aboutsummaryrefslogtreecommitdiff
path: root/src/where.c
diff options
context:
space:
mode:
Diffstat (limited to 'src/where.c')
-rw-r--r--src/where.c293
1 files changed, 247 insertions, 46 deletions
diff --git a/src/where.c b/src/where.c
index e95fd57f3..58e35384b 100644
--- a/src/where.c
+++ b/src/where.c
@@ -772,6 +772,79 @@ static void markTermAsChild(WhereClause *pWC, int iChild, int iParent){
pWC->a[iParent].nChild++;
}
+/*
+** Return the N-th AND-connected subterm of pTerm. Or if pTerm is not
+** a conjunction, then return just pTerm when N==0. If N is exceeds
+** the number of available subterms, return NULL.
+*/
+static WhereTerm *whereNthSubterm(WhereTerm *pTerm, int N){
+ if( pTerm->eOperator!=WO_AND ){
+ return N==0 ? pTerm : 0;
+ }
+ if( N<pTerm->u.pAndInfo->wc.nTerm ){
+ return &pTerm->u.pAndInfo->wc.a[N];
+ }
+ return 0;
+}
+
+/*
+** Subterms pOne and pTwo are contained within WHERE clause pWC. The
+** two subterms are in disjunction - they are OR-ed together.
+**
+** If these two terms are both of the form: "A op B" with the same
+** A and B values but different operators and if the operators are
+** compatible (if one is = and the other is <, for example) then
+** add a new virtual AND term to pWC that is the combination of the
+** two.
+**
+** Some examples:
+**
+** x<y OR x=y --> x<=y
+** x=y OR x=y --> x=y
+** x<=y OR x<y --> x<=y
+**
+** The following is NOT generated:
+**
+** x<y OR x>y --> x!=y
+*/
+static void whereCombineDisjuncts(
+ SrcList *pSrc, /* the FROM clause */
+ WhereClause *pWC, /* The complete WHERE clause */
+ WhereTerm *pOne, /* First disjunct */
+ WhereTerm *pTwo /* Second disjunct */
+){
+ u16 eOp = pOne->eOperator | pTwo->eOperator;
+ sqlite3 *db; /* Database connection (for malloc) */
+ Expr *pNew; /* New virtual expression */
+ int op; /* Operator for the combined expression */
+ int idxNew; /* Index in pWC of the next virtual term */
+
+ if( (pOne->eOperator & (WO_EQ|WO_LT|WO_LE|WO_GT|WO_GE))==0 ) return;
+ if( (pTwo->eOperator & (WO_EQ|WO_LT|WO_LE|WO_GT|WO_GE))==0 ) return;
+ if( (eOp & (WO_EQ|WO_LT|WO_LE))!=eOp
+ && (eOp & (WO_EQ|WO_GT|WO_GE))!=eOp ) return;
+ assert( pOne->pExpr->pLeft!=0 && pOne->pExpr->pRight!=0 );
+ assert( pTwo->pExpr->pLeft!=0 && pTwo->pExpr->pRight!=0 );
+ if( sqlite3ExprCompare(pOne->pExpr->pLeft, pTwo->pExpr->pLeft, -1) ) return;
+ if( sqlite3ExprCompare(pOne->pExpr->pRight, pTwo->pExpr->pRight, -1) )return;
+ /* If we reach this point, it means the two subterms can be combined */
+ if( (eOp & (eOp-1))!=0 ){
+ if( eOp & (WO_LT|WO_LE) ){
+ eOp = WO_LE;
+ }else{
+ assert( eOp & (WO_GT|WO_GE) );
+ eOp = WO_GE;
+ }
+ }
+ db = pWC->pWInfo->pParse->db;
+ pNew = sqlite3ExprDup(db, pOne->pExpr, 0);
+ if( pNew==0 ) return;
+ for(op=TK_EQ; eOp!=(WO_EQ<<(op-TK_EQ)); op++){ assert( op<TK_GE ); }
+ pNew->op = op;
+ idxNew = whereClauseInsert(pWC, pNew, TERM_VIRTUAL|TERM_DYNAMIC);
+ exprAnalyze(pSrc, pWC, idxNew);
+}
+
#if !defined(SQLITE_OMIT_OR_OPTIMIZATION) && !defined(SQLITE_OMIT_SUBQUERY)
/*
** Analyze a term that consists of two or more OR-connected
@@ -796,6 +869,7 @@ static void markTermAsChild(WhereClause *pWC, int iChild, int iParent){
** (C) t1.x=t2.y OR (t1.x=t2.z AND t1.y=15)
** (D) x=expr1 OR (y>11 AND y<22 AND z LIKE '*hello*')
** (E) (p.a=1 AND q.b=2 AND r.c=3) OR (p.x=4 AND q.y=5 AND r.z=6)
+** (F) x>A OR (x=A AND y>=B)
**
** CASE 1:
**
@@ -812,6 +886,16 @@ static void markTermAsChild(WhereClause *pWC, int iChild, int iParent){
**
** CASE 2:
**
+** If there are exactly two disjuncts one side has x>A and the other side
+** has x=A (for the same x and A) then add a new virtual conjunct term to the
+** WHERE clause of the form "x>=A". Example:
+**
+** x>A OR (x=A AND y>B) adds: x>=A
+**
+** The added conjunct can sometimes be helpful in query planning.
+**
+** CASE 3:
+**
** If all subterms are indexable by a single table T, then set
**
** WhereTerm.eOperator = WO_OR
@@ -938,12 +1022,26 @@ static void exprAnalyzeOrTerm(
}
/*
- ** Record the set of tables that satisfy case 2. The set might be
+ ** Record the set of tables that satisfy case 3. The set might be
** empty.
*/
pOrInfo->indexable = indexable;
pTerm->eOperator = indexable==0 ? 0 : WO_OR;
+ /* For a two-way OR, attempt to implementation case 2.
+ */
+ if( indexable && pOrWc->nTerm==2 ){
+ int iOne = 0;
+ WhereTerm *pOne;
+ while( (pOne = whereNthSubterm(&pOrWc->a[0],iOne++))!=0 ){
+ int iTwo = 0;
+ WhereTerm *pTwo;
+ while( (pTwo = whereNthSubterm(&pOrWc->a[1],iTwo++))!=0 ){
+ whereCombineDisjuncts(pSrc, pWC, pOne, pTwo);
+ }
+ }
+ }
+
/*
** chngToIN holds a set of tables that *might* satisfy case 1. But
** we have to do some additional checking to see if case 1 really
@@ -1073,7 +1171,7 @@ static void exprAnalyzeOrTerm(
}else{
sqlite3ExprListDelete(db, pList);
}
- pTerm->eOperator = WO_NOOP; /* case 1 trumps case 2 */
+ pTerm->eOperator = WO_NOOP; /* case 1 trumps case 3 */
}
}
}
@@ -1268,7 +1366,7 @@ static void exprAnalyze(
Expr *pNewExpr2;
int idxNew1;
int idxNew2;
- Token sCollSeqName; /* Name of collating sequence */
+ const char *zCollSeqName; /* Name of collating sequence */
const u16 wtFlags = TERM_LIKEOPT | TERM_VIRTUAL | TERM_DYNAMIC;
pLeft = pExpr->x.pList->a[1].pExpr;
@@ -1304,11 +1402,10 @@ static void exprAnalyze(
}
*pC = c + 1;
}
- sCollSeqName.z = noCase ? "NOCASE" : "BINARY";
- sCollSeqName.n = 6;
+ zCollSeqName = noCase ? "NOCASE" : "BINARY";
pNewExpr1 = sqlite3ExprDup(db, pLeft, 0);
pNewExpr1 = sqlite3PExpr(pParse, TK_GE,
- sqlite3ExprAddCollateToken(pParse,pNewExpr1,&sCollSeqName),
+ sqlite3ExprAddCollateString(pParse,pNewExpr1,zCollSeqName),
pStr1, 0);
transferJoinMarkings(pNewExpr1, pExpr);
idxNew1 = whereClauseInsert(pWC, pNewExpr1, wtFlags);
@@ -1316,7 +1413,7 @@ static void exprAnalyze(
exprAnalyze(pSrc, pWC, idxNew1);
pNewExpr2 = sqlite3ExprDup(db, pLeft, 0);
pNewExpr2 = sqlite3PExpr(pParse, TK_LT,
- sqlite3ExprAddCollateToken(pParse,pNewExpr2,&sCollSeqName),
+ sqlite3ExprAddCollateString(pParse,pNewExpr2,zCollSeqName),
pStr2, 0);
transferJoinMarkings(pNewExpr2, pExpr);
idxNew2 = whereClauseInsert(pWC, pNewExpr2, wtFlags);
@@ -1933,11 +2030,14 @@ static int vtabBestIndex(Parse *pParse, Table *pTab, sqlite3_index_info *p){
** Estimate the location of a particular key among all keys in an
** index. Store the results in aStat as follows:
**
-** aStat[0] Est. number of rows less than pVal
-** aStat[1] Est. number of rows equal to pVal
+** aStat[0] Est. number of rows less than pRec
+** aStat[1] Est. number of rows equal to pRec
**
** Return the index of the sample that is the smallest sample that
-** is greater than or equal to pRec.
+** is greater than or equal to pRec. Note that this index is not an index
+** into the aSample[] array - it is an index into a virtual set of samples
+** based on the contents of aSample[] and the number of fields in record
+** pRec.
*/
static int whereKeyStats(
Parse *pParse, /* Database connection */
@@ -1948,67 +2048,158 @@ static int whereKeyStats(
){
IndexSample *aSample = pIdx->aSample;
int iCol; /* Index of required stats in anEq[] etc. */
+ int i; /* Index of first sample >= pRec */
+ int iSample; /* Smallest sample larger than or equal to pRec */
int iMin = 0; /* Smallest sample not yet tested */
- int i = pIdx->nSample; /* Smallest sample larger than or equal to pRec */
int iTest; /* Next sample to test */
int res; /* Result of comparison operation */
+ int nField; /* Number of fields in pRec */
+ tRowcnt iLower = 0; /* anLt[] + anEq[] of largest sample pRec is > */
#ifndef SQLITE_DEBUG
UNUSED_PARAMETER( pParse );
#endif
assert( pRec!=0 );
- iCol = pRec->nField - 1;
assert( pIdx->nSample>0 );
- assert( pRec->nField>0 && iCol<pIdx->nSampleCol );
+ assert( pRec->nField>0 && pRec->nField<=pIdx->nSampleCol );
+
+ /* Do a binary search to find the first sample greater than or equal
+ ** to pRec. If pRec contains a single field, the set of samples to search
+ ** is simply the aSample[] array. If the samples in aSample[] contain more
+ ** than one fields, all fields following the first are ignored.
+ **
+ ** If pRec contains N fields, where N is more than one, then as well as the
+ ** samples in aSample[] (truncated to N fields), the search also has to
+ ** consider prefixes of those samples. For example, if the set of samples
+ ** in aSample is:
+ **
+ ** aSample[0] = (a, 5)
+ ** aSample[1] = (a, 10)
+ ** aSample[2] = (b, 5)
+ ** aSample[3] = (c, 100)
+ ** aSample[4] = (c, 105)
+ **
+ ** Then the search space should ideally be the samples above and the
+ ** unique prefixes [a], [b] and [c]. But since that is hard to organize,
+ ** the code actually searches this set:
+ **
+ ** 0: (a)
+ ** 1: (a, 5)
+ ** 2: (a, 10)
+ ** 3: (a, 10)
+ ** 4: (b)
+ ** 5: (b, 5)
+ ** 6: (c)
+ ** 7: (c, 100)
+ ** 8: (c, 105)
+ ** 9: (c, 105)
+ **
+ ** For each sample in the aSample[] array, N samples are present in the
+ ** effective sample array. In the above, samples 0 and 1 are based on
+ ** sample aSample[0]. Samples 2 and 3 on aSample[1] etc.
+ **
+ ** Often, sample i of each block of N effective samples has (i+1) fields.
+ ** Except, each sample may be extended to ensure that it is greater than or
+ ** equal to the previous sample in the array. For example, in the above,
+ ** sample 2 is the first sample of a block of N samples, so at first it
+ ** appears that it should be 1 field in size. However, that would make it
+ ** smaller than sample 1, so the binary search would not work. As a result,
+ ** it is extended to two fields. The duplicates that this creates do not
+ ** cause any problems.
+ */
+ nField = pRec->nField;
+ iCol = 0;
+ iSample = pIdx->nSample * nField;
do{
- iTest = (iMin+i)/2;
- res = sqlite3VdbeRecordCompare(aSample[iTest].n, aSample[iTest].p, pRec);
+ int iSamp; /* Index in aSample[] of test sample */
+ int n; /* Number of fields in test sample */
+
+ iTest = (iMin+iSample)/2;
+ iSamp = iTest / nField;
+ if( iSamp>0 ){
+ /* The proposed effective sample is a prefix of sample aSample[iSamp].
+ ** Specifically, the shortest prefix of at least (1 + iTest%nField)
+ ** fields that is greater than the previous effective sample. */
+ for(n=(iTest % nField) + 1; n<nField; n++){
+ if( aSample[iSamp-1].anLt[n-1]!=aSample[iSamp].anLt[n-1] ) break;
+ }
+ }else{
+ n = iTest + 1;
+ }
+
+ pRec->nField = n;
+ res = sqlite3VdbeRecordCompare(aSample[iSamp].n, aSample[iSamp].p, pRec);
if( res<0 ){
+ iLower = aSample[iSamp].anLt[n-1] + aSample[iSamp].anEq[n-1];
+ iMin = iTest+1;
+ }else if( res==0 && n<nField ){
+ iLower = aSample[iSamp].anLt[n-1];
iMin = iTest+1;
+ res = -1;
}else{
- i = iTest;
+ iSample = iTest;
+ iCol = n-1;
}
- }while( res && iMin<i );
+ }while( res && iMin<iSample );
+ i = iSample / nField;
#ifdef SQLITE_DEBUG
/* The following assert statements check that the binary search code
** above found the right answer. This block serves no purpose other
** than to invoke the asserts. */
- if( res==0 ){
- /* If (res==0) is true, then sample $i must be equal to pRec */
- assert( i<pIdx->nSample );
- assert( 0==sqlite3VdbeRecordCompare(aSample[i].n, aSample[i].p, pRec)
- || pParse->db->mallocFailed );
- }else{
- /* Otherwise, pRec must be smaller than sample $i and larger than
- ** sample ($i-1). */
- assert( i==pIdx->nSample
- || sqlite3VdbeRecordCompare(aSample[i].n, aSample[i].p, pRec)>0
- || pParse->db->mallocFailed );
- assert( i==0
- || sqlite3VdbeRecordCompare(aSample[i-1].n, aSample[i-1].p, pRec)<0
- || pParse->db->mallocFailed );
+ if( pParse->db->mallocFailed==0 ){
+ if( res==0 ){
+ /* If (res==0) is true, then pRec must be equal to sample i. */
+ assert( i<pIdx->nSample );
+ assert( iCol==nField-1 );
+ pRec->nField = nField;
+ assert( 0==sqlite3VdbeRecordCompare(aSample[i].n, aSample[i].p, pRec)
+ || pParse->db->mallocFailed
+ );
+ }else{
+ /* Unless i==pIdx->nSample, indicating that pRec is larger than
+ ** all samples in the aSample[] array, pRec must be smaller than the
+ ** (iCol+1) field prefix of sample i. */
+ assert( i<=pIdx->nSample && i>=0 );
+ pRec->nField = iCol+1;
+ assert( i==pIdx->nSample
+ || sqlite3VdbeRecordCompare(aSample[i].n, aSample[i].p, pRec)>0
+ || pParse->db->mallocFailed );
+
+ /* if i==0 and iCol==0, then record pRec is smaller than all samples
+ ** in the aSample[] array. Otherwise, if (iCol>0) then pRec must
+ ** be greater than or equal to the (iCol) field prefix of sample i.
+ ** If (i>0), then pRec must also be greater than sample (i-1). */
+ if( iCol>0 ){
+ pRec->nField = iCol;
+ assert( sqlite3VdbeRecordCompare(aSample[i].n, aSample[i].p, pRec)<=0
+ || pParse->db->mallocFailed );
+ }
+ if( i>0 ){
+ pRec->nField = nField;
+ assert( sqlite3VdbeRecordCompare(aSample[i-1].n, aSample[i-1].p, pRec)<0
+ || pParse->db->mallocFailed );
+ }
+ }
}
#endif /* ifdef SQLITE_DEBUG */
- /* At this point, aSample[i] is the first sample that is greater than
- ** or equal to pVal. Or if i==pIdx->nSample, then all samples are less
- ** than pVal. If aSample[i]==pVal, then res==0.
- */
if( res==0 ){
+ /* Record pRec is equal to sample i */
+ assert( iCol==nField-1 );
aStat[0] = aSample[i].anLt[iCol];
aStat[1] = aSample[i].anEq[iCol];
}else{
- tRowcnt iLower, iUpper, iGap;
- if( i==0 ){
- iLower = 0;
- iUpper = aSample[0].anLt[iCol];
+ /* At this point, the (iCol+1) field prefix of aSample[i] is the first
+ ** sample that is greater than pRec. Or, if i==pIdx->nSample then pRec
+ ** is larger than all samples in the array. */
+ tRowcnt iUpper, iGap;
+ if( i>=pIdx->nSample ){
+ iUpper = sqlite3LogEstToInt(pIdx->aiRowLogEst[0]);
}else{
- i64 nRow0 = sqlite3LogEstToInt(pIdx->aiRowLogEst[0]);
- iUpper = i>=pIdx->nSample ? nRow0 : aSample[i].anLt[iCol];
- iLower = aSample[i-1].anEq[iCol] + aSample[i-1].anLt[iCol];
+ iUpper = aSample[i].anLt[iCol];
}
- aStat[1] = pIdx->aAvgEq[iCol];
+
if( iLower>=iUpper ){
iGap = 0;
}else{
@@ -2020,7 +2211,11 @@ static int whereKeyStats(
iGap = iGap/3;
}
aStat[0] = iLower + iGap;
+ aStat[1] = pIdx->aAvgEq[iCol];
}
+
+ /* Restore the pRec->nField value before returning. */
+ pRec->nField = nField;
return i;
}
#endif /* SQLITE_ENABLE_STAT3_OR_STAT4 */
@@ -5807,10 +6002,10 @@ static int wherePathSolver(WhereInfo *pWInfo, LogEst nRowEst){
/* Seed the search with a single WherePath containing zero WhereLoops.
**
- ** TUNING: Do not let the number of iterations go above 25. If the cost
- ** of computing an automatic index is not paid back within the first 25
+ ** TUNING: Do not let the number of iterations go above 28. If the cost
+ ** of computing an automatic index is not paid back within the first 28
** rows, then do not use the automatic index. */
- aFrom[0].nRow = MIN(pParse->nQueryLoop, 46); assert( 46==sqlite3LogEst(25) );
+ aFrom[0].nRow = MIN(pParse->nQueryLoop, 48); assert( 48==sqlite3LogEst(28) );
nFrom = 1;
assert( aFrom[0].isOrdered==0 );
if( nOrderBy ){
@@ -6610,6 +6805,12 @@ WhereInfo *sqlite3WhereBegin(
if( op ){
sqlite3VdbeAddOp3(v, op, iIndexCur, pIx->tnum, iDb);
sqlite3VdbeSetP4KeyInfo(pParse, pIx);
+ if( (pLoop->wsFlags & WHERE_CONSTRAINT)!=0
+ && (pLoop->wsFlags & (WHERE_COLUMN_RANGE|WHERE_SKIPSCAN))==0
+ && (pWInfo->wctrlFlags&WHERE_ORDERBY_MIN)==0
+ ){
+ sqlite3VdbeChangeP5(v, OPFLAG_SEEKEQ); /* Hint to COMDB2 */
+ }
VdbeComment((v, "%s", pIx->zName));
}
}
generated by cgit v1.2.3 (git 2.25.1) at 2025-07-22 03:07:08 +0000