/ Check-in [a58aafdb]
Login

Many hyperlinks are disabled.
Use anonymous login to enable hyperlinks.

Overview
Comment:Fix the LIKE optimization even when comparing mixed-case BLOBs.
Downloads: Tarball | ZIP archive | SQL archive
Timelines: family | ancestors | descendants | both | like-opt-fix
Files: files | file ages | folders
SHA1: a58aafdb4e1422b6a8ffc07a67984928bbedf919
User & Date: drh 2015-03-06 19:47:38
Context
2015-03-06
20:49
Test cases added. Comments fixed. Proposed solution for ticket [05f43be8fdda9fbd9]. check-in: 6b993bd5 user: drh tags: like-opt-fix
19:47
Fix the LIKE optimization even when comparing mixed-case BLOBs. check-in: a58aafdb user: drh tags: like-opt-fix
16:45
The LIKE optimization must be applied twice, once for strings and a second time for BLOBs. Ticket [05f43be8fdda9f]. This check-in is a proof-of-concept of how that might be done. check-in: 5757e803 user: drh tags: like-opt-fix
Changes
Hide Diffs Unified Diffs Ignore Whitespace Patch

Changes to src/where.c.

1260
1261
1262
1263
1264
1265
1266

1267

1268
1269














1270
1271
1272
1273
1274
1275
1276
....
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
....
2471
2472
2473
2474
2475
2476
2477













2478
2479

2480
2481
2482
2483
2484
2485
2486
2487
2488
2489
2490
2491



2492
2493
2494
2495
2496
2497
2498
....
3803
3804
3805
3806
3807
3808
3809

3810
3811
3812
3813
3814
3815
3816
3817
3818
3819
3820
3821
3822
3823





3824

3825
3826
3827
3828
3829
3830
3831
    Expr *pLeft;       /* LHS of LIKE/GLOB operator */
    Expr *pStr2;       /* Copy of pStr1 - RHS of LIKE/GLOB operator */
    Expr *pNewExpr1;
    Expr *pNewExpr2;
    int idxNew1;
    int idxNew2;
    Token sCollSeqName;  /* Name of collating sequence */



    pLeft = pExpr->x.pList->a[1].pExpr;
    pStr2 = sqlite3ExprDup(db, pStr1, 0);














    if( !db->mallocFailed ){
      u8 c, *pC;       /* Last character before the first wildcard */
      pC = (u8*)&pStr2->u.zToken[sqlite3Strlen30(pStr2->u.zToken)-1];
      c = *pC;
      if( noCase ){
        /* The point is to increment the last character before the first
        ** wildcard.  But if we increment '@', that will push it into the
................................................................................
        c = sqlite3UpperToLower[c];
      }
      *pC = c + 1;
    }
    sCollSeqName.z = noCase ? "NOCASE" : "BINARY";
    sCollSeqName.n = 6;
    pNewExpr1 = sqlite3ExprDup(db, pLeft, 0);
    pNewExpr1 = sqlite3PExpr(pParse, TK_GE, 
           sqlite3ExprAddCollateToken(pParse,pNewExpr1,&sCollSeqName),
           pStr1, 0);
    transferJoinMarkings(pNewExpr1, pExpr);
    idxNew1 = whereClauseInsert(pWC, pNewExpr1, 
                                TERM_LIKEOPT|TERM_VIRTUAL|TERM_DYNAMIC);
    testcase( idxNew1==0 );
    exprAnalyze(pSrc, pWC, idxNew1);
    pNewExpr2 = sqlite3ExprDup(db, pLeft, 0);
    pNewExpr2 = sqlite3PExpr(pParse, TK_LT,
           sqlite3ExprAddCollateToken(pParse,pNewExpr2,&sCollSeqName),
           pStr2, 0);
    transferJoinMarkings(pNewExpr2, pExpr);
    idxNew2 = whereClauseInsert(pWC, pNewExpr2,
                                TERM_LIKEOPT|TERM_VIRTUAL|TERM_DYNAMIC);
    testcase( idxNew2==0 );
    exprAnalyze(pSrc, pWC, idxNew2);
    pTerm = &pWC->a[idxTerm];
    if( isComplete ){
      markTermAsChild(pWC, idxNew1, idxTerm);
      markTermAsChild(pWC, idxNew2, idxTerm);
    }
................................................................................
** Disabling a term causes that term to not be tested in the inner loop
** of the join.  Disabling is an optimization.  When terms are satisfied
** by indices, we disable them to prevent redundant tests in the inner
** loop.  We would get the correct results if nothing were ever disabled,
** but joins might run a little slower.  The trick is to disable as much
** as we can without disabling too much.  If we disabled in (1), we'd get
** the wrong answer.  See ticket #813.













*/
static void disableTerm(WhereLevel *pLevel, WhereTerm *pTerm){

  if( pTerm
      && (pTerm->wtFlags & TERM_CODED)==0
      && (pLevel->iLeftJoin==0 || ExprHasProperty(pTerm->pExpr, EP_FromJoin))
      && (pLevel->notReady & pTerm->prereqAll)==0
  ){
    pTerm->wtFlags |= TERM_CODED;
    if( pTerm->iParent>=0 ){
      WhereTerm *pOther = &pTerm->pWC->a[pTerm->iParent];
      if( (--pOther->nChild)==0 ){
        disableTerm(pLevel, pOther);
      }
    }



  }
}

/*
** Code an OP_Affinity opcode to apply the column affinity string zAff
** to the n registers starting at base. 
**
................................................................................
#endif

  /* Insert code to test every subexpression that can be completely
  ** computed using the current set of tables.
  */
  for(pTerm=pWC->a, j=pWC->nTerm; j>0; j--, pTerm++){
    Expr *pE;

    testcase( pTerm->wtFlags & TERM_VIRTUAL );
    testcase( pTerm->wtFlags & TERM_CODED );
    if( pTerm->wtFlags & (TERM_VIRTUAL|TERM_CODED) ) continue;
    if( (pTerm->prereqAll & pLevel->notReady)!=0 ){
      testcase( pWInfo->untestedTerms==0
               && (pWInfo->wctrlFlags & WHERE_ONETABLE_ONLY)!=0 );
      pWInfo->untestedTerms = 1;
      continue;
    }
    pE = pTerm->pExpr;
    assert( pE!=0 );
    if( pLevel->iLeftJoin && !ExprHasProperty(pE, EP_FromJoin) ){
      continue;
    }





    sqlite3ExprIfFalse(pParse, pE, addrCont, SQLITE_JUMPIFNULL);

    pTerm->wtFlags |= TERM_CODED;
  }

  /* Insert code to test for implied constraints based on transitivity
  ** of the "==" operator.
  **
  ** Example: If the WHERE clause contains "t1.a=t2.b" and "t2.b=123"







>

>


>
>
>
>
>
>
>
>
>
>
>
>
>
>







 







|



|
<







|
<







 







>
>
>
>
>
>
>
>
>
>
>
>
>


>
|




|
|
|
|
|
|
|
>
>
>







 







>














>
>
>
>
>

>







1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
....
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309

1310
1311
1312
1313
1314
1315
1316
1317

1318
1319
1320
1321
1322
1323
1324
....
2485
2486
2487
2488
2489
2490
2491
2492
2493
2494
2495
2496
2497
2498
2499
2500
2501
2502
2503
2504
2505
2506
2507
2508
2509
2510
2511
2512
2513
2514
2515
2516
2517
2518
2519
2520
2521
2522
2523
2524
2525
2526
2527
2528
2529
....
3834
3835
3836
3837
3838
3839
3840
3841
3842
3843
3844
3845
3846
3847
3848
3849
3850
3851
3852
3853
3854
3855
3856
3857
3858
3859
3860
3861
3862
3863
3864
3865
3866
3867
3868
3869
    Expr *pLeft;       /* LHS of LIKE/GLOB operator */
    Expr *pStr2;       /* Copy of pStr1 - RHS of LIKE/GLOB operator */
    Expr *pNewExpr1;
    Expr *pNewExpr2;
    int idxNew1;
    int idxNew2;
    Token sCollSeqName;  /* Name of collating sequence */
    const u16 wtFlags = TERM_LIKEOPT | TERM_VIRTUAL | TERM_DYNAMIC;

    pTerm->wtFlags |= TERM_LIKE;
    pLeft = pExpr->x.pList->a[1].pExpr;
    pStr2 = sqlite3ExprDup(db, pStr1, 0);

    /* Convert the lower bound to upper-case and the upper bound to
    ** lower-case (upper-case is less than lower-case in ASCII) so that
    ** the range constraints also work for BLOBs
    */
    if( noCase && !pParse->db->mallocFailed ){
      int i;
      char c;
      for(i=0; (c = pStr1->u.zToken[i])!=0; i++){
        pStr1->u.zToken[i] = sqlite3Toupper(c);
        pStr2->u.zToken[i] = sqlite3Tolower(c);
      }
    }

    if( !db->mallocFailed ){
      u8 c, *pC;       /* Last character before the first wildcard */
      pC = (u8*)&pStr2->u.zToken[sqlite3Strlen30(pStr2->u.zToken)-1];
      c = *pC;
      if( noCase ){
        /* The point is to increment the last character before the first
        ** wildcard.  But if we increment '@', that will push it into the
................................................................................
        c = sqlite3UpperToLower[c];
      }
      *pC = c + 1;
    }
    sCollSeqName.z = noCase ? "NOCASE" : "BINARY";
    sCollSeqName.n = 6;
    pNewExpr1 = sqlite3ExprDup(db, pLeft, 0);
    pNewExpr1 = sqlite3PExpr(pParse, TK_GE,
           sqlite3ExprAddCollateToken(pParse,pNewExpr1,&sCollSeqName),
           pStr1, 0);
    transferJoinMarkings(pNewExpr1, pExpr);
    idxNew1 = whereClauseInsert(pWC, pNewExpr1, wtFlags);

    testcase( idxNew1==0 );
    exprAnalyze(pSrc, pWC, idxNew1);
    pNewExpr2 = sqlite3ExprDup(db, pLeft, 0);
    pNewExpr2 = sqlite3PExpr(pParse, TK_LT,
           sqlite3ExprAddCollateToken(pParse,pNewExpr2,&sCollSeqName),
           pStr2, 0);
    transferJoinMarkings(pNewExpr2, pExpr);
    idxNew2 = whereClauseInsert(pWC, pNewExpr2, wtFlags);

    testcase( idxNew2==0 );
    exprAnalyze(pSrc, pWC, idxNew2);
    pTerm = &pWC->a[idxTerm];
    if( isComplete ){
      markTermAsChild(pWC, idxNew1, idxTerm);
      markTermAsChild(pWC, idxNew2, idxTerm);
    }
................................................................................
** Disabling a term causes that term to not be tested in the inner loop
** of the join.  Disabling is an optimization.  When terms are satisfied
** by indices, we disable them to prevent redundant tests in the inner
** loop.  We would get the correct results if nothing were ever disabled,
** but joins might run a little slower.  The trick is to disable as much
** as we can without disabling too much.  If we disabled in (1), we'd get
** the wrong answer.  See ticket #813.
**
** If all the children of a term are disabled, then that term is also
** automatically disabled.  In this way, terms get disabled if derived
** virtual terms are tested first.  For example:
**
**      x GLOB 'abc*' AND x>='abc' AND x<'acd'
**      \___________/     \______/     \_____/
**         parent          child1       child2
**
** Only the parent term was in the original WHERE clause.  The child1
** and child2 terms were added by the LIKE optimization.  If both of
** the virtual child terms are valid, then testing of the parent can be 
** skipped.
*/
static void disableTerm(WhereLevel *pLevel, WhereTerm *pTerm){
  int nLoop = 0;
  while( pTerm
      && (pTerm->wtFlags & TERM_CODED)==0
      && (pLevel->iLeftJoin==0 || ExprHasProperty(pTerm->pExpr, EP_FromJoin))
      && (pLevel->notReady & pTerm->prereqAll)==0
  ){
    if( nLoop && (pTerm->wtFlags & TERM_LIKE)!=0 ){
      pTerm->wtFlags |= TERM_LIKECOND;
    }else{
      pTerm->wtFlags |= TERM_CODED;
    }
    if( pTerm->iParent<0 ) break;
    pTerm = &pTerm->pWC->a[pTerm->iParent];
    pTerm->nChild--;
    if( pTerm->nChild!=0 ) break;
    nLoop++;
  }
}

/*
** Code an OP_Affinity opcode to apply the column affinity string zAff
** to the n registers starting at base. 
**
................................................................................
#endif

  /* Insert code to test every subexpression that can be completely
  ** computed using the current set of tables.
  */
  for(pTerm=pWC->a, j=pWC->nTerm; j>0; j--, pTerm++){
    Expr *pE;
    int skipLikeAddr = 0;
    testcase( pTerm->wtFlags & TERM_VIRTUAL );
    testcase( pTerm->wtFlags & TERM_CODED );
    if( pTerm->wtFlags & (TERM_VIRTUAL|TERM_CODED) ) continue;
    if( (pTerm->prereqAll & pLevel->notReady)!=0 ){
      testcase( pWInfo->untestedTerms==0
               && (pWInfo->wctrlFlags & WHERE_ONETABLE_ONLY)!=0 );
      pWInfo->untestedTerms = 1;
      continue;
    }
    pE = pTerm->pExpr;
    assert( pE!=0 );
    if( pLevel->iLeftJoin && !ExprHasProperty(pE, EP_FromJoin) ){
      continue;
    }
    if( pTerm->wtFlags & TERM_LIKECOND ){
      assert( pLevel->iLikeRepCntr>0 );
      skipLikeAddr = sqlite3VdbeAddOp1(v, OP_IfZero, pLevel->iLikeRepCntr);
      VdbeCoverage(v);
    }
    sqlite3ExprIfFalse(pParse, pE, addrCont, SQLITE_JUMPIFNULL);
    if( skipLikeAddr ) sqlite3VdbeJumpHere(v, skipLikeAddr);
    pTerm->wtFlags |= TERM_CODED;
  }

  /* Insert code to test for implied constraints based on transitivity
  ** of the "==" operator.
  **
  ** Example: If the WHERE clause contains "t1.a=t2.b" and "t2.b=123"

Changes to src/whereInt.h.

273
274
275
276
277
278
279
280


281
282
283
284
285
286
287
#define TERM_ANDINFO    0x20   /* Need to free the WhereTerm.u.pAndInfo obj */
#define TERM_OR_OK      0x40   /* Used during OR-clause processing */
#ifdef SQLITE_ENABLE_STAT3_OR_STAT4
#  define TERM_VNULL    0x80   /* Manufactured x>NULL or x<=NULL term */
#else
#  define TERM_VNULL    0x00   /* Disabled if not using stat3 */
#endif
#define TERM_LIKEOPT    0x100  /* Used by the LIKE optimization */



/*
** An instance of the WhereScan object is used as an iterator for locating
** terms in the WHERE clause that are useful to the query planner.
*/
struct WhereScan {
  WhereClause *pOrigWC;      /* Original, innermost WhereClause */







|
>
>







273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
#define TERM_ANDINFO    0x20   /* Need to free the WhereTerm.u.pAndInfo obj */
#define TERM_OR_OK      0x40   /* Used during OR-clause processing */
#ifdef SQLITE_ENABLE_STAT3_OR_STAT4
#  define TERM_VNULL    0x80   /* Manufactured x>NULL or x<=NULL term */
#else
#  define TERM_VNULL    0x00   /* Disabled if not using stat3 */
#endif
#define TERM_LIKEOPT    0x100  /* Virtual terms from the LIKE optimization */
#define TERM_LIKECOND   0x200  /* Conditionally this LIKE operator term */
#define TERM_LIKE       0x400  /* The original LIKE operator */

/*
** An instance of the WhereScan object is used as an iterator for locating
** terms in the WHERE clause that are useful to the query planner.
*/
struct WhereScan {
  WhereClause *pOrigWC;      /* Original, innermost WhereClause */