From: dan Date: Thu, 13 Dec 2012 16:37:10 +0000 (+0000) Subject: Attempt to further generalize the min/max optimization so that, if an appropriate... X-Git-Tag: version-3.7.16~100^2 X-Git-Url: http://git.ipfire.org/?a=commitdiff_plain;h=refs%2Fheads%2Fminmax-opt;p=thirdparty%2Fsqlite.git Attempt to further generalize the min/max optimization so that, if an appropriate index exists, it can be used by any aggregate query that contains only a single aggregate of the form max(colname) or min(colname) and does not contain a GROUP BY clause. FossilOrigin-Name: 7280e14cd8f55896451847b432171e8750a07c81 --- diff --git a/manifest b/manifest index aef5a7ffec..a4ed9e2e1c 100644 --- a/manifest +++ b/manifest @@ -1,5 +1,5 @@ -C Allow\sthe\serror\smessage\sfrom\s"PRAGMA\sintegrity_check"\sto\sbe\slonger\sthan\n20,000\sbytes. -D 2012-12-12T14:30:03.521 +C Attempt\sto\sfurther\sgeneralize\sthe\smin/max\soptimization\sso\sthat,\sif\san\sappropriate\sindex\sexists,\sit\scan\sbe\sused\sby\sany\saggregate\squery\sthat\scontains\sonly\sa\ssingle\saggregate\sof\sthe\sform\smax(colname)\sor\smin(colname)\sand\sdoes\snot\scontain\sa\sGROUP\sBY\sclause. +D 2012-12-13T16:37:10.621 F Makefile.arm-wince-mingw32ce-gcc d6df77f1f48d690bd73162294bbba7f59507c72f F Makefile.in 690d441a758cbffd13e814dc2724a721a6ebd400 F Makefile.linux-gcc 91d710bdc4998cb015f39edf3cb314ec4f4d7e23 @@ -174,7 +174,7 @@ F src/printf.c 4a9f882f1c1787a8b494a2987765acf9d97ac21f F src/random.c cd4a67b3953b88019f8cd4ccd81394a8ddfaba50 F src/resolve.c cdd546d62da7763119ea1fa455a898959e03457f F src/rowset.c 64655f1a627c9c212d9ab497899e7424a34222e0 -F src/select.c e6daa524bbdfa98f4abdb8cb281498f0047d3161 +F src/select.c f6c73171209b4af4d6b4258b6740415790415e2e F src/shell.c e392dd1ccbb77cc1d75a8367a89b473c24bea019 F src/sqlite.h.in 39cc33bb08897c748fe3383c29ccf56585704177 F src/sqlite3.rc fea433eb0a59f4c9393c8e6d76a6e2596b1fe0c0 @@ -613,7 +613,7 @@ F test/memdb.test 708a028d6d373e5b3842e4bdc8ba80998c9a4da6 F test/memleak.test 10b9c6c57e19fc68c32941495e9ba1c50123f6e2 F test/memsubsys1.test a8f9e37567453a5d1d9d37ec102d4d88ab6be33f F test/memsubsys2.test 3a1c1a9de48e5726faa85108b02459fae8cb9ee9 -F test/minmax.test c61518429e66e228efc79661fbd2dc3e4924ec44 +F test/minmax.test 42fbad0e81afaa6e0de41c960329f2b2c3526efd F test/minmax2.test b44bae787fc7b227597b01b0ca5575c7cb54d3bc F test/minmax3.test cc1e8b010136db0d01a6f2a29ba5a9f321034354 F test/minmax4.test 536a3360470633a177e42fbc19660d146b51daef @@ -1025,7 +1025,10 @@ F tool/vdbe-compress.tcl f12c884766bd14277f4fcedcae07078011717381 F tool/warnings-clang.sh f6aa929dc20ef1f856af04a730772f59283631d4 F tool/warnings.sh fbc018d67fd7395f440c28f33ef0f94420226381 F tool/win/sqlite.vsix 97894c2790eda7b5bce3cc79cb2a8ec2fde9b3ac -P cd0b37c52658bfdf992b1e3dc467bae1835a94ae -R c074324819c1cb0b1b063013401b3d51 -U drh -Z 0c15681c6bbc86d62fafd9ce06645760 +P 120c82d56ecb574f46db01679317b8062d8779ed +R 39203e605a8f25e92556c6b1c71c785c +T *branch * minmax-opt +T *sym-minmax-opt * +T -sym-trunk * +U dan +Z ab1716797ba85b3e90d83237db257cf3 diff --git a/manifest.uuid b/manifest.uuid index 3f1ad31ac9..47b9352d93 100644 --- a/manifest.uuid +++ b/manifest.uuid @@ -1 +1 @@ -120c82d56ecb574f46db01679317b8062d8779ed \ No newline at end of file +7280e14cd8f55896451847b432171e8750a07c81 \ No newline at end of file diff --git a/src/select.c b/src/select.c index a90877dc61..2568cf7081 100644 --- a/src/select.c +++ b/src/select.c @@ -3160,34 +3160,43 @@ static int flattenSubquery( #endif /* !defined(SQLITE_OMIT_SUBQUERY) || !defined(SQLITE_OMIT_VIEW) */ /* -** Analyze the SELECT statement passed as an argument to see if it -** is a min() or max() query. Return WHERE_ORDERBY_MIN or WHERE_ORDERBY_MAX if -** it is, or 0 otherwise. At present, a query is considered to be -** a min()/max() query if: +** Based on the contents of the AggInfo structure indicated by the first +** argument, this function checks if the following are true: ** -** 1. There is a single object in the FROM clause. +** * the query contains just a single aggregate function, +** * the aggregate function is either min() or max(), and +** * the argument to the aggregate function is a column value. ** -** 2. There is a single expression in the result set, and it is -** either min(x) or max(x), where x is a column reference. +** If all of the above are true, then WHERE_ORDERBY_MIN or WHERE_ORDERBY_MAX +** is returned as appropriate. Also, *ppMinMax is set to point to the +** list of arguments passed to the aggregate before returning. +** +** Or, if the conditions above are not met, *ppMinMax is set to 0 and +** WHERE_ORDERBY_NORMAL is returned. */ -static u8 minMaxQuery(Select *p){ - Expr *pExpr; - ExprList *pEList = p->pEList; - - if( pEList->nExpr!=1 ) return WHERE_ORDERBY_NORMAL; - pExpr = pEList->a[0].pExpr; - if( pExpr->op!=TK_AGG_FUNCTION ) return 0; - if( NEVER(ExprHasProperty(pExpr, EP_xIsSelect)) ) return 0; - pEList = pExpr->x.pList; - if( pEList==0 || pEList->nExpr!=1 ) return 0; - if( pEList->a[0].pExpr->op!=TK_AGG_COLUMN ) return WHERE_ORDERBY_NORMAL; - assert( !ExprHasProperty(pExpr, EP_IntValue) ); - if( sqlite3StrICmp(pExpr->u.zToken,"min")==0 ){ - return WHERE_ORDERBY_MIN; - }else if( sqlite3StrICmp(pExpr->u.zToken,"max")==0 ){ - return WHERE_ORDERBY_MAX; +static u8 minMaxQuery(AggInfo *pAggInfo, ExprList **ppMinMax){ + int eRet = WHERE_ORDERBY_NORMAL; /* Return value */ + + *ppMinMax = 0; + if( pAggInfo->nFunc==1 ){ + Expr *pExpr = pAggInfo->aFunc[0].pExpr; /* Aggregate function */ + ExprList *pEList = pExpr->x.pList; /* Arguments to agg function */ + + assert( pExpr->op==TK_AGG_FUNCTION ); + if( pEList && pEList->nExpr==1 && pEList->a[0].pExpr->op==TK_AGG_COLUMN ){ + const char *zFunc = pExpr->u.zToken; + if( sqlite3StrICmp(zFunc, "min")==0 ){ + eRet = WHERE_ORDERBY_MIN; + *ppMinMax = pEList; + }else if( sqlite3StrICmp(zFunc, "max")==0 ){ + eRet = WHERE_ORDERBY_MAX; + *ppMinMax = pEList; + } + } } - return WHERE_ORDERBY_NORMAL; + + assert( *ppMinMax==0 || (*ppMinMax)->nExpr==1 ); + return eRet; } /* @@ -4527,11 +4536,17 @@ int sqlite3Select( ** Refer to code and comments in where.c for details. */ ExprList *pMinMax = 0; - u8 flag = minMaxQuery(p); + u8 flag = WHERE_ORDERBY_NORMAL; + + assert( p->pGroupBy==0 ); + assert( flag==0 ); + if( p->pHaving==0 ){ + flag = minMaxQuery(&sAggInfo, &pMinMax); + } + assert( flag==0 || (pMinMax!=0 && pMinMax->nExpr==1) ); + if( flag ){ - assert( !ExprHasProperty(p->pEList->a[0].pExpr, EP_xIsSelect) ); - assert( p->pEList->a[0].pExpr->x.pList->nExpr==1 ); - pMinMax = sqlite3ExprListDup(db, p->pEList->a[0].pExpr->x.pList,0); + pMinMax = sqlite3ExprListDup(db, pMinMax, 0); pDel = pMinMax; if( pMinMax && !db->mallocFailed ){ pMinMax->a[0].sortOrder = flag!=WHERE_ORDERBY_MIN ?1:0; diff --git a/test/minmax.test b/test/minmax.test index 3621375ee7..fb9bbb383a 100644 --- a/test/minmax.test +++ b/test/minmax.test @@ -17,6 +17,7 @@ set testdir [file dirname $argv0] source $testdir/tester.tcl +set ::testprefix minmax do_test minmax-1.0 { execsql { @@ -536,7 +537,96 @@ do_test minmax-12.17 { } } {5} +#------------------------------------------------------------------------- +reset_db +proc do_test_13 {op name sql1 sql2 res} { + set ::sqlite_search_count 0 + uplevel [list do_execsql_test $name.1 $sql1 $res] + set a $::sqlite_search_count + + set ::sqlite_search_count 0 + uplevel [list do_execsql_test $name.2 $sql2 $res] + set b $::sqlite_search_count + + uplevel [list do_test $name.3 [list expr "$a $op $b"] 1] +} + +# Run a test named $name. Check that SQL statements $sql1 and $sql2 both +# return the same result, but that $sql2 increments the $sqlite_search_count +# variable more often (indicating that it is visiting more rows to determine +# the result). +# +proc do_test_13_opt {name sql1 sql2 res} { + uplevel [list do_test_13 < $name $sql1 $sql2 $res] +} + +# Like [do_test_13_noopt], except this time check that the $sqlite_search_count +# variable is incremented the same number of times by both SQL statements. +# +proc do_test_13_noopt {name sql1 sql2 res} { + uplevel [list do_test_13 == $name $sql1 $sql2 $res] +} + +do_execsql_test 13.1 { + CREATE TABLE t1(a, b, c); + INSERT INTO t1 VALUES('a', 1, 1); + INSERT INTO t1 VALUES('b', 6, 6); + INSERT INTO t1 VALUES('c', 5, 5); + INSERT INTO t1 VALUES('a', 4, 4); + INSERT INTO t1 VALUES('a', 5, 5); + INSERT INTO t1 VALUES('c', 6, 6); + INSERT INTO t1 VALUES('b', 4, 4); + INSERT INTO t1 VALUES('c', 7, 7); + INSERT INTO t1 VALUES('b', 2, 2); + INSERT INTO t1 VALUES('b', 3, 3); + INSERT INTO t1 VALUES('a', 3, 3); + INSERT INTO t1 VALUES('b', 5, 5); + INSERT INTO t1 VALUES('c', 4, 4); + INSERT INTO t1 VALUES('c', 3, 3); + INSERT INTO t1 VALUES('a', 2, 2); + SELECT * FROM t1 ORDER BY a, b, c; +} {a 1 1 a 2 2 a 3 3 a 4 4 a 5 5 + b 2 2 b 3 3 b 4 4 b 5 5 b 6 6 + c 3 3 c 4 4 c 5 5 c 6 6 c 7 7 +} +do_execsql_test 13.2 { CREATE INDEX i1 ON t1(a, b, c) } + +do_test_13_opt 13.3 { + SELECT min(b) FROM t1 WHERE a='b' +} { + SELECT min(c) FROM t1 WHERE a='b' +} {2} + +do_test_13_opt 13.4 { + SELECT a, min(b) FROM t1 WHERE a='b' +} { + SELECT a, min(c) FROM t1 WHERE a='b' +} {b 2} + +do_test_13_opt 13.4 { + SELECT a||c, max(b)+4 FROM t1 WHERE a='c' +} { + SELECT a||c, max(c)+4 FROM t1 WHERE a='c' +} {c7 11} + +do_test_13_noopt 13.5 { + SELECT a||c, max(b+1) FROM t1 WHERE a='c' +} { + SELECT a||c, max(c+1) FROM t1 WHERE a='c' +} {c7 8} + +do_test_13_noopt 13.6 { + SELECT count(b) FROM t1 WHERE a='c' +} { + SELECT count(c) FROM t1 WHERE a='c' +} {5} + +do_test_13_noopt 13.7 { + SELECT min(b), count(b) FROM t1 WHERE a='a'; +} { + SELECT min(c), count(c) FROM t1 WHERE a='a'; +} {1 5} finish_test