SQLite
Check-in [cbfc0f6d49b6607bb4eb45bfe4c258e39ac27403]
Not logged in
Overview
SHA1 Hash:cbfc0f6d49b6607bb4eb45bfe4c258e39ac27403
Date: 2014-04-07 18:44:00
User: dan
Comment:Experimental multi-threaded sorting changes to begin merging PMAs before SorterRewind() is called.
Tags And Properties
Changes
Hide Diffs Unified Diffs Ignore Whitespace Patch

Changes to src/shell.c

3531
3532
3533
3534
3535
3536
3537
3538
3539
3540
3541
3542
3543
3544
3545
  memcpy(data->separator,"|", 2);
  data->showHeader = 0;
  sqlite3_config(SQLITE_CONFIG_URI, 1);
  sqlite3_config(SQLITE_CONFIG_LOG, shellLog, data);
  sqlite3_snprintf(sizeof(mainPrompt), mainPrompt,"sqlite> ");
  sqlite3_snprintf(sizeof(continuePrompt), continuePrompt,"   ...> ");
  sqlite3_config(SQLITE_CONFIG_MULTITHREAD);
  sqlite3_config(SQLITE_CONFIG_WORKER_THREADS, 3);
}

/*
** Output text to the console in a font that attracts extra attention.
*/
#ifdef _WIN32
static void printBold(const char *zText){







|







3531
3532
3533
3534
3535
3536
3537
3538
3539
3540
3541
3542
3543
3544
3545
  memcpy(data->separator,"|", 2);
  data->showHeader = 0;
  sqlite3_config(SQLITE_CONFIG_URI, 1);
  sqlite3_config(SQLITE_CONFIG_LOG, shellLog, data);
  sqlite3_snprintf(sizeof(mainPrompt), mainPrompt,"sqlite> ");
  sqlite3_snprintf(sizeof(continuePrompt), continuePrompt,"   ...> ");
  sqlite3_config(SQLITE_CONFIG_MULTITHREAD);
  sqlite3_config(SQLITE_CONFIG_WORKER_THREADS, 4);
}

/*
** Output text to the console in a font that attracts extra attention.
*/
#ifdef _WIN32
static void printBold(const char *zText){

Changes to src/vdbesort.c

93
94
95
96
97
98
99



100

101
102

103
104
105
106




























107
108
109
110
111
112
113
...
137
138
139
140
141
142
143
144
145
146
147

148
149
150
151
152
153
154
155
156
157
158
159
160

161
162
163
164
165
166
167
...
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
...
492
493
494
495
496
497
498

499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
...
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
...
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577

578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
...
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
...
670
671
672
673
674
675
676

677
678
679
680
681
682
683
...
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
...
706
707
708
709
710
711
712
713

714
715
716
717
718





719
720
721
722
723
724
725
726



727
728


729


730
731
732
733
734
735
736
737
738
739
740


741
742
743
744
745
746
747

748
749
750
751
752
753
754
...
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819

820
821
822
823
824
825
826
...
842
843
844
845
846
847
848
849

850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
...
872
873
874
875
876
877
878
879
880
881
882
883




884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
....
1013
1014
1015
1016
1017
1018
1019
1020




1021

1022
1023
1024
1025
1026

1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105


1106
1107
1108
1109
1110
1111
1112
....
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
....
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172


1173
1174
1175
1176
1177
1178
1179


1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191


1192




1193
1194
1195
1196
1197
1198







1199
1200
1201



1202
1203
1204
1205
1206





1207
1208

1209
1210
1211





1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224


1225
1226



1227









1228
1229
1230

1231

1232
1233
1234
1235
1236
1237

1238
1239
1240
1241








1242
1243

1244






1245
1246




1247
1248







1249
1250

1251
1252

1253



1254
1255
1256
1257
1258
1259
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
















1293
1294
1295


















1296
1297
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
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371



1372
1373
1374
1375
1376
1377
1378
....
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418

1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490


1491
1492
1493
1494
1495
1496

1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524

1525
1526


1527
1528
1529

1530
1531
1532
1533
1534
1535
1536
1537
1538
1539

1540
1541
1542
1543
1544
1545
1546
1547
1548
....
1550
1551
1552
1553
1554
1555
1556
1557







1558
1559
1560
1561
1562
1563
1564
1565
1566


1567
1568
1569



1570
1571
1572
1573
1574

1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586

1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
....
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
....
1665
1666
1667
1668
1669
1670
1671
1672
1673
1674
1675
1676




1677
1678


1679
1680
1681
1682
1683
1684
1685
1686
1687
1688
1689
1690
1691
*/
typedef struct MergeEngine MergeEngine;     /* Merge PMAs together */
typedef struct PmaReader PmaReader;         /* Incrementally read one PMA */
typedef struct PmaWriter PmaWriter;         /* Incrementally write on PMA */
typedef struct SorterRecord SorterRecord;   /* A record being sorted */
typedef struct SortSubtask SortSubtask;     /* A sub-task in the sort process */






/*
** Candidate values for SortSubtask.eWork

*/
#define SORT_SUBTASK_SORT   1     /* Sort records on pList */
#define SORT_SUBTASK_TO_PMA 2     /* Xfer pList to Packed-Memory-Array pTemp1 */
#define SORT_SUBTASK_CONS   3     /* Consolidate multiple PMAs */





























/*
** Sorting is divided up into smaller subtasks.  Each subtask is controlled
** by an instance of this object. A Subtask might run in either the main thread
** or in a background thread.
**
** Exactly VdbeSorter.nTask instances of this object are allocated
................................................................................
**     the temp file if it is not already open.
**
**   SORT_SUBTASK_CONS:
**     Merge existing PMAs until SortSubtask.nConsolidate or fewer
**     remain in temp file SortSubtask.pTemp1.
*/
struct SortSubtask {
  SQLiteThread *pThread;          /* Thread handle, or NULL */
  int bDone;                      /* Set to true by pTask when finished */

  sqlite3 *db;                    /* Database connection */

  KeyInfo *pKeyInfo;              /* How to compare records */
  UnpackedRecord *pUnpacked;      /* Space to unpack a record */
  int pgsz;                       /* Main database page size */

  u8 eWork;                       /* One of the SORT_SUBTASK_* constants */
  int nConsolidate;               /* For SORT_SUBTASK_CONS, max final PMAs */
  SorterRecord *pList;            /* List of records for pTask to sort */
  int nInMemory;                  /* Expected size of PMA based on pList */
  u8 *aListMemory;                /* Records memory (or NULL) */

  int nPMA;                       /* Number of PMAs currently in pTemp1 */
  i64 iTemp1Off;                  /* Offset to write to in pTemp1 */
  sqlite3_file *pTemp1;           /* File to write PMAs to, or NULL */

};


/*
** The MergeEngine object is used to combine two or more smaller PMAs into
** one big PMA using a merge operation.  Separate PMAs all need to be
** combined into one big PMA in order to be able to step through the sorted
................................................................................
};

/*
** Main sorter structure. A single instance of this is allocated for each 
** sorter cursor created by the VDBE.
*/
struct VdbeSorter {
  int nInMemory;                  /* Current size of pRecord list as PMA */
  int mnPmaSize;                  /* Minimum PMA size, in bytes */
  int mxPmaSize;                  /* Maximum PMA size, in bytes.  0==no limit */
  int bUsePMA;                    /* True if one or more PMAs created */
  SorterRecord *pRecord;          /* Head of in-memory record list */
  MergeEngine *pMerger;           /* For final merge of PMAs (by caller) */ 
  u8 *aMemory;                    /* Block of memory to alloc records from */
  int iMemory;                    /* Offset of first free byte in aMemory */
  int nMemory;                    /* Size of aMemory allocation in bytes */
  int iPrev;                      /* Previous thread used to flush PMA */
  int nTask;                      /* Size of aTask[] array */
  SortSubtask aTask[1];           /* One or more subtasks */
};

/*
** An instance of the following object is used to read records out of a
** PMA, in sorted order.  The next key to be read is cached in nKey/aKey.
................................................................................
** Initialize iterator pIter to scan through the PMA stored in file pFile
** starting at offset iStart and ending at offset iEof-1. This function 
** leaves the iterator pointing to the first key in the PMA (or EOF if the 
** PMA is empty).
*/
static int vdbePmaReaderInit(
  SortSubtask *pTask,             /* Thread context */

  i64 iStart,                     /* Start offset in pTask->pTemp1 */
  PmaReader *pIter,               /* Iterator to populate */
  i64 *pnByte                     /* IN/OUT: Increment this value by PMA size */
){
  int rc = SQLITE_OK;
  int nBuf = pTask->pgsz;
  void *pMap = 0;                 /* Mapping of temp file */

  assert( pTask->iTemp1Off>iStart );
  assert( pIter->aAlloc==0 );
  assert( pIter->aBuffer==0 );
  pIter->pFile = pTask->pTemp1;
  pIter->iReadOff = iStart;
  pIter->nAlloc = 128;
  pIter->aAlloc = (u8*)sqlite3Malloc(pIter->nAlloc);
  if( pIter->aAlloc ){
    /* Try to xFetch() a mapping of the entire temp file. If this is possible,
    ** the PMA will be read via the mapping. Otherwise, use xRead().  */
    if( pTask->iTemp1Off<=(i64)(pTask->db->nMaxSorterMmap) ){
      rc = sqlite3OsFetch(pIter->pFile, 0, pTask->iTemp1Off, &pMap);
    }
  }else{
    rc = SQLITE_NOMEM;
  }

  if( rc==SQLITE_OK ){
    if( pMap ){
................................................................................
      pIter->aBuffer = (u8*)sqlite3Malloc(nBuf);
      if( !pIter->aBuffer ){
        rc = SQLITE_NOMEM;
      }else{
        int iBuf = iStart % nBuf;
        if( iBuf ){
          int nRead = nBuf - iBuf;
          if( (iStart + nRead) > pTask->iTemp1Off ){
            nRead = (int)(pTask->iTemp1Off - iStart);
          }
          rc = sqlite3OsRead(
              pTask->pTemp1, &pIter->aBuffer[iBuf], nRead, iStart
              );
          assert( rc!=SQLITE_IOERR_SHORT_READ );
        }
      }
    }
  }

  if( rc==SQLITE_OK ){
    u64 nByte;                    /* Size of PMA in bytes */
    pIter->iEof = pTask->iTemp1Off;
    rc = vdbePmaReadVarint(pIter, &nByte);
    pIter->iEof = pIter->iReadOff + nByte;
    *pnByte += nByte;
  }

  if( rc==SQLITE_OK ){
    rc = vdbePmaReaderNext(pIter);
................................................................................
  }
  return rc;
}


/*
** Compare key1 (buffer pKey1, size nKey1 bytes) with key2 (buffer pKey2, 
** size nKey2 bytes). Use (pTask->pKeyInfo) for the collation sequences
** used by the comparison. Return the result of the comparison.
**
** Before returning, object (pTask->pUnpacked) is populated with the
** unpacked version of key2. Or, if pKey2 is passed a NULL pointer, then it 
** is assumed that the (pTask->pUnpacked) structure already contains the 
** unpacked key to use as key2.
**
** If an OOM error is encountered, (pTask->pUnpacked->error_rc) is set
** to SQLITE_NOMEM.
*/
static int vdbeSorterCompare(
  SortSubtask *pTask,             /* Subtask context (for pKeyInfo) */

  const void *pKey1, int nKey1,   /* Left side of comparison */
  const void *pKey2, int nKey2    /* Right side of comparison */
){
  UnpackedRecord *r2 = pTask->pUnpacked;
  if( pKey2 ){
    sqlite3VdbeRecordUnpack(pTask->pKeyInfo, nKey2, pKey2, r2);
  }
  return sqlite3VdbeRecordCompare(nKey1, pKey1, r2, 0);
}

/*
** This function is called to compare two iterator keys when merging 
** multiple b-tree segments. Parameter iOut is the index of the aTree[] 
** value to recalculate.
*/
static int vdbeSorterDoCompare(
  SortSubtask *pTask, 
  MergeEngine *pMerger, 
  int iOut
){
  int i1;
  int i2;
  int iRes;
  PmaReader *p1;
................................................................................

  if( p1->pFile==0 ){
    iRes = i2;
  }else if( p2->pFile==0 ){
    iRes = i1;
  }else{
    int res;
    assert( pTask->pUnpacked!=0 );  /* allocated in vdbeSortSubtaskMain() */
    res = vdbeSorterCompare(
        pTask, p1->aKey, p1->nKey, p2->aKey, p2->nKey
    );
    if( res<=0 ){
      iRes = i1;
    }else{
      iRes = i2;
    }
  }
................................................................................

    pSorter->nTask = nWorker + 1;
    for(i=0; i<pSorter->nTask; i++){
      SortSubtask *pTask = &pSorter->aTask[i];
      pTask->pKeyInfo = pKeyInfo;
      pTask->pgsz = pgsz;
      pTask->db = db;

    }

    if( !sqlite3TempInMemory(db) ){
      pSorter->mnPmaSize = SORTER_MIN_WORKING * pgsz;
      mxCache = db->aDb[0].pSchema->cache_size;
      if( mxCache<SORTER_MIN_WORKING ) mxCache = SORTER_MIN_WORKING;
      pSorter->mxPmaSize = mxCache * pgsz;
................................................................................

      /* If the application is using memsys3 or memsys5, use a separate 
      ** allocation for each sort-key in memory. Otherwise, use a single big
      ** allocation at pSorter->aMemory for all sort-keys.  */
      if( sqlite3GlobalConfig.pHeap==0 ){
        assert( pSorter->iMemory==0 );
        pSorter->nMemory = pgsz;
        pSorter->aMemory = (u8*)sqlite3Malloc(pgsz);
        if( !pSorter->aMemory ) rc = SQLITE_NOMEM;
      }
    }
  }

  return rc;
}

................................................................................
  for(p=pRecord; p; p=pNext){
    pNext = p->u.pNext;
    sqlite3DbFree(db, p);
  }
}

/*
** Free all resources owned by the object indicated by argument pTask. All 

** fields of *pTask are zeroed before returning.
*/
static void vdbeSortSubtaskCleanup(sqlite3 *db, SortSubtask *pTask){
  sqlite3DbFree(db, pTask->pUnpacked);
  pTask->pUnpacked = 0;





  if( pTask->aListMemory==0 ){
    vdbeSorterRecordFree(0, pTask->pList);
  }else{
    sqlite3_free(pTask->aListMemory);
    pTask->aListMemory = 0;
  }
  pTask->pList = 0;
  if( pTask->pTemp1 ){



    sqlite3OsCloseFree(pTask->pTemp1);
    pTask->pTemp1 = 0;


  }


}

/*
** Join all threads.  
*/
#if SQLITE_MAX_WORKER_THREADS>0
static int vdbeSorterJoinAll(VdbeSorter *pSorter, int rcin){
  int rc = rcin;
  int i;
  for(i=0; i<pSorter->nTask; i++){
    SortSubtask *pTask = &pSorter->aTask[i];


    if( pTask->pThread ){
      void *pRet;
      int rc2 = sqlite3ThreadJoin(pTask->pThread, &pRet);
      pTask->pThread = 0;
      pTask->bDone = 0;
      if( rc==SQLITE_OK ) rc = rc2;
      if( rc==SQLITE_OK ) rc = SQLITE_PTR_TO_INT(pRet);

    }
  }
  return rc;
}
#else
# define vdbeSorterJoinAll(x,rcin) (rcin)
#endif
................................................................................
  (void)vdbeSorterJoinAll(pSorter, SQLITE_OK);
  vdbeMergeEngineFree(pSorter->pMerger);
  pSorter->pMerger = 0;
  for(i=0; i<pSorter->nTask; i++){
    SortSubtask *pTask = &pSorter->aTask[i];
    vdbeSortSubtaskCleanup(db, pTask);
  }
  if( pSorter->aMemory==0 ){
    vdbeSorterRecordFree(0, pSorter->pRecord);
  }
  pSorter->pRecord = 0;
  pSorter->nInMemory = 0;
  pSorter->bUsePMA = 0;
  pSorter->iMemory = 0;
}

/*
** Free any cursor components allocated by sqlite3VdbeSorterXXX routines.
*/
void sqlite3VdbeSorterClose(sqlite3 *db, VdbeCursor *pCsr){
  VdbeSorter *pSorter = pCsr->pSorter;
  if( pSorter ){
    sqlite3VdbeSorterReset(db, pSorter);
    vdbeMergeEngineFree(pSorter->pMerger);
    sqlite3_free(pSorter->aMemory);

    sqlite3DbFree(db, pSorter);
    pCsr->pSorter = 0;
  }
}

/*
** Allocate space for a file-handle and open a temporary file. If successful,
................................................................................
}

/*
** Merge the two sorted lists p1 and p2 into a single list.
** Set *ppOut to the head of the new list.
*/
static void vdbeSorterMerge(
  SortSubtask *pTask,             /* Calling thread context */

  SorterRecord *p1,               /* First list to merge */
  SorterRecord *p2,               /* Second list to merge */
  SorterRecord **ppOut            /* OUT: Head of merged list */
){
  SorterRecord *pFinal = 0;
  SorterRecord **pp = &pFinal;
  void *pVal2 = p2 ? SRVAL(p2) : 0;

  while( p1 && p2 ){
    int res;
    res = vdbeSorterCompare(pTask, SRVAL(p1), p1->nVal, pVal2, p2->nVal);
    if( res<=0 ){
      *pp = p1;
      pp = &p1->u.pNext;
      p1 = p1->u.pNext;
      pVal2 = 0;
    }else{
      *pp = p2;
................................................................................
    }
  }
  *pp = p1 ? p1 : p2;
  *ppOut = pFinal;
}

/*
** Sort the linked list of records headed at pTask->pList. Return 
** SQLITE_OK if successful, or an SQLite error code (i.e. SQLITE_NOMEM) if 
** an error occurs.
*/
static int vdbeSorterSort(SortSubtask *pTask){




  int i;
  SorterRecord **aSlot;
  SorterRecord *p;

  aSlot = (SorterRecord **)sqlite3MallocZero(64 * sizeof(SorterRecord *));
  if( !aSlot ){
    return SQLITE_NOMEM;
  }

  p = pTask->pList;
  while( p ){
    SorterRecord *pNext;
    if( pTask->aListMemory ){
      if( (u8*)p==pTask->aListMemory ){
        pNext = 0;
      }else{
        assert( p->u.iNext<sqlite3MallocSize(pTask->aListMemory) );
        pNext = (SorterRecord*)&pTask->aListMemory[p->u.iNext];
      }
    }else{
      pNext = p->u.pNext;
    }

    p->u.pNext = 0;
    for(i=0; aSlot[i]; i++){
      vdbeSorterMerge(pTask, p, aSlot[i], &p);
      aSlot[i] = 0;
    }
    aSlot[i] = p;
    p = pNext;
  }

  p = 0;
  for(i=0; i<64; i++){
    vdbeSorterMerge(pTask, p, aSlot[i], &p);
  }
  pTask->pList = p;

  sqlite3_free(aSlot);
  return SQLITE_OK;
}

/*
** Initialize a PMA-writer object.
................................................................................
** is guaranteed to be nByte bytes or smaller in size. This function
** attempts to extend the file to nByte bytes in size and to ensure that
** the VFS has memory mapped it.
**
** Whether or not the file does end up memory mapped of course depends on
** the specific VFS implementation.
*/
static void vdbeSorterExtendFile(sqlite3 *db, sqlite3_file *pFile, i64 nByte){




  if( nByte<=(i64)(db->nMaxSorterMmap) ){

    int rc = sqlite3OsTruncate(pFile, nByte);
    if( rc==SQLITE_OK ){
      void *p = 0;
      sqlite3OsFetch(pFile, 0, nByte, &p);
      sqlite3OsUnfetch(pFile, 0, p);

    }
  }
}
#else
# define vdbeSorterExtendFile(x,y,z) SQLITE_OK
#endif


/*
** Write the current contents of the in-memory linked-list to a PMA. Return
** SQLITE_OK if successful, or an SQLite error code otherwise.
**
** The format of a PMA is:
**
**     * A varint. This varint contains the total number of bytes of content
**       in the PMA (not including the varint itself).
**
**     * One or more records packed end-to-end in order of ascending keys. 
**       Each record consists of a varint followed by a blob of data (the 
**       key). The varint is the number of bytes in the blob of data.
*/
static int vdbeSorterListToPMA(SortSubtask *pTask){
  int rc = SQLITE_OK;             /* Return code */
  PmaWriter writer;               /* Object used to write to the file */

  memset(&writer, 0, sizeof(PmaWriter));
  assert( pTask->nInMemory>0 );

  /* If the first temporary PMA file has not been opened, open it now. */
  if( pTask->pTemp1==0 ){
    rc = vdbeSorterOpenTempFile(pTask->db->pVfs, &pTask->pTemp1);
    assert( rc!=SQLITE_OK || pTask->pTemp1 );
    assert( pTask->iTemp1Off==0 );
    assert( pTask->nPMA==0 );
  }

  /* Try to get the file to memory map */
  if( rc==SQLITE_OK ){
    vdbeSorterExtendFile(pTask->db, 
        pTask->pTemp1, pTask->iTemp1Off + pTask->nInMemory + 9
    );
  }

  if( rc==SQLITE_OK ){
    SorterRecord *p;
    SorterRecord *pNext = 0;

    vdbePmaWriterInit(pTask->pTemp1, &writer, pTask->pgsz,
                      pTask->iTemp1Off);
    pTask->nPMA++;
    vdbePmaWriteVarint(&writer, pTask->nInMemory);
    for(p=pTask->pList; p; p=pNext){
      pNext = p->u.pNext;
      vdbePmaWriteVarint(&writer, p->nVal);
      vdbePmaWriteBlob(&writer, SRVAL(p), p->nVal);
      if( pTask->aListMemory==0 ) sqlite3_free(p);
    }
    pTask->pList = p;
    rc = vdbePmaWriterFinish(&writer, &pTask->iTemp1Off);
  }

  assert( pTask->pList==0 || rc!=SQLITE_OK );
  return rc;
}

/*
** Advance the MergeEngine iterator passed as the second argument to
** the next entry. Set *pbEof to true if this means the iterator has 
** reached EOF.
**
** Return SQLITE_OK if successful or an error code if an error occurs.
*/
static int vdbeSorterNext(
  SortSubtask *pTask, 
  MergeEngine *pMerger, 
  int *pbEof
){
  int rc;
  int iPrev = pMerger->aTree[1];/* Index of iterator to advance */



  /* Advance the current iterator */
  rc = vdbePmaReaderNext(&pMerger->aIter[iPrev]);

  /* Update contents of aTree[] */
  if( rc==SQLITE_OK ){
    int i;                      /* Index of aTree[] to recalculate */
................................................................................
      /* Compare pIter1 and pIter2. Store the result in variable iRes. */
      int iRes;
      if( pIter1->pFile==0 ){
        iRes = +1;
      }else if( pIter2->pFile==0 ){
        iRes = -1;
      }else{
        iRes = vdbeSorterCompare(pTask, 
            pIter1->aKey, pIter1->nKey, pKey2, pIter2->nKey
        );
      }

      /* If pIter1 contained the smaller value, set aTree[i] to its index.
      ** Then set pIter2 to the next iterator to compare to pIter1. In this
      ** case there is no cache of pIter2 in pTask->pUnpacked, so set
      ** pKey2 to point to the record belonging to pIter2.
      **
      ** Alternatively, if pIter2 contains the smaller of the two values,
      ** set aTree[i] to its index and update pIter1. If vdbeSorterCompare()
      ** was actually called above, then pTask->pUnpacked now contains
      ** a value equivalent to pIter2. So set pKey2 to NULL to prevent
      ** vdbeSorterCompare() from decoding pIter2 again.
      **
      ** If the two values were equal, then the value from the oldest
      ** PMA should be considered smaller. The VdbeSorter.aIter[] array
      ** is sorted from oldest to newest, so pIter1 contains older values
      ** than pIter2 iff (pIter1<pIter2).  */
................................................................................
    }
    *pbEof = (pMerger->aIter[pMerger->aTree[1]].pFile==0);
  }

  return rc;
}

/*
** The main routine for sorter-thread operations.
*/
static void *vdbeSortSubtaskMain(void *pCtx){
  int rc = SQLITE_OK;
  SortSubtask *pTask = (SortSubtask*)pCtx;



  assert( pTask->eWork==SORT_SUBTASK_SORT
       || pTask->eWork==SORT_SUBTASK_TO_PMA
       || pTask->eWork==SORT_SUBTASK_CONS
  );
  assert( pTask->bDone==0 );



  if( pTask->pUnpacked==0 ){
    char *pFree;
    pTask->pUnpacked = sqlite3VdbeAllocUnpackedRecord(
        pTask->pKeyInfo, 0, 0, &pFree
    );
    assert( pTask->pUnpacked==(UnpackedRecord*)pFree );
    if( pFree==0 ){
      rc = SQLITE_NOMEM;
      goto thread_out;
    }
    pTask->pUnpacked->nField = pTask->pKeyInfo->nField;
    pTask->pUnpacked->errCode = 0;


  }





  if( pTask->eWork==SORT_SUBTASK_CONS ){
    assert( pTask->pList==0 );
    while( pTask->nPMA>pTask->nConsolidate && rc==SQLITE_OK ){
      int nIter = MIN(pTask->nPMA, SORTER_MAX_MERGE_COUNT);
      sqlite3_file *pTemp2 = 0;     /* Second temp file to use */







      MergeEngine *pMerger;         /* Object for reading/merging PMA data */
      i64 iReadOff = 0;             /* Offset in pTemp1 to read from */
      i64 iWriteOff = 0;            /* Offset in pTemp2 to write to */



      int i;
      
      /* Allocate a merger object to merge PMAs together. */
      pMerger = vdbeMergeEngineNew(nIter);
      if( pMerger==0 ){





        rc = SQLITE_NOMEM;
        break;

      }

      /* Open a second temp file to write merged data to */





      rc = vdbeSorterOpenTempFile(pTask->db->pVfs, &pTemp2);
      if( rc==SQLITE_OK ){
        vdbeSorterExtendFile(pTask->db, pTemp2, pTask->iTemp1Off);
      }else{
        vdbeMergeEngineFree(pMerger);
        break;
      }

      /* This loop runs once for each output PMA. Each output PMA is made
      ** of data merged from up to SORTER_MAX_MERGE_COUNT input PMAs. */
      for(i=0; rc==SQLITE_OK && i<pTask->nPMA; i+=SORTER_MAX_MERGE_COUNT){
        PmaWriter writer;         /* Object for writing data to pTemp2 */
        i64 nOut = 0;             /* Bytes of data in output PMA */


        int bEof = 0;
        int rc2;













        /* Configure the merger object to read and merge data from the next 
        ** SORTER_MAX_MERGE_COUNT PMAs in pTemp1 (or from all remaining PMAs,
        ** if that is fewer). */

        int iIter;

        for(iIter=0; iIter<SORTER_MAX_MERGE_COUNT; iIter++){
          PmaReader *pIter = &pMerger->aIter[iIter];
          rc = vdbePmaReaderInit(pTask, iReadOff, pIter, &nOut);
          iReadOff = pIter->iEof;
          if( iReadOff>=pTask->iTemp1Off || rc!=SQLITE_OK ) break;
        }

        for(iIter=pMerger->nTree-1; rc==SQLITE_OK && iIter>0; iIter--){
          rc = vdbeSorterDoCompare(pTask, pMerger, iIter);
        }









        vdbePmaWriterInit(pTemp2, &writer, pTask->pgsz, iWriteOff);
        vdbePmaWriteVarint(&writer, nOut);

        while( rc==SQLITE_OK && bEof==0 ){






          PmaReader *pIter = &pMerger->aIter[ pMerger->aTree[1] ];
          assert( pIter->pFile!=0 );        /* pIter is not at EOF */




          vdbePmaWriteVarint(&writer, pIter->nKey);
          vdbePmaWriteBlob(&writer, pIter->aKey, pIter->nKey);







          rc = vdbeSorterNext(pTask, pMerger, &bEof);
        }

        rc2 = vdbePmaWriterFinish(&writer, &iWriteOff);
        if( rc==SQLITE_OK ) rc = rc2;

      }




      vdbeMergeEngineFree(pMerger);
      sqlite3OsCloseFree(pTask->pTemp1);
      pTask->pTemp1 = pTemp2;
      pTask->nPMA = (i / SORTER_MAX_MERGE_COUNT);
      pTask->iTemp1Off = iWriteOff;
    }
  }else{
    /* Sort the pTask->pList list */
    rc = vdbeSorterSort(pTask);

    /* If required, write the list out to a PMA. */
    if( rc==SQLITE_OK && pTask->eWork==SORT_SUBTASK_TO_PMA ){








#ifdef SQLITE_DEBUG
      i64 nExpect = pTask->nInMemory
        + sqlite3VarintLen(pTask->nInMemory)
        + pTask->iTemp1Off;

















#endif
      rc = vdbeSorterListToPMA(pTask);
      assert( rc!=SQLITE_OK || (nExpect==pTask->iTemp1Off) );
    }


  }

 thread_out:
  pTask->bDone = 1;

  if( rc==SQLITE_OK && pTask->pUnpacked->errCode ){
    assert( pTask->pUnpacked->errCode==SQLITE_NOMEM );
    rc = SQLITE_NOMEM;
  }
  return SQLITE_INT_TO_PTR(rc);
}

/*
** Run the activity scheduled by the object passed as the only argument
** in the current thread.
*/
static int vdbeSorterRunTask(SortSubtask *pTask){
  int rc = SQLITE_PTR_TO_INT( vdbeSortSubtaskMain((void*)pTask) );

  assert( pTask->bDone );
















  pTask->bDone = 0;
  return rc;
}



















/*
** Flush the current contents of VdbeSorter.pRecord to a new PMA, possibly
** using a background thread.
**
** If argument bFg is non-zero, the operation always uses the calling thread.
*/
static int vdbeSorterFlushPMA(sqlite3 *db, const VdbeCursor *pCsr, int bFg){
  VdbeSorter *pSorter = pCsr->pSorter;
  int rc = SQLITE_OK;
  int i;
  SortSubtask *pTask = 0;    /* Thread context used to create new PMA */
  int nWorker = (pSorter->nTask-1);

  pSorter->bUsePMA = 1;
  for(i=0; i<nWorker; i++){
    int iTest = (pSorter->iPrev + i + 1) % nWorker;
    pTask = &pSorter->aTask[iTest];
#if SQLITE_MAX_WORKER_THREADS>0
    if( pTask->bDone ){
      void *pRet;
      assert( pTask->pThread );
      rc = sqlite3ThreadJoin(pTask->pThread, &pRet);
      pTask->pThread = 0;
      pTask->bDone = 0;
      if( rc==SQLITE_OK ){
        rc = SQLITE_PTR_TO_INT(pRet);
      }
    }
#endif
    if( pTask->pThread==0 ) break;
    pTask = 0;
  }
  if( pTask==0 ){
    pTask = &pSorter->aTask[nWorker];
  }
  pSorter->iPrev = (pTask - pSorter->aTask);

  if( rc==SQLITE_OK ){
    assert( pTask->pThread==0 && pTask->bDone==0 );
    pTask->eWork = SORT_SUBTASK_TO_PMA;
    pTask->pList = pSorter->pRecord;
    pTask->nInMemory = pSorter->nInMemory;
    pSorter->nInMemory = 0;
    pSorter->pRecord = 0;

    if( pSorter->aMemory ){
      u8 *aMem = pTask->aListMemory;
      pTask->aListMemory = pSorter->aMemory;
      pSorter->aMemory = aMem;
    }

#if SQLITE_MAX_WORKER_THREADS>0
    if( !bFg && pTask!=&pSorter->aTask[nWorker] ){
      /* Launch a background thread for this operation */
      void *pCtx = (void*)pTask;
      assert( pSorter->aMemory==0 || pTask->aListMemory!=0 );
      if( pTask->aListMemory ){
        if( pSorter->aMemory==0 ){
          pSorter->aMemory = sqlite3Malloc(pSorter->nMemory);
          if( pSorter->aMemory==0 ) return SQLITE_NOMEM;
        }else{
          pSorter->nMemory = sqlite3MallocSize(pSorter->aMemory);
        }
      }
      rc = sqlite3ThreadCreate(&pTask->pThread, vdbeSortSubtaskMain, pCtx);
    }else
#endif
    {
      /* Use the foreground thread for this operation */
      rc = vdbeSorterRunTask(pTask);
      if( rc==SQLITE_OK ){
        u8 *aMem = pTask->aListMemory;
        pTask->aListMemory = pSorter->aMemory;
        pSorter->aMemory = aMem;
        assert( pTask->pList==0 );



      }
    }
  }

  return rc;
}

................................................................................
  **
  **   * The total memory allocated for the in-memory list is greater 
  **     than (page-size * 10) and sqlite3HeapNearlyFull() returns true.
  */
  nReq = pVal->n + sizeof(SorterRecord);
  nPMA = pVal->n + sqlite3VarintLen(pVal->n);
  if( pSorter->mxPmaSize ){
    if( pSorter->aMemory ){
      bFlush = pSorter->iMemory && (pSorter->iMemory+nReq) > pSorter->mxPmaSize;
    }else{

      bFlush = (
          (pSorter->nInMemory > pSorter->mxPmaSize)
       || (pSorter->nInMemory > pSorter->mnPmaSize && sqlite3HeapNearlyFull())
      );
    }
    if( bFlush ){
      rc = vdbeSorterFlushPMA(db, pCsr, 0);
      pSorter->nInMemory = 0;
      pSorter->iMemory = 0;
      assert( rc!=SQLITE_OK || pSorter->pRecord==0 );
    }
  }

  pSorter->nInMemory += nPMA;

  if( pSorter->aMemory ){
    int nMin = pSorter->iMemory + nReq;

    if( nMin>pSorter->nMemory ){
      u8 *aNew;
      int nNew = pSorter->nMemory * 2;
      while( nNew < nMin ) nNew = nNew*2;
      if( nNew > pSorter->mxPmaSize ) nNew = pSorter->mxPmaSize;
      if( nNew < nMin ) nNew = nMin;

      aNew = sqlite3Realloc(pSorter->aMemory, nNew);
      if( !aNew ) return SQLITE_NOMEM;
      pSorter->pRecord = (SorterRecord*)(
          aNew + ((u8*)pSorter->pRecord - pSorter->aMemory)
      );
      pSorter->aMemory = aNew;
      pSorter->nMemory = nNew;
    }

    pNew = (SorterRecord*)&pSorter->aMemory[pSorter->iMemory];
    pSorter->iMemory += ROUND8(nReq);
    pNew->u.iNext = (u8*)(pSorter->pRecord) - pSorter->aMemory;
  }else{
    pNew = (SorterRecord *)sqlite3Malloc(nReq);
    if( pNew==0 ){
      return SQLITE_NOMEM;
    }
    pNew->u.pNext = pSorter->pRecord;
  }

  memcpy(SRVAL(pNew), pVal->z, pVal->n);
  pNew->nVal = pVal->n;
  pSorter->pRecord = pNew;

  return rc;
}

/*
** Return the total number of PMAs in all temporary files.
*/
static int vdbeSorterCountPMA(VdbeSorter *pSorter){
  int nPMA = 0;
  int i;
  for(i=0; i<pSorter->nTask; i++){
    nPMA += pSorter->aTask[i].nPMA;
  }
  return nPMA;
}

/*
** Once the sorter has been populated by calls to sqlite3VdbeSorterWrite,
** this function is called to prepare for iterating through the records
** in sorted order.
*/
int sqlite3VdbeSorterRewind(sqlite3 *db, const VdbeCursor *pCsr, int *pbEof){
  VdbeSorter *pSorter = pCsr->pSorter;
  int rc = SQLITE_OK;             /* Return code */



  assert( pSorter );

  /* If no data has been written to disk, then do not do so now. Instead,
  ** sort the VdbeSorter.pRecord list. The vdbe layer will read data directly
  ** from the in-memory list.  */

  if( pSorter->bUsePMA==0 ){
    if( pSorter->pRecord ){
      SortSubtask *pTask = &pSorter->aTask[0];
      *pbEof = 0;
      pTask->pList = pSorter->pRecord;
      pTask->eWork = SORT_SUBTASK_SORT;
      assert( pTask->aListMemory==0 );
      pTask->aListMemory = pSorter->aMemory;
      rc = vdbeSorterRunTask(pTask);
      pTask->aListMemory = 0;
      pSorter->pRecord = pTask->pList;
      pTask->pList = 0;
    }else{
      *pbEof = 1;
    }
    return rc;
  }

  /* Write the current in-memory list to a PMA. */
  if( pSorter->pRecord ){
    rc = vdbeSorterFlushPMA(db, pCsr, 1);
  }

  /* Join all threads */
  rc = vdbeSorterJoinAll(pSorter, rc);

  /* If there are more than SORTER_MAX_MERGE_COUNT PMAs on disk, merge
  ** some of them together so that this is no longer the case. */

  if( vdbeSorterCountPMA(pSorter)>SORTER_MAX_MERGE_COUNT ){
    int i;


    for(i=0; rc==SQLITE_OK && i<pSorter->nTask; i++){
      SortSubtask *pTask = &pSorter->aTask[i];
      if( pTask->pTemp1 ){

        pTask->nConsolidate = SORTER_MAX_MERGE_COUNT / pSorter->nTask;
        pTask->eWork = SORT_SUBTASK_CONS;

#if SQLITE_MAX_WORKER_THREADS>0
        if( i<(pSorter->nTask-1) ){
          void *pCtx = (void*)pTask;
          rc = sqlite3ThreadCreate(&pTask->pThread, vdbeSortSubtaskMain, pCtx);
        }else
#endif
        {

          rc = vdbeSorterRunTask(pTask);
        }
      }
    }
  }

  /* Join all threads */
  rc = vdbeSorterJoinAll(pSorter, rc);

................................................................................
  ** and merge all remaining PMAs.  */
  assert( pSorter->pMerger==0 );
  if( rc==SQLITE_OK ){
    int nIter = 0;                /* Number of iterators used */
    int i;
    MergeEngine *pMerger;
    for(i=0; i<pSorter->nTask; i++){
      nIter += pSorter->aTask[i].nPMA;







    }

    pSorter->pMerger = pMerger = vdbeMergeEngineNew(nIter);
    if( pMerger==0 ){
      rc = SQLITE_NOMEM;
    }else{
      int iIter = 0;
      int iThread = 0;
      for(iThread=0; iThread<pSorter->nTask; iThread++){


        int iPMA;
        i64 iReadOff = 0;
        SortSubtask *pTask = &pSorter->aTask[iThread];



        for(iPMA=0; iPMA<pTask->nPMA && rc==SQLITE_OK; iPMA++){
          i64 nDummy = 0;
          PmaReader *pIter = &pMerger->aIter[iIter++];
          rc = vdbePmaReaderInit(pTask, iReadOff, pIter, &nDummy);
          iReadOff = pIter->iEof;

        }
      }

      for(i=pMerger->nTree-1; rc==SQLITE_OK && i>0; i--){
        rc = vdbeSorterDoCompare(&pSorter->aTask[0], pMerger, i);
      }
    }
  }

  if( rc==SQLITE_OK ){
    *pbEof = (pSorter->pMerger->aIter[pSorter->pMerger->aTree[1]].pFile==0);
  }

  return rc;
}

/*
** Advance to the next element in the sorter.
*/
int sqlite3VdbeSorterNext(sqlite3 *db, const VdbeCursor *pCsr, int *pbEof){
  VdbeSorter *pSorter = pCsr->pSorter;
  int rc;                         /* Return code */

  if( pSorter->pMerger ){
    rc = vdbeSorterNext(&pSorter->aTask[0], pSorter->pMerger, pbEof);
  }else{
    SorterRecord *pFree = pSorter->pRecord;
    pSorter->pRecord = pFree->u.pNext;
    pFree->u.pNext = 0;
    if( pSorter->aMemory==0 ) vdbeSorterRecordFree(db, pFree);
    *pbEof = !pSorter->pRecord;
    rc = SQLITE_OK;
  }
  return rc;
}

/*
** Return a pointer to a buffer owned by the sorter that contains the 
................................................................................
  void *pKey;
  if( pSorter->pMerger ){
    PmaReader *pIter;
    pIter = &pSorter->pMerger->aIter[ pSorter->pMerger->aTree[1] ];
    *pnKey = pIter->nKey;
    pKey = pIter->aKey;
  }else{
    *pnKey = pSorter->pRecord->nVal;
    pKey = SRVAL(pSorter->pRecord);
  }
  return pKey;
}

/*
** Copy the current sorter key into the memory cell pOut.
*/
................................................................................
int sqlite3VdbeSorterCompare(
  const VdbeCursor *pCsr,         /* Sorter cursor */
  Mem *pVal,                      /* Value to compare to current sorter key */
  int nIgnore,                    /* Ignore this many fields at the end */
  int *pRes                       /* OUT: Result of comparison */
){
  VdbeSorter *pSorter = pCsr->pSorter;
  UnpackedRecord *r2 = pSorter->aTask[0].pUnpacked;
  KeyInfo *pKeyInfo = pCsr->pKeyInfo;
  int i;
  void *pKey; int nKey;           /* Sorter key to compare pVal with */





  assert( r2->nField>=pKeyInfo->nField-nIgnore );
  r2->nField = pKeyInfo->nField-nIgnore;



  pKey = vdbeSorterRowkey(pSorter, &nKey);
  sqlite3VdbeRecordUnpack(pKeyInfo, nKey, pKey, r2);
  for(i=0; i<r2->nField; i++){
    if( r2->aMem[i].flags & MEM_Null ){
      *pRes = -1;
      return SQLITE_OK;
    }
  }

  *pRes = sqlite3VdbeRecordCompare(pVal->n, pVal->z, r2, 0);
  return SQLITE_OK;
}







>
>
>

>

<
>

<
<
<
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>







 







<
|
<

>

<

<
<
|
<
<
<
<
<
<
<
>







 







<



|
|
|
|
|
|







 







>








|


|






|
|







 







|
|


|
|








|







 







|


|

|


|



|
>



<

|










|







 







|
|
|







 







>







 







|
|







 







|
>
|


<
|
>
>
>
>
>
|
|
|
|
<
|
<
<
>
>
>
|
<
>
>

>
>











>
>
|
|
|
|
|
|
|
>







 







|
|

|
|












|
>







 







|
>










|







 







|



|
>
>
>
>









|


|
|


|
|







|








|

|







 







|
>
>
>
>
|
>
|


|
|
>







<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<








|
|




>
>







 







|






|




|







 







|
|
|
|
|
|
>
>
|
<
<
|
<
<
|
>
>
|
|
|
|
|
|
|
|
|
|
|
|
>
>
|
>
>
>
>

<
<
<
<
<
>
>
>
>
>
>
>
|
<
<
>
>
>
|
|
<
<
<
>
>
>
>
>
|
<
>
|
|
<
>
>
>
>
>
|
|
<
<
<
<
|

<
<
<
<
<
>
>
|
<
>
>
>
|
>
>
>
>
>
>
>
>
>
|
<
<
>

>
|

|

<

>

|

|
>
>
>
>
>
>
>
>
|
|
>
|
>
>
>
>
>
>
|
|
>
>
>
>
|
|
>
>
>
>
>
>
>
|
|
>
|
<
>
|
>
>
>
|
|
<
<
<
<
|
<
<
<

<
|
>
>
>
>
>
>
>
>
|
<
<
<
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>

<
<
|
>
>


|
<
>
|
|


|






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


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











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







 







|


>

|
|




|

|



|

|









|

|
|

|



|

|





|




|




<
<
<
<
<
<
<
<
<
<
<









>
>




|
|
>

|

<
|
|
|
|
<
<
<
<







|
|





|
<
>
|
|
>
>
|
|
|
>
|
<
<

|
|
|
|

|
>
|
<







 







|
>
>
>
>
>
>
>







<
|
>
>
|
|
<
>
>
>
|
|
|
|
|
>




|







>











|

|
|

|
|







 







|
|







 







|




>
>
>
>
|
|
>
>













93
94
95
96
97
98
99
100
101
102
103
104
105

106
107



108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
...
166
167
168
169
170
171
172

173

174
175
176

177


178







179
180
181
182
183
184
185
186
...
250
251
252
253
254
255
256

257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
...
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
...
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
...
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600

601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
...
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
...
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
...
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
...
726
727
728
729
730
731
732
733
734
735
736
737

738
739
740
741
742
743
744
745
746
747

748


749
750
751
752

753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
...
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
...
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
...
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
....
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076


























































1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
....
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
....
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162


1163


1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186





1187
1188
1189
1190
1191
1192
1193
1194


1195
1196
1197
1198
1199



1200
1201
1202
1203
1204
1205

1206
1207
1208

1209
1210
1211
1212
1213
1214
1215




1216
1217





1218
1219
1220

1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234


1235
1236
1237
1238
1239
1240
1241

1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
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



1293

1294
1295
1296
1297
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
1325
1326
1327

1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
....
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559











1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580

1581
1582
1583
1584




1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599

1600
1601
1602
1603
1604
1605
1606
1607
1608
1609


1610
1611
1612
1613
1614
1615
1616
1617
1618

1619
1620
1621
1622
1623
1624
1625
....
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637
1638
1639
1640
1641
1642
1643
1644
1645
1646
1647
1648

1649
1650
1651
1652
1653

1654
1655
1656
1657
1658
1659
1660
1661
1662
1663
1664
1665
1666
1667
1668
1669
1670
1671
1672
1673
1674
1675
1676
1677
1678
1679
1680
1681
1682
1683
1684
1685
1686
1687
1688
1689
1690
1691
1692
1693
1694
1695
1696
1697
1698
1699
1700
....
1707
1708
1709
1710
1711
1712
1713
1714
1715
1716
1717
1718
1719
1720
1721
1722
....
1754
1755
1756
1757
1758
1759
1760
1761
1762
1763
1764
1765
1766
1767
1768
1769
1770
1771
1772
1773
1774
1775
1776
1777
1778
1779
1780
1781
1782
1783
1784
1785
1786
*/
typedef struct MergeEngine MergeEngine;     /* Merge PMAs together */
typedef struct PmaReader PmaReader;         /* Incrementally read one PMA */
typedef struct PmaWriter PmaWriter;         /* Incrementally write on PMA */
typedef struct SorterRecord SorterRecord;   /* A record being sorted */
typedef struct SortSubtask SortSubtask;     /* A sub-task in the sort process */

typedef struct SortList SortList;
typedef struct SortFile SortFile;
typedef struct SortLevel SortLevel;


/*

** A file containing zero or more PMAs.
*/



struct SortFile {
  sqlite3_file *pFd;              /* File handle */
  i64 iOff;                       /* Current write offset */
  i64 nByte;                      /* Actual size of file */
  int nPMA;                       /* Number of PMA currently in file */
};

/*
** A list of records.
*/
struct SortList {
  SorterRecord *pRecord;          /* List of records for pTask to sort */
  int nInMemory;                  /* Expected size of PMA based on pList */
  u8 *aMemory;                    /* Records memory (or NULL) */
};

struct SortLevel {
  SortSubtask *pTask;             /* Sorter task this level is a part of */
  SQLiteThread *pThread;          /* Thread handle, or NULL */
  int bDone;                      /* Set to true by pThread when finished */
  union {
    SortFile f;                   /* Input for level 1 and greater */
    SortList l;                   /* Input for level 0 */
  } in;
  SortFile out;                   /* Level storage */
  SortLevel *pNext;               /* Next level (containing larger PMAs) */
  UnpackedRecord *pUnpacked;      /* Space to unpack a record */
};

/*
** Sorting is divided up into smaller subtasks.  Each subtask is controlled
** by an instance of this object. A Subtask might run in either the main thread
** or in a background thread.
**
** Exactly VdbeSorter.nTask instances of this object are allocated
................................................................................
**     the temp file if it is not already open.
**
**   SORT_SUBTASK_CONS:
**     Merge existing PMAs until SortSubtask.nConsolidate or fewer
**     remain in temp file SortSubtask.pTemp1.
*/
struct SortSubtask {

  int iId;                        /* Sub-task id */

  sqlite3 *db;                    /* Database connection */
  VdbeSorter *pSorter;            /* Sorter that owns this object */
  KeyInfo *pKeyInfo;              /* How to compare records */

  int pgsz;                       /* Main database page size */


  int nConsolidate;               /* For consolidation, max final PMAs */







  SortLevel *pLevel;              /* PMA level 0 */
};


/*
** The MergeEngine object is used to combine two or more smaller PMAs into
** one big PMA using a merge operation.  Separate PMAs all need to be
** combined into one big PMA in order to be able to step through the sorted
................................................................................
};

/*
** Main sorter structure. A single instance of this is allocated for each 
** sorter cursor created by the VDBE.
*/
struct VdbeSorter {

  int mnPmaSize;                  /* Minimum PMA size, in bytes */
  int mxPmaSize;                  /* Maximum PMA size, in bytes.  0==no limit */
  int bUsePMA;                    /* True if one or more PMAs created */
  MergeEngine *pMerger;           /* For final merge of PMAs (by caller) */ 
  UnpackedRecord *pUnpacked;      /* Used by sqlite3VdbeSorterCompare */
  int iMemory;                    /* Offset of free byte in list.aMemory */
  int nMemory;                    /* Size of list.aMemory allocation in bytes */
  SortList list;                  /* In memory records */
  int iPrev;                      /* Previous PMA flushed via task iPrev */
  int nTask;                      /* Size of aTask[] array */
  SortSubtask aTask[1];           /* One or more subtasks */
};

/*
** An instance of the following object is used to read records out of a
** PMA, in sorted order.  The next key to be read is cached in nKey/aKey.
................................................................................
** Initialize iterator pIter to scan through the PMA stored in file pFile
** starting at offset iStart and ending at offset iEof-1. This function 
** leaves the iterator pointing to the first key in the PMA (or EOF if the 
** PMA is empty).
*/
static int vdbePmaReaderInit(
  SortSubtask *pTask,             /* Thread context */
  SortFile *pFile,                /* File to read from */
  i64 iStart,                     /* Start offset in pTask->pTemp1 */
  PmaReader *pIter,               /* Iterator to populate */
  i64 *pnByte                     /* IN/OUT: Increment this value by PMA size */
){
  int rc = SQLITE_OK;
  int nBuf = pTask->pgsz;
  void *pMap = 0;                 /* Mapping of temp file */

  assert( pFile->iOff>iStart );
  assert( pIter->aAlloc==0 );
  assert( pIter->aBuffer==0 );
  pIter->pFile = pFile->pFd;
  pIter->iReadOff = iStart;
  pIter->nAlloc = 128;
  pIter->aAlloc = (u8*)sqlite3Malloc(pIter->nAlloc);
  if( pIter->aAlloc ){
    /* Try to xFetch() a mapping of the entire temp file. If this is possible,
    ** the PMA will be read via the mapping. Otherwise, use xRead().  */
    if( pFile->iOff<=(i64)(pTask->db->nMaxSorterMmap) ){
      rc = sqlite3OsFetch(pIter->pFile, 0, pFile->iOff, &pMap);
    }
  }else{
    rc = SQLITE_NOMEM;
  }

  if( rc==SQLITE_OK ){
    if( pMap ){
................................................................................
      pIter->aBuffer = (u8*)sqlite3Malloc(nBuf);
      if( !pIter->aBuffer ){
        rc = SQLITE_NOMEM;
      }else{
        int iBuf = iStart % nBuf;
        if( iBuf ){
          int nRead = nBuf - iBuf;
          if( (iStart + nRead) > pFile->iOff ){
            nRead = (int)(pFile->iOff - iStart);
          }
          rc = sqlite3OsRead(
              pIter->pFile, &pIter->aBuffer[iBuf], nRead, iStart
          );
          assert( rc!=SQLITE_IOERR_SHORT_READ );
        }
      }
    }
  }

  if( rc==SQLITE_OK ){
    u64 nByte;                    /* Size of PMA in bytes */
    pIter->iEof = pFile->iOff;
    rc = vdbePmaReadVarint(pIter, &nByte);
    pIter->iEof = pIter->iReadOff + nByte;
    *pnByte += nByte;
  }

  if( rc==SQLITE_OK ){
    rc = vdbePmaReaderNext(pIter);
................................................................................
  }
  return rc;
}


/*
** Compare key1 (buffer pKey1, size nKey1 bytes) with key2 (buffer pKey2, 
** size nKey2 bytes). Use pKeyInfo for the collation sequences
** used by the comparison. Return the result of the comparison.
**
** Before returning, object pUnpacked is populated with the
** unpacked version of key2. Or, if pKey2 is passed a NULL pointer, then it 
** is assumed that the pUnpacked structure already contains the 
** unpacked key to use as key2.
**
** If an OOM error is encountered, (pUnpacked->error_rc) is set
** to SQLITE_NOMEM.
*/
static int vdbeSorterCompare(
  KeyInfo *pKeyInfo,
  UnpackedRecord *r2,
  const void *pKey1, int nKey1,   /* Left side of comparison */
  const void *pKey2, int nKey2    /* Right side of comparison */
){

  if( pKey2 ){
    sqlite3VdbeRecordUnpack(pKeyInfo, nKey2, pKey2, r2);
  }
  return sqlite3VdbeRecordCompare(nKey1, pKey1, r2, 0);
}

/*
** This function is called to compare two iterator keys when merging 
** multiple b-tree segments. Parameter iOut is the index of the aTree[] 
** value to recalculate.
*/
static int vdbeSorterDoCompare(
  SortLevel *pLvl, 
  MergeEngine *pMerger, 
  int iOut
){
  int i1;
  int i2;
  int iRes;
  PmaReader *p1;
................................................................................

  if( p1->pFile==0 ){
    iRes = i2;
  }else if( p2->pFile==0 ){
    iRes = i1;
  }else{
    int res;
    assert( pLvl->pUnpacked!=0 );  /* allocated in vdbeSorterThread() */
    res = vdbeSorterCompare(pLvl->pTask->pKeyInfo, pLvl->pUnpacked,
        p1->aKey, p1->nKey, p2->aKey, p2->nKey
    );
    if( res<=0 ){
      iRes = i1;
    }else{
      iRes = i2;
    }
  }
................................................................................

    pSorter->nTask = nWorker + 1;
    for(i=0; i<pSorter->nTask; i++){
      SortSubtask *pTask = &pSorter->aTask[i];
      pTask->pKeyInfo = pKeyInfo;
      pTask->pgsz = pgsz;
      pTask->db = db;
      pTask->pSorter = pSorter;
    }

    if( !sqlite3TempInMemory(db) ){
      pSorter->mnPmaSize = SORTER_MIN_WORKING * pgsz;
      mxCache = db->aDb[0].pSchema->cache_size;
      if( mxCache<SORTER_MIN_WORKING ) mxCache = SORTER_MIN_WORKING;
      pSorter->mxPmaSize = mxCache * pgsz;
................................................................................

      /* If the application is using memsys3 or memsys5, use a separate 
      ** allocation for each sort-key in memory. Otherwise, use a single big
      ** allocation at pSorter->aMemory for all sort-keys.  */
      if( sqlite3GlobalConfig.pHeap==0 ){
        assert( pSorter->iMemory==0 );
        pSorter->nMemory = pgsz;
        pSorter->list.aMemory = (u8*)sqlite3Malloc(pgsz);
        if( !pSorter->list.aMemory ) rc = SQLITE_NOMEM;
      }
    }
  }

  return rc;
}

................................................................................
  for(p=pRecord; p; p=pNext){
    pNext = p->u.pNext;
    sqlite3DbFree(db, p);
  }
}

/*
** Free all resources owned by the object indicated by argument pTask. 
** This does not include joining any outstanding threads. All fields of 
** *pTask are zeroed before returning.
*/
static void vdbeSortSubtaskCleanup(sqlite3 *db, SortSubtask *pTask){

  SortLevel *pLvl;
  SortLevel *pNext;
  for(pLvl=pTask->pLevel; pLvl; pLvl=pNext){
    pNext = pLvl->pNext;
    assert( pLvl->pThread==0 );
    if( pLvl==pTask->pLevel ){
      if( pLvl->in.l.aMemory==0 ){
        vdbeSorterRecordFree(0, pLvl->in.l.pRecord);
      }else{
        sqlite3_free(pLvl->in.l.aMemory);

      }


    }else{
      if( pLvl->in.f.pFd ) sqlite3OsCloseFree(pLvl->in.f.pFd);
    }
    if( pLvl->out.pFd ) sqlite3OsCloseFree(pLvl->out.pFd);

    sqlite3DbFree(db, pLvl->pUnpacked);
    sqlite3_free(pLvl);
  }
  pTask->pLevel = 0;
  pTask->nConsolidate = 0;
}

/*
** Join all threads.  
*/
#if SQLITE_MAX_WORKER_THREADS>0
static int vdbeSorterJoinAll(VdbeSorter *pSorter, int rcin){
  int rc = rcin;
  int i;
  for(i=0; i<pSorter->nTask; i++){
    SortSubtask *pTask = &pSorter->aTask[i];
    SortLevel *pLvl;
    for(pLvl=pTask->pLevel; pLvl; pLvl=pLvl->pNext){
      if( pLvl->pThread ){
        void *pRet;
        int rc2 = sqlite3ThreadJoin(pLvl->pThread, &pRet);
        pLvl->pThread = 0;
        pLvl->bDone = 0;
        if( rc==SQLITE_OK ) rc = rc2;
        if( rc==SQLITE_OK ) rc = SQLITE_PTR_TO_INT(pRet);
      }
    }
  }
  return rc;
}
#else
# define vdbeSorterJoinAll(x,rcin) (rcin)
#endif
................................................................................
  (void)vdbeSorterJoinAll(pSorter, SQLITE_OK);
  vdbeMergeEngineFree(pSorter->pMerger);
  pSorter->pMerger = 0;
  for(i=0; i<pSorter->nTask; i++){
    SortSubtask *pTask = &pSorter->aTask[i];
    vdbeSortSubtaskCleanup(db, pTask);
  }
  if( pSorter->list.aMemory==0 ){
    vdbeSorterRecordFree(0, pSorter->list.pRecord);
  }
  pSorter->list.pRecord = 0;
  pSorter->list.nInMemory = 0;
  pSorter->bUsePMA = 0;
  pSorter->iMemory = 0;
}

/*
** Free any cursor components allocated by sqlite3VdbeSorterXXX routines.
*/
void sqlite3VdbeSorterClose(sqlite3 *db, VdbeCursor *pCsr){
  VdbeSorter *pSorter = pCsr->pSorter;
  if( pSorter ){
    sqlite3VdbeSorterReset(db, pSorter);
    vdbeMergeEngineFree(pSorter->pMerger);
    sqlite3_free(pSorter->list.aMemory);
    sqlite3DbFree(db, pSorter->pUnpacked);
    sqlite3DbFree(db, pSorter);
    pCsr->pSorter = 0;
  }
}

/*
** Allocate space for a file-handle and open a temporary file. If successful,
................................................................................
}

/*
** Merge the two sorted lists p1 and p2 into a single list.
** Set *ppOut to the head of the new list.
*/
static void vdbeSorterMerge(
  KeyInfo *pKeyInfo,
  UnpackedRecord *r2,
  SorterRecord *p1,               /* First list to merge */
  SorterRecord *p2,               /* Second list to merge */
  SorterRecord **ppOut            /* OUT: Head of merged list */
){
  SorterRecord *pFinal = 0;
  SorterRecord **pp = &pFinal;
  void *pVal2 = p2 ? SRVAL(p2) : 0;

  while( p1 && p2 ){
    int res;
    res = vdbeSorterCompare(pKeyInfo, r2, SRVAL(p1), p1->nVal, pVal2, p2->nVal);
    if( res<=0 ){
      *pp = p1;
      pp = &p1->u.pNext;
      p1 = p1->u.pNext;
      pVal2 = 0;
    }else{
      *pp = p2;
................................................................................
    }
  }
  *pp = p1 ? p1 : p2;
  *ppOut = pFinal;
}

/*
** Sort the linked list of records headed at pLvl->in.l.pRecord. Return 
** SQLITE_OK if successful, or an SQLite error code (i.e. SQLITE_NOMEM) if 
** an error occurs.
*/
static int vdbeSorterSort(
  SortList *pList, 
  KeyInfo *pKeyInfo, 
  UnpackedRecord *pUnpacked
){
  int i;
  SorterRecord **aSlot;
  SorterRecord *p;

  aSlot = (SorterRecord **)sqlite3MallocZero(64 * sizeof(SorterRecord *));
  if( !aSlot ){
    return SQLITE_NOMEM;
  }

  p = pList->pRecord;
  while( p ){
    SorterRecord *pNext;
    if( pList->aMemory ){
      if( (u8*)p==pList->aMemory ){
        pNext = 0;
      }else{
        assert( p->u.iNext<sqlite3MallocSize(pList->aMemory) );
        pNext = (SorterRecord*)&pList->aMemory[p->u.iNext];
      }
    }else{
      pNext = p->u.pNext;
    }

    p->u.pNext = 0;
    for(i=0; aSlot[i]; i++){
      vdbeSorterMerge(pKeyInfo, pUnpacked, p, aSlot[i], &p);
      aSlot[i] = 0;
    }
    aSlot[i] = p;
    p = pNext;
  }

  p = 0;
  for(i=0; i<64; i++){
    vdbeSorterMerge(pKeyInfo, pUnpacked, p, aSlot[i], &p);
  }
  pList->pRecord = p;

  sqlite3_free(aSlot);
  return SQLITE_OK;
}

/*
** Initialize a PMA-writer object.
................................................................................
** is guaranteed to be nByte bytes or smaller in size. This function
** attempts to extend the file to nByte bytes in size and to ensure that
** the VFS has memory mapped it.
**
** Whether or not the file does end up memory mapped of course depends on
** the specific VFS implementation.
*/
static void vdbeSorterExtendFile(
  sqlite3 *db, 
  SortFile *pFile,
  i64 nByte
){
  if( nByte<=(i64)(db->nMaxSorterMmap) && nByte>pFile->nByte ){
    sqlite3_file *pFd = pFile->pFd;
    int rc = sqlite3OsTruncate(pFd, nByte);
    if( rc==SQLITE_OK ){
      void *p = 0;
      sqlite3OsFetch(pFd, 0, nByte, &p);
      sqlite3OsUnfetch(pFd, 0, p);
      pFile->nByte = nByte;
    }
  }
}
#else
# define vdbeSorterExtendFile(x,y,z) SQLITE_OK
#endif



























































/*
** Advance the MergeEngine iterator passed as the second argument to
** the next entry. Set *pbEof to true if this means the iterator has 
** reached EOF.
**
** Return SQLITE_OK if successful or an error code if an error occurs.
*/
static int vdbeSorterNext(
  SortLevel *pLvl,
  MergeEngine *pMerger,
  int *pbEof
){
  int rc;
  int iPrev = pMerger->aTree[1];/* Index of iterator to advance */
  KeyInfo *pKeyInfo = pLvl->pTask->pKeyInfo;
  UnpackedRecord *r2 = pLvl->pUnpacked;

  /* Advance the current iterator */
  rc = vdbePmaReaderNext(&pMerger->aIter[iPrev]);

  /* Update contents of aTree[] */
  if( rc==SQLITE_OK ){
    int i;                      /* Index of aTree[] to recalculate */
................................................................................
      /* Compare pIter1 and pIter2. Store the result in variable iRes. */
      int iRes;
      if( pIter1->pFile==0 ){
        iRes = +1;
      }else if( pIter2->pFile==0 ){
        iRes = -1;
      }else{
        iRes = vdbeSorterCompare(pKeyInfo, r2, 
            pIter1->aKey, pIter1->nKey, pKey2, pIter2->nKey
        );
      }

      /* If pIter1 contained the smaller value, set aTree[i] to its index.
      ** Then set pIter2 to the next iterator to compare to pIter1. In this
      ** case there is no cache of pIter2 in pLvl->pUnpacked, so set
      ** pKey2 to point to the record belonging to pIter2.
      **
      ** Alternatively, if pIter2 contains the smaller of the two values,
      ** set aTree[i] to its index and update pIter1. If vdbeSorterCompare()
      ** was actually called above, then pLvl->pUnpacked now contains
      ** a value equivalent to pIter2. So set pKey2 to NULL to prevent
      ** vdbeSorterCompare() from decoding pIter2 again.
      **
      ** If the two values were equal, then the value from the oldest
      ** PMA should be considered smaller. The VdbeSorter.aIter[] array
      ** is sorted from oldest to newest, so pIter1 contains older values
      ** than pIter2 iff (pIter1<pIter2).  */
................................................................................
    }
    *pbEof = (pMerger->aIter[pMerger->aTree[1]].pFile==0);
  }

  return rc;
}

static UnpackedRecord *vdbeSorterAllocUnpackedRecord(KeyInfo *pKeyInfo){
  char *pFree;
  UnpackedRecord *pRet;
  pRet = sqlite3VdbeAllocUnpackedRecord(pKeyInfo, 0, 0, &pFree);
  assert( pRet==(UnpackedRecord*)pFree );
  if( pRet ){
    pRet->nField = pKeyInfo->nField;
    pRet->errCode = 0;
  }


  return pRet;


}

#if 0
static void vdbeSorterWorkDebug(SortLevel *pLvl, const char *zEvent){
  i64 t;
  SortLevel *p;
  SortSubtask *pTask = pLvl->pTask;
  int iTask = (pTask - pTask->pSorter->aTask);
  int iLvl = 0;
  for(p=pTask->pLevel; p!=pLvl; p=p->pNext) iLvl++;
  sqlite3OsCurrentTimeInt64(pTask->db->pVfs, &t);
  fprintf(stderr, "%lld:%d.%d %s\n", t, iTask, iLvl, zEvent);
}
static void vdbeSorterRewindDebug(sqlite3 *db, const char *zEvent){
  i64 t;
  sqlite3OsCurrentTimeInt64(db->pVfs, &t);
  fprintf(stderr, "%lld:X %s\n", t, zEvent);
}
#else
# define vdbeSorterWorkDebug(x,y)
# define vdbeSorterRewindDebug(x,y)
#endif






/*
** Merge the data currently stored in (pLevel->in), if any, into a new PMA 
** stored within (pLevel->out).
*/
static int vdbeSorterWorkLevel(SortLevel *pLvl){
  int rc = SQLITE_OK;             /* Return code */
  SortSubtask *pTask = pLvl->pTask;
  MergeEngine *pMerger = 0;


  SortFile *pOut = &pLvl->out;    /* Write new PMA here */
  i64 nOut = 0;                   /* Expected size of new PMA */
  PmaWriter writer;               /* Used to write new PMA to pOut */
  int bEof = 0;




  vdbeSorterWorkDebug(pLvl, "enter");
 
  if( pLvl->pUnpacked==0 ){
    pLvl->pUnpacked = vdbeSorterAllocUnpackedRecord(pTask->pKeyInfo);
    if( pLvl->pUnpacked==0 ){
      rc = SQLITE_NOMEM;

      goto work_level_out;
    }
  }


  if( pLvl->out.pFd==0 ){
    assert( pLvl->out.iOff==0 );
    assert( pLvl->out.nByte==0 );
    assert( pLvl->out.nPMA==0 );
    rc = vdbeSorterOpenTempFile(pTask->db->pVfs, &pLvl->out.pFd);
    if( rc!=SQLITE_OK ) goto work_level_out;




  }






  if( pLvl==pTask->pLevel ){
    if( pLvl->in.l.pRecord==0 ){
      bEof = 1;

    }else{
      rc = vdbeSorterSort(&pLvl->in.l, pTask->pKeyInfo, pLvl->pUnpacked);
      nOut = pLvl->in.l.nInMemory;
    }
  }else{
    int nPMA = pLvl->in.f.nPMA;
    if( nPMA==0 ){
      bEof = 1;
    }else{
      pMerger = vdbeMergeEngineNew(nPMA);
      if( pMerger==0 ){
        rc = SQLITE_NOMEM;
      }else{
        /* Configure the merger object to read and merge data from all 


         ** PMAs at pLvl.  */
        int iIter;
        i64 iReadOff = 0;
        for(iIter=0; iIter<nPMA && rc==SQLITE_OK; iIter++){
          PmaReader *pIter = &pMerger->aIter[iIter];
          rc = vdbePmaReaderInit(pTask, &pLvl->in.f, iReadOff, pIter, &nOut);
          iReadOff = pIter->iEof;

        }

        for(iIter=pMerger->nTree-1; rc==SQLITE_OK && iIter>0; iIter--){
          rc = vdbeSorterDoCompare(pLvl, pMerger, iIter);
        }
      }
    }
  }
  if( rc!=SQLITE_OK ) goto work_level_out;

  /* If mmap is to be used, pre-extend and map the temp file. */
  vdbeSorterExtendFile(pTask->db, &pLvl->out, pLvl->out.iOff + nOut + 9);

  if( bEof==0 ){
    vdbePmaWriterInit(pOut->pFd, &writer, pTask->pgsz, pOut->iOff);
    vdbePmaWriteVarint(&writer, nOut);

    while( rc==SQLITE_OK && bEof==0 ){
      u8 *aKey;                     /* Next key to write to output */
      int nKey;                     /* Size of aKey[] in bytes */
      if( pMerger==0 ){
        aKey = SRVAL(pLvl->in.l.pRecord);
        nKey = pLvl->in.l.pRecord->nVal;
      }else{
        PmaReader *pIter = &pMerger->aIter[ pMerger->aTree[1] ];
        assert( pIter->pFile );  /* pIter is not at EOF */
        aKey = pIter->aKey;
        nKey = pIter->nKey;
      }

      vdbePmaWriteVarint(&writer, nKey);
      vdbePmaWriteBlob(&writer, aKey, nKey);

      if( pMerger==0 ){
        SorterRecord *pNext = pLvl->in.l.pRecord->u.pNext;
        if( pLvl->in.l.aMemory==0 ) sqlite3_free(pLvl->in.l.pRecord);
        pLvl->in.l.pRecord = pNext;
        bEof = (pNext==0);
      }else{
        rc = vdbeSorterNext(pLvl, pMerger, &bEof);
      }
    }
    rc = vdbePmaWriterFinish(&writer, &pOut->iOff);

    pOut->nPMA++;

    if( rc==SQLITE_OK && pMerger ){
      sqlite3OsCloseFree(pLvl->in.f.pFd);
      pLvl->in.f.pFd = 0;
    }
    vdbeMergeEngineFree(pMerger);




  }





  if( rc==SQLITE_OK && (
      (pOut->nPMA>=SORTER_MAX_MERGE_COUNT)
   || (pTask->nConsolidate && pLvl->pNext)
   || (pTask->nConsolidate && pTask->nConsolidate<pOut->nPMA)
  )){
    SortLevel *pNext = pLvl->pNext;
    if( pNext==0 ){
      pNext = (SortLevel*)sqlite3_malloc(sizeof(SortLevel));
      if( pNext==0 ){
        rc = SQLITE_NOMEM;



        goto work_level_out;
      }
      memset(pNext, 0, sizeof(SortLevel));
      pLvl->pNext = pNext;
      pNext->pTask = pTask;
    }

    /* If there is a thread running on the next level, block on it. */
#if SQLITE_MAX_WORKER_THREADS>0
    if( pNext->pThread ){
      void *pRet;
      rc = sqlite3ThreadJoin(pNext->pThread, &pRet);
      pNext->pThread = 0;
      pNext->bDone = 0;
      if( rc==SQLITE_OK ) rc = SQLITE_PTR_TO_INT(pRet);
      if( rc!=SQLITE_OK ) goto work_level_out;
    }
#endif



    pNext->in.f = pLvl->out;
    memset(&pLvl->out, 0, sizeof(pLvl->out));
  }

 work_level_out:

  vdbeSorterWorkDebug(pLvl, "exit");
  if( rc==SQLITE_OK && pLvl->pUnpacked->errCode ){
    assert( pLvl->pUnpacked->errCode==SQLITE_NOMEM );
    rc = SQLITE_NOMEM;
  }
  return rc;
}

/*
** Run the activity scheduled by the object passed as the only argument
** in the current thread.
*/
static int vdbeSorterRun(SortLevel *pLvl){
  int rc;

  assert( pLvl->bDone==0 );
  assert( pLvl->pThread==0 );
  while( 1 ){
    rc = vdbeSorterWorkLevel(pLvl);
    if( rc==SQLITE_OK && pLvl->pTask->pLevel==pLvl && pLvl->in.l.aMemory ){
      assert( pLvl->pTask->pSorter->list.aMemory==0 );
      assert( pLvl->in.l.pRecord==0 );
      pLvl->pTask->pSorter->list.aMemory = pLvl->in.l.aMemory;
      pLvl->in.l.aMemory = 0;
    }

    if( rc!=SQLITE_OK || pLvl->out.nPMA>0 ) break;
    pLvl = pLvl->pNext;
    assert( pLvl->bDone==0 );
    assert( pLvl->pThread==0 );
  }

  pLvl->bDone = 0;
  return rc;
}

#if SQLITE_MAX_WORKER_THREADS>0
static void *vdbeSorterThread(void *pCtx){
  int rc;
  SortLevel *pLvl = (SortLevel*)pCtx;

  rc = vdbeSorterWorkLevel(pLvl);
  if( rc==SQLITE_OK && pLvl->out.nPMA==0 ){
    SortLevel *pNext = pLvl->pNext;
    void *pCtx = (void*)pNext;
    assert( pNext->pThread==0 );
    rc = sqlite3ThreadCreate(&pNext->pThread, vdbeSorterThread, pCtx);
  }

  pLvl->bDone = 1;
  return SQLITE_INT_TO_PTR(rc);
}
#endif

/*
** Flush the current contents of VdbeSorter.pRecord to a new PMA, possibly
** using a background thread.
**
** If argument bFg is non-zero, the operation always uses the calling thread.
*/
static int vdbeSorterFlushPMA(sqlite3 *db, const VdbeCursor *pCsr, int bFg){
  VdbeSorter *pSorter = pCsr->pSorter;
  int rc = SQLITE_OK;
  int i;
  SortSubtask *pTask = 0;    /* Sub-task new PMA is written to */
  SortLevel *pLevel;         /* Level to write to */

  /* Set the use-temp-files flag. */
  pSorter->bUsePMA = 1;

  /* Select one of the sub-tasks to flush this PMA. In single threaded
  ** mode (pSorter->nTask==1), this is always aTask[0]. In multi-threaded mode,
  ** it may be any of the pSorter->nTask sub-tasks.  */
  for(i=0; i<pSorter->nTask; i++){
    pTask = &pSorter->aTask[i];
    if( pTask->pLevel==0 
     || pTask->pLevel->pThread==0 
     || pTask->pLevel->bDone 
    ){
      break;
    }
  }

  /* If the first level for this task has not been allocated, allocate it. */
  if( pTask->pLevel==0 ){
    SortLevel *pNew = (SortLevel*)sqlite3_malloc(sizeof(SortLevel));
    if( pNew==0 ){
      rc = SQLITE_NOMEM;
    }else{
      memset(pNew, 0, sizeof(SortLevel));
      pNew->pTask = pTask;
      pTask->pLevel = pNew;
    }
  }
  pLevel = pTask->pLevel;

  /* If there is a background thread using the selected task, wait for
  ** it to finish. */
#if SQLITE_MAX_WORKER_THREADS>0
  if( rc==SQLITE_OK && pLevel->pThread ){
    void *pRet = 0;
    rc = sqlite3ThreadJoin(pLevel->pThread, &pRet);
    pLevel->pThread = 0;
    pLevel->bDone = 0;
    if( rc==SQLITE_OK ) rc = SQLITE_PTR_TO_INT(pRet);
  }
#endif

  if( rc==SQLITE_OK ){
    u8 *aNewMem = 0;
    if( pSorter->list.aMemory && pSorter->nTask>1 ){
      aNewMem = pLevel->in.l.aMemory;
      if( aNewMem==0 ){
        aNewMem = sqlite3_malloc(pSorter->mxPmaSize);
        if( aNewMem==0 ) rc = SQLITE_NOMEM;
      }
    }
    assert( pLevel->in.l.pRecord==0 );
    pLevel->in.l = pSorter->list;
    pSorter->list.pRecord = 0;
    pSorter->list.nInMemory = 0;
    pSorter->list.aMemory = aNewMem;
    if( rc==SQLITE_OK ){
#if SQLITE_MAX_WORKER_THREADS>0
      if( pSorter->nTask>1 ){
        void *pCtx = (void*)pLevel;
        rc = sqlite3ThreadCreate(&pLevel->pThread, vdbeSorterThread, pCtx);
        pSorter->nMemory = aNewMem ? sqlite3MallocSize(aNewMem) : 0;
      }else
#endif
      {
        rc = vdbeSorterRun(pLevel);
      }
    }
  }

  return rc;
}

................................................................................
  **
  **   * The total memory allocated for the in-memory list is greater 
  **     than (page-size * 10) and sqlite3HeapNearlyFull() returns true.
  */
  nReq = pVal->n + sizeof(SorterRecord);
  nPMA = pVal->n + sqlite3VarintLen(pVal->n);
  if( pSorter->mxPmaSize ){
    if( pSorter->list.aMemory ){
      bFlush = pSorter->iMemory && (pSorter->iMemory+nReq) > pSorter->mxPmaSize;
    }else{
      int nInMemory = pSorter->list.nInMemory;
      bFlush = (
          (nInMemory > pSorter->mxPmaSize)
       || (nInMemory > pSorter->mnPmaSize && sqlite3HeapNearlyFull())
      );
    }
    if( bFlush ){
      rc = vdbeSorterFlushPMA(db, pCsr, 0);
      pSorter->list.nInMemory = 0;
      pSorter->iMemory = 0;
      assert( rc!=SQLITE_OK || pSorter->list.pRecord==0 );
    }
  }

  pSorter->list.nInMemory += nPMA;

  if( pSorter->list.aMemory ){
    int nMin = pSorter->iMemory + nReq;

    if( nMin>pSorter->nMemory ){
      u8 *aNew;
      int nNew = pSorter->nMemory * 2;
      while( nNew < nMin ) nNew = nNew*2;
      if( nNew > pSorter->mxPmaSize ) nNew = pSorter->mxPmaSize;
      if( nNew < nMin ) nNew = nMin;

      aNew = sqlite3Realloc(pSorter->list.aMemory, nNew);
      if( !aNew ) return SQLITE_NOMEM;
      pSorter->list.pRecord = (SorterRecord*)(
          aNew + ((u8*)pSorter->list.pRecord - pSorter->list.aMemory)
      );
      pSorter->list.aMemory = aNew;
      pSorter->nMemory = nNew;
    }

    pNew = (SorterRecord*)&pSorter->list.aMemory[pSorter->iMemory];
    pSorter->iMemory += ROUND8(nReq);
    pNew->u.iNext = (u8*)(pSorter->list.pRecord) - pSorter->list.aMemory;
  }else{
    pNew = (SorterRecord *)sqlite3Malloc(nReq);
    if( pNew==0 ){
      return SQLITE_NOMEM;
    }
    pNew->u.pNext = pSorter->list.pRecord;
  }

  memcpy(SRVAL(pNew), pVal->z, pVal->n);
  pNew->nVal = pVal->n;
  pSorter->list.pRecord = pNew;

  return rc;
}













/*
** Once the sorter has been populated by calls to sqlite3VdbeSorterWrite,
** this function is called to prepare for iterating through the records
** in sorted order.
*/
int sqlite3VdbeSorterRewind(sqlite3 *db, const VdbeCursor *pCsr, int *pbEof){
  VdbeSorter *pSorter = pCsr->pSorter;
  int rc = SQLITE_OK;             /* Return code */
  int nTask = 0;
  int i;

  assert( pSorter );

  /* If no data has been written to disk, then do not do so now. Instead,
  ** sort the VdbeSorter.list.pRecord list. The vdbe layer will read data 
  ** directly from the in-memory list.  */
  *pbEof = 0;
  if( pSorter->bUsePMA==0 ){
    if( pSorter->list.pRecord ){
      SortSubtask *pTask = &pSorter->aTask[0];

      UnpackedRecord *pUnpack = vdbeSorterAllocUnpackedRecord(pTask->pKeyInfo);
      if( pUnpack==0 ) return SQLITE_NOMEM;
      rc = vdbeSorterSort(&pSorter->list, pTask->pKeyInfo, pUnpack);
      sqlite3DbFree(db, pUnpack);




    }else{
      *pbEof = 1;
    }
    return rc;
  }

  /* Write the current in-memory list to a PMA. */
  if( pSorter->list.pRecord ){
    rc = vdbeSorterFlushPMA(db, pCsr, 0);
  }

  /* Join all threads */
  rc = vdbeSorterJoinAll(pSorter, rc);

  vdbeSorterRewindDebug(db, "rewind");


  for(i=0; i<pSorter->nTask; i++){
    if( pSorter->aTask[i].pLevel ) nTask++;
  }

  for(i=0; rc==SQLITE_OK && i<pSorter->nTask; i++){
    SortSubtask *pTask = &pSorter->aTask[i];
    if( pTask->pLevel ){
      SortLevel *pLvl = pTask->pLevel;
      pTask->nConsolidate = (SORTER_MAX_MERGE_COUNT / nTask);


#if SQLITE_MAX_WORKER_THREADS>0
      if( i<(pSorter->nTask-1) ){
        void *pCtx = (void*)pLvl;
        rc = sqlite3ThreadCreate(&pLvl->pThread, vdbeSorterThread, pCtx);
      }else
#endif
      {
        assert( pLvl->pThread==0 );
        rc = vdbeSorterRun(pLvl);

      }
    }
  }

  /* Join all threads */
  rc = vdbeSorterJoinAll(pSorter, rc);

................................................................................
  ** and merge all remaining PMAs.  */
  assert( pSorter->pMerger==0 );
  if( rc==SQLITE_OK ){
    int nIter = 0;                /* Number of iterators used */
    int i;
    MergeEngine *pMerger;
    for(i=0; i<pSorter->nTask; i++){
      SortSubtask *pTask = &pSorter->aTask[i];
      if( pTask->pLevel ){
        SortLevel *pLvl;
        for(pLvl=pTask->pLevel; pLvl->pNext; pLvl=pLvl->pNext){
          assert( pLvl->out.nPMA==0 );
        }
        nIter += pLvl->out.nPMA;
      }
    }

    pSorter->pMerger = pMerger = vdbeMergeEngineNew(nIter);
    if( pMerger==0 ){
      rc = SQLITE_NOMEM;
    }else{
      int iIter = 0;

      for(i=0; i<pSorter->nTask; i++){
        SortSubtask *pTask = &pSorter->aTask[i];
        if( pTask->pLevel ){
          int iPMA;
          i64 iReadOff = 0;

          SortLevel *pLvl;
          for(pLvl=pTask->pLevel; pLvl->pNext; pLvl=pLvl->pNext);

          for(iPMA=0; iPMA<pLvl->out.nPMA && rc==SQLITE_OK; iPMA++){
            i64 nDummy = 0;
            PmaReader *pIter = &pMerger->aIter[iIter++];
            rc = vdbePmaReaderInit(pTask, &pLvl->out, iReadOff, pIter, &nDummy);
            iReadOff = pIter->iEof;
          }
        }
      }

      for(i=pMerger->nTree-1; rc==SQLITE_OK && i>0; i--){
        rc = vdbeSorterDoCompare(pSorter->aTask[0].pLevel, pMerger, i);
      }
    }
  }

  if( rc==SQLITE_OK ){
    *pbEof = (pSorter->pMerger->aIter[pSorter->pMerger->aTree[1]].pFile==0);
  }
  vdbeSorterRewindDebug(db, "rewinddone");
  return rc;
}

/*
** Advance to the next element in the sorter.
*/
int sqlite3VdbeSorterNext(sqlite3 *db, const VdbeCursor *pCsr, int *pbEof){
  VdbeSorter *pSorter = pCsr->pSorter;
  int rc;                         /* Return code */

  if( pSorter->pMerger ){
    rc = vdbeSorterNext(pSorter->aTask[0].pLevel, pSorter->pMerger, pbEof);
  }else{
    SorterRecord *pFree = pSorter->list.pRecord;
    pSorter->list.pRecord = pFree->u.pNext;
    pFree->u.pNext = 0;
    if( pSorter->list.aMemory==0 ) vdbeSorterRecordFree(db, pFree);
    *pbEof = !pSorter->list.pRecord;
    rc = SQLITE_OK;
  }
  return rc;
}

/*
** Return a pointer to a buffer owned by the sorter that contains the 
................................................................................
  void *pKey;
  if( pSorter->pMerger ){
    PmaReader *pIter;
    pIter = &pSorter->pMerger->aIter[ pSorter->pMerger->aTree[1] ];
    *pnKey = pIter->nKey;
    pKey = pIter->aKey;
  }else{
    *pnKey = pSorter->list.pRecord->nVal;
    pKey = SRVAL(pSorter->list.pRecord);
  }
  return pKey;
}

/*
** Copy the current sorter key into the memory cell pOut.
*/
................................................................................
int sqlite3VdbeSorterCompare(
  const VdbeCursor *pCsr,         /* Sorter cursor */
  Mem *pVal,                      /* Value to compare to current sorter key */
  int nIgnore,                    /* Ignore this many fields at the end */
  int *pRes                       /* OUT: Result of comparison */
){
  VdbeSorter *pSorter = pCsr->pSorter;
  UnpackedRecord *r2 = pSorter->pUnpacked;
  KeyInfo *pKeyInfo = pCsr->pKeyInfo;
  int i;
  void *pKey; int nKey;           /* Sorter key to compare pVal with */

  if( r2==0 ){
    r2 = vdbeSorterAllocUnpackedRecord(pSorter->aTask[0].pKeyInfo);
    if( r2==0 ) return SQLITE_NOMEM;
    pSorter->pUnpacked = r2;
    assert( r2->nField>=pKeyInfo->nField-nIgnore );
    r2->nField = pKeyInfo->nField-nIgnore;
  }
  assert( r2->nField==pKeyInfo->nField-nIgnore );

  pKey = vdbeSorterRowkey(pSorter, &nKey);
  sqlite3VdbeRecordUnpack(pKeyInfo, nKey, pKey, r2);
  for(i=0; i<r2->nField; i++){
    if( r2->aMem[i].flags & MEM_Null ){
      *pRes = -1;
      return SQLITE_OK;
    }
  }

  *pRes = sqlite3VdbeRecordCompare(pVal->n, pVal->z, r2, 0);
  return SQLITE_OK;
}