SQLite

**      to sqlite3ThreadJoin() is likely to block. Cases that are likely to
**      block provoke debugging output.
**
** In both cases, the effects of the main thread seeing (bDone==0) even
** after the thread has finished are not dire. So we don't worry about
** memory barriers and such here.
*/
typedef int (*SorterCompare)(SortSubtask*,int*,const void*,int,const void*,int);
struct SortSubtask {
  SQLiteThread *pThread;          /* Background thread, if any */
  int bDone;                      /* Set if thread is finished but not joined */
  VdbeSorter *pSorter;            /* Sorter that owns this sub-task */
  UnpackedRecord *pUnpacked;      /* Space to unpack a record */
  SorterList list;                /* List for thread to write to a PMA */
  int nPMA;                       /* Number of PMAs currently in file */
  SorterCompare xCompare;         /* Compare function to use */
  SorterFile file;                /* Temp file for level-0 PMAs */
  SorterFile file2;               /* Space for other PMAs */
};


/*
** Main sorter structure. A single instance of this is allocated for each 
** sorter cursor created by the VDBE.
**
** mxKeysize:
**   As records are added to the sorter by calls to sqlite3VdbeSorterWrite(),

  if( rc==SQLITE_OK ){
    rc = vdbePmaReaderNext(pReadr);
  }
  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.
**
** If IN/OUT parameter *pbKey2Cached is true when this function is called,
** it is assumed that (pTask->pUnpacked) contains the unpacked version
** of key2. If it is false, (pTask->pUnpacked) is populated with the unpacked
** version of key2 and *pbKey2Cached set to true before returning.


**
** If an OOM error is encountered, (pTask->pUnpacked->error_rc) is set
** to SQLITE_NOMEM.
*/
static int vdbeSorterCompare(
  SortSubtask *pTask,             /* Subtask context (for pKeyInfo) */
  int *pbKey2Cached,              /* True if pTask->pUnpacked is pKey2 */
  const void *pKey1, int nKey1,   /* Left side of comparison */
  const void *pKey2, int nKey2    /* Right side of comparison */
){
  UnpackedRecord *r2 = pTask->pUnpacked;
  if( !*pbKey2Cached ){
    sqlite3VdbeRecordUnpack(pTask->pSorter->pKeyInfo, nKey2, pKey2, r2);
    *pbKey2Cached = 1;
  }
  return sqlite3VdbeRecordCompare(nKey1, pKey1, r2);
}

/*
** A specially optimized version of vdbeSorterCompare() that assumes that
** the first field of each key is a TEXT value and that the collation
** sequence to compare them with is BINARY.
*/
static int vdbeSorterCompareText(
  SortSubtask *pTask,             /* Subtask context (for pKeyInfo) */
  int *pbKey2Cached,              /* True if pTask->pUnpacked is pKey2 */
  const void *pKey1, int nKey1,   /* Left side of comparison */
  const void *pKey2, int nKey2    /* Right side of comparison */
){
  const u8 * const p1 = (const u8 * const)pKey1;
  const u8 * const p2 = (const u8 * const)pKey2;
  const u8 * const v1 = &p1[ p1[0] ];   /* Pointer to value 1 */
  const u8 * const v2 = &p2[ p2[0] ];   /* Pointer to value 2 */

  int n1;
  int n2;
  int res;

  getVarint32(&p1[1], n1); n1 = (n1 - 13) / 2;
  getVarint32(&p2[1], n2); n2 = (n2 - 13) / 2;
  res = memcmp(v1, v2, MIN(n1, n2));
  if( res==0 ){
    res = n1 - n2;
  }

  if( res==0 ){
    if( pTask->pSorter->pKeyInfo->nField>1 ){
      res = vdbeSorterCompare(pTask, pbKey2Cached, pKey1, nKey1, pKey2, nKey2);
    }
  }else{
    if( pTask->pSorter->pKeyInfo->aSortOrder[0] ){
      res = res * -1;
    }
  }

  return res;
}

/*
** A specially optimized version of vdbeSorterCompare() that assumes that
** the first field of each key is an INTEGER value.
*/
static int vdbeSorterCompareInt(
  SortSubtask *pTask,             /* Subtask context (for pKeyInfo) */
  int *pbKey2Cached,              /* True if pTask->pUnpacked is pKey2 */
  const void *pKey1, int nKey1,   /* Left side of comparison */
  const void *pKey2, int nKey2    /* Right side of comparison */
){
  const u8 * const p1 = (const u8 * const)pKey1;
  const u8 * const p2 = (const u8 * const)pKey2;
  const int s1 = p1[1];                 /* Left hand serial type */
  const int s2 = p2[1];                 /* Right hand serial type */
  const u8 * const v1 = &p1[ p1[0] ];   /* Pointer to value 1 */
  const u8 * const v2 = &p2[ p2[0] ];   /* Pointer to value 2 */
  int res;                              /* Return value */

  assert( (s1>0 && s1<7) || s1==8 || s1==9 );
  assert( (s2>0 && s2<7) || s2==8 || s2==9 );

  if( s1>7 && s2>7 ){
    res = s1 - s2;
  }else{
    if( s1==s2 ){
      if( (*v1 ^ *v2) & 0x80 ){
        /* The two values have different signs */
        res = (*v1 & 0x80) ? -1 : +1;
      }else{
        /* The two values have the same sign. Compare using memcmp(). */
        static const u8 aLen[] = {0, 1, 2, 3, 4, 6, 8 };
        int i;
        res = 0;
        for(i=0; i<aLen[s1]; i++){
          if( (res = v1[i] - v2[i]) ) break;
        }
      }
    }else{
      if( s2>7 ){
        res = +1;
      }else if( s1>7 ){
        res = -1;
      }else{
        res = s1 - s2;
      }

      if( res>0 ){
        if( *v1 & 0x80 ) res = -1;
      }else if( res<0 ){
        if( *v2 & 0x80 ) res = +1;
      }
    }
  }

  if( res==0 ){
    if( pTask->pSorter->pKeyInfo->nField>1 ){
      res = vdbeSorterCompare(pTask, pbKey2Cached, pKey1, nKey1, pKey2, nKey2);
    }
  }else if( pTask->pSorter->pKeyInfo->aSortOrder[0] ){
    res = res * -1;
  }

  return res;
}

/*
** Initialize the temporary index cursor just opened as a sorter cursor.
**
** Usually, the sorter module uses the value of (pCsr->pKeyInfo->nField)
** to determine the number of fields that should be compared from the

        assert( pSorter->iMemory==0 );
        pSorter->nMemory = pgsz;
        pSorter->list.aMemory = (u8*)sqlite3Malloc(pgsz);
        if( !pSorter->list.aMemory ) rc = SQLITE_NOMEM;
      }
    }

    if( (pKeyInfo->nField+pKeyInfo->nXField)<13 
     && (pKeyInfo->aColl[0]==0 || pKeyInfo->aColl[0]==db->pDfltColl)
    ){
      pSorter->typeMask = SORTER_TYPE_INTEGER | SORTER_TYPE_TEXT;
    }
  }

  return rc;
}
#undef nWorker   /* Defined at the top of this function */

  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;
  int bCached = 0;

  while( p1 && p2 ){
    int res;
    res = pTask->xCompare(
        pTask, &bCached, SRVAL(p1), p1->nVal, SRVAL(p2), p2->nVal
    );

    if( res<=0 ){
      *pp = p1;
      pp = &p1->u.pNext;
      p1 = p1->u.pNext;

    }else{
      *pp = p2;
      pp = &p2->u.pNext;
      p2 = p2->u.pNext;

      bCached = 0;
    }
  }
  *pp = p1 ? p1 : p2;
  *ppOut = pFinal;
}

/*
** Return the SorterCompare function to compare values collected by the
** sorter object passed as the only argument.
*/
static SorterCompare vdbeSorterGetCompare(VdbeSorter *p){
  if( p->typeMask==SORTER_TYPE_INTEGER ){
    return vdbeSorterCompareInt;
  }else if( p->typeMask==SORTER_TYPE_TEXT ){
    return vdbeSorterCompareText; 
  }
  return vdbeSorterCompare;
}

/*
** 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, SorterList *pList){
  int i;
  SorterRecord **aSlot;
  SorterRecord *p;
  int rc;

  rc = vdbeSortAllocUnpacked(pTask);
  if( rc!=SQLITE_OK ) return rc;

  p = pList->pList;

  pTask->xCompare = vdbeSorterGetCompare(pTask->pSorter);






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

  while( p ){

  rc = vdbePmaReaderNext(&pMerger->aReadr[iPrev]);

  /* Update contents of aTree[] */
  if( rc==SQLITE_OK ){
    int i;                      /* Index of aTree[] to recalculate */
    PmaReader *pReadr1;         /* First PmaReader to compare */
    PmaReader *pReadr2;         /* Second PmaReader to compare */
    int bCached = 0;

    /* Find the first two PmaReaders to compare. The one that was just
    ** advanced (iPrev) and the one next to it in the array.  */
    pReadr1 = &pMerger->aReadr[(iPrev & 0xFFFE)];
    pReadr2 = &pMerger->aReadr[(iPrev | 0x0001)];


    for(i=(pMerger->nTree+iPrev)/2; i>0; i=i/2){
      /* Compare pReadr1 and pReadr2. Store the result in variable iRes. */
      int iRes;
      if( pReadr1->pFd==0 ){
        iRes = +1;
      }else if( pReadr2->pFd==0 ){
        iRes = -1;
      }else{
        iRes = pTask->xCompare(pTask, &bCached,
            pReadr1->aKey, pReadr1->nKey, pReadr2->aKey, pReadr2->nKey
        );
      }

      /* If pReadr1 contained the smaller value, set aTree[i] to its index.
      ** Then set pReadr2 to the next PmaReader to compare to pReadr1. In this
      ** case there is no cache of pReadr2 in pTask->pUnpacked, so set
      ** pKey2 to point to the record belonging to pReadr2.

      ** If the two values were equal, then the value from the oldest
      ** PMA should be considered smaller. The VdbeSorter.aReadr[] array
      ** is sorted from oldest to newest, so pReadr1 contains older values
      ** than pReadr2 iff (pReadr1<pReadr2).  */
      if( iRes<0 || (iRes==0 && pReadr1<pReadr2) ){
        pMerger->aTree[i] = (int)(pReadr1 - pMerger->aReadr);
        pReadr2 = &pMerger->aReadr[ pMerger->aTree[i ^ 0x0001] ];
        bCached = 0;
      }else{
        bCached = (pReadr1->pFd!=0);
        pMerger->aTree[i] = (int)(pReadr2 - pMerger->aReadr);
        pReadr1 = &pMerger->aReadr[ pMerger->aTree[i ^ 0x0001] ];
      }
    }
    *pbEof = (pMerger->aReadr[pMerger->aTree[1]].pFd==0);
  }

  int nReq;                       /* Bytes of memory required */
  int nPMA;                       /* Bytes of PMA space required */
  int t;                          /* serial type of first record field */

  getVarint32((const u8*)&pVal->z[1], t);
  if( t>0 && t<10 && t!=7 ){
    pSorter->typeMask &= SORTER_TYPE_INTEGER;
  }else if( t>10 && (t & 0x01) ){
    pSorter->typeMask &= SORTER_TYPE_TEXT;
  }else{
    pSorter->typeMask = 0;
  }

  assert( pSorter );

  p2 = &pMerger->aReadr[i2];

  if( p1->pFd==0 ){
    iRes = i2;
  }else if( p2->pFd==0 ){
    iRes = i1;
  }else{
    SortSubtask *pTask = pMerger->pTask;
    int bCached = 0;
    int res;
    assert( pTask->pUnpacked!=0 );  /* from vdbeSortSubtaskMain() */
    res = pTask->xCompare(
        pTask, &bCached, p1->aKey, p1->nKey, p2->aKey, p2->nKey
    );
    if( res<=0 ){
      iRes = i1;
    }else{
      iRes = i2;
    }
  }

*/
static int vdbeSorterSetupMerge(VdbeSorter *pSorter){
  int rc;                         /* Return code */
  SortSubtask *pTask0 = &pSorter->aTask[0];
  MergeEngine *pMain = 0;
#if SQLITE_MAX_WORKER_THREADS
  sqlite3 *db = pTask0->pSorter->db;



  int i;
  SorterCompare xCompare = vdbeSorterGetCompare(pSorter);
  for(i=0; i<pSorter->nTask; i++){
    pSorter->aTask[i].xCompare = xCompare;
  }
#endif

  rc = vdbeSorterMergeTreeBuild(pSorter, &pMain);
  if( rc==SQLITE_OK ){