/ Check-in [bc2de809]
Login

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

Overview
Comment:Fix for ticket [b2fa5424e6fcb15]: Better define the format of the sqlite_stat4 file for WITHOUT ROWID tables and make sure the ANALYZE command generates a file in the appropriate format. Use the sqlite_stat4 data to enable the use of WHERE terms that cover all indexed columns plus some prefix of columns in the primary key.
Downloads: Tarball | ZIP archive | SQL archive
Timelines: family | ancestors | descendants | both | trunk
Files: files | file ages | folders
SHA1:bc2de8095fa9c385db0adf22ca55b0298a33c284
User & Date: drh 2014-06-30 18:57:53
Context
2014-06-30
19:07
Attempt to use STAT4 information to estimate the selectivity of WHERE clause terms when using the skip-scan optimization. check-in: d09ca6d5 user: drh tags: trunk
18:57
Fix for ticket [b2fa5424e6fcb15]: Better define the format of the sqlite_stat4 file for WITHOUT ROWID tables and make sure the ANALYZE command generates a file in the appropriate format. Use the sqlite_stat4 data to enable the use of WHERE terms that cover all indexed columns plus some prefix of columns in the primary key. check-in: bc2de809 user: drh tags: trunk
18:02
Fix a problem in where.c with using the stat4 sample data of an index on a WITHOUT ROWID table. Closed-Leaf check-in: 053a210e user: dan tags: stat4-without-rowid
11:14
Add makefile targets for various diagnostic tools, such as showstat4. Fix harmless compiler warnings in diagnostic tools. check-in: 6f86d89b user: drh tags: trunk
Changes
Hide Diffs Unified Diffs Ignore Whitespace Patch

Changes to src/analyze.c.

242
243
244
245
246
247
248

249
250
251
252
253
254
255
...
277
278
279
280
281
282
283
284

285
286
287
288
289
290
291
...
363
364
365
366
367
368
369
370



371

372




373
374
375
376
377
378
379
380
381
382
383
384

385
386
387
388
389
390
391
392
393
394
395
396



397
398
399
400
401
402
403
...
411
412
413
414
415
416
417

418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
...
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
...
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
...
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
...
992
993
994
995
996
997
998
999
1000





1001
1002
1003
1004
1005
1006
1007
1008
1009
1010

1011
1012
1013
1014
1015
1016
1017
....
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066

1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
....
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
....
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
  }

  /* Open the sqlite_stat[134] tables for writing. */
  for(i=0; aTable[i].zCols; i++){
    assert( i<ArraySize(aTable) );
    sqlite3VdbeAddOp4Int(v, OP_OpenWrite, iStatCur+i, aRoot[i], iDb, 3);
    sqlite3VdbeChangeP5(v, aCreateTbl[i]);

  }
}

/*
** Recommended number of samples for sqlite_stat4
*/
#ifndef SQLITE_STAT4_SAMPLES
................................................................................
  int iCol;                       /* If !isPSample, the reason for inclusion */
  u32 iHash;                      /* Tiebreaker hash */
#endif
};                                                    
struct Stat4Accum {
  tRowcnt nRow;             /* Number of rows in the entire table */
  tRowcnt nPSample;         /* How often to do a periodic sample */
  int nCol;                 /* Number of columns in index + rowid */

  int mxSample;             /* Maximum number of samples to accumulate */
  Stat4Sample current;      /* Current row as a Stat4Sample */
  u32 iPrn;                 /* Pseudo-random number used for sampling */
  Stat4Sample *aBest;       /* Array of nCol best samples */
  int iMin;                 /* Index in a[] of entry with minimum score */
  int nSample;              /* Current number of samples */
  int iGet;                 /* Index of current sample accessed by stat_get() */
................................................................................
  for(i=0; i<p->mxSample; i++) sampleClear(p->db, p->a+i);
  sampleClear(p->db, &p->current);
#endif
  sqlite3DbFree(p->db, p);
}

/*
** Implementation of the stat_init(N,C) SQL function. The two parameters



** are the number of rows in the table or index (C) and the number of columns

** in the index (N).  The second argument (C) is only used for STAT3 and STAT4.




**
** This routine allocates the Stat4Accum object in heap memory. The return 
** value is a pointer to the the Stat4Accum object encoded as a blob (i.e. 
** the size of the blob is sizeof(void*) bytes). 
*/
static void statInit(
  sqlite3_context *context,
  int argc,
  sqlite3_value **argv
){
  Stat4Accum *p;
  int nCol;                       /* Number of columns in index being sampled */

  int nColUp;                     /* nCol rounded up for alignment */
  int n;                          /* Bytes of space to allocate */
  sqlite3 *db;                    /* Database connection */
#ifdef SQLITE_ENABLE_STAT3_OR_STAT4
  int mxSample = SQLITE_STAT4_SAMPLES;
#endif

  /* Decode the three function arguments */
  UNUSED_PARAMETER(argc);
  nCol = sqlite3_value_int(argv[0]);
  assert( nCol>1 );               /* >1 because it includes the rowid column */
  nColUp = sizeof(tRowcnt)<8 ? (nCol+1)&~1 : nCol;




  /* Allocate the space required for the Stat4Accum object */
  n = sizeof(*p) 
    + sizeof(tRowcnt)*nColUp                  /* Stat4Accum.anEq */
    + sizeof(tRowcnt)*nColUp                  /* Stat4Accum.anDLt */
#ifdef SQLITE_ENABLE_STAT3_OR_STAT4
    + sizeof(tRowcnt)*nColUp                  /* Stat4Accum.anLt */
................................................................................
    sqlite3_result_error_nomem(context);
    return;
  }

  p->db = db;
  p->nRow = 0;
  p->nCol = nCol;

  p->current.anDLt = (tRowcnt*)&p[1];
  p->current.anEq = &p->current.anDLt[nColUp];

#ifdef SQLITE_ENABLE_STAT3_OR_STAT4
  {
    u8 *pSpace;                     /* Allocated space not yet assigned */
    int i;                          /* Used to iterate through p->aSample[] */

    p->iGet = -1;
    p->mxSample = mxSample;
    p->nPSample = (tRowcnt)(sqlite3_value_int64(argv[1])/(mxSample/3+1) + 1);
    p->current.anLt = &p->current.anEq[nColUp];
    p->iPrn = nCol*0x689e962d ^ sqlite3_value_int(argv[1])*0xd0944565;
  
    /* Set up the Stat4Accum.a[] and aBest[] arrays */
    p->a = (struct Stat4Sample*)&p->current.anLt[nColUp];
    p->aBest = &p->a[mxSample];
    pSpace = (u8*)(&p->a[mxSample+nCol]);
    for(i=0; i<(mxSample+nCol); i++){
      p->a[i].anEq = (tRowcnt *)pSpace; pSpace += (sizeof(tRowcnt) * nColUp);
................................................................................
  }
#endif

  /* Return a pointer to the allocated object to the caller */
  sqlite3_result_blob(context, p, sizeof(p), stat4Destructor);
}
static const FuncDef statInitFuncdef = {
  1+IsStat34,      /* nArg */
  SQLITE_UTF8,     /* funcFlags */
  0,               /* pUserData */
  0,               /* pNext */
  statInit,        /* xFunc */
  0,               /* xStep */
  0,               /* xFinalize */
  "stat_init",     /* zName */
................................................................................

  /* The three function arguments */
  Stat4Accum *p = (Stat4Accum*)sqlite3_value_blob(argv[0]);
  int iChng = sqlite3_value_int(argv[1]);

  UNUSED_PARAMETER( argc );
  UNUSED_PARAMETER( context );
  assert( p->nCol>1 );        /* Includes rowid field */
  assert( iChng<p->nCol );

  if( p->nRow==0 ){
    /* This is the first call to this function. Do initialization. */
    for(i=0; i<p->nCol; i++) p->current.anEq[i] = 1;
  }else{
    /* Second and subsequent calls get processed here */
................................................................................
    ** rows, then each estimate is computed as:
    **
    **        I = (K+D-1)/D
    */
    char *z;
    int i;

    char *zRet = sqlite3MallocZero(p->nCol * 25);
    if( zRet==0 ){
      sqlite3_result_error_nomem(context);
      return;
    }

    sqlite3_snprintf(24, zRet, "%llu", (u64)p->nRow);
    z = zRet + sqlite3Strlen30(zRet);
    for(i=0; i<(p->nCol-1); i++){
      u64 nDistinct = p->current.anDLt[i] + 1;
      u64 iVal = (p->nRow + nDistinct - 1) / nDistinct;
      sqlite3_snprintf(24, z, " %llu", iVal);
      z += sqlite3Strlen30(z);
      assert( p->current.anEq[i] );
    }
    assert( z[0]=='\0' && z>zRet );
................................................................................
    int addrRewind;               /* Address of "OP_Rewind iIdxCur" */
    int addrGotoChng0;            /* Address of "Goto addr_chng_0" */
    int addrNextRow;              /* Address of "next_row:" */
    const char *zIdxName;         /* Name of the index */

    if( pOnlyIdx && pOnlyIdx!=pIdx ) continue;
    if( pIdx->pPartIdxWhere==0 ) needTableCnt = 0;
    VdbeNoopComment((v, "Begin analysis of %s", pIdx->zName));
    nCol = pIdx->nKeyCol;





    aGotoChng = sqlite3DbMallocRaw(db, sizeof(int)*(nCol+1));
    if( aGotoChng==0 ) continue;

    /* Populate the register containing the index name. */
    if( IsPrimaryKeyIndex(pIdx) && !HasRowid(pTab) ){
      zIdxName = pTab->zName;
    }else{
      zIdxName = pIdx->zName;
    }
    sqlite3VdbeAddOp4(v, OP_String8, 0, regIdxname, 0, zIdxName, 0);


    /*
    ** Pseudo-code for loop that calls stat_push():
    **
    **   Rewind csr
    **   if eof(csr) goto end_of_scan;
    **   regChng = 0
................................................................................
    ** 
    **    (1) the number of columns in the index including the rowid,
    **    (2) the number of rows in the index,
    **
    ** The second argument is only used for STAT3 and STAT4
    */
#ifdef SQLITE_ENABLE_STAT3_OR_STAT4
    sqlite3VdbeAddOp2(v, OP_Count, iIdxCur, regStat4+2);
#endif
    sqlite3VdbeAddOp2(v, OP_Integer, nCol+1, regStat4+1);

    sqlite3VdbeAddOp3(v, OP_Function, 0, regStat4+1, regStat4);
    sqlite3VdbeChangeP4(v, -1, (char*)&statInitFuncdef, P4_FUNCDEF);
    sqlite3VdbeChangeP5(v, 1+IsStat34);

    /* Implementation of the following:
    **
    **   Rewind csr
    **   if eof(csr) goto end_of_scan;
    **   regChng = 0
    **   goto next_push_0;
................................................................................
      int regSample = regStat1+3;
      int regCol = regStat1+4;
      int regSampleRowid = regCol + nCol;
      int addrNext;
      int addrIsNull;
      u8 seekOp = HasRowid(pTab) ? OP_NotExists : OP_NotFound;

      pParse->nMem = MAX(pParse->nMem, regCol+nCol+1);

      addrNext = sqlite3VdbeCurrentAddr(v);
      callStatGet(v, regStat4, STAT_GET_ROWID, regSampleRowid);
      addrIsNull = sqlite3VdbeAddOp1(v, OP_IsNull, regSampleRowid);
      VdbeCoverage(v);
      callStatGet(v, regStat4, STAT_GET_NEQ, regEq);
      callStatGet(v, regStat4, STAT_GET_NLT, regLt);
................................................................................
      sqlite3ExprCodeGetColumnOfTable(v, pTab, iTabCur, 
                                      pIdx->aiColumn[0], regSample);
#else
      for(i=0; i<nCol; i++){
        i16 iCol = pIdx->aiColumn[i];
        sqlite3ExprCodeGetColumnOfTable(v, pTab, iTabCur, iCol, regCol+i);
      }
      sqlite3VdbeAddOp3(v, OP_MakeRecord, regCol, nCol+1, regSample);
#endif
      sqlite3VdbeAddOp3(v, OP_MakeRecord, regTabname, 6, regTemp);
      sqlite3VdbeAddOp2(v, OP_NewRowid, iStatCur+1, regNewRowid);
      sqlite3VdbeAddOp3(v, OP_Insert, iStatCur+1, regTemp, regNewRowid);
      sqlite3VdbeAddOp2(v, OP_Goto, 1, addrNext); /* P1==1 for end-of-loop */
      sqlite3VdbeJumpHere(v, addrIsNull);
    }







>







 







|
>







 







|
>
>
>
|
>
|
>
>
>
>












>










|

>
>
>







 







>










|

|







 







|







 







|







 







|







|







 







|
|
>
>
>
>
>




<
<
<
<
<

>







 







|

|
>


|







 







|







 







|







242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
...
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
...
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
...
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
...
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
...
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
...
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
....
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024





1025
1026
1027
1028
1029
1030
1031
1032
1033
....
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
....
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
....
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
  }

  /* Open the sqlite_stat[134] tables for writing. */
  for(i=0; aTable[i].zCols; i++){
    assert( i<ArraySize(aTable) );
    sqlite3VdbeAddOp4Int(v, OP_OpenWrite, iStatCur+i, aRoot[i], iDb, 3);
    sqlite3VdbeChangeP5(v, aCreateTbl[i]);
    VdbeComment((v, aTable[i].zName));
  }
}

/*
** Recommended number of samples for sqlite_stat4
*/
#ifndef SQLITE_STAT4_SAMPLES
................................................................................
  int iCol;                       /* If !isPSample, the reason for inclusion */
  u32 iHash;                      /* Tiebreaker hash */
#endif
};                                                    
struct Stat4Accum {
  tRowcnt nRow;             /* Number of rows in the entire table */
  tRowcnt nPSample;         /* How often to do a periodic sample */
  int nCol;                 /* Number of columns in index + pk/rowid */
  int nKeyCol;              /* Number of index columns w/o the pk/rowid */
  int mxSample;             /* Maximum number of samples to accumulate */
  Stat4Sample current;      /* Current row as a Stat4Sample */
  u32 iPrn;                 /* Pseudo-random number used for sampling */
  Stat4Sample *aBest;       /* Array of nCol best samples */
  int iMin;                 /* Index in a[] of entry with minimum score */
  int nSample;              /* Current number of samples */
  int iGet;                 /* Index of current sample accessed by stat_get() */
................................................................................
  for(i=0; i<p->mxSample; i++) sampleClear(p->db, p->a+i);
  sampleClear(p->db, &p->current);
#endif
  sqlite3DbFree(p->db, p);
}

/*
** Implementation of the stat_init(N,K,C) SQL function. The three parameters
** are:
**     N:    The number of columns in the index including the rowid/pk
**     K:    The number of columns in the index excluding the rowid/pk
**     C:    The number of rows in the index
**
** C is only used for STAT3 and STAT4.
**
** For ordinary rowid tables, N==K+1.  But for WITHOUT ROWID tables,
** N=K+P where P is the number of columns in the primary key.  For the
** covering index that implements the original WITHOUT ROWID table, N==K.
**
** This routine allocates the Stat4Accum object in heap memory. The return 
** value is a pointer to the the Stat4Accum object encoded as a blob (i.e. 
** the size of the blob is sizeof(void*) bytes). 
*/
static void statInit(
  sqlite3_context *context,
  int argc,
  sqlite3_value **argv
){
  Stat4Accum *p;
  int nCol;                       /* Number of columns in index being sampled */
  int nKeyCol;                    /* Number of key columns */
  int nColUp;                     /* nCol rounded up for alignment */
  int n;                          /* Bytes of space to allocate */
  sqlite3 *db;                    /* Database connection */
#ifdef SQLITE_ENABLE_STAT3_OR_STAT4
  int mxSample = SQLITE_STAT4_SAMPLES;
#endif

  /* Decode the three function arguments */
  UNUSED_PARAMETER(argc);
  nCol = sqlite3_value_int(argv[0]);
  assert( nCol>0 );
  nColUp = sizeof(tRowcnt)<8 ? (nCol+1)&~1 : nCol;
  nKeyCol = sqlite3_value_int(argv[1]);
  assert( nKeyCol<=nCol );
  assert( nKeyCol>0 );

  /* Allocate the space required for the Stat4Accum object */
  n = sizeof(*p) 
    + sizeof(tRowcnt)*nColUp                  /* Stat4Accum.anEq */
    + sizeof(tRowcnt)*nColUp                  /* Stat4Accum.anDLt */
#ifdef SQLITE_ENABLE_STAT3_OR_STAT4
    + sizeof(tRowcnt)*nColUp                  /* Stat4Accum.anLt */
................................................................................
    sqlite3_result_error_nomem(context);
    return;
  }

  p->db = db;
  p->nRow = 0;
  p->nCol = nCol;
  p->nKeyCol = nKeyCol;
  p->current.anDLt = (tRowcnt*)&p[1];
  p->current.anEq = &p->current.anDLt[nColUp];

#ifdef SQLITE_ENABLE_STAT3_OR_STAT4
  {
    u8 *pSpace;                     /* Allocated space not yet assigned */
    int i;                          /* Used to iterate through p->aSample[] */

    p->iGet = -1;
    p->mxSample = mxSample;
    p->nPSample = (tRowcnt)(sqlite3_value_int64(argv[2])/(mxSample/3+1) + 1);
    p->current.anLt = &p->current.anEq[nColUp];
    p->iPrn = nCol*0x689e962d ^ sqlite3_value_int(argv[2])*0xd0944565;
  
    /* Set up the Stat4Accum.a[] and aBest[] arrays */
    p->a = (struct Stat4Sample*)&p->current.anLt[nColUp];
    p->aBest = &p->a[mxSample];
    pSpace = (u8*)(&p->a[mxSample+nCol]);
    for(i=0; i<(mxSample+nCol); i++){
      p->a[i].anEq = (tRowcnt *)pSpace; pSpace += (sizeof(tRowcnt) * nColUp);
................................................................................
  }
#endif

  /* Return a pointer to the allocated object to the caller */
  sqlite3_result_blob(context, p, sizeof(p), stat4Destructor);
}
static const FuncDef statInitFuncdef = {
  2+IsStat34,      /* nArg */
  SQLITE_UTF8,     /* funcFlags */
  0,               /* pUserData */
  0,               /* pNext */
  statInit,        /* xFunc */
  0,               /* xStep */
  0,               /* xFinalize */
  "stat_init",     /* zName */
................................................................................

  /* The three function arguments */
  Stat4Accum *p = (Stat4Accum*)sqlite3_value_blob(argv[0]);
  int iChng = sqlite3_value_int(argv[1]);

  UNUSED_PARAMETER( argc );
  UNUSED_PARAMETER( context );
  assert( p->nCol>0 );
  assert( iChng<p->nCol );

  if( p->nRow==0 ){
    /* This is the first call to this function. Do initialization. */
    for(i=0; i<p->nCol; i++) p->current.anEq[i] = 1;
  }else{
    /* Second and subsequent calls get processed here */
................................................................................
    ** rows, then each estimate is computed as:
    **
    **        I = (K+D-1)/D
    */
    char *z;
    int i;

    char *zRet = sqlite3MallocZero( (p->nKeyCol+1)*25 );
    if( zRet==0 ){
      sqlite3_result_error_nomem(context);
      return;
    }

    sqlite3_snprintf(24, zRet, "%llu", (u64)p->nRow);
    z = zRet + sqlite3Strlen30(zRet);
    for(i=0; i<p->nKeyCol; i++){
      u64 nDistinct = p->current.anDLt[i] + 1;
      u64 iVal = (p->nRow + nDistinct - 1) / nDistinct;
      sqlite3_snprintf(24, z, " %llu", iVal);
      z += sqlite3Strlen30(z);
      assert( p->current.anEq[i] );
    }
    assert( z[0]=='\0' && z>zRet );
................................................................................
    int addrRewind;               /* Address of "OP_Rewind iIdxCur" */
    int addrGotoChng0;            /* Address of "Goto addr_chng_0" */
    int addrNextRow;              /* Address of "next_row:" */
    const char *zIdxName;         /* Name of the index */

    if( pOnlyIdx && pOnlyIdx!=pIdx ) continue;
    if( pIdx->pPartIdxWhere==0 ) needTableCnt = 0;
    if( !HasRowid(pTab) && IsPrimaryKeyIndex(pIdx) ){
      nCol = pIdx->nKeyCol;
      zIdxName = pTab->zName;
    }else{
      nCol = pIdx->nColumn;
      zIdxName = pIdx->zName;
    }
    aGotoChng = sqlite3DbMallocRaw(db, sizeof(int)*(nCol+1));
    if( aGotoChng==0 ) continue;

    /* Populate the register containing the index name. */





    sqlite3VdbeAddOp4(v, OP_String8, 0, regIdxname, 0, zIdxName, 0);
    VdbeComment((v, "Analysis for %s.%s", pTab->zName, zIdxName));

    /*
    ** Pseudo-code for loop that calls stat_push():
    **
    **   Rewind csr
    **   if eof(csr) goto end_of_scan;
    **   regChng = 0
................................................................................
    ** 
    **    (1) the number of columns in the index including the rowid,
    **    (2) the number of rows in the index,
    **
    ** The second argument is only used for STAT3 and STAT4
    */
#ifdef SQLITE_ENABLE_STAT3_OR_STAT4
    sqlite3VdbeAddOp2(v, OP_Count, iIdxCur, regStat4+3);
#endif
    sqlite3VdbeAddOp2(v, OP_Integer, nCol, regStat4+1);
    sqlite3VdbeAddOp2(v, OP_Integer, pIdx->nKeyCol, regStat4+2);
    sqlite3VdbeAddOp3(v, OP_Function, 0, regStat4+1, regStat4);
    sqlite3VdbeChangeP4(v, -1, (char*)&statInitFuncdef, P4_FUNCDEF);
    sqlite3VdbeChangeP5(v, 2+IsStat34);

    /* Implementation of the following:
    **
    **   Rewind csr
    **   if eof(csr) goto end_of_scan;
    **   regChng = 0
    **   goto next_push_0;
................................................................................
      int regSample = regStat1+3;
      int regCol = regStat1+4;
      int regSampleRowid = regCol + nCol;
      int addrNext;
      int addrIsNull;
      u8 seekOp = HasRowid(pTab) ? OP_NotExists : OP_NotFound;

      pParse->nMem = MAX(pParse->nMem, regCol+nCol);

      addrNext = sqlite3VdbeCurrentAddr(v);
      callStatGet(v, regStat4, STAT_GET_ROWID, regSampleRowid);
      addrIsNull = sqlite3VdbeAddOp1(v, OP_IsNull, regSampleRowid);
      VdbeCoverage(v);
      callStatGet(v, regStat4, STAT_GET_NEQ, regEq);
      callStatGet(v, regStat4, STAT_GET_NLT, regLt);
................................................................................
      sqlite3ExprCodeGetColumnOfTable(v, pTab, iTabCur, 
                                      pIdx->aiColumn[0], regSample);
#else
      for(i=0; i<nCol; i++){
        i16 iCol = pIdx->aiColumn[i];
        sqlite3ExprCodeGetColumnOfTable(v, pTab, iTabCur, iCol, regCol+i);
      }
      sqlite3VdbeAddOp3(v, OP_MakeRecord, regCol, nCol, regSample);
#endif
      sqlite3VdbeAddOp3(v, OP_MakeRecord, regTabname, 6, regTemp);
      sqlite3VdbeAddOp2(v, OP_NewRowid, iStatCur+1, regNewRowid);
      sqlite3VdbeAddOp3(v, OP_Insert, iStatCur+1, regTemp, regNewRowid);
      sqlite3VdbeAddOp2(v, OP_Goto, 1, addrNext); /* P1==1 for end-of-loop */
      sqlite3VdbeJumpHere(v, addrIsNull);
    }

Changes to src/where.c.

540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
....
2197
2198
2199
2200
2201
2202
2203
2204
2205
2206
2207
2208
2209
2210
2211
2212
2213
2214
2215
2216
2217
2218
2219
2220
2221
2222
2223
2224
....
2641
2642
2643
2644
2645
2646
2647
2648
2649
2650
2651
2652
2653
2654
2655
2656
2657
2658
2659
2660
2661
2662
2663
2664
2665
2666
2667
2668
2669
2670
2671
2672
....
4160
4161
4162
4163
4164
4165
4166
4167
4168
4169
4170
4171
4172
4173
4174
4175
4176
4177
4178
4179
....
4355
4356
4357
4358
4359
4360
4361
4362
4363
4364
4365
4366
4367
4368
4369
....
4502
4503
4504
4505
4506
4507
4508

4509
4510
4511
4512
4513
4514
4515

  /* memset(pScan, 0, sizeof(*pScan)); */
  pScan->pOrigWC = pWC;
  pScan->pWC = pWC;
  if( pIdx && iColumn>=0 ){
    pScan->idxaff = pIdx->pTable->aCol[iColumn].affinity;
    for(j=0; pIdx->aiColumn[j]!=iColumn; j++){
      if( NEVER(j>=pIdx->nKeyCol) ) return 0;
    }
    pScan->zCollName = pIdx->azColl[j];
  }else{
    pScan->idxaff = 0;
    pScan->zCollName = 0;
  }
  pScan->opMask = opMask;
................................................................................
  UnpackedRecord *pRec = pBuilder->pRec;
  u8 aff;                   /* Column affinity */
  int rc;                   /* Subfunction return code */
  tRowcnt a[2];             /* Statistics */
  int bOk;

  assert( nEq>=1 );
  assert( nEq<=(p->nKeyCol+1) );
  assert( p->aSample!=0 );
  assert( p->nSample>0 );
  assert( pBuilder->nRecValid<nEq );

  /* If values are not available for all fields of the index to the left
  ** of this one, no estimate can be made. Return SQLITE_NOTFOUND. */
  if( pBuilder->nRecValid<(nEq-1) ){
    return SQLITE_NOTFOUND;
  }

  /* This is an optimization only. The call to sqlite3Stat4ProbeSetValue()
  ** below would return the same value.  */
  if( nEq>p->nKeyCol ){
    *pnRow = 1;
    return SQLITE_OK;
  }

  aff = p->pTable->aCol[p->aiColumn[nEq-1]].affinity;
  rc = sqlite3Stat4ProbeSetValue(pParse, p, &pRec, pExpr, aff, nEq-1, &bOk);
  pBuilder->pRec = pRec;
................................................................................
  if( nEq==0 && (pLoop->wsFlags & (WHERE_BTM_LIMIT|WHERE_TOP_LIMIT))==0 ){
    return 0;
  }
  sqlite3StrAccumInit(&txt, 0, 0, SQLITE_MAX_LENGTH);
  txt.db = db;
  sqlite3StrAccumAppend(&txt, " (", 2);
  for(i=0; i<nEq; i++){
    char *z = (i==pIndex->nKeyCol ) ? "rowid" : aCol[aiColumn[i]].zName;
    if( i>=nSkip ){
      explainAppendTerm(&txt, i, z, "=");
    }else{
      if( i ) sqlite3StrAccumAppend(&txt, " AND ", 5);
      sqlite3StrAccumAppend(&txt, "ANY(", 4);
      sqlite3StrAccumAppendAll(&txt, z);
      sqlite3StrAccumAppend(&txt, ")", 1);
    }
  }

  j = i;
  if( pLoop->wsFlags&WHERE_BTM_LIMIT ){
    char *z = (j==pIndex->nKeyCol ) ? "rowid" : aCol[aiColumn[j]].zName;
    explainAppendTerm(&txt, i++, z, ">");
  }
  if( pLoop->wsFlags&WHERE_TOP_LIMIT ){
    char *z = (j==pIndex->nKeyCol ) ? "rowid" : aCol[aiColumn[j]].zName;
    explainAppendTerm(&txt, i, z, "<");
  }
  sqlite3StrAccumAppend(&txt, ")", 1);
  return sqlite3StrAccumFinish(&txt);
}

/*
................................................................................
  }else if( pProbe->tnum<=0 || (pSrc->jointype & JT_LEFT)!=0 ){
    opMask = WO_EQ|WO_IN|WO_GT|WO_GE|WO_LT|WO_LE;
  }else{
    opMask = WO_EQ|WO_IN|WO_ISNULL|WO_GT|WO_GE|WO_LT|WO_LE;
  }
  if( pProbe->bUnordered ) opMask &= ~(WO_GT|WO_GE|WO_LT|WO_LE);

  assert( pNew->u.btree.nEq<=pProbe->nKeyCol );
  if( pNew->u.btree.nEq < pProbe->nKeyCol ){
    iCol = pProbe->aiColumn[pNew->u.btree.nEq];
  }else{
    iCol = -1;
  }
  pTerm = whereScanInit(&scan, pBuilder->pWC, pSrc->iCursor, iCol,
                        opMask, pProbe);
  saved_nEq = pNew->u.btree.nEq;
  saved_nSkip = pNew->u.btree.nSkip;
  saved_nLTerm = pNew->nLTerm;
  saved_wsFlags = pNew->wsFlags;
  saved_prereq = pNew->prereq;
................................................................................
    if( pNew->wsFlags & WHERE_COLUMN_RANGE ){
      pNew->nOut = saved_nOut;
    }else{
      pNew->nOut = nOutUnadjusted;
    }

    if( (pNew->wsFlags & WHERE_TOP_LIMIT)==0
     && pNew->u.btree.nEq<(pProbe->nKeyCol + (pProbe->zName!=0))
    ){
      whereLoopAddBtreeIndex(pBuilder, pSrc, pProbe, nInMul+nIn);
    }
    pNew->nOut = saved_nOut;
#ifdef SQLITE_ENABLE_STAT3_OR_STAT4
    pBuilder->nRecValid = nRecValid;
#endif
................................................................................
    /* There is no INDEXED BY clause.  Create a fake Index object in local
    ** variable sPk to represent the rowid primary key index.  Make this
    ** fake index the first in a chain of Index objects with all of the real
    ** indices to follow */
    Index *pFirst;                  /* First of real indices on the table */
    memset(&sPk, 0, sizeof(Index));
    sPk.nKeyCol = 1;

    sPk.aiColumn = &aiColumnPk;
    sPk.aiRowLogEst = aiRowEstPk;
    sPk.onError = OE_Replace;
    sPk.pTable = pTab;
    sPk.szIdxRow = pTab->szTabRow;
    aiRowEstPk[0] = pTab->nRowLogEst;
    aiRowEstPk[1] = 0;







|







 







|












|







 







|












|



|







 







|
<
|
<
<
|







 







|







 







>







540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
....
2197
2198
2199
2200
2201
2202
2203
2204
2205
2206
2207
2208
2209
2210
2211
2212
2213
2214
2215
2216
2217
2218
2219
2220
2221
2222
2223
2224
....
2641
2642
2643
2644
2645
2646
2647
2648
2649
2650
2651
2652
2653
2654
2655
2656
2657
2658
2659
2660
2661
2662
2663
2664
2665
2666
2667
2668
2669
2670
2671
2672
....
4160
4161
4162
4163
4164
4165
4166
4167

4168


4169
4170
4171
4172
4173
4174
4175
4176
....
4352
4353
4354
4355
4356
4357
4358
4359
4360
4361
4362
4363
4364
4365
4366
....
4499
4500
4501
4502
4503
4504
4505
4506
4507
4508
4509
4510
4511
4512
4513

  /* memset(pScan, 0, sizeof(*pScan)); */
  pScan->pOrigWC = pWC;
  pScan->pWC = pWC;
  if( pIdx && iColumn>=0 ){
    pScan->idxaff = pIdx->pTable->aCol[iColumn].affinity;
    for(j=0; pIdx->aiColumn[j]!=iColumn; j++){
      if( NEVER(j>pIdx->nColumn) ) return 0;
    }
    pScan->zCollName = pIdx->azColl[j];
  }else{
    pScan->idxaff = 0;
    pScan->zCollName = 0;
  }
  pScan->opMask = opMask;
................................................................................
  UnpackedRecord *pRec = pBuilder->pRec;
  u8 aff;                   /* Column affinity */
  int rc;                   /* Subfunction return code */
  tRowcnt a[2];             /* Statistics */
  int bOk;

  assert( nEq>=1 );
  assert( nEq<=p->nColumn );
  assert( p->aSample!=0 );
  assert( p->nSample>0 );
  assert( pBuilder->nRecValid<nEq );

  /* If values are not available for all fields of the index to the left
  ** of this one, no estimate can be made. Return SQLITE_NOTFOUND. */
  if( pBuilder->nRecValid<(nEq-1) ){
    return SQLITE_NOTFOUND;
  }

  /* This is an optimization only. The call to sqlite3Stat4ProbeSetValue()
  ** below would return the same value.  */
  if( nEq>=p->nColumn ){
    *pnRow = 1;
    return SQLITE_OK;
  }

  aff = p->pTable->aCol[p->aiColumn[nEq-1]].affinity;
  rc = sqlite3Stat4ProbeSetValue(pParse, p, &pRec, pExpr, aff, nEq-1, &bOk);
  pBuilder->pRec = pRec;
................................................................................
  if( nEq==0 && (pLoop->wsFlags & (WHERE_BTM_LIMIT|WHERE_TOP_LIMIT))==0 ){
    return 0;
  }
  sqlite3StrAccumInit(&txt, 0, 0, SQLITE_MAX_LENGTH);
  txt.db = db;
  sqlite3StrAccumAppend(&txt, " (", 2);
  for(i=0; i<nEq; i++){
    char *z = aiColumn[i] < 0 ? "rowid" : aCol[aiColumn[i]].zName;
    if( i>=nSkip ){
      explainAppendTerm(&txt, i, z, "=");
    }else{
      if( i ) sqlite3StrAccumAppend(&txt, " AND ", 5);
      sqlite3StrAccumAppend(&txt, "ANY(", 4);
      sqlite3StrAccumAppendAll(&txt, z);
      sqlite3StrAccumAppend(&txt, ")", 1);
    }
  }

  j = i;
  if( pLoop->wsFlags&WHERE_BTM_LIMIT ){
    char *z = aiColumn[j] < 0 ? "rowid" : aCol[aiColumn[j]].zName;
    explainAppendTerm(&txt, i++, z, ">");
  }
  if( pLoop->wsFlags&WHERE_TOP_LIMIT ){
    char *z = aiColumn[j] < 0 ? "rowid" : aCol[aiColumn[j]].zName;
    explainAppendTerm(&txt, i, z, "<");
  }
  sqlite3StrAccumAppend(&txt, ")", 1);
  return sqlite3StrAccumFinish(&txt);
}

/*
................................................................................
  }else if( pProbe->tnum<=0 || (pSrc->jointype & JT_LEFT)!=0 ){
    opMask = WO_EQ|WO_IN|WO_GT|WO_GE|WO_LT|WO_LE;
  }else{
    opMask = WO_EQ|WO_IN|WO_ISNULL|WO_GT|WO_GE|WO_LT|WO_LE;
  }
  if( pProbe->bUnordered ) opMask &= ~(WO_GT|WO_GE|WO_LT|WO_LE);

  assert( pNew->u.btree.nEq<pProbe->nColumn );

  iCol = pProbe->aiColumn[pNew->u.btree.nEq];



  pTerm = whereScanInit(&scan, pBuilder->pWC, pSrc->iCursor, iCol,
                        opMask, pProbe);
  saved_nEq = pNew->u.btree.nEq;
  saved_nSkip = pNew->u.btree.nSkip;
  saved_nLTerm = pNew->nLTerm;
  saved_wsFlags = pNew->wsFlags;
  saved_prereq = pNew->prereq;
................................................................................
    if( pNew->wsFlags & WHERE_COLUMN_RANGE ){
      pNew->nOut = saved_nOut;
    }else{
      pNew->nOut = nOutUnadjusted;
    }

    if( (pNew->wsFlags & WHERE_TOP_LIMIT)==0
     && pNew->u.btree.nEq<pProbe->nColumn
    ){
      whereLoopAddBtreeIndex(pBuilder, pSrc, pProbe, nInMul+nIn);
    }
    pNew->nOut = saved_nOut;
#ifdef SQLITE_ENABLE_STAT3_OR_STAT4
    pBuilder->nRecValid = nRecValid;
#endif
................................................................................
    /* There is no INDEXED BY clause.  Create a fake Index object in local
    ** variable sPk to represent the rowid primary key index.  Make this
    ** fake index the first in a chain of Index objects with all of the real
    ** indices to follow */
    Index *pFirst;                  /* First of real indices on the table */
    memset(&sPk, 0, sizeof(Index));
    sPk.nKeyCol = 1;
    sPk.nColumn = 1;
    sPk.aiColumn = &aiColumnPk;
    sPk.aiRowLogEst = aiRowEstPk;
    sPk.onError = OE_Replace;
    sPk.pTable = pTab;
    sPk.szIdxRow = pTab->szTabRow;
    aiRowEstPk[0] = pTab->nRowLogEst;
    aiRowEstPk[1] = 0;

Changes to test/analyze9.test.

948
949
950
951
952
953
954
955


































































956


for {set i 0} {$i<16} {incr i} {
    set val "$i $i $i $i"
    do_execsql_test 20.3.$i {
      SELECT count(*) FROM sqlite_stat4 
      WHERE lrange(test_decode(sample), 0, 3)=$val
    } {1}
}



































































finish_test










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

>
>
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
for {set i 0} {$i<16} {incr i} {
    set val "$i $i $i $i"
    do_execsql_test 20.3.$i {
      SELECT count(*) FROM sqlite_stat4 
      WHERE lrange(test_decode(sample), 0, 3)=$val
    } {1}
}

#-------------------------------------------------------------------------
#
reset_db

do_execsql_test 21.0 {
  CREATE TABLE t2(a, b);
  CREATE INDEX i2 ON t2(a);
}

do_test 21.1 {
  for {set i 1} {$i < 100} {incr i} {
    execsql { 
      INSERT INTO t2 VALUES(CASE WHEN $i < 80 THEN 'one' ELSE 'two' END, $i) 
    }
  }
  execsql ANALYZE
} {}

# Condition (a='one') matches 80% of the table. (rowid<10) reduces this to
# 10%, but (rowid<50) only reduces it to 50%. So in the first case below
# the index is used. In the second, it is not. 
#
do_eqp_test 21.2 {
  SELECT * FROM t2 WHERE a='one' AND rowid < 10
} {/*USING INDEX i2 (a=? AND rowid<?)*/}
do_eqp_test 21.3 {
  SELECT * FROM t2 WHERE a='one' AND rowid < 50
} {/*USING INTEGER PRIMARY KEY*/}

#-------------------------------------------------------------------------
#
reset_db
do_execsql_test 22.0 {
  CREATE TABLE t3(a, b, c, d, PRIMARY KEY(a, b)) WITHOUT ROWID;
}
do_execsql_test 22.1 {
  WITH r(x) AS (
    SELECT 1
    UNION ALL
    SELECT x+1 FROM r WHERE x<=100
  )

  INSERT INTO t3 SELECT
    CASE WHEN (x>45 AND x<96) THEN 'B' ELSE 'A' END,  /* Column "a" */
    x,                                                /* Column "b" */
    CASE WHEN (x<51) THEN 'one' ELSE 'two' END,       /* Column "c" */
    x                                                 /* Column "d" */
  FROM r;

  CREATE INDEX i3 ON t3(c);
  CREATE INDEX i4 ON t3(d);
  ANALYZE;
}

# Expression (c='one' AND a='B') matches 5 table rows. But (c='one' AND a=A')
# matches 45. Expression (d<?) matches 20. Neither index is a covering index.
#
# Therefore, with stat4 data, SQLite prefers (c='one' AND a='B') over (d<20),
# and (d<20) over (c='one' AND a='A').
foreach {tn where res} {
  1 "c='one' AND a='B' AND d < 20"   {/*INDEX i3 (c=? AND a=?)*/}
  2 "c='one' AND a='A' AND d < 20"   {/*INDEX i4 (d<?)*/}
} {
  do_eqp_test 22.2.$tn "SELECT * FROM t3 WHERE $where" $res
}

finish_test


Changes to test/without_rowid1.test.

209
210
211
212
213
214
215
216
































































217
  CREATE TABLE t42(x);
  INSERT INTO t42 VALUES('xyz');
  SELECT t42.rowid FROM t41, t42;
} {1}
do_execsql_test 4.2 {
  SELECT t42.rowid FROM t42, t41;
} {1}
  
































































finish_test







|
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>

209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
  CREATE TABLE t42(x);
  INSERT INTO t42 VALUES('xyz');
  SELECT t42.rowid FROM t41, t42;
} {1}
do_execsql_test 4.2 {
  SELECT t42.rowid FROM t42, t41;
} {1}


#--------------------------------------------------------------------------
# The following tests verify that the trailing PK fields added to each
# entry in an index on a WITHOUT ROWID table are used correctly.
#
do_execsql_test 5.0 {
  CREATE TABLE t45(a PRIMARY KEY, b, c) WITHOUT ROWID;
  CREATE INDEX i45 ON t45(b);

  INSERT INTO t45 VALUES(2, 'one', 'x');
  INSERT INTO t45 VALUES(4, 'one', 'x');
  INSERT INTO t45 VALUES(6, 'one', 'x');
  INSERT INTO t45 VALUES(8, 'one', 'x');
  INSERT INTO t45 VALUES(10, 'one', 'x');

  INSERT INTO t45 VALUES(1, 'two', 'x');
  INSERT INTO t45 VALUES(3, 'two', 'x');
  INSERT INTO t45 VALUES(5, 'two', 'x');
  INSERT INTO t45 VALUES(7, 'two', 'x');
  INSERT INTO t45 VALUES(9, 'two', 'x');
}

do_eqp_test 5.1 {
  SELECT * FROM t45 WHERE b=? AND a>?
} {/*USING INDEX i45 (b=? AND a>?)*/}

do_execsql_test 5.2 {
  SELECT * FROM t45 WHERE b='two' AND a>4
} {5 two x 7 two x 9 two x}

do_execsql_test 5.3 {
  SELECT * FROM t45 WHERE b='one' AND a<8
} { 2 one x 4 one x 6 one x }

do_execsql_test 5.4 {
  CREATE TABLE t46(a, b, c, d, PRIMARY KEY(a, b)) WITHOUT ROWID;
  WITH r(x) AS (
    SELECT 1 UNION ALL SELECT x+1 FROM r WHERE x<100
  )
  INSERT INTO t46 SELECT x / 20, x % 20, x % 10, x FROM r;
}

set queries {
  1    2    "c = 5 AND a = 1"          {/*i46 (c=? AND a=?)*/}
  2    6    "c = 4 AND a < 3"          {/*i46 (c=? AND a<?)*/}
  3    4    "c = 2 AND a >= 3"         {/*i46 (c=? AND a>?)*/}
  4    1    "c = 2 AND a = 1 AND b<10" {/*i46 (c=? AND a=? AND b<?)*/}
  5    1    "c = 0 AND a = 0 AND b>5"  {/*i46 (c=? AND a=? AND b>?)*/}
}

foreach {tn cnt where eqp} $queries {
  do_execsql_test 5.5.$tn.1 "SELECT count(*) FROM t46 WHERE $where" $cnt
}

do_execsql_test 5.6 {
  CREATE INDEX i46 ON t46(c);
}

foreach {tn cnt where eqp} $queries {
  do_execsql_test 5.7.$tn.1 "SELECT count(*) FROM t46 WHERE $where" $cnt
  do_eqp_test 5.7.$tn.2  "SELECT count(*) FROM t46 WHERE $where" $eqp
}

  
finish_test