);
5523 sSort.sortFlags |= SORTFLAG_UseSorter;
5526 /* Open an ephemeral index to use for the distinct set.
5528 if( p->selFlags & SF_Distinct ){
5529 sDistinct.tabTnct = pParse->nTab++;
5530 sDistinct.addrTnct = sqlite3VdbeAddOp4(v, OP_OpenEphemeral,
5531 sDistinct.tabTnct, 0, 0,
5532 (char*)keyInfoFromExprList(pParse, p->pEList,0,0),
5533 P4_KEYINFO);
5534 sqlite3VdbeChangeP5(v, BTREE_UNORDERED);
5535 sDistinct.eTnctType = WHERE_DISTINCT_UNORDERED;
5536 }else{
5537 sDistinct.eTnctType = WHERE_DISTINCT_NOOP;
5540 if( !isAgg && pGroupBy==0 ){
5541 /* No aggregate functions and no GROUP BY clause */
5542 u16 wctrlFlags = (sDistinct.isTnct ? WHERE_WANT_DISTINCT : 0);
5543 assert( WHERE_USE_LIMIT==SF_FixedLimit );
5544 wctrlFlags |= p->selFlags & SF_FixedLimit;
5546 /* Begin the database scan. */
5547 pWInfo = sqlite3WhereBegin(pParse, pTabList, pWhere, sSort.pOrderBy,
5548 p->pEList, wctrlFlags, p->nSelectRow);
5549 if( pWInfo==0 ) goto select_end;
5550 if( sqlite3WhereOutputRowCount(pWInfo) < p->nSelectRow ){
5551 p->nSelectRow = sqlite3WhereOutputRowCount(pWInfo);
5553 if( sDistinct.isTnct && sqlite3WhereIsDistinct(pWInfo) ){
5554 sDistinct.eTnctType = sqlite3WhereIsDistinct(pWInfo);
5556 if( sSort.pOrderBy ){
5557 sSort.nOBSat = sqlite3WhereIsOrdered(pWInfo);
5558 sSort.bOrderedInnerLoop = sqlite3WhereOrderedInnerLoop(pWInfo);
5559 if( sSort.nOBSat==sSort.pOrderBy->nExpr ){
5560 sSort.pOrderBy = 0;
5564 /* If sorting index that was created by a prior OP_OpenEphemeral
5565 ** instruction ended up not being needed, then change the OP_OpenEphemeral
5566 ** into an OP_Noop.
5568 if( sSort.addrSortIndex>=0 && sSort.pOrderBy==0 ){
5569 sqlite3VdbeChangeToNoop(v, sSort.addrSortIndex);
5572 /* Use the standard inner loop. */
5573 assert( p->pEList==pEList );
5574 selectInnerLoop(pParse, p, -1, &sSort, &sDistinct, pDest,
5575 sqlite3WhereContinueLabel(pWInfo),
5576 sqlite3WhereBreakLabel(pWInfo));
5578 /* End the database scan loop.
5580 sqlite3WhereEnd(pWInfo);
5581 }else{
5582 /* This case when there exist aggregate functions or a GROUP BY clause
5583 ** or both */
5584 NameContext sNC; /* Name context for processing aggregate information */
5585 int iAMem; /* First Mem address for storing current GROUP BY */
5586 int iBMem; /* First Mem address for previous GROUP BY */
5587 int iUseFlag; /* Mem address holding flag indicating that at least
5588 ** one row of the input to the aggregator has been
5589 ** processed */
5590 int iAbortFlag; /* Mem address which causes query abort if positive */
5591 int groupBySort; /* Rows come from source in GROUP BY order */
5592 int addrEnd; /* End of processing for this SELECT */
5593 int sortPTab = 0; /* Pseudotable used to decode sorting results */
5594 int sortOut = 0; /* Output register from the sorter */
5595 int orderByGrp = 0; /* True if the GROUP BY and ORDER BY are the same */
5597 /* Remove any and all aliases between the result set and the
5598 ** GROUP BY clause.
5600 if( pGroupBy ){
5601 int k; /* Loop counter */
5602 struct ExprList_item *pItem; /* For looping over expression in a list */
5604 for(k=p->pEList->nExpr, pItem=p->pEList->a; k>0; k--, pItem++){
5605 pItem->u.x.iAlias = 0;
5607 for(k=pGroupBy->nExpr, pItem=pGroupBy->a; k>0; k--, pItem++){
5608 pItem->u.x.iAlias = 0;
5610 assert( 66==sqlite3LogEst(100) );
5611 if( p->nSelectRow>66 ) p->nSelectRow = 66;
5612 }else{
5613 assert( 0==sqlite3LogEst(1) );
5614 p->nSelectRow = 0;
5617 /* If there is both a GROUP BY and an ORDER BY clause and they are
5618 ** identical, then it may be possible to disable the ORDER BY clause
5619 ** on the grounds that the GROUP BY will cause elements to come out
5620 ** in the correct order. It also may not - the GROUP BY might use a
5621 ** database index that causes rows to be grouped together as required
5622 ** but not actually sorted. Either way, record the fact that the
5623 ** ORDER BY and GROUP BY clauses are the same by setting the orderByGrp
5624 ** variable. */
5625 if( sqlite3ExprListCompare(pGroupBy, sSort.pOrderBy, -1)==0 ){
5626 orderByGrp = 1;
5629 /* Create a label to jump to when we want to abort the query */
5630 addrEnd = sqlite3VdbeMakeLabel(v);
5632 /* Convert TK_COLUMN nodes into TK_AGG_COLUMN and make entries in
5633 ** sAggInfo for all TK_AGG_FUNCTION nodes in expressions of the
5634 ** SELECT statement.
5636 memset(&sNC, 0, sizeof(sNC));
5637 sNC.pParse = pParse;
5638 sNC.pSrcList = pTabList;
5639 sNC.pAggInfo = &sAggInfo;
5640 sAggInfo.mnReg = pParse->nMem+1;
5641 sAggInfo.nSortingColumn = pGroupBy ? pGroupBy->nExpr : 0;
5642 sAggInfo.pGroupBy = pGroupBy;
5643 sqlite3ExprAnalyzeAggList(&sNC, pEList);
5644 sqlite3ExprAnalyzeAggList(&sNC, sSort.pOrderBy);
5645 if( pHaving ){
5646 if( pGroupBy ){
5647 assert( pWhere==p->pWhere );
5648 havingToWhere(pParse, pGroupBy, pHaving, &p->pWhere);
5649 pWhere = p->pWhere;
5651 sqlite3ExprAnalyzeAggregates(&sNC, pHaving);
5653 sAggInfo.nAccumulator = sAggInfo.nColumn;
5654 for(i=0; i<sAggInfo.nFunc; i++){
5655 assert( !ExprHasProperty(sAggInfo.aFunc[i].pExpr, EP_xIsSelect) );
5656 sNC.ncFlags |= NC_InAggFunc;
5657 sqlite3ExprAnalyzeAggList(&sNC, sAggInfo.aFunc[i].pExpr->x.pList);
5658 sNC.ncFlags &= ~NC_InAggFunc;
5660 sAggInfo.mxReg = pParse->nMem;
5661 if( db->mallocFailed ) goto select_end;
5662 #if SELECTTRACE_ENABLED
5663 if( sqlite3SelectTrace & 0x400 ){
5664 int ii;
5665 SELECTTRACE(0x400,pParse,p,("After aggregate analysis:\n"));
5666 sqlite3TreeViewSelect(0, p, 0);
5667 for(ii=0; ii<sAggInfo.nColumn; ii++){
5668 sqlite3DebugPrintf("agg-column[%d] iMem=%d\n",
5669 ii, sAggInfo.aCol[ii].iMem);
5670 sqlite3TreeViewExpr(0, sAggInfo.aCol[ii].pExpr, 0);
5672 for(ii=0; ii<sAggInfo.nFunc; ii++){
5673 sqlite3DebugPrintf("agg-func[%d]: iMem=%d\n",
5674 ii, sAggInfo.aFunc[ii].iMem);
5675 sqlite3TreeViewExpr(0, sAggInfo.aFunc[ii].pExpr, 0);
5678 #endif
5681 /* Processing for aggregates with GROUP BY is very different and
5682 ** much more complex than aggregates without a GROUP BY.
5684 if( pGroupBy ){
5685 KeyInfo *pKeyInfo; /* Keying information for the group by clause */
5686 int addr1; /* A-vs-B comparision jump */
5687 int addrOutputRow; /* Start of subroutine that outputs a result row */
5688 int regOutputRow; /* Return address register for output subroutine */
5689 int addrSetAbort; /* Set the abort flag and return */
5690 int addrTopOfLoop; /* Top of the input loop */
5691 int addrSortingIdx; /* The OP_OpenEphemeral for the sorting index */
5692 int addrReset; /* Subroutine for resetting the accumulator */
5693 int regReset; /* Return address register for reset subroutine */
5695 /* If there is a GROUP BY clause we might need a sorting index to
5696 ** implement it. Allocate that sorting index now. If it turns out
5697 ** that we do not need it after all, the OP_SorterOpen instruction
5698 ** will be converted into a Noop.
5700 sAggInfo.sortingIdx = pParse->nTab++;
5701 pKeyInfo = keyInfoFromExprList(pParse, pGroupBy, 0, sAggInfo.nColumn);
5702 addrSortingIdx = sqlite3VdbeAddOp4(v, OP_SorterOpen,
5703 sAggInfo.sortingIdx, sAggInfo.nSortingColumn,
5704 0, (char*)pKeyInfo, P4_KEYINFO);
5706 /* Initialize memory locations used by GROUP BY aggregate processing
5708 iUseFlag = ++pParse->nMem;
5709 iAbortFlag = ++pParse->nMem;
5710 regOutputRow = ++pParse->nMem;
5711 addrOutputRow = sqlite3VdbeMakeLabel(v);
5712 regReset = ++pParse->nMem;
5713 addrReset = sqlite3VdbeMakeLabel(v);
5714 iAMem = pParse->nMem + 1;
5715 pParse->nMem += pGroupBy->nExpr;
5716 iBMem = pParse->nMem + 1;
5717 pParse->nMem += pGroupBy->nExpr;
5718 sqlite3VdbeAddOp2(v, OP_Integer, 0, iAbortFlag);
5719 VdbeComment((v, "clear abort flag"));
5720 sqlite3VdbeAddOp2(v, OP_Integer, 0, iUseFlag);
5721 VdbeComment((v, "indicate accumulator empty"));
5722 sqlite3VdbeAddOp3(v, OP_Null, 0, iAMem, iAMem+pGroupBy->nExpr-1);
5724 /* Begin a loop that will extract all source rows in GROUP BY order.
5725 ** This might involve two separate loops with an OP_Sort in between, or
5726 ** it might be a single loop that uses an index to extract information
5727 ** in the right order to begin with.
5729 sqlite3VdbeAddOp2(v, OP_Gosub, regReset, addrReset);
5730 pWInfo = sqlite3WhereBegin(pParse, pTabList, pWhere, pGroupBy, 0,
5731 WHERE_GROUPBY | (orderByGrp ? WHERE_SORTBYGROUP : 0), 0
5733 if( pWInfo==0 ) goto select_end;
5734 if( sqlite3WhereIsOrdered(pWInfo)==pGroupBy->nExpr ){
5735 /* The optimizer is able to deliver rows in group by order so
5736 ** we do not have to sort. The OP_OpenEphemeral table will be
5737 ** cancelled later because we still need to use the pKeyInfo
5739 groupBySort = 0;
5740 }else{
5741 /* Rows are coming out in undetermined order. We have to push
5742 ** each row into a sorting index, terminate the first loop,
5743 ** then loop over the sorting index in order to get the output
5744 ** in sorted order
5746 int regBase;
5747 int regRecord;
5748 int nCol;
5749 int nGroupBy;
5751 explainTempTable(pParse,
5752 (sDistinct.isTnct && (p->selFlags&SF_Distinct)==0) ?
5753 "DISTINCT" : "GROUP BY");
5755 groupBySort = 1;
5756 nGroupBy = pGroupBy->nExpr;
5757 nCol = nGroupBy;
5758 j = nGroupBy;
5759 for(i=0; i<sAggInfo.nColumn; i++){
5760 if( sAggInfo.aCol[i].iSorterColumn>=j ){
5761 nCol++;
5762 j++;
5765 regBase = sqlite3GetTempRange(pParse, nCol);
5766 sqlite3ExprCacheClear(pParse);
5767 sqlite3ExprCodeExprList(pParse, pGroupBy, regBase, 0, 0);
5768 j = nGroupBy;
5769 for(i=0; i<sAggInfo.nColumn; i++){
5770 struct AggInfo_col *pCol = &sAggInfo.aCol[i];
5771 if( pCol->iSorterColumn>=j ){
5772 int r1 = j + regBase;
5773 sqlite3ExprCodeGetColumnToReg(pParse,
5774 pCol->pTab, pCol->iColumn, pCol->iTable, r1);
5775 j++;
5778 regRecord = sqlite3GetTempReg(pParse);
5779 sqlite3VdbeAddOp3(v, OP_MakeRecord, regBase, nCol, regRecord);
5780 sqlite3VdbeAddOp2(v, OP_SorterInsert, sAggInfo.sortingIdx, regRecord);
5781 sqlite3ReleaseTempReg(pParse, regRecord);
5782 sqlite3ReleaseTempRange(pParse, regBase, nCol);
5783 sqlite3WhereEnd(pWInfo);
5784 sAggInfo.sortingIdxPTab = sortPTab = pParse->nTab++;
5785 sortOut = sqlite3GetTempReg(pParse);
5786 sqlite3VdbeAddOp3(v, OP_OpenPseudo, sortPTab, sortOut, nCol);
5787 sqlite3VdbeAddOp2(v, OP_SorterSort, sAggInfo.sortingIdx, addrEnd);
5788 VdbeComment((v, "GROUP BY sort")); VdbeCoverage(v);
5789 sAggInfo.useSortingIdx = 1;
5790 sqlite3ExprCacheClear(pParse);
5794 /* If the index or temporary table used by the GROUP BY sort
5795 ** will naturally deliver rows in the order required by the ORDER BY
5796 ** clause, cancel the ephemeral table open coded earlier.
5798 ** This is an optimization - the correct answer should result regardless.
5799 ** Use the SQLITE_GroupByOrder flag with SQLITE_TESTCTRL_OPTIMIZER to
5800 ** disable this optimization for testing purposes. */
5801 if( orderByGrp && OptimizationEnabled(db, SQLITE_GroupByOrder)
5802 && (groupBySort || sqlite3WhereIsSorted(pWInfo))
5804 sSort.pOrderBy = 0;
5805 sqlite3VdbeChangeToNoop(v, sSort.addrSortIndex);
5808 /* Evaluate the current GROUP BY terms and store in b0, b1, b2...
5809 ** (b0 is memory location iBMem+0, b1 is iBMem+1, and so forth)
5810 ** Then compare the current GROUP BY terms against the GROUP BY terms
5811 ** from the previous row currently stored in a0, a1, a2...
5813 addrTopOfLoop = sqlite3VdbeCurrentAddr(v);
5814 sqlite3ExprCacheClear(pParse);
5815 if( groupBySort ){
5816 sqlite3VdbeAddOp3(v, OP_SorterData, sAggInfo.sortingIdx,
5817 sortOut, sortPTab);
5819 for(j=0; j<pGroupBy->nExpr; j++){
5820 if( groupBySort ){
5821 sqlite3VdbeAddOp3(v, OP_Column, sortPTab, j, iBMem+j);
5822 }else{
5823 sAggInfo.directMode = 1;
5824 sqlite3ExprCode(pParse, pGroupBy->a[j].pExpr, iBMem+j);
5827 sqlite3VdbeAddOp4(v, OP_Compare, iAMem, iBMem, pGroupBy->nExpr,
5828 (char*)sqlite3KeyInfoRef(pKeyInfo), P4_KEYINFO);
5829 addr1 = sqlite3VdbeCurrentAddr(v);
5830 sqlite3VdbeAddOp3(v, OP_Jump, addr1+1, 0, addr1+1); VdbeCoverage(v);
5832 /* Generate code that runs whenever the GROUP BY changes.
5833 ** Changes in the GROUP BY are detected by the previous code
5834 ** block. If there were no changes, this block is skipped.
5836 ** This code copies current group by terms in b0,b1,b2,...
5837 ** over to a0,a1,a2. It then calls the output subroutine
5838 ** and resets the aggregate accumulator registers in preparation
5839 ** for the next GROUP BY batch.
5841 sqlite3ExprCodeMove(pParse, iBMem, iAMem, pGroupBy->nExpr);
5842 sqlite3VdbeAddOp2(v, OP_Gosub, regOutputRow, addrOutputRow);
5843 VdbeComment((v, "output one row"));
5844 sqlite3VdbeAddOp2(v, OP_IfPos, iAbortFlag, addrEnd); VdbeCoverage(v);
5845 VdbeComment((v, "check abort flag"));
5846 sqlite3VdbeAddOp2(v, OP_Gosub, regReset, addrReset);
5847 VdbeComment((v, "reset accumulator"));
5849 /* Update the aggregate accumulators based on the content of
5850 ** the current row
5852 sqlite3VdbeJumpHere(v, addr1);
5853 updateAccumulator(pParse, &sAggInfo);
5854 sqlite3VdbeAddOp2(v, OP_Integer, 1, iUseFlag);
5855 VdbeComment((v, "indicate data in accumulator"));
5857 /* End of the loop
5859 if( groupBySort ){
5860 sqlite3VdbeAddOp2(v, OP_SorterNext, sAggInfo.sortingIdx, addrTopOfLoop);
5861 VdbeCoverage(v);
5862 }else{
5863 sqlite3WhereEnd(pWInfo);
5864 sqlite3VdbeChangeToNoop(v, addrSortingIdx);
5867 /* Output the final row of result
5869 sqlite3VdbeAddOp2(v, OP_Gosub, regOutputRow, addrOutputRow);
5870 VdbeComment((v, "output final row"));
5872 /* Jump over the subroutines
5874 sqlite3VdbeGoto(v, addrEnd);
5876 /* Generate a subroutine that outputs a single row of the result
5877 ** set. This subroutine first looks at the iUseFlag. If iUseFlag
5878 ** is less than or equal to zero, the subroutine is a no-op. If
5879 ** the processing calls for the query to abort, this subroutine
5880 ** increments the iAbortFlag memory location before returning in
5881 ** order to signal the caller to abort.
5883 addrSetAbort = sqlite3VdbeCurrentAddr(v);
5884 sqlite3VdbeAddOp2(v, OP_Integer, 1, iAbortFlag);
5885 VdbeComment((v, "set abort flag"));
5886 sqlite3VdbeAddOp1(v, OP_Return, regOutputRow);
5887 sqlite3VdbeResolveLabel(v, addrOutputRow);
5888 addrOutputRow = sqlite3VdbeCurrentAddr(v);
5889 sqlite3VdbeAddOp2(v, OP_IfPos, iUseFlag, addrOutputRow+2);
5890 VdbeCoverage(v);
5891 VdbeComment((v, "Groupby result generator entry point"));
5892 sqlite3VdbeAddOp1(v, OP_Return, regOutputRow);
5893 finalizeAggFunctions(pParse, &sAggInfo);
5894 sqlite3ExprIfFalse(pParse, pHaving, addrOutputRow+1, SQLITE_JUMPIFNULL);
5895 selectInnerLoop(pParse, p, -1, &sSort,
5896 &sDistinct, pDest,
5897 addrOutputRow+1, addrSetAbort);
5898 sqlite3VdbeAddOp1(v, OP_Return, regOutputRow);
5899 VdbeComment((v, "end groupby result generator"));
5901 /* Generate a subroutine that will reset the group-by accumulator
5903 sqlite3VdbeResolveLabel(v, addrReset);
5904 resetAccumulator(pParse, &sAggInfo);
5905 sqlite3VdbeAddOp1(v, OP_Return, regReset);
5907 } /* endif pGroupBy. Begin aggregate queries without GROUP BY: */
5908 else {
5909 ExprList *pDel = 0;
5910 #ifndef SQLITE_OMIT_BTREECOUNT
5911 Table *pTab;
5912 if( (pTab = isSimpleCount(p, &sAggInfo))!=0 ){
5913 /* If isSimpleCount() returns a pointer to a Table structure, then
5914 ** the SQL statement is of the form:
5916 ** SELECT count(*) FROM <tbl>
5918 ** where the Table structure returned represents table <tbl>.
5920 ** This statement is so common that it is optimized specially. The
5921 ** OP_Count instruction is executed either on the intkey table that
5922 ** contains the data for table <tbl> or on one of its indexes. It
5923 ** is better to execute the op on an index, as indexes are almost
5924 ** always spread across less pages than their corresponding tables.
5926 const int iDb = sqlite3SchemaToIndex(pParse->db, pTab->pSchema);
5927 const int iCsr = pParse->nTab++; /* Cursor to scan b-tree */
5928 Index *pIdx; /* Iterator variable */
5929 KeyInfo *pKeyInfo = 0; /* Keyinfo for scanned index */
5930 Index *pBest = 0; /* Best index found so far */
5931 int iRoot = pTab->tnum; /* Root page of scanned b-tree */
5933 sqlite3CodeVerifySchema(pParse, iDb);
5934 sqlite3TableLock(pParse, iDb, pTab->tnum, 0, pTab->zName);
5936 /* Search for the index that has the lowest scan cost.
5938 ** (2011-04-15) Do not do a full scan of an unordered index.
5940 ** (2013-10-03) Do not count the entries in a partial index.
5942 ** In practice the KeyInfo structure will not be used. It is only
5943 ** passed to keep OP_OpenRead happy.
5945 if( !HasRowid(pTab) ) pBest = sqlite3PrimaryKeyIndex(pTab);
5946 for(pIdx=pTab->pIndex; pIdx; pIdx=pIdx->pNext){
5947 if( pIdx->bUnordered==0
5948 && pIdx->szIdxRow<pTab->szTabRow
5949 && pIdx->pPartIdxWhere==0
5950 && (!pBest || pIdx->szIdxRow<pBest->szIdxRow)
5952 pBest = pIdx;
5955 if( pBest ){
5956 iRoot = pBest->tnum;
5957 pKeyInfo = sqlite3KeyInfoOfIndex(pParse, pBest);
5960 /* Open a read-only cursor, execute the OP_Count, close the cursor. */
5961 sqlite3VdbeAddOp4Int(v, OP_OpenRead, iCsr, iRoot, iDb, 1);
5962 if( pKeyInfo ){
5963 sqlite3VdbeChangeP4(v, -1, (char *)pKeyInfo, P4_KEYINFO);
5965 sqlite3VdbeAddOp2(v, OP_Count, iCsr, sAggInfo.aFunc[0].iMem);
5966 sqlite3VdbeAddOp1(v, OP_Close, iCsr);
5967 explainSimpleCount(pParse, pTab, pBest);
5968 }else
5969 #endif /* SQLITE_OMIT_BTREECOUNT */
5971 /* Check if the query is of one of the following forms:
5973 ** SELECT min(x) FROM ...
5974 ** SELECT max(x) FROM ...
5976 ** If it is, then ask the code in where.c to attempt to sort results
5977 ** as if there was an "ORDER BY x" or "ORDER BY x DESC" clause.
5978 ** If where.c is able to produce results sorted in this order, then
5979 ** add vdbe code to break out of the processing loop after the
5980 ** first iteration (since the first iteration of the loop is
5981 ** guaranteed to operate on the row with the minimum or maximum
5982 ** value of x, the only row required).
5984 ** A special flag must be passed to sqlite3WhereBegin() to slightly
5985 ** modify behavior as follows:
5987 ** + If the query is a "SELECT min(x)", then the loop coded by
5988 ** where.c should not iterate over any values with a NULL value
5989 ** for x.
5991 ** + The optimizer code in where.c (the thing that decides which
5992 ** index or indices to use) should place a different priority on
5993 ** satisfying the 'ORDER BY' clause than it does in other cases.
5994 ** Refer to code and comments in where.c for details.
5996 ExprList *pMinMax = 0;
5997 u8 flag = WHERE_ORDERBY_NORMAL;
5999 assert( p->pGroupBy==0 );
6000 assert( flag==0 );
6001 if( p->pHaving==0 ){
6002 flag = minMaxQuery(&sAggInfo, &pMinMax);
6004 assert( flag==0 || (pMinMax!=0 && pMinMax->nExpr==1) );
6006 if( flag ){
6007 pMinMax = sqlite3ExprListDup(db, pMinMax, 0);
6008 pDel = pMinMax;
6009 assert( db->mallocFailed || pMinMax!=0 );
6010 if( !db->mallocFailed ){
6011 pMinMax->a[0].sortOrder = flag!=WHERE_ORDERBY_MIN ?1:0;
6012 pMinMax->a[0].pExpr->op = TK_COLUMN;
6016 /* This case runs if the aggregate has no GROUP BY clause. The
6017 ** processing is much simpler since there is only a single row
6018 ** of output.
6020 resetAccumulator(pParse, &sAggInfo);
6021 pWInfo = sqlite3WhereBegin(pParse, pTabList, pWhere, pMinMax, 0,flag,0);
6022 if( pWInfo==0 ){
6023 sqlite3ExprListDelete(db, pDel);
6024 goto select_end;
6026 updateAccumulator(pParse, &sAggInfo);
6027 assert( pMinMax==0 || pMinMax->nExpr==1 );
6028 if( sqlite3WhereIsOrdered(pWInfo)>0 ){
6029 sqlite3VdbeGoto(v, sqlite3WhereBreakLabel(pWInfo));
6030 VdbeComment((v, "%s() by index",
6031 (flag==WHERE_ORDERBY_MIN?"min":"max")));
6033 sqlite3WhereEnd(pWInfo);
6034 finalizeAggFunctions(pParse, &sAggInfo);
6037 sSort.pOrderBy = 0;
6038 sqlite3ExprIfFalse(pParse, pHaving, addrEnd, SQLITE_JUMPIFNULL);
6039 selectInnerLoop(pParse, p, -1, 0, 0,
6040 pDest, addrEnd, addrEnd);
6041 sqlite3ExprListDelete(db, pDel);
6043 sqlite3VdbeResolveLabel(v, addrEnd);
6045 } /* endif aggregate query */
6047 if( sDistinct.eTnctType==WHERE_DISTINCT_UNORDERED ){
6048 explainTempTable(pParse, "DISTINCT");
6051 /* If there is an ORDER BY clause, then we need to sort the results
6052 ** and send them to the callback one by one.
6054 if( sSort.pOrderBy ){
6055 explainTempTable(pParse,
6056 sSort.nOBSat>0 ? "RIGHT PART OF ORDER BY":"ORDER BY");
6057 generateSortTail(pParse, p, &sSort, pEList->nExpr, pDest);
6060 /* Jump here to skip this query
6062 sqlite3VdbeResolveLabel(v, iEnd);
6064 /* The SELECT has been coded. If there is an error in the Parse structure,
6065 ** set the return code to 1. Otherwise 0. */
6066 rc = (pParse->nErr>0);
6068 /* Control jumps to here if an error is encountered above, or upon
6069 ** successful coding of the SELECT.
6071 select_end:
6072 explainSetInteger(pParse->iSelectId, iRestoreSelectId);
6074 sqlite3DbFree(db, sAggInfo.aCol);
6075 sqlite3DbFree(db, sAggInfo.aFunc);
6076 #if SELECTTRACE_ENABLED
6077 SELECTTRACE(1,pParse,p,("end processing\n"));
6078 pParse->nSelectIndent--;
6079 #endif
6080 return rc;