/ Check-in [7c914e39]
Login

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

Overview
Comment:Refactoring the OP_Column opcode for improved performance and maintainability.
Downloads: Tarball | ZIP archive | SQL archive
Timelines: family | ancestors | descendants | both | OP_Column-refactor
Files: files | file ages | folders
SHA1: 7c914e3997d2b28164a2fa7eb4398262b6ddb4b2
User & Date: drh 2013-11-20 17:25:55
Context
2013-11-20
19:28
Further performance tweaks to OP_Column. check-in: 0e3f5df6 user: drh tags: OP_Column-refactor
17:25
Refactoring the OP_Column opcode for improved performance and maintainability. check-in: 7c914e39 user: drh tags: OP_Column-refactor
02:53
Simplifications to the VdbeCursor object. check-in: 5562cd34 user: drh tags: trunk
Changes
Hide Diffs Unified Diffs Ignore Whitespace Patch

Changes to src/vdbe.c.

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
....
2249
2250
2251
2252
2253
2254
2255
2256
2257
2258
2259
2260
2261
2262
2263
2264
2265
2266
2267
2268
2269
2270

2271
2272
2273
2274
2275
2276
2277
2278
2279
2280
2281

2282
2283
2284
2285
2286
2287
2288
2289
2290
2291
2292
2293
2294
2295
2296
2297
2298
2299
2300
2301
2302
2303
2304




2305
2306
2307
2308
2309
2310


2311
2312




2313






2314

2315
2316
2317



2318
2319
2320
2321
2322
2323
2324
2325

2326
2327
2328
2329
2330

2331



2332
2333
2334
2335
2336
2337
2338
2339
2340
2341
2342
2343
2344
2345
2346
2347
2348
2349
2350
2351
2352
2353
2354
2355
2356
2357
2358
2359
2360
2361
2362
2363
2364
2365
2366
2367
2368
2369
2370
2371
2372
2373
2374
2375
2376
2377
2378
2379
2380
2381
2382
2383
2384
2385
2386
2387
2388



2389
2390
2391
2392
2393
2394
2395
2396
2397
2398
2399
2400
2401
2402
2403
2404
2405
2406
2407
2408
2409
2410
2411
2412
2413
2414
2415
2416
2417
2418
2419
2420
2421
2422
2423
2424
2425
2426
2427
2428
2429
2430
2431
2432
2433
2434
2435
2436


2437
2438
2439
2440




2441
2442
2443
2444


2445
2446
2447
2448
2449
2450
2451
2452
2453
2454
2455
2456


2457



2458
2459
2460
2461
2462
2463
2464
2465
2466
2467
2468
2469
2470
2471
2472
2473
2474
2475
2476
2477
2478
2479
2480



2481
2482

2483
2484
2485
2486
2487
2488
2489

2490
2491

2492
2493
2494
2495
2496
2497
2498
2499
2500
2501
2502
2503
2504
2505
2506
2507
2508
2509
2510
2511
2512
2513
2514
2515
2516
2517
2518
2519
2520
2521
2522

2523
2524
2525
2526
2527
2528
2529
2530
2531
2532
2533
2534
2535
2536
2537
2538
....
3308
3309
3310
3311
3312
3313
3314


3315
3316
3317
3318
3319
3320
3321
....
3368
3369
3370
3371
3372
3373
3374

3375
3376
3377
3378
3379
3380
3381
....
3413
3414
3415
3416
3417
3418
3419


3420
3421
3422
3423
3424
3425
3426
....
3444
3445
3446
3447
3448
3449
3450

3451
3452
3453
3454
3455
3456
3457
....
4417
4418
4419
4420
4421
4422
4423

4424
4425
4426
4427
4428
4429
4430
  Mem *pMem = &p->aMem[p->nMem-iCur];

  int nByte;
  VdbeCursor *pCx = 0;
  nByte = 
      ROUND8(sizeof(VdbeCursor)) + 
      (isBtreeCursor?sqlite3BtreeCursorSize():0) + 
      2*nField*sizeof(u32);

  assert( iCur<p->nCursor );
  if( p->apCsr[iCur] ){
    sqlite3VdbeFreeCursor(p, p->apCsr[iCur]);
    p->apCsr[iCur] = 0;
  }
  if( SQLITE_OK==sqlite3VdbeMemGrow(pMem, nByte, 0) ){
    p->apCsr[iCur] = pCx = (VdbeCursor*)pMem->z;
    memset(pCx, 0, sizeof(VdbeCursor));
    pCx->iDb = iDb;
    pCx->nField = nField;
    if( nField ){
      pCx->aType = (u32 *)&pMem->z[ROUND8(sizeof(VdbeCursor))];

    }
    if( isBtreeCursor ){
      pCx->pCursor = (BtCursor*)
          &pMem->z[ROUND8(sizeof(VdbeCursor))+2*nField*sizeof(u32)];
      sqlite3BtreeCursorZero(pCx->pCursor);
    }
  }
  return pCx;
}

/*
................................................................................
**
** If the OPFLAG_LENGTHARG and OPFLAG_TYPEOFARG bits are set on P5 when
** the result is guaranteed to only be used as the argument of a length()
** or typeof() function, respectively.  The loading of large blobs can be
** skipped for length() and all content loading can be skipped for typeof().
*/
case OP_Column: {
  u32 payloadSize;   /* Number of bytes in the record */
  i64 payloadSize64; /* Number of bytes in the record */
  int p1;            /* P1 value of the opcode */
  int p2;            /* column number to retrieve */
  VdbeCursor *pC;    /* The VDBE cursor */
  char *zRec;        /* Pointer to complete record-data */
  BtCursor *pCrsr;   /* The BTree cursor */
  u32 *aType;        /* aType[i] holds the numeric type of the i-th column */
  u32 *aOffset;      /* aOffset[i] is offset to start of data for i-th column */
  int nField;        /* number of fields in the record */
  int len;           /* The length of the serialized data for the column */
  int i;             /* Loop counter */
  char *zData;       /* Part of the record being decoded */
  Mem *pDest;        /* Where to write the extracted value */
  Mem sMem;          /* For storing the record being decoded */

  u8 *zIdx;          /* Index into header */
  u8 *zEndHdr;       /* Pointer to first byte after the header */
  u32 offset;        /* Offset into the data */
  u32 szField;       /* Number of bytes in the content of a field */
  int szHdr;         /* Size of the header size field at start of record */
  int avail;         /* Number of bytes of available data */
  u32 t;             /* A type code from the record header */
  Mem *pReg;         /* PseudoTable input register */


  p1 = pOp->p1;

  p2 = pOp->p2;
  pC = 0;
  memset(&sMem, 0, sizeof(sMem));
  assert( p1<p->nCursor );
  assert( pOp->p3>0 && pOp->p3<=(p->nMem-p->nCursor) );
  pDest = &aMem[pOp->p3];
  memAboutToChange(p, pDest);
  zRec = 0;

  /* This block sets the variable payloadSize to be the total number of
  ** bytes in the record.
  **
  ** zRec is set to be the complete text of the record if it is available.
  ** The complete record text is always available for pseudo-tables
  ** If the record is stored in a cursor, the complete record text
  ** might be available in the  pC->aRow cache.  Or it might not be.
  ** If the data is unavailable,  zRec is set to NULL.
  **
  ** We also compute the number of columns in the record.  For cursors,
  ** the number of columns is stored in the VdbeCursor.nField element.
  */
  pC = p->apCsr[p1];
  assert( pC!=0 );




#ifndef SQLITE_OMIT_VIRTUALTABLE
  assert( pC->pVtabCursor==0 );
#endif
  pCrsr = pC->pCursor;
  if( pCrsr!=0 ){
    /* The record is stored in a B-Tree */


    rc = sqlite3VdbeCursorMoveto(pC);
    if( rc ) goto abort_due_to_error;




    if( pC->nullRow ){






      payloadSize = 0;

    }else if( pC->cacheStatus==p->cacheCtr ){
      payloadSize = pC->payloadSize;
      zRec = (char*)pC->aRow;



    }else if( pC->isIndex ){
      assert( sqlite3BtreeCursorIsValid(pCrsr) );
      VVA_ONLY(rc =) sqlite3BtreeKeySize(pCrsr, &payloadSize64);
      assert( rc==SQLITE_OK );   /* True because of CursorMoveto() call above */
      /* sqlite3BtreeParseCellPtr() uses getVarint32() to extract the
      ** payload size, so it is impossible for payloadSize64 to be
      ** larger than 32 bits. */
      assert( (payloadSize64 & SQLITE_MAX_U32)==(u64)payloadSize64 );

      payloadSize = (u32)payloadSize64;
    }else{
      assert( sqlite3BtreeCursorIsValid(pCrsr) );
      VVA_ONLY(rc =) sqlite3BtreeDataSize(pCrsr, &payloadSize);
      assert( rc==SQLITE_OK );   /* DataSize() cannot fail */

    }



  }else{
    assert( pC->pseudoTableReg>0 );
    pReg = &aMem[pC->pseudoTableReg];
    if( pC->multiPseudo ){
      sqlite3VdbeMemShallowCopy(pDest, pReg+p2, MEM_Ephem);
      Deephemeralize(pDest);
      goto op_column_out;
    }
    assert( pReg->flags & MEM_Blob );
    assert( memIsValid(pReg) );
    payloadSize = pReg->n;
    zRec = pReg->z;
    pC->cacheStatus = (pOp->p5&OPFLAG_CLEARCACHE) ? CACHE_STALE : p->cacheCtr;
    assert( payloadSize==0 || zRec!=0 );
  }

  /* If payloadSize is 0, then just store a NULL.  This can happen because of
  ** nullRow or because of a corrupt database. */
  if( payloadSize==0 ){
    MemSetTypeFlag(pDest, MEM_Null);
    goto op_column_out;
  }
  assert( db->aLimit[SQLITE_LIMIT_LENGTH]>=0 );
  if( payloadSize > (u32)db->aLimit[SQLITE_LIMIT_LENGTH] ){
    goto too_big;
  }

  nField = pC->nField;
  assert( p2<nField );

  /* Read and parse the table header.  Store the results of the parse
  ** into the record header cache fields of the cursor.
  */
  aType = pC->aType;
  if( pC->cacheStatus==p->cacheCtr ){
    aOffset = pC->aOffset;
  }else{
    assert(aType);
    avail = 0;
    pC->aOffset = aOffset = &aType[nField];
    pC->payloadSize = payloadSize;
    pC->cacheStatus = p->cacheCtr;

    /* Figure out how many bytes are in the header */
    if( zRec ){
      zData = zRec;
    }else{
      if( pC->isIndex ){
        zData = (char*)sqlite3BtreeKeyFetch(pCrsr, &avail);
      }else{
        zData = (char*)sqlite3BtreeDataFetch(pCrsr, &avail);
      }
      /* If KeyFetch()/DataFetch() managed to get the entire payload,
      ** save the payload in the pC->aRow cache.  That will save us from
      ** having to make additional calls to fetch the content portion of
      ** the record.
      */



      assert( avail>=0 );
      if( payloadSize <= (u32)avail ){
        zRec = zData;
        pC->aRow = (u8*)zData;
      }else{
        pC->aRow = 0;
      }
    }
    /* The following assert is true in all cases except when
    ** the database file has been corrupted externally.
    **    assert( zRec!=0 || avail>=payloadSize || avail>=9 ); */
    szHdr = getVarint32((u8*)zData, offset);

    /* Make sure a corrupt database has not given us an oversize header.
    ** Do this now to avoid an oversize memory allocation.
    **
    ** Type entries can be between 1 and 5 bytes each.  But 4 and 5 byte
    ** types use so much data space that there can only be 4096 and 32 of
    ** them, respectively.  So the maximum header length results from a
    ** 3-byte type for each of the maximum of 32768 columns plus three
    ** extra bytes for the header length itself.  32768*3 + 3 = 98307.
    */
    if( offset > 98307 ){
      rc = SQLITE_CORRUPT_BKPT;
      goto op_column_out;
    }

    /* Compute in len the number of bytes of data we need to read in order
    ** to get nField type values.  offset is an upper bound on this.  But
    ** nField might be significantly less than the true number of columns
    ** in the table, and in that case, 5*nField+3 might be smaller than offset.
    ** We want to minimize len in order to limit the size of the memory
    ** allocation, especially if a corrupt database file has caused offset
    ** to be oversized. Offset is limited to 98307 above.  But 98307 might
    ** still exceed Robson memory allocation limits on some configurations.
    ** On systems that cannot tolerate large memory allocations, nField*5+3
    ** will likely be much smaller since nField will likely be less than
    ** 20 or so.  This insures that Robson memory allocation limits are
    ** not exceeded even for corrupt database files.
    */
    len = nField*5 + 3;
    if( len > (int)offset ) len = (int)offset;

    /* The KeyFetch() or DataFetch() above are fast and will get the entire
    ** record header in most cases.  But they will fail to get the complete
    ** record header if the record header does not fit on a single page
    ** in the B-Tree.  When that happens, use sqlite3VdbeMemFromBtree() to
    ** acquire the complete header text.


    */
    if( !zRec && avail<len ){
      sMem.flags = 0;
      sMem.db = 0;




      rc = sqlite3VdbeMemFromBtree(pCrsr, 0, len, pC->isIndex, &sMem);
      if( rc!=SQLITE_OK ){
        goto op_column_out;
      }


      zData = sMem.z;
    }
    zEndHdr = (u8 *)&zData[len];
    zIdx = (u8 *)&zData[szHdr];

    /* Scan the header and use it to fill in the aType[] and aOffset[]
    ** arrays.  aType[i] will contain the type integer for the i-th
    ** column and aOffset[i] will contain the offset from the beginning
    ** of the record to the start of the data for the i-th column
    */
    for(i=0; i<nField; i++){
      if( zIdx<zEndHdr ){


        aOffset[i] = offset;



        if( zIdx[0]<0x80 ){
          t = zIdx[0];
          zIdx++;
        }else{
          zIdx += sqlite3GetVarint32(zIdx, &t);
        }
        aType[i] = t;
        szField = sqlite3VdbeSerialTypeLen(t);
        offset += szField;
        if( offset<szField ){  /* True if offset overflows */
          zIdx = &zEndHdr[1];  /* Forces SQLITE_CORRUPT return below */
          break;
        }
      }else{
        /* If i is less that nField, then there are fewer fields in this
        ** record than SetNumColumns indicated there are columns in the
        ** table. Set the offset for any extra columns not present in
        ** the record to 0. This tells code below to store the default value
        ** for the column instead of deserializing a value from the record.
        */
        aOffset[i] = 0;
      }
    }



    sqlite3VdbeMemRelease(&sMem);
    sMem.flags = MEM_Null;


    /* If we have read more header data than was contained in the header,
    ** or if the end of the last field appears to be past the end of the
    ** record, or if the end of the last field appears to be before the end
    ** of the record (when all fields present), then we must be dealing 
    ** with a corrupt database.
    */

    if( (zIdx > zEndHdr) || (offset > payloadSize)
         || (zIdx==zEndHdr && offset!=payloadSize) ){

      rc = SQLITE_CORRUPT_BKPT;
      goto op_column_out;
    }
  }

  /* Get the column information. If aOffset[p2] is non-zero, then 
  ** deserialize the value from the record. If aOffset[p2] is zero,
  ** then there are not enough fields in the record to satisfy the
  ** request.  In this case, set the value NULL or to P4 if P4 is
  ** a pointer to a Mem object.
  */
  if( aOffset[p2] ){
    assert( rc==SQLITE_OK );
    if( zRec ){
      /* This is the common case where the whole row fits on a single page */
      VdbeMemRelease(pDest);
      sqlite3VdbeSerialGet((u8 *)&zRec[aOffset[p2]], aType[p2], pDest);
    }else{
      /* This branch happens only when the row overflows onto multiple pages */
      t = aType[p2];
      if( (pOp->p5 & (OPFLAG_LENGTHARG|OPFLAG_TYPEOFARG))!=0
       && ((t>=12 && (t&1)==0) || (pOp->p5 & OPFLAG_TYPEOFARG)!=0)
      ){
        /* Content is irrelevant for the typeof() function and for
        ** the length(X) function if X is a blob.  So we might as well use
        ** bogus content rather than reading content from disk.  NULL works
        ** for text and blob and whatever is in the payloadSize64 variable
        ** will work for everything else. */
        zData = t<12 ? (char*)&payloadSize64 : 0;
      }else{
        len = sqlite3VdbeSerialTypeLen(t);

        sqlite3VdbeMemMove(&sMem, pDest);
        rc = sqlite3VdbeMemFromBtree(pCrsr, aOffset[p2], len,  pC->isIndex,
                                     &sMem);
        if( rc!=SQLITE_OK ){
          goto op_column_out;
        }
        zData = sMem.z;
      }
      sqlite3VdbeSerialGet((u8*)zData, t, pDest);
    }
    pDest->enc = encoding;
  }else{
    if( pOp->p4type==P4_MEM ){
      sqlite3VdbeMemShallowCopy(pDest, pOp->p4.pMem, MEM_Static);
    }else{
      MemSetTypeFlag(pDest, MEM_Null);
................................................................................
    assert( pKeyInfo->enc==ENC(db) );
    assert( pKeyInfo->db==db );
    nField = pKeyInfo->nField+pKeyInfo->nXField;
  }else if( pOp->p4type==P4_INT32 ){
    nField = pOp->p4.i;
  }
  assert( pOp->p1>=0 );


  pCur = allocateCursor(p, pOp->p1, nField, iDb, 1);
  if( pCur==0 ) goto no_mem;
  pCur->nullRow = 1;
  pCur->isOrdered = 1;
  rc = sqlite3BtreeCursor(pX, p2, wrFlag, pKeyInfo, pCur->pCursor);
  pCur->pKeyInfo = pKeyInfo;
  assert( OPFLAG_BULKCSR==BTREE_BULKLOAD );
................................................................................
  static const int vfsFlags = 
      SQLITE_OPEN_READWRITE |
      SQLITE_OPEN_CREATE |
      SQLITE_OPEN_EXCLUSIVE |
      SQLITE_OPEN_DELETEONCLOSE |
      SQLITE_OPEN_TRANSIENT_DB;
  assert( pOp->p1>=0 );

  pCx = allocateCursor(p, pOp->p1, pOp->p2, -1, 1);
  if( pCx==0 ) goto no_mem;
  pCx->nullRow = 1;
  rc = sqlite3BtreeOpen(db->pVfs, 0, db, &pCx->pBt, 
                        BTREE_OMIT_JOURNAL | BTREE_SINGLE | pOp->p5, vfsFlags);
  if( rc==SQLITE_OK ){
    rc = sqlite3BtreeBeginTrans(pCx->pBt, 1);
................................................................................
** This opcode works like OP_OpenEphemeral except that it opens
** a transient index that is specifically designed to sort large
** tables using an external merge-sort algorithm.
*/
case OP_SorterOpen: {
  VdbeCursor *pCx;



  pCx = allocateCursor(p, pOp->p1, pOp->p2, -1, 1);
  if( pCx==0 ) goto no_mem;
  pCx->pKeyInfo = pOp->p4.pKeyInfo;
  assert( pCx->pKeyInfo->db==db );
  assert( pCx->pKeyInfo->enc==ENC(db) );
  pCx->isSorter = 1;
  rc = sqlite3VdbeSorterInit(db, pCx);
................................................................................
** P3 is the number of fields in the records that will be stored by
** the pseudo-table.
*/
case OP_OpenPseudo: {
  VdbeCursor *pCx;

  assert( pOp->p1>=0 );

  pCx = allocateCursor(p, pOp->p1, pOp->p3, -1, 0);
  if( pCx==0 ) goto no_mem;
  pCx->nullRow = 1;
  pCx->pseudoTableReg = pOp->p2;
  pCx->isTable = 1;
  pCx->isIndex = 0;
  pCx->multiPseudo = pOp->p5;
................................................................................
  VdbeCursor *pC;

  assert( pOp->p1>=0 && pOp->p1<p->nCursor );
  pC = p->apCsr[pOp->p1];
  assert( pC!=0 );
  pC->nullRow = 1;
  pC->rowidIsValid = 0;

  assert( pC->pCursor || pC->pVtabCursor );
  if( pC->pCursor ){
    sqlite3BtreeClearCursor(pC->pCursor);
  }
  break;
}








|













>



|







 







<




<






<


>
|
|


<




<

>

|
<
<



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


>
>
>
>




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

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










|



|
<
<
<
<
<
<
<
<
<
<
<
<
<
<
<

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



>
>
|

<
<

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







>
|
|
>











|

|


|











|


>






|

|







 







>
>







 







>







 







>
>







 







>







 







>







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
....
2250
2251
2252
2253
2254
2255
2256

2257
2258
2259
2260

2261
2262
2263
2264
2265
2266

2267
2268
2269
2270
2271
2272
2273

2274
2275
2276
2277

2278
2279
2280
2281


2282
2283
2284














2285
2286
2287
2288
2289
2290
2291
2292
2293
2294
2295

2296
2297
2298
2299
2300
2301
2302
2303
2304
2305
2306
2307
2308
2309
2310
2311
2312
2313


2314
2315
2316
2317
2318
2319
2320
2321
2322
2323
2324
2325
2326
2327
2328
2329
2330
2331
2332
2333
2334
2335
2336








2337




2338








2339
2340
2341
2342


2343











2344















2345
2346
2347
2348


2349

2350
2351





2352
2353
2354
2355
2356
2357
2358
2359
2360
2361
2362
2363
2364
2365
2366















2367





2368
2369
2370



2371
2372
2373
2374
2375
2376
2377
2378
2379
2380
2381
2382


2383







2384
2385
2386
2387
2388
2389
2390
2391
2392
2393
2394
2395
2396
2397
2398
2399
2400
2401
2402







2403
2404

2405
2406
2407
2408
2409
2410
2411
2412
2413
2414
2415
2416
2417
2418
2419
2420
2421
2422
2423
2424
2425
2426
2427
2428
2429
2430
2431
2432
2433
2434
2435
2436
2437
2438
2439
2440
2441
2442
2443
2444
2445
2446
2447
2448
2449
2450
2451
2452
2453
2454
2455
2456
2457
2458
2459
2460
2461
2462
2463
2464
2465
2466
2467
2468
2469
....
3239
3240
3241
3242
3243
3244
3245
3246
3247
3248
3249
3250
3251
3252
3253
3254
....
3301
3302
3303
3304
3305
3306
3307
3308
3309
3310
3311
3312
3313
3314
3315
....
3347
3348
3349
3350
3351
3352
3353
3354
3355
3356
3357
3358
3359
3360
3361
3362
....
3380
3381
3382
3383
3384
3385
3386
3387
3388
3389
3390
3391
3392
3393
3394
....
4354
4355
4356
4357
4358
4359
4360
4361
4362
4363
4364
4365
4366
4367
4368
  Mem *pMem = &p->aMem[p->nMem-iCur];

  int nByte;
  VdbeCursor *pCx = 0;
  nByte = 
      ROUND8(sizeof(VdbeCursor)) + 
      (isBtreeCursor?sqlite3BtreeCursorSize():0) + 
      (2*nField+2)*sizeof(u32);

  assert( iCur<p->nCursor );
  if( p->apCsr[iCur] ){
    sqlite3VdbeFreeCursor(p, p->apCsr[iCur]);
    p->apCsr[iCur] = 0;
  }
  if( SQLITE_OK==sqlite3VdbeMemGrow(pMem, nByte, 0) ){
    p->apCsr[iCur] = pCx = (VdbeCursor*)pMem->z;
    memset(pCx, 0, sizeof(VdbeCursor));
    pCx->iDb = iDb;
    pCx->nField = nField;
    if( nField ){
      pCx->aType = (u32 *)&pMem->z[ROUND8(sizeof(VdbeCursor))];
      pCx->aOffset = pCx->aType + nField;
    }
    if( isBtreeCursor ){
      pCx->pCursor = (BtCursor*)
          &pMem->z[ROUND8(sizeof(VdbeCursor))+(2*nField+2)*sizeof(u32)];
      sqlite3BtreeCursorZero(pCx->pCursor);
    }
  }
  return pCx;
}

/*
................................................................................
**
** If the OPFLAG_LENGTHARG and OPFLAG_TYPEOFARG bits are set on P5 when
** the result is guaranteed to only be used as the argument of a length()
** or typeof() function, respectively.  The loading of large blobs can be
** skipped for length() and all content loading can be skipped for typeof().
*/
case OP_Column: {

  i64 payloadSize64; /* Number of bytes in the record */
  int p1;            /* P1 value of the opcode */
  int p2;            /* column number to retrieve */
  VdbeCursor *pC;    /* The VDBE cursor */

  BtCursor *pCrsr;   /* The BTree cursor */
  u32 *aType;        /* aType[i] holds the numeric type of the i-th column */
  u32 *aOffset;      /* aOffset[i] is offset to start of data for i-th column */
  int nField;        /* number of fields in the record */
  int len;           /* The length of the serialized data for the column */
  int i;             /* Loop counter */

  Mem *pDest;        /* Where to write the extracted value */
  Mem sMem;          /* For storing the record being decoded */
  const u8 *zData;   /* Part of the record being decoded */
  const u8 *zHdr;    /* Next unparsed byte of the header */
  const u8 *zEndHdr; /* Pointer to first byte after the header */
  u32 offset;        /* Offset into the data */
  u32 szField;       /* Number of bytes in the content of a field */

  int avail;         /* Number of bytes of available data */
  u32 t;             /* A type code from the record header */
  Mem *pReg;         /* PseudoTable input register */


  p1 = pOp->p1;
  assert( p1<p->nCursor );
  p2 = pOp->p2;
  sMem.zMalloc = 0;


  assert( pOp->p3>0 && pOp->p3<=(p->nMem-p->nCursor) );
  pDest = &aMem[pOp->p3];
  memAboutToChange(p, pDest);














  pC = p->apCsr[p1];
  assert( pC!=0 );
  nField = pC->nField;
  assert( p2<nField );
  aType = pC->aType;
  aOffset = pC->aOffset;
#ifndef SQLITE_OMIT_VIRTUALTABLE
  assert( pC->pVtabCursor==0 );
#endif
  pCrsr = pC->pCursor;
  assert( pCrsr!=0 || pC->pseudoTableReg>0 );


  /* If the cursor cache is stale, bring it up-to-date */
  rc = sqlite3VdbeCursorMoveto(pC);
  if( rc ) goto abort_due_to_error;
  if( pC->cacheStatus!=p->cacheCtr || (pOp->p5&OPFLAG_CLEARCACHE)!=0 ){
    if( pCrsr==0 ){
      assert( pC->pseudoTableReg>0 );
      pReg = &aMem[pC->pseudoTableReg];
      if( pC->multiPseudo ){
        sqlite3VdbeMemShallowCopy(pDest, pReg+p2, MEM_Ephem);
        Deephemeralize(pDest);
        goto op_column_out;
      }
      assert( pReg->flags & MEM_Blob );
      assert( memIsValid(pReg) );
      pC->payloadSize = pC->szRow = avail = pReg->n;
      pC->aRow = (u8*)pReg->z;
    }else if( pC->nullRow ){


      MemSetTypeFlag(pDest, MEM_Null);
      goto op_column_out;
    }else{
      if( pC->isIndex ){
        assert( sqlite3BtreeCursorIsValid(pCrsr) );
        VVA_ONLY(rc =) sqlite3BtreeKeySize(pCrsr, &payloadSize64);
        assert( rc==SQLITE_OK ); /* True because of CursorMoveto() call above */
        /* sqlite3BtreeParseCellPtr() uses getVarint32() to extract the
        ** payload size, so it is impossible for payloadSize64 to be
        ** larger than 32 bits. */
        assert( (payloadSize64 & SQLITE_MAX_U32)==(u64)payloadSize64 );
        pC->aRow = sqlite3BtreeKeyFetch(pCrsr, &avail);
        pC->payloadSize = (u32)payloadSize64;
      }else{
        assert( sqlite3BtreeCursorIsValid(pCrsr) );
        VVA_ONLY(rc =) sqlite3BtreeDataSize(pCrsr, &pC->payloadSize);
        assert( rc==SQLITE_OK );   /* DataSize() cannot fail */
        pC->aRow = sqlite3BtreeDataFetch(pCrsr, &avail);
      }
      assert( avail<=65536 );  /* Maximum page size is 64KiB */
      if( pC->payloadSize <= (u32)avail ){
        pC->szRow = pC->payloadSize;
      }else{








        pC->szRow = avail;




      }








      if( pC->payloadSize > (u32)db->aLimit[SQLITE_LIMIT_LENGTH] ){
        goto too_big;
      }
    }


      











    pC->cacheStatus = p->cacheCtr;















    pC->iHdrOffset = getVarint32(pC->aRow, offset);
    pC->nHdrParsed = 0;
    aOffset[0] = offset;
    if( avail<offset ){


      pC->aRow = 0;

      pC->szRow = 0;
    }






    /* Make sure a corrupt database has not given us an oversize header.
    ** Do this now to avoid an oversize memory allocation.
    **
    ** Type entries can be between 1 and 5 bytes each.  But 4 and 5 byte
    ** types use so much data space that there can only be 4096 and 32 of
    ** them, respectively.  So the maximum header length results from a
    ** 3-byte type for each of the maximum of 32768 columns plus three
    ** extra bytes for the header length itself.  32768*3 + 3 = 98307.
    */
    if( offset > 98307 || offset > pC->payloadSize ){
      rc = SQLITE_CORRUPT_BKPT;
      goto op_column_out;
    }
  }





















  /* Make sure at least the first p2+1 entries of the header have been
  ** parsed and valid information is in aOffset[] and aType[].
  */



  if( pC->nHdrParsed<=p2 && pC->iHdrOffset<aOffset[0] ){
    /* Make sure zData points to enough of the record to cover the header. */
    if( pC->aRow==0 ){
      memset(&sMem, 0, sizeof(sMem));
      rc = sqlite3VdbeMemFromBtree(pCrsr, 0, pC->aOffset[0], pC->isIndex,&sMem);
      if( rc!=SQLITE_OK ){
        goto op_column_out;
      }
      zData = (u8*)sMem.z;
    }else{
      zData = pC->aRow;
    }










    /* Fill in aType[i] and aOffset[i] values through the p2-th field. */
    i = pC->nHdrParsed;
    offset = aOffset[i];
    zHdr = zData + pC->iHdrOffset;
    zEndHdr = zData + pC->aOffset[0];
    for(; i<=p2 && zHdr<zEndHdr; i++){
      if( zHdr[0]<0x80 ){
        t = zHdr[0];
        zHdr++;
      }else{
        zHdr += sqlite3GetVarint32(zHdr, &t);
      }
      aType[i] = t;
      szField = sqlite3VdbeSerialTypeLen(t);
      offset += szField;
      if( offset<szField ){  /* True if offset overflows */
        zHdr = &zEndHdr[1];  /* Forces SQLITE_CORRUPT return below */
        break;
      }







      aOffset[i+1] = offset;
    }

    pC->nHdrParsed = i;
    pC->iHdrOffset = (u32)(zHdr - zData);
    if( pC->aRow==0 ){
      sqlite3VdbeMemRelease(&sMem);
      sMem.flags = MEM_Null;
    }

    /* If we have read more header data than was contained in the header,
    ** or if the end of the last field appears to be past the end of the
    ** record, or if the end of the last field appears to be before the end
    ** of the record (when all fields present), then we must be dealing 
    ** with a corrupt database.
    */
    if( (zHdr > zEndHdr)
     || (offset > pC->payloadSize)
     || (zHdr==zEndHdr && offset!=pC->payloadSize)
    ){
      rc = SQLITE_CORRUPT_BKPT;
      goto op_column_out;
    }
  }

  /* Get the column information. If aOffset[p2] is non-zero, then 
  ** deserialize the value from the record. If aOffset[p2] is zero,
  ** then there are not enough fields in the record to satisfy the
  ** request.  In this case, set the value NULL or to P4 if P4 is
  ** a pointer to a Mem object.
  */
  if( p2<pC->nHdrParsed ){
    assert( rc==SQLITE_OK );
    if( pC->szRow>=aOffset[p2+1] ){
      /* This is the common case where the whole row fits on a single page */
      VdbeMemRelease(pDest);
      sqlite3VdbeSerialGet(pC->aRow+aOffset[p2], aType[p2], pDest);
    }else{
      /* This branch happens only when the row overflows onto multiple pages */
      t = aType[p2];
      if( (pOp->p5 & (OPFLAG_LENGTHARG|OPFLAG_TYPEOFARG))!=0
       && ((t>=12 && (t&1)==0) || (pOp->p5 & OPFLAG_TYPEOFARG)!=0)
      ){
        /* Content is irrelevant for the typeof() function and for
        ** the length(X) function if X is a blob.  So we might as well use
        ** bogus content rather than reading content from disk.  NULL works
        ** for text and blob and whatever is in the payloadSize64 variable
        ** will work for everything else. */
        zData = t<12 ? (u8*)&payloadSize64 : 0;
      }else{
        len = sqlite3VdbeSerialTypeLen(t);
        memset(&sMem, 0, sizeof(sMem));
        sqlite3VdbeMemMove(&sMem, pDest);
        rc = sqlite3VdbeMemFromBtree(pCrsr, aOffset[p2], len,  pC->isIndex,
                                     &sMem);
        if( rc!=SQLITE_OK ){
          goto op_column_out;
        }
        zData = (u8*)sMem.z;
      }
      sqlite3VdbeSerialGet(zData, t, pDest);
    }
    pDest->enc = encoding;
  }else{
    if( pOp->p4type==P4_MEM ){
      sqlite3VdbeMemShallowCopy(pDest, pOp->p4.pMem, MEM_Static);
    }else{
      MemSetTypeFlag(pDest, MEM_Null);
................................................................................
    assert( pKeyInfo->enc==ENC(db) );
    assert( pKeyInfo->db==db );
    nField = pKeyInfo->nField+pKeyInfo->nXField;
  }else if( pOp->p4type==P4_INT32 ){
    nField = pOp->p4.i;
  }
  assert( pOp->p1>=0 );
  assert( nField>=0 );
  testcase( nField==0 );  /* Table with INTEGER PRIMARY KEY and nothing else */
  pCur = allocateCursor(p, pOp->p1, nField, iDb, 1);
  if( pCur==0 ) goto no_mem;
  pCur->nullRow = 1;
  pCur->isOrdered = 1;
  rc = sqlite3BtreeCursor(pX, p2, wrFlag, pKeyInfo, pCur->pCursor);
  pCur->pKeyInfo = pKeyInfo;
  assert( OPFLAG_BULKCSR==BTREE_BULKLOAD );
................................................................................
  static const int vfsFlags = 
      SQLITE_OPEN_READWRITE |
      SQLITE_OPEN_CREATE |
      SQLITE_OPEN_EXCLUSIVE |
      SQLITE_OPEN_DELETEONCLOSE |
      SQLITE_OPEN_TRANSIENT_DB;
  assert( pOp->p1>=0 );
  assert( pOp->p2>=0 );
  pCx = allocateCursor(p, pOp->p1, pOp->p2, -1, 1);
  if( pCx==0 ) goto no_mem;
  pCx->nullRow = 1;
  rc = sqlite3BtreeOpen(db->pVfs, 0, db, &pCx->pBt, 
                        BTREE_OMIT_JOURNAL | BTREE_SINGLE | pOp->p5, vfsFlags);
  if( rc==SQLITE_OK ){
    rc = sqlite3BtreeBeginTrans(pCx->pBt, 1);
................................................................................
** This opcode works like OP_OpenEphemeral except that it opens
** a transient index that is specifically designed to sort large
** tables using an external merge-sort algorithm.
*/
case OP_SorterOpen: {
  VdbeCursor *pCx;

  assert( pOp->p1>=0 );
  assert( pOp->p2>=0 );
  pCx = allocateCursor(p, pOp->p1, pOp->p2, -1, 1);
  if( pCx==0 ) goto no_mem;
  pCx->pKeyInfo = pOp->p4.pKeyInfo;
  assert( pCx->pKeyInfo->db==db );
  assert( pCx->pKeyInfo->enc==ENC(db) );
  pCx->isSorter = 1;
  rc = sqlite3VdbeSorterInit(db, pCx);
................................................................................
** P3 is the number of fields in the records that will be stored by
** the pseudo-table.
*/
case OP_OpenPseudo: {
  VdbeCursor *pCx;

  assert( pOp->p1>=0 );
  assert( pOp->p3>=0 );
  pCx = allocateCursor(p, pOp->p1, pOp->p3, -1, 0);
  if( pCx==0 ) goto no_mem;
  pCx->nullRow = 1;
  pCx->pseudoTableReg = pOp->p2;
  pCx->isTable = 1;
  pCx->isIndex = 0;
  pCx->multiPseudo = pOp->p5;
................................................................................
  VdbeCursor *pC;

  assert( pOp->p1>=0 && pOp->p1<p->nCursor );
  pC = p->apCsr[pOp->p1];
  assert( pC!=0 );
  pC->nullRow = 1;
  pC->rowidIsValid = 0;
  pC->cacheStatus = CACHE_STALE;
  assert( pC->pCursor || pC->pVtabCursor );
  if( pC->pCursor ){
    sqlite3BtreeClearCursor(pC->pCursor);
  }
  break;
}

Changes to src/vdbeInt.h.

89
90
91
92
93
94
95
96




97
98
99
100
101
102
103
104
105
106
  ** CACHE_STALE and so setting cacheStatus=CACHE_STALE guarantees that
  ** the cache is out of date.
  **
  ** aRow might point to (ephemeral) data for the current row, or it might
  ** be NULL.
  */
  u32 cacheStatus;      /* Cache is valid if this matches Vdbe.cacheCtr */
  int payloadSize;      /* Total number of bytes in the record */




  u32 *aType;           /* Type values for all entries in the record */
  u32 *aOffset;         /* Cached offsets to the start of each columns data */
  u8 *aRow;             /* Data for the current row, if all on one page */
};
typedef struct VdbeCursor VdbeCursor;

/*
** When a sub-program is executed (OP_Program), a structure of this type
** is allocated to store the current value of the program counter, as
** well as the current memory cell array and various other frame specific







|
>
>
>
>


|







89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
  ** CACHE_STALE and so setting cacheStatus=CACHE_STALE guarantees that
  ** the cache is out of date.
  **
  ** aRow might point to (ephemeral) data for the current row, or it might
  ** be NULL.
  */
  u32 cacheStatus;      /* Cache is valid if this matches Vdbe.cacheCtr */
  u32 payloadSize;      /* Total number of bytes in the record */
  u16 nHdrParsed;       /* Number of header fields parsed so far */
  u16 nFieldPresent;    /* Number of fields in the record */
  u32 szRow;            /* Byte available in aRow */
  u32 iHdrOffset;       /* Offset to next unparsed byte of the header */
  u32 *aType;           /* Type values for all entries in the record */
  u32 *aOffset;         /* Cached offsets to the start of each columns data */
  const u8 *aRow;       /* Data for the current row, if all on one page */
};
typedef struct VdbeCursor VdbeCursor;

/*
** When a sub-program is executed (OP_Program), a structure of this type
** is allocated to store the current value of the program counter, as
** well as the current memory cell array and various other frame specific

Changes to src/vdbeaux.c.

2654
2655
2656
2657
2658
2659
2660
2661
2662
2663
2664
2665
2666
2667
2668
    if( res!=0 ) return SQLITE_CORRUPT_BKPT;
    p->rowidIsValid = 1;
#ifdef SQLITE_TEST
    sqlite3_search_count++;
#endif
    p->deferredMoveto = 0;
    p->cacheStatus = CACHE_STALE;
  }else if( ALWAYS(p->pCursor) ){
    int hasMoved;
    int rc = sqlite3BtreeCursorHasMoved(p->pCursor, &hasMoved);
    if( rc ) return rc;
    if( hasMoved ){
      p->cacheStatus = CACHE_STALE;
      p->nullRow = 1;
    }







|







2654
2655
2656
2657
2658
2659
2660
2661
2662
2663
2664
2665
2666
2667
2668
    if( res!=0 ) return SQLITE_CORRUPT_BKPT;
    p->rowidIsValid = 1;
#ifdef SQLITE_TEST
    sqlite3_search_count++;
#endif
    p->deferredMoveto = 0;
    p->cacheStatus = CACHE_STALE;
  }else if( p->pCursor ){
    int hasMoved;
    int rc = sqlite3BtreeCursorHasMoved(p->pCursor, &hasMoved);
    if( rc ) return rc;
    if( hasMoved ){
      p->cacheStatus = CACHE_STALE;
      p->nullRow = 1;
    }