SQLite4  Check-in [a89abc2117]

  if( rc!=LSM_OK ){
    testPrintError("lsm_open(): rc=%d\n", rc);
  }else{
    rc = lsm_open(pDb, zDb);
    if( rc!=LSM_OK ){
      testPrintError("lsm_open(): rc=%d\n", rc);
    }else{
      int n = -1;
      lsm_config(pDb, LSM_CONFIG_BLOCK_SIZE, &n);
      n = n*2;
      lsm_config(pDb, LSM_CONFIG_AUTOCHECKPOINT, &n);

      rc = lsm_work(pDb, nMerge, nWork, 0);
      if( rc!=LSM_OK ){
        testPrintError("lsm_work(): rc=%d\n", rc);
      }
    }
  }
  if( rc==LSM_OK ){

  Page *p;
  int iHash;
  int rc = LSM_OK;

  /* In most cases iReal is the same as iPg. Except, if pRedirect is not
  ** NULL, and the block containing iPg has been redirected, then iReal
  ** is the page number after redirection.  */
  Pgno iReal = lsmFsRedirectPage(pFS, pRedirect, iPg);


















  assert( iPg>=fsFirstPageOnBlock(pFS, 1) );
  assert( iReal>=fsFirstPageOnBlock(pFS, 1) );
  *ppPg = 0;

  assert( pFS->bUseMmap==0 || pFS->pCompress==0 );
  if( pFS->bUseMmap ){

    if( fsPageToBlock(pFS, iPg)==iBlk && (iRet==0 || iPg<iRet) ){
      iRet = iPg;
    }
  }
  return iRet;
}

#ifndef NDEBUG
static int fsPageRedirects(FileSystem *pFS, Segment *p, Pgno iPg){
  return (iPg!=0 && iPg!=lsmFsRedirectPage(pFS, p->pRedirect, iPg));
}
static int fsSegmentRedirects(FileSystem *pFS, Segment *p){
  return (
      fsPageRedirects(pFS, p, p->iFirst)
   || fsPageRedirects(pFS, p, p->iRoot)
   || fsPageRedirects(pFS, p, p->iLastPg)
  );
}
#endif

/*
** Argument aPgno is an array of nPgno page numbers. All pages belong to
** the segment pRun. This function gobbles from the start of the run to the
** first page that appears in aPgno[] (i.e. so that the aPgno[] entry is
** the new first page of the run).
*/
void lsmFsGobble(
  lsm_db *pDb,
  Segment *pRun, 
  Pgno *aPgno,
  int nPgno
){
  int rc = LSM_OK;
  FileSystem *pFS = pDb->pFS;
  Snapshot *pSnapshot = pDb->pWorker;
  int iBlk;

  assert( pRun->nSize>0 );
  assert( 0==fsSegmentRedirects(pFS, pRun) );
  assert( nPgno>0 && 0==fsPageRedirects(pFS, pRun, aPgno[0]) );

  iBlk = fsPageToBlock(pFS, pRun->iFirst);
  pRun->nSize += (pRun->iFirst - fsFirstPageOnBlock(pFS, iBlk));

  while( rc==LSM_OK ){
    int iNext = 0;
    Pgno iFirst = firstOnBlock(pFS, iBlk, aPgno, nPgno);
    if( iFirst ){

** caller using lsmFsPageRelease().
*/
int lsmFsDbPageNext(Segment *pRun, Page *pPg, int eDir, Page **ppNext){
  int rc = LSM_OK;
  FileSystem *pFS = pPg->pFS;
  Pgno iPg = pPg->iPg;

  assert( 0==fsSegmentRedirects(pFS, pRun) );
  if( pFS->pCompress ){
    int nSpace = pPg->nCompress + 2*3;

    /* TODO: Fix redirects */
    assert( 0 );

    do {

        assert( pPg->flags & PAGE_HASPREV );
        iPg = fsLastPageOnBlock(pFS, lsmGetU32(&pPg->aData[-4]));
      }else{
        iPg--;
      }
    }else{
      if( pRun ){

        if( iPg==pRun->iLastPg ){
          *ppNext = 0;
          return LSM_OK;
        }
      }

      if( fsIsLast(pFS, iPg) ){
        int iBlk = fsRedirectBlock(

/*
** Return true if page is currently writable.
*/
int lsmFsPageWritable(Page *pPg){
  return (pPg->flags & PAGE_DIRTY) ? 1 : 0;
}

static void fsMovePage(
  FileSystem *pFS, 
  Segment *pSeg, 
  int iTo, 
  int iFrom, 
  Pgno *piPg
){
  Pgno iPg = *piPg;
  if( iFrom==fsPageToBlock(pFS, iPg) ){
    const int nPagePerBlock = (pFS->nBlocksize / pFS->nPagesize);
    *piPg = iPg - (Pgno)(iFrom - iTo) * nPagePerBlock;
  }
}

/*
** Copy the contents of block iFrom to block iTo. 
**
** It is safe to assume that there are no outstanding references to pages 
** on block iTo. And that block iFrom is not currently being written. In
** other words, the data can be read and written directly.
*/
int lsmFsMoveBlock(FileSystem *pFS, Segment *pSeg, int iTo, int iFrom){
  Snapshot *p = pFS->pDb->pWorker;
  int rc = LSM_OK;

  i64 iFromOff = (i64)(iFrom-1) * pFS->nBlocksize;
  i64 iToOff = (i64)(iTo-1) * pFS->nBlocksize;
  
  assert( iTo!=1 );
  assert( iFrom>iTo );

  if( pFS->bUseMmap ){
    fsGrowMapping(pFS, (i64)iFrom * pFS->nBlocksize, &rc);
    if( rc==LSM_OK ){
      u8 *aMap = (u8 *)(pFS->pMap);


      memcpy(&aMap[iToOff], &aMap[iFromOff], pFS->nBlocksize);
    }
  }else{
    int nSz = pFS->nPagesize;
    u8 *aData = (u8 *)lsmMallocRc(pFS->pEnv, nSz, &rc);
    if( rc==LSM_OK ){
      const int nPagePerBlock = (pFS->nBlocksize / pFS->nPagesize);
      int i;
      for(i=0; rc==LSM_OK && i<nPagePerBlock; i++){
        i64 iOff = iFromOff + i*nSz;
        rc = lsmEnvRead(pFS->pEnv, pFS->fdDb, iOff, aData, nSz);
        if( rc==LSM_OK ){
          iOff = iToOff + i*nSz;
          rc = lsmEnvWrite(pFS->pEnv, pFS->fdDb, iOff, aData, nSz);
        }
      }
    }
  }

  /* Update append-point list if necessary */
  if( rc==LSM_OK ){
    int i;
    for(i=0; i<LSM_APPLIST_SZ; i++){
      if( fsPageToBlock(pFS, p->aiAppend[i])==iFrom ){
        const int nPagePerBlock = (pFS->nBlocksize / pFS->nPagesize);
        p->aiAppend[i] -= (i64)(iFrom-iTo) * nPagePerBlock;
      }
    }
  }

  /* Update the Segment structure itself */
  fsMovePage(pFS, pSeg, iTo, iFrom, &pSeg->iFirst);
  fsMovePage(pFS, pSeg, iTo, iFrom, &pSeg->iLastPg);
  fsMovePage(pFS, pSeg, iTo, iFrom, &pSeg->iRoot);

  return rc;
}

/*
** Append raw data to a segment. This function is only used in compressed
** database mode.
*/

  int nUsed,
  u8 *aUsed
){
  if( pSeg ){
    if( pSeg && pSeg->nSize>0 ){
      Redirect *pRedir = pSeg->pRedirect;
      int rc;


      int iBlk;                   /* Current block (during iteration) */
      int iLastBlk;               /* Last real block of segment */
      int bLastIsLastOnBlock;     /* True iLast is the last on its block */

      assert( 0==fsSegmentRedirects(pFS, pSeg) );
      iBlk = fsPageToBlock(pFS, pSeg->iFirst);
      iLastBlk = fsPageToBlock(pFS, pSeg->iLastPg);



      bLastIsLastOnBlock = (fsLastPageOnBlock(pFS, iLastBlk)==pSeg->iLastPg);
      assert( iBlk>0 );

      /* If the first page of this run is also the first page of its first
      ** block, set the flag to indicate that the first page of iBlk is 
      ** in use.  */
      if( fsFirstPageOnBlock(pFS, iBlk)==pSeg->iFirst ){
        assert( (aUsed[iBlk-1] & INTEGRITY_CHECK_FIRST_PG)==0 );
        aUsed[iBlk-1] |= INTEGRITY_CHECK_FIRST_PG;
      }

      do {
        /* iBlk is a part of this sorted run. */
        aUsed[iBlk-1] |= INTEGRITY_CHECK_USED;

  for(i=0; i<nRhs; i++){
    pCsr->aPtr[i].pSeg = &pLvl->aRhs[i];
    pCsr->aPtr[i].pLevel = pLvl;
  }

  return LSM_OK;
}

static void multiCursorAddOne(MultiCursor *pCsr, Level *pLvl, int *pRc){
  if( *pRc==LSM_OK ){
    int iPtr = pCsr->nPtr;
    int i;
    pCsr->aPtr[iPtr].pLevel = pLvl;
    pCsr->aPtr[iPtr].pSeg = &pLvl->lhs;
    iPtr++;
    for(i=0; i<pLvl->nRight; i++){
      pCsr->aPtr[iPtr].pLevel = pLvl;
      pCsr->aPtr[iPtr].pSeg = &pLvl->aRhs[i];
      iPtr++;
    }

    if( pLvl->nRight && pLvl->pSplitKey==0 ){
      sortedSplitkey(pCsr->pDb, pLvl, pRc);
    }
    pCsr->nPtr = iPtr;
  }
}

static int multiCursorAddAll(MultiCursor *pCsr, Snapshot *pSnap){
  Level *pLvl;
  int nPtr = 0;
  int iPtr = 0;
  int rc = LSM_OK;

  for(pLvl=pSnap->pLevel; pLvl; pLvl=pLvl->pNext){
    /* If the LEVEL_INCOMPLETE flag is set, then this function is being
    ** called (indirectly) from within a sortedNewToplevel() call to
    ** construct pLvl. In this case ignore pLvl - this cursor is going to
    ** be used to retrieve a freelist entry from the LSM, and the partially
    ** complete level may confuse it.  */
    if( pLvl->flags & LEVEL_INCOMPLETE ) continue;
    nPtr += (1 + pLvl->nRight);
  }

  assert( pCsr->aPtr==0 );
  pCsr->aPtr = lsmMallocZeroRc(pCsr->pDb->pEnv, sizeof(SegmentPtr) * nPtr, &rc);


  for(pLvl=pSnap->pLevel; pLvl; pLvl=pLvl->pNext){

    if( (pLvl->flags & LEVEL_INCOMPLETE)==0 ){










      multiCursorAddOne(pCsr, pLvl, &rc);
    }
  }

  return rc;
}

static int multiCursorInit(MultiCursor *pCsr, Snapshot *pSnap){

  int eTree,                      /* One of the TREE_XXX constants */
  int *pnWrite                    /* OUT: Number of database pages written */
){
  int rc = LSM_OK;                /* Return Code */
  MultiCursor *pCsr = 0;
  Level *pNext = 0;               /* The current top level */
  Level *pNew;                    /* The new level itself */
  Segment *pLinked = 0;           /* Delete separators from this segment */
  Level *pDel = 0;                /* Delete this entire level */
  int nWrite = 0;                 /* Number of database pages written */
  Freelist freelist;

  if( eTree!=TREE_NONE ){
    rc = lsmShmCacheChunks(pDb, pDb->treehdr.nChunk);
  }

  pCsr = multiCursorNew(pDb, &rc);
  if( pCsr ){
    pCsr->pDb = pDb;
    rc = multiCursorVisitFreelist(pCsr);
    if( rc==LSM_OK ){
      rc = multiCursorAddTree(pCsr, pDb->pWorker, eTree);
    }
    if( rc==LSM_OK && pNext && pNext->pMerge==0 ){

      if( (pNext->flags & LEVEL_FREELIST_ONLY) ){
        pDel = pNext;
        pCsr->aPtr = lsmMallocZeroRc(pDb->pEnv, sizeof(SegmentPtr), &rc);
        multiCursorAddOne(pCsr, pNext, &rc);
      }else if( eTree!=TREE_NONE && pNext->lhs.iRoot ){
        pLinked = &pNext->lhs;
        rc = btreeCursorNew(pDb, pLinked, &pCsr->pBtCsr);
      }
    }

    /* If this will be the only segment in the database, discard any delete
    ** markers present in the in-memory tree.  */
    if( pNext==0 ){
      multiCursorIgnoreDelete(pCsr);
    }

  }

  if( rc!=LSM_OK || pNew->lhs.iFirst==0 ){
    assert( rc!=LSM_OK || pDb->pWorker->freelist.nEntry==0 );
    lsmDbSnapshotSetLevel(pDb->pWorker, pNext);
    sortedFreeLevel(pDb->pEnv, pNew);
  }else{
    if( pLinked ){
      pLinked->iRoot = 0;
    }else if( pDel ){
      assert( pNew->pNext==pDel );
      pNew->pNext = pDel->pNext;
      lsmFsSortedDelete(pDb->pFS, pDb->pWorker, 1, &pDel->lhs);
      sortedFreeLevel(pDb->pEnv, pDel);
    }

#if 0
    lsmSortedDumpStructure(pDb, pDb->pWorker, 0, 0, "new-toplevel");
#endif

    if( freelist.nEntry ){
      Freelist *p = &pDb->pWorker->freelist;

  ** with the largest number of right-hand segments. Work on it. */
  for(pLevel=pTopLevel; pLevel; pLevel=pLevel->pNext){
    if( pLevel->nRight==0 && pThis && pLevel->iAge==pThis->iAge ){
      nThis++;
    }else{
      if( nThis>nBest ){
        if( (pLevel->iAge!=pThis->iAge+1)
         || (pLevel->nRight==0 && sortedCountLevels(pLevel)<=pDb->nMerge)
        ){
          pBest = pThis;
          nBest = nThis;
        }
      }
      if( pLevel->nRight ){
        if( pLevel->nRight>nBest ){
          nBest = pLevel->nRight;

  /* Find the first free block in the database, ignoring block 1. Block
  ** 1 is tricky as it is smaller than the other blocks.  */
  rc = lsmBlockAllocate(pDb, iFrom, &iTo);
  if( rc!=LSM_OK || iTo==0 ) return rc;
  assert( iTo!=1 && iTo<iFrom );

  rc = lsmFsMoveBlock(pDb->pFS, &pLvl->lhs, iTo, iFrom);
  if( rc==LSM_OK ){
    if( p->redirect.a==0 ){
      int nByte = sizeof(struct RedirectEntry) * LSM_MAX_BLOCK_REDIRECTS;
      p->redirect.a = lsmMallocZeroRc(pDb->pEnv, nByte, &rc);
    }
    if( rc==LSM_OK ){
      memmove(&p->redirect.a[1], &p->redirect.a[0], 

){
  Snapshot *pWorker;              /* Worker snapshot */
  int rc = LSM_OK;                /* Return code */
  int bDirty = 0;
  int nMax = nPage;               /* Maximum pages to write to disk */
  int nRem = nPage;
  int bCkpt = 0;

  assert( nPage>0 );

  /* Open the worker 'transaction'. It will be closed before this function
  ** returns.  */
  assert( pDb->pWorker==0 );
  rc = lsmBeginWork(pDb);
  if( rc!=LSM_OK ) return rc;
  pWorker = pDb->pWorker;

  if( rc==LSM_OK && bDirty ){
    lsmFinishWork(pDb, 0, &rc);
  }else{
    int rcdummy = LSM_BUSY;
    lsmFinishWork(pDb, 0, &rcdummy);
    *pnWrite = 0;

  }
  assert( pDb->pWorker==0 );
  return rc;
}

static int doLsmWork(lsm_db *pDb, int nMerge, int nPage, int *pnWrite){
  int rc;