/ Check-in [0cda241a]
Login

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

Overview
Comment:Refactoring of DISTINCT code. Change the name of the local variable "distinct" to "distinctTab". Generate cleaner code w/o unnecessary P4 and P5 values on the OP_Null for WHERE_DISTINCT_ORDERED.
Downloads: Tarball | ZIP archive | SQL archive
Timelines: family | ancestors | descendants | both | trunk
Files: files | file ages | folders
SHA1:0cda241a2bcb3c6f2ae6c48f522780bc4eddfc02
User & Date: drh 2012-09-20 14:26:22
Context
2012-09-20
15:41
Continuing incremental enhancements of SELECT code generation: Remove the Select.affinity field. Use SelectDest.affSdst instead. check-in: cf40b7b5 user: drh tags: trunk
14:26
Refactoring of DISTINCT code. Change the name of the local variable "distinct" to "distinctTab". Generate cleaner code w/o unnecessary P4 and P5 values on the OP_Null for WHERE_DISTINCT_ORDERED. check-in: 0cda241a user: drh tags: trunk
2012-09-19
21:15
Tighter VDBE code for the WHERE_DISTINCT_ORDERED case of DISTINCT keyword handling. check-in: 94b48064 user: drh tags: trunk
Changes
Hide Diffs Unified Diffs Ignore Whitespace Patch

Changes to src/select.c.

537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
...
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
...
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
....
3783
3784
3785
3786
3787
3788
3789
3790
3791
3792
3793
3794
3795
3796
3797
....
4014
4015
4016
4017
4018
4019
4020
4021
4022
4023
4024
4025
4026
4027
4028
4029
4030
4031
4032
4033
4034
....
4045
4046
4047
4048
4049
4050
4051

4052
4053
4054
4055
4056
4057
4058
4059
4060
4061
4062
4063
4064
4065
....
4069
4070
4071
4072
4073
4074
4075
4076
4077
4078
4079
4080
4081
4082
4083
....
4086
4087
4088
4089
4090
4091
4092
4093
4094
4095
4096
4097
4098
4099
4100
4101
4102
4103
4104
4105
4106
....
4360
4361
4362
4363
4364
4365
4366
4367
4368
4369
4370
4371
4372
4373
4374
....
4501
4502
4503
4504
4505
4506
4507
4508
4509
4510
4511
4512
4513
4514
4515
static void selectInnerLoop(
  Parse *pParse,          /* The parser context */
  Select *p,              /* The complete select statement being coded */
  ExprList *pEList,       /* List of values being extracted */
  int srcTab,             /* Pull data from this table */
  int nColumn,            /* Number of columns in the source table */
  ExprList *pOrderBy,     /* If not NULL, sort results using this key */
  int distinct,           /* If >=0, make sure results are distinct */
  SelectDest *pDest,      /* How to dispose of the results */
  int iContinue,          /* Jump here to continue with next row */
  int iBreak              /* Jump here to break out of the inner loop */
){
  Vdbe *v = pParse->pVdbe;
  int i;
  int hasDistinct;        /* True if the DISTINCT keyword is present */
................................................................................
  int eDest = pDest->eDest;   /* How to dispose of results */
  int iParm = pDest->iSDParm; /* First argument to disposal method */
  int nResultCol;             /* Number of result columns */

  assert( v );
  if( NEVER(v==0) ) return;
  assert( pEList!=0 );
  hasDistinct = distinct>=0;
  if( pOrderBy==0 && !hasDistinct ){
    codeOffset(v, p, iContinue);
  }

  /* Pull the requested columns.
  */
  if( nColumn>0 ){
................................................................................
  /* If the DISTINCT keyword was present on the SELECT statement
  ** and this row has been seen before, then do not make this row
  ** part of the result.
  */
  if( hasDistinct ){
    assert( pEList!=0 );
    assert( pEList->nExpr==nColumn );
    codeDistinct(pParse, distinct, iContinue, nColumn, regResult);
    if( pOrderBy==0 ){
      codeOffset(v, p, iContinue);
    }
  }

  switch( eDest ){
    /* In this mode, write each query result to the key of the temporary
................................................................................
  ExprList *pEList;      /* List of columns to extract. */
  SrcList *pTabList;     /* List of tables to select from */
  Expr *pWhere;          /* The WHERE clause.  May be NULL */
  ExprList *pOrderBy;    /* The ORDER BY clause.  May be NULL */
  ExprList *pGroupBy;    /* The GROUP BY clause.  May be NULL */
  Expr *pHaving;         /* The HAVING clause.  May be NULL */
  int isDistinct;        /* True if the DISTINCT keyword is present */
  int distinct;          /* Table to use for the distinct set */
  int rc = 1;            /* Value to return from this function */
  int addrSortIndex;     /* Address of an OP_OpenEphemeral instruction */
  int addrDistinctIndex; /* Address of an OP_OpenEphemeral instruction */
  AggInfo sAggInfo;      /* Information used by aggregate queries */
  int iEnd;              /* Address of the end of the query */
  sqlite3 *db;           /* The database connection */

................................................................................
    p->selFlags |= SF_UseSorter;
  }

  /* Open a virtual index to use for the distinct set.
  */
  if( p->selFlags & SF_Distinct ){
    KeyInfo *pKeyInfo;
    distinct = pParse->nTab++;
    pKeyInfo = keyInfoFromExprList(pParse, p->pEList);
    addrDistinctIndex = sqlite3VdbeAddOp4(v, OP_OpenEphemeral, distinct, 0, 0,
        (char*)pKeyInfo, P4_KEYINFO_HANDOFF);
    sqlite3VdbeChangeP5(v, BTREE_UNORDERED);
  }else{
    distinct = addrDistinctIndex = -1;
  }

  /* Aggregate and non-aggregate queries are handled differently */
  if( !isAgg && pGroupBy==0 ){
    ExprList *pDist = (isDistinct ? p->pEList : 0);

    /* Begin the database scan. */
................................................................................
      p->addrOpenEphm[2] = -1;
    }

    if( pWInfo->eDistinct ){
      VdbeOp *pOp;                /* No longer required OpenEphemeral instr. */
     
      assert( addrDistinctIndex>=0 );

      pOp = sqlite3VdbeGetOp(v, addrDistinctIndex);

      assert( isDistinct );
      assert( pWInfo->eDistinct==WHERE_DISTINCT_ORDERED 
           || pWInfo->eDistinct==WHERE_DISTINCT_UNIQUE 
      );
      distinct = -1;
      if( pWInfo->eDistinct==WHERE_DISTINCT_ORDERED ){
        int iJump;
        int iExpr;
        int nExpr = pEList->nExpr;
        int iBase = pParse->nMem+1;
        int iBase2 = iBase + nExpr;
        pParse->nMem += (pEList->nExpr*2);
................................................................................
        ** previous value.  This will cause the OP_Ne below to always
        ** fail on the first iteration of the loop even if the first
        ** row is all NULLs.
        */
        pOp->opcode = OP_Null;
        pOp->p1 = 1;
        pOp->p2 = iBase2;
        pOp->p3 = iBase2 + nExpr - 1;

        sqlite3ExprCodeExprList(pParse, pEList, iBase, 1);
        iJump = sqlite3VdbeCurrentAddr(v) + pEList->nExpr;
        for(iExpr=0; iExpr<nExpr; iExpr++){
          CollSeq *pColl = sqlite3ExprCollSeq(pParse, pEList->a[iExpr].pExpr);
          if( iExpr<nExpr-1 ){
            sqlite3VdbeAddOp3(v, OP_Ne, iBase+iExpr, iJump, iBase2+iExpr);
................................................................................
                              iBase2+iExpr);
          }
          sqlite3VdbeChangeP4(v, -1, (const char *)pColl, P4_COLLSEQ);
          sqlite3VdbeChangeP5(v, SQLITE_NULLEQ);
        }
        assert( sqlite3VdbeCurrentAddr(v)==iJump );
        sqlite3VdbeAddOp3(v, OP_Move, iBase, iBase2, pEList->nExpr);
      }else{
        pOp->opcode = OP_Noop;
      }
    }

    /* Use the standard inner loop. */
    selectInnerLoop(pParse, p, pEList, 0, 0, pOrderBy, distinct, pDest,
                    pWInfo->iContinue, pWInfo->iBreak);

    /* End the database scan loop.
    */
    sqlite3WhereEnd(pWInfo);
  }else{
    /* This is the processing for aggregate queries */
................................................................................
      addrOutputRow = sqlite3VdbeCurrentAddr(v);
      sqlite3VdbeAddOp2(v, OP_IfPos, iUseFlag, addrOutputRow+2);
      VdbeComment((v, "Groupby result generator entry point"));
      sqlite3VdbeAddOp1(v, OP_Return, regOutputRow);
      finalizeAggFunctions(pParse, &sAggInfo);
      sqlite3ExprIfFalse(pParse, pHaving, addrOutputRow+1, SQLITE_JUMPIFNULL);
      selectInnerLoop(pParse, p, p->pEList, 0, 0, pOrderBy,
                      distinct, pDest,
                      addrOutputRow+1, addrSetAbort);
      sqlite3VdbeAddOp1(v, OP_Return, regOutputRow);
      VdbeComment((v, "end groupby result generator"));

      /* Generate a subroutine that will reset the group-by accumulator
      */
      sqlite3VdbeResolveLabel(v, addrReset);
................................................................................
                      pDest, addrEnd, addrEnd);
      sqlite3ExprListDelete(db, pDel);
    }
    sqlite3VdbeResolveLabel(v, addrEnd);
    
  } /* endif aggregate query */

  if( distinct>=0 ){
    explainTempTable(pParse, "DISTINCT");
  }

  /* If there is an ORDER BY clause, then we need to sort the results
  ** and send them to the callback one by one.
  */
  if( pOrderBy ){







|







 







|







 







|







 







|







 







|

|
|


|







 







>






|







 







<







 







<
<




|







 







|







 







|







537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
...
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
...
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
....
3783
3784
3785
3786
3787
3788
3789
3790
3791
3792
3793
3794
3795
3796
3797
....
4014
4015
4016
4017
4018
4019
4020
4021
4022
4023
4024
4025
4026
4027
4028
4029
4030
4031
4032
4033
4034
....
4045
4046
4047
4048
4049
4050
4051
4052
4053
4054
4055
4056
4057
4058
4059
4060
4061
4062
4063
4064
4065
4066
....
4070
4071
4072
4073
4074
4075
4076

4077
4078
4079
4080
4081
4082
4083
....
4086
4087
4088
4089
4090
4091
4092


4093
4094
4095
4096
4097
4098
4099
4100
4101
4102
4103
4104
....
4358
4359
4360
4361
4362
4363
4364
4365
4366
4367
4368
4369
4370
4371
4372
....
4499
4500
4501
4502
4503
4504
4505
4506
4507
4508
4509
4510
4511
4512
4513
static void selectInnerLoop(
  Parse *pParse,          /* The parser context */
  Select *p,              /* The complete select statement being coded */
  ExprList *pEList,       /* List of values being extracted */
  int srcTab,             /* Pull data from this table */
  int nColumn,            /* Number of columns in the source table */
  ExprList *pOrderBy,     /* If not NULL, sort results using this key */
  int distinctTab,        /* If >=0, make sure results are distinct */
  SelectDest *pDest,      /* How to dispose of the results */
  int iContinue,          /* Jump here to continue with next row */
  int iBreak              /* Jump here to break out of the inner loop */
){
  Vdbe *v = pParse->pVdbe;
  int i;
  int hasDistinct;        /* True if the DISTINCT keyword is present */
................................................................................
  int eDest = pDest->eDest;   /* How to dispose of results */
  int iParm = pDest->iSDParm; /* First argument to disposal method */
  int nResultCol;             /* Number of result columns */

  assert( v );
  if( NEVER(v==0) ) return;
  assert( pEList!=0 );
  hasDistinct = distinctTab>=0;
  if( pOrderBy==0 && !hasDistinct ){
    codeOffset(v, p, iContinue);
  }

  /* Pull the requested columns.
  */
  if( nColumn>0 ){
................................................................................
  /* If the DISTINCT keyword was present on the SELECT statement
  ** and this row has been seen before, then do not make this row
  ** part of the result.
  */
  if( hasDistinct ){
    assert( pEList!=0 );
    assert( pEList->nExpr==nColumn );
    codeDistinct(pParse, distinctTab, iContinue, nColumn, regResult);
    if( pOrderBy==0 ){
      codeOffset(v, p, iContinue);
    }
  }

  switch( eDest ){
    /* In this mode, write each query result to the key of the temporary
................................................................................
  ExprList *pEList;      /* List of columns to extract. */
  SrcList *pTabList;     /* List of tables to select from */
  Expr *pWhere;          /* The WHERE clause.  May be NULL */
  ExprList *pOrderBy;    /* The ORDER BY clause.  May be NULL */
  ExprList *pGroupBy;    /* The GROUP BY clause.  May be NULL */
  Expr *pHaving;         /* The HAVING clause.  May be NULL */
  int isDistinct;        /* True if the DISTINCT keyword is present */
  int distinctTab;       /* Table to use for the distinct set */
  int rc = 1;            /* Value to return from this function */
  int addrSortIndex;     /* Address of an OP_OpenEphemeral instruction */
  int addrDistinctIndex; /* Address of an OP_OpenEphemeral instruction */
  AggInfo sAggInfo;      /* Information used by aggregate queries */
  int iEnd;              /* Address of the end of the query */
  sqlite3 *db;           /* The database connection */

................................................................................
    p->selFlags |= SF_UseSorter;
  }

  /* Open a virtual index to use for the distinct set.
  */
  if( p->selFlags & SF_Distinct ){
    KeyInfo *pKeyInfo;
    distinctTab = pParse->nTab++;
    pKeyInfo = keyInfoFromExprList(pParse, p->pEList);
    addrDistinctIndex = sqlite3VdbeAddOp4(v, OP_OpenEphemeral, distinctTab,
                                   0, 0, (char*)pKeyInfo, P4_KEYINFO_HANDOFF);
    sqlite3VdbeChangeP5(v, BTREE_UNORDERED);
  }else{
    distinctTab = addrDistinctIndex = -1;
  }

  /* Aggregate and non-aggregate queries are handled differently */
  if( !isAgg && pGroupBy==0 ){
    ExprList *pDist = (isDistinct ? p->pEList : 0);

    /* Begin the database scan. */
................................................................................
      p->addrOpenEphm[2] = -1;
    }

    if( pWInfo->eDistinct ){
      VdbeOp *pOp;                /* No longer required OpenEphemeral instr. */
     
      assert( addrDistinctIndex>=0 );
      sqlite3VdbeChangeToNoop(v, addrDistinctIndex);
      pOp = sqlite3VdbeGetOp(v, addrDistinctIndex);

      assert( isDistinct );
      assert( pWInfo->eDistinct==WHERE_DISTINCT_ORDERED 
           || pWInfo->eDistinct==WHERE_DISTINCT_UNIQUE 
      );
      distinctTab = -1;
      if( pWInfo->eDistinct==WHERE_DISTINCT_ORDERED ){
        int iJump;
        int iExpr;
        int nExpr = pEList->nExpr;
        int iBase = pParse->nMem+1;
        int iBase2 = iBase + nExpr;
        pParse->nMem += (pEList->nExpr*2);
................................................................................
        ** previous value.  This will cause the OP_Ne below to always
        ** fail on the first iteration of the loop even if the first
        ** row is all NULLs.
        */
        pOp->opcode = OP_Null;
        pOp->p1 = 1;
        pOp->p2 = iBase2;


        sqlite3ExprCodeExprList(pParse, pEList, iBase, 1);
        iJump = sqlite3VdbeCurrentAddr(v) + pEList->nExpr;
        for(iExpr=0; iExpr<nExpr; iExpr++){
          CollSeq *pColl = sqlite3ExprCollSeq(pParse, pEList->a[iExpr].pExpr);
          if( iExpr<nExpr-1 ){
            sqlite3VdbeAddOp3(v, OP_Ne, iBase+iExpr, iJump, iBase2+iExpr);
................................................................................
                              iBase2+iExpr);
          }
          sqlite3VdbeChangeP4(v, -1, (const char *)pColl, P4_COLLSEQ);
          sqlite3VdbeChangeP5(v, SQLITE_NULLEQ);
        }
        assert( sqlite3VdbeCurrentAddr(v)==iJump );
        sqlite3VdbeAddOp3(v, OP_Move, iBase, iBase2, pEList->nExpr);


      }
    }

    /* Use the standard inner loop. */
    selectInnerLoop(pParse, p, pEList, 0, 0, pOrderBy, distinctTab, pDest,
                    pWInfo->iContinue, pWInfo->iBreak);

    /* End the database scan loop.
    */
    sqlite3WhereEnd(pWInfo);
  }else{
    /* This is the processing for aggregate queries */
................................................................................
      addrOutputRow = sqlite3VdbeCurrentAddr(v);
      sqlite3VdbeAddOp2(v, OP_IfPos, iUseFlag, addrOutputRow+2);
      VdbeComment((v, "Groupby result generator entry point"));
      sqlite3VdbeAddOp1(v, OP_Return, regOutputRow);
      finalizeAggFunctions(pParse, &sAggInfo);
      sqlite3ExprIfFalse(pParse, pHaving, addrOutputRow+1, SQLITE_JUMPIFNULL);
      selectInnerLoop(pParse, p, p->pEList, 0, 0, pOrderBy,
                      distinctTab, pDest,
                      addrOutputRow+1, addrSetAbort);
      sqlite3VdbeAddOp1(v, OP_Return, regOutputRow);
      VdbeComment((v, "end groupby result generator"));

      /* Generate a subroutine that will reset the group-by accumulator
      */
      sqlite3VdbeResolveLabel(v, addrReset);
................................................................................
                      pDest, addrEnd, addrEnd);
      sqlite3ExprListDelete(db, pDel);
    }
    sqlite3VdbeResolveLabel(v, addrEnd);
    
  } /* endif aggregate query */

  if( distinctTab>=0 ){
    explainTempTable(pParse, "DISTINCT");
  }

  /* If there is an ORDER BY clause, then we need to sort the results
  ** and send them to the callback one by one.
  */
  if( pOrderBy ){

Changes to src/sqliteInt.h.

2055
2056
2057
2058
2059
2060
2061
2062
2063
2064
2065
2066
2067
2068
2069
** addrOpenEphm[] entries contain the address of OP_OpenEphemeral opcodes.
** These addresses must be stored so that we can go back and fill in
** the P4_KEYINFO and P2 parameters later.  Neither the KeyInfo nor
** the number of columns in P2 can be computed at the same time
** as the OP_OpenEphm instruction is coded because not
** enough information about the compound query is known at that point.
** The KeyInfo for addrOpenTran[0] and [1] contains collating sequences
** for the result set.  The KeyInfo for addrOpenTran[2] contains collating
** sequences for the ORDER BY clause.
*/
struct Select {
  ExprList *pEList;      /* The fields of the result */
  u8 op;                 /* One of: TK_UNION TK_ALL TK_INTERSECT TK_EXCEPT */
  char affinity;         /* MakeRecord with this affinity for SRT_Set */
  u16 selFlags;          /* Various SF_* values */







|







2055
2056
2057
2058
2059
2060
2061
2062
2063
2064
2065
2066
2067
2068
2069
** addrOpenEphm[] entries contain the address of OP_OpenEphemeral opcodes.
** These addresses must be stored so that we can go back and fill in
** the P4_KEYINFO and P2 parameters later.  Neither the KeyInfo nor
** the number of columns in P2 can be computed at the same time
** as the OP_OpenEphm instruction is coded because not
** enough information about the compound query is known at that point.
** The KeyInfo for addrOpenTran[0] and [1] contains collating sequences
** for the result set.  The KeyInfo for addrOpenEphm[2] contains collating
** sequences for the ORDER BY clause.
*/
struct Select {
  ExprList *pEList;      /* The fields of the result */
  u8 op;                 /* One of: TK_UNION TK_ALL TK_INTERSECT TK_EXCEPT */
  char affinity;         /* MakeRecord with this affinity for SRT_Set */
  u16 selFlags;          /* Various SF_* values */