aboutsummaryrefslogtreecommitdiff
path: root/src/where.c
diff options
context:
space:
mode:
Diffstat (limited to 'src/where.c')
-rw-r--r--src/where.c52
1 files changed, 34 insertions, 18 deletions
diff --git a/src/where.c b/src/where.c
index e1e1e1d52..5b990fc10 100644
--- a/src/where.c
+++ b/src/where.c
@@ -701,7 +701,7 @@ static int isLikeOrGlob(
** value of the variable means there is no need to invoke the LIKE
** function, then no OP_Variable will be added to the program.
** This causes problems for the sqlite3_bind_parameter_name()
- ** API. To workaround them, add a dummy OP_Variable here.
+ ** API. To work around them, add a dummy OP_Variable here.
*/
int r1 = sqlite3GetTempReg(pParse);
sqlite3ExprCodeTarget(pParse, pRight, r1);
@@ -821,7 +821,7 @@ static void transferJoinMarkings(Expr *pDerived, Expr *pBase){
** appropriate for indexing exist.
**
** All examples A through E above satisfy case 2. But if a term
-** also statisfies case 1 (such as B) we know that the optimizer will
+** also satisfies case 1 (such as B) we know that the optimizer will
** always prefer case 1, so in that case we pretend that case 2 is not
** satisfied.
**
@@ -979,7 +979,7 @@ static void exprAnalyzeOrTerm(
}
if( (chngToIN & getMask(&pWInfo->sMaskSet, pOrTerm->leftCursor))==0 ){
/* This term must be of the form t1.a==t2.b where t2 is in the
- ** chngToIN set but t1 is not. This term will be either preceeded
+ ** chngToIN set but t1 is not. This term will be either preceded
** or follwed by an inverted copy (t2.b==t1.a). Skip this term
** and use its inversion. */
testcase( pOrTerm->wtFlags & TERM_COPIED );
@@ -1390,7 +1390,7 @@ static void exprAnalyze(
}
/*
-** This function searches pList for a entry that matches the iCol-th column
+** This function searches pList for an entry that matches the iCol-th column
** of index pIdx.
**
** If such an expression is found, its index in pList->a[] is returned. If
@@ -2140,7 +2140,7 @@ static int whereRangeSkipScanEst(
** number of rows that the index scan is expected to visit without
** considering the range constraints. If nEq is 0, this is the number of
** rows in the index. Assuming no error occurs, *pnOut is adjusted (reduced)
-** to account for the range contraints pLower and pUpper.
+** to account for the range constraints pLower and pUpper.
**
** In the absence of sqlite_stat4 ANALYZE data, or if such data cannot be
** used, a single range inequality reduces the search space by a factor of 4.
@@ -3417,7 +3417,7 @@ static Bitmask codeOneLoopStart(
** B: <after the loop>
**
** Added 2014-05-26: If the table is a WITHOUT ROWID table, then
- ** use an ephermeral index instead of a RowSet to record the primary
+ ** use an ephemeral index instead of a RowSet to record the primary
** keys of the rows we have already seen.
**
*/
@@ -3468,7 +3468,7 @@ static Bitmask codeOneLoopStart(
}
/* Initialize the rowset register to contain NULL. An SQL NULL is
- ** equivalent to an empty rowset. Or, create an ephermeral index
+ ** equivalent to an empty rowset. Or, create an ephemeral index
** capable of holding primary keys in the case of a WITHOUT ROWID.
**
** Also initialize regReturn to contain the address of the instruction
@@ -4221,14 +4221,16 @@ static int whereLoopInsert(WhereLoopBuilder *pBuilder, WhereLoop *pTemplate){
** the number of output rows by a factor of 10 and each additional term
** reduces the number of output rows by sqrt(2).
*/
-static void whereLoopOutputAdjust(WhereClause *pWC, WhereLoop *pLoop){
+static void whereLoopOutputAdjust(
+ WhereClause *pWC, /* The WHERE clause */
+ WhereLoop *pLoop, /* The loop to adjust downward */
+ LogEst nRow /* Number of rows in the entire table */
+){
WhereTerm *pTerm, *pX;
Bitmask notAllowed = ~(pLoop->prereq|pLoop->maskSelf);
int i, j;
+ int nEq = 0; /* Number of = constraints not within likely()/unlikely() */
- if( !OptimizationEnabled(pWC->pWInfo->pParse->db, SQLITE_AdjustOutEst) ){
- return;
- }
for(i=pWC->nTerm, pTerm=pWC->a; i>0; i--, pTerm++){
if( (pTerm->wtFlags & TERM_VIRTUAL)!=0 ) break;
if( (pTerm->prereqAll & pLoop->maskSelf)==0 ) continue;
@@ -4240,9 +4242,21 @@ static void whereLoopOutputAdjust(WhereClause *pWC, WhereLoop *pLoop){
if( pX->iParent>=0 && (&pWC->a[pX->iParent])==pTerm ) break;
}
if( j<0 ){
- pLoop->nOut += (pTerm->truthProb<=0 ? pTerm->truthProb : -1);
+ if( pTerm->truthProb<=0 ){
+ pLoop->nOut += pTerm->truthProb;
+ }else{
+ pLoop->nOut--;
+ if( pTerm->eOperator&WO_EQ ) nEq++;
+ }
}
}
+ /* TUNING: If there is at least one equality constraint in the WHERE
+ ** clause that does not have a likelihood() explicitly assigned to it
+ ** then do not let the estimated number of output rows exceed half
+ ** the number of rows in the table. */
+ if( nEq && pLoop->nOut>nRow-10 ){
+ pLoop->nOut = nRow - 10;
+ }
}
/*
@@ -4288,6 +4302,7 @@ static int whereLoopAddBtreeIndex(
LogEst saved_nOut; /* Original value of pNew->nOut */
int iCol; /* Index of the column in the table */
int rc = SQLITE_OK; /* Return code */
+ LogEst rSize; /* Number of rows in the table */
LogEst rLogSize; /* Logarithm of table size */
WhereTerm *pTop = 0, *pBtm = 0; /* Top and bottom range constraints */
@@ -4317,7 +4332,8 @@ static int whereLoopAddBtreeIndex(
saved_prereq = pNew->prereq;
saved_nOut = pNew->nOut;
pNew->rSetup = 0;
- rLogSize = estLog(pProbe->aiRowLogEst[0]);
+ rSize = pProbe->aiRowLogEst[0];
+ rLogSize = estLog(rSize);
/* Consider using a skip-scan if there are no WHERE clause constraints
** available for the left-most terms of the index, and if the average
@@ -4494,7 +4510,7 @@ static int whereLoopAddBtreeIndex(
nOutUnadjusted = pNew->nOut;
pNew->rRun += nInMul + nIn;
pNew->nOut += nInMul + nIn;
- whereLoopOutputAdjust(pBuilder->pWC, pNew);
+ whereLoopOutputAdjust(pBuilder->pWC, pNew, rSize);
rc = whereLoopInsert(pBuilder, pNew);
if( pNew->wsFlags & WHERE_COLUMN_RANGE ){
@@ -4707,7 +4723,7 @@ static int whereLoopAddBtree(
ApplyCostMultiplier(pNew->rSetup, pTab->costMult);
/* TUNING: Each index lookup yields 20 rows in the table. This
** is more than the usual guess of 10 rows, since we have no way
- ** of knowning how selective the index will ultimately be. It would
+ ** of knowing how selective the index will ultimately be. It would
** not be unreasonable to make this value much larger. */
pNew->nOut = 43; assert( 43==sqlite3LogEst(20) );
pNew->rRun = sqlite3LogEstAdd(rLogSize,pNew->nOut);
@@ -4748,7 +4764,7 @@ static int whereLoopAddBtree(
/* TUNING: Cost of full table scan is (N*3.0). */
pNew->rRun = rSize + 16;
ApplyCostMultiplier(pNew->rRun, pTab->costMult);
- whereLoopOutputAdjust(pWC, pNew);
+ whereLoopOutputAdjust(pWC, pNew, rSize);
rc = whereLoopInsert(pBuilder, pNew);
pNew->nOut = rSize;
if( rc ) break;
@@ -4784,7 +4800,7 @@ static int whereLoopAddBtree(
pNew->rRun = sqlite3LogEstAdd(pNew->rRun, rSize+16);
}
ApplyCostMultiplier(pNew->rRun, pTab->costMult);
- whereLoopOutputAdjust(pWC, pNew);
+ whereLoopOutputAdjust(pWC, pNew, rSize);
rc = whereLoopInsert(pBuilder, pNew);
pNew->nOut = rSize;
if( rc ) break;
@@ -5137,7 +5153,7 @@ static int whereLoopAddAll(WhereLoopBuilder *pBuilder){
** strict. With GROUP BY and DISTINCT the only requirement is that
** equivalent rows appear immediately adjacent to one another. GROUP BY
** and DISTINCT do not require rows to appear in any particular order as long
-** as equivelent rows are grouped together. Thus for GROUP BY and DISTINCT
+** as equivalent rows are grouped together. Thus for GROUP BY and DISTINCT
** the pOrderBy terms can be matched in any order. With ORDER BY, the
** pOrderBy terms must be matched in strict left-to-right order.
*/
generated by cgit v1.2.3 (git 2.25.1) at 2025-07-19 07:39:03 +0000