SQLite

typedef unsigned short int u16;
typedef unsigned int u32;
typedef lsm_i64 i64;
typedef unsigned long long int u64;
#endif

/* A page number is a 64-bit integer. */
typedef i64 LsmPgno;

#ifdef LSM_DEBUG
int lsmErrorBkpt(int);
#else
# define lsmErrorBkpt(x) (x)
#endif

  void **apShm;                   /* Shared memory chunks */
  ShmHeader *pShmhdr;             /* Live shared-memory header */
  TreeHeader treehdr;             /* Local copy of tree-header */
  u32 aSnapshot[LSM_META_PAGE_SIZE / sizeof(u32)];
};

struct Segment {
  LsmPgno iFirst;                  /* First page of this run */
  LsmPgno iLastPg;                 /* Last page of this run */
  LsmPgno iRoot;                   /* Root page number (if any) */
  int nSize;                       /* Size of this run in pages */

  Redirect *pRedirect;             /* Block redirects (or NULL) */
};

/*
** iSplitTopic/pSplitKey/nSplitKey:

** access to the associated Level struct.
**
** iOutputOff:
**   The byte offset to write to next within the last page of the 
**   output segment.
*/
struct MergeInput {
  LsmPgno iPg;                    /* Page on which next input is stored */
  int iCell;                      /* Cell containing next input to merge */
};
struct Merge {
  int nInput;                     /* Number of input runs being merged */
  MergeInput *aInput;             /* Array nInput entries in size */
  MergeInput splitkey;            /* Location in file of current splitkey */
  int nSkip;                      /* Number of separators entries to skip */
  int iOutputOff;                 /* Write offset on output page */
  LsmPgno iCurrentPtr;            /* Current pointer value */
};

/* 
** The first argument to this macro is a pointer to a Segment structure.
** Returns true if the structure instance indicates that the separators
** array is valid.
*/

  u32 iCmpId;                     /* Id of compression scheme */
  Level *pLevel;                  /* Pointer to level 0 of snapshot (or NULL) */
  i64 iId;                        /* Snapshot id */
  i64 iLogOff;                    /* Log file offset */
  Redirect redirect;              /* Block redirection array */

  /* Used by worker snapshots only */
  int nBlock;                        /* Number of blocks in database file */
  LsmPgno aiAppend[LSM_APPLIST_SZ];  /* Append point list */
  Freelist freelist;                 /* Free block list */
  u32 nWrite;                        /* Total number of pages written to disk */
};
#define LSM_INITIAL_SNAPSHOT_ID 11

/*
** Functions from file "lsm_ckpt.c".
*/
int lsmCheckpointWrite(lsm_db *, u32 *);

int lsmFsPageSize(FileSystem *);
void lsmFsSetPageSize(FileSystem *, int);

int lsmFsFileid(lsm_db *pDb, void **ppId, int *pnId);

/* Creating, populating, gobbling and deleting sorted runs. */
void lsmFsGobble(lsm_db *, Segment *, LsmPgno *, int);
int lsmFsSortedDelete(FileSystem *, Snapshot *, int, Segment *);
int lsmFsSortedFinish(FileSystem *, Segment *);
int lsmFsSortedAppend(FileSystem *, Snapshot *, Level *, int, Page **);
int lsmFsSortedPadding(FileSystem *, Snapshot *, Segment *);

/* Functions to retrieve the lsm_env pointer from a FileSystem or Page object */
lsm_env *lsmFsEnv(FileSystem *);
lsm_env *lsmPageEnv(Page *);
FileSystem *lsmPageFS(Page *);

int lsmFsSectorSize(FileSystem *);

void lsmSortedSplitkey(lsm_db *, Level *, int *);

/* Reading sorted run content. */
int lsmFsDbPageLast(FileSystem *pFS, Segment *pSeg, Page **ppPg);
int lsmFsDbPageGet(FileSystem *, Segment *, LsmPgno, Page **);
int lsmFsDbPageNext(Segment *, Page *, int eDir, Page **);

u8 *lsmFsPageData(Page *, int *);
int lsmFsPageRelease(Page *);
int lsmFsPagePersist(Page *);
void lsmFsPageRef(Page *);
LsmPgno lsmFsPageNumber(Page *);

int lsmFsNRead(FileSystem *);
int lsmFsNWrite(FileSystem *);

int lsmFsMetaPageGet(FileSystem *, int, int, MetaPage **);
int lsmFsMetaPageRelease(MetaPage *);
u8 *lsmFsMetaPageData(MetaPage *, int *);

#ifdef LSM_DEBUG
int lsmFsDbPageIsLast(Segment *pSeg, Page *pPg);
int lsmFsIntegrityCheck(lsm_db *);
#endif

LsmPgno lsmFsRedirectPage(FileSystem *, Redirect *, LsmPgno);

int lsmFsPageWritable(Page *);

/* Functions to read, write and sync the log file. */
int lsmFsWriteLog(FileSystem *pFS, i64 iOff, LsmString *pStr);
int lsmFsSyncLog(FileSystem *pFS);
int lsmFsReadLog(FileSystem *pFS, i64 iOff, int nRead, LsmString *pStr);
int lsmFsTruncateLog(FileSystem *pFS, i64 nByte);
int lsmFsTruncateDb(FileSystem *pFS, i64 nByte);
int lsmFsCloseAndDeleteLog(FileSystem *pFS);

LsmFile *lsmFsDeferClose(FileSystem *pFS);

/* And to sync the db file */
int lsmFsSyncDb(FileSystem *, int);

void lsmFsFlushWaiting(FileSystem *, int *);

/* Used by lsm_info(ARRAY_STRUCTURE) and lsm_config(MMAP) */
int lsmInfoArrayStructure(lsm_db *pDb, int bBlock, LsmPgno iFirst, char **pz);
int lsmInfoArrayPages(lsm_db *pDb, LsmPgno iFirst, char **pzOut);
int lsmConfigMmap(lsm_db *pDb, int *piParam);

int lsmEnvOpen(lsm_env *, const char *, int, lsm_file **);
int lsmEnvClose(lsm_env *pEnv, lsm_file *pFile);
int lsmEnvLock(lsm_env *pEnv, lsm_file *pFile, int iLock, int eLock);
int lsmEnvTestLock(lsm_env *pEnv, lsm_file *pFile, int iLock, int nLock, int);

int lsmEnvShmMap(lsm_env *, lsm_file *, int, int, void **); 
void lsmEnvShmBarrier(lsm_env *);
void lsmEnvShmUnmap(lsm_env *, lsm_file *, int);

void lsmEnvSleep(lsm_env *, int);

int lsmFsReadSyncedId(lsm_db *db, int, i64 *piVal);

int lsmFsSegmentContainsPg(FileSystem *pFS, Segment *, LsmPgno, int *);

void lsmFsPurgeCache(FileSystem *);

/*
** End of functions from "lsm_file.c".
**************************************************************************/

/* 
** Functions from file "lsm_sorted.c".
*/
int lsmInfoPageDump(lsm_db *, LsmPgno, int, char **);
void lsmSortedCleanup(lsm_db *);
int lsmSortedAutoWork(lsm_db *, int nUnit);

int lsmSortedWalkFreelist(lsm_db *, int, int (*)(void *, int, i64), void *);

int lsmSaveWorker(lsm_db *, int);

static void ckptExportAppendlist(
  lsm_db *db,                     /* Database connection */
  CkptBuffer *p,                  /* Checkpoint buffer to write to */
  int *piOut,                     /* IN/OUT: Offset within checkpoint buffer */
  int *pRc                        /* IN/OUT: Error code */
){
  int i;
  LsmPgno *aiAppend = db->pWorker->aiAppend;

  for(i=0; i<LSM_APPLIST_SZ; i++){
    ckptAppend64(p, piOut, aiAppend[i], pRc);
  }
};

static int ckptExportSnapshot( 

**   The lsmFsSortedAppend() function sets the pSeg pointer to point to the
**   segment that the new page will be a part of. It is unset by
**   lsmFsPagePersist() after the page is written to disk.
*/
struct Page {
  u8 *aData;                      /* Buffer containing page data */
  int nData;                      /* Bytes of usable data at aData[] */
  LsmPgno iPg;                    /* Page number */
  int nRef;                       /* Number of outstanding references */
  int flags;                      /* Combination of PAGE_XXX flags */
  Page *pHashNext;                /* Next page in hash table slot */
  Page *pLruNext;                 /* Next page in LRU list */
  Page *pLruPrev;                 /* Previous page in LRU list */
  FileSystem *pFS;                /* File system that owns this page */

#else
# define IOERR_WRAPPER(rc) (rc)
#endif

#ifdef NDEBUG
# define assert_lists_are_ok(x)
#else
static Page *fsPageFindInHash(FileSystem *pFS, LsmPgno iPg, int *piHash);

static void assert_lists_are_ok(FileSystem *pFS){
#if 0
  Page *p;

  assert( pFS->nMapLimit>=0 );

  return LSM_OK;
}

/*
** Return true if page iReal of the database should be accessed using mmap.
** False otherwise.
*/
static int fsMmapPage(FileSystem *pFS, LsmPgno iReal){
  return ((i64)iReal*pFS->nPagesize <= pFS->nMapLimit);
}

/*
** Given that there are currently nHash slots in the hash table, return 
** the hash key for file iFile, page iPg.
*/
static int fsHashKey(int nHash, LsmPgno iPg){
  return (iPg % nHash);
}

/*
** This is a helper function for lsmFsOpen(). It opens a single file on
** disk (either the database or log file).
*/

** Return the page number of the first page on block iBlock. Blocks are
** numbered starting from 1.
**
** For a compressed database, page numbers are byte offsets. The first
** page on each block is the byte offset immediately following the 4-byte
** "previous block" pointer at the start of each block.
*/
static LsmPgno fsFirstPageOnBlock(FileSystem *pFS, int iBlock){
  LsmPgno iPg;
  if( pFS->pCompress ){
    if( iBlock==1 ){
      iPg = pFS->nMetasize * 2 + 4;
    }else{
      iPg = pFS->nBlocksize * (LsmPgno)(iBlock-1) + 4;
    }
  }else{
    const int nPagePerBlock = (pFS->nBlocksize / pFS->nPagesize);
    if( iBlock==1 ){
      iPg = 1 + ((pFS->nMetasize*2 + pFS->nPagesize - 1) / pFS->nPagesize);
    }else{
      iPg = 1 + (iBlock-1) * nPagePerBlock;
    }
  }
  return iPg;
}

/*
** Return the page number of the last page on block iBlock. Blocks are
** numbered starting from 1.
**
** For a compressed database, page numbers are byte offsets. The first
** page on each block is the byte offset of the byte immediately before 
** the 4-byte "next block" pointer at the end of each block.
*/
static LsmPgno fsLastPageOnBlock(FileSystem *pFS, int iBlock){
  if( pFS->pCompress ){
    return pFS->nBlocksize * (LsmPgno)iBlock - 1 - 4;
  }else{
    const int nPagePerBlock = (pFS->nBlocksize / pFS->nPagesize);
    return iBlock * nPagePerBlock;
  }
}

/*
** Return the block number of the block that page iPg is located on. 
** Blocks are numbered starting from 1.
*/
static int fsPageToBlock(FileSystem *pFS, LsmPgno iPg){
  if( pFS->pCompress ){
    return (int)((iPg / pFS->nBlocksize) + 1);
  }else{
    return (int)(1 + ((iPg-1) / (pFS->nBlocksize / pFS->nPagesize)));
  }
}

/*
** Return true if page iPg is the last page on its block.
**
** This function is only called in non-compressed database mode.
*/
static int fsIsLast(FileSystem *pFS, LsmPgno iPg){
  const int nPagePerBlock = (pFS->nBlocksize / pFS->nPagesize);
  assert( !pFS->pCompress );
  return ( iPg && (iPg % nPagePerBlock)==0 );
}

/*
** Return true if page iPg is the first page on its block.
**
** This function is only called in non-compressed database mode.
*/
static int fsIsFirst(FileSystem *pFS, LsmPgno iPg){
  const int nPagePerBlock = (pFS->nBlocksize / pFS->nPagesize);
  assert( !pFS->pCompress );
  return ( (iPg % nPagePerBlock)==1
        || (iPg<nPagePerBlock && iPg==fsFirstPageOnBlock(pFS, 1))
  );
}

  }
  return pPage->aData;
}

/*
** Return the page number of a page.
*/
LsmPgno lsmFsPageNumber(Page *pPage){
  /* assert( (pPage->flags & PAGE_DIRTY)==0 ); */
  return pPage ? pPage->iPg : 0;
}

/*
** Page pPg is currently part of the LRU list belonging to pFS. Remove
** it from the list. pPg->pLruNext and pPg->pLruPrev are cleared by this

/*
** Search the hash-table for page iPg. If an entry is round, return a pointer
** to it. Otherwise, return NULL.
**
** Either way, if argument piHash is not NULL set *piHash to the hash slot
** number that page iPg would be stored in before returning.
*/
static Page *fsPageFindInHash(FileSystem *pFS, LsmPgno iPg, int *piHash){
  Page *p;                        /* Return value */
  int iHash = fsHashKey(pFS->nHash, iPg);

  if( piHash ) *piHash = iHash;
  for(p=pFS->apHash[iHash]; p; p=p->pHashNext){
    if( p->iPg==iPg) break;
  }

}

/*
** If page iPg has been redirected according to the redirections in the
** object passed as the second argument, return the destination page to
** which it is redirected. Otherwise, return a copy of iPg.
*/
LsmPgno lsmFsRedirectPage(FileSystem *pFS, Redirect *pRedir, LsmPgno iPg){
  LsmPgno iReal = iPg;

  if( pRedir ){
    const int nPagePerBlock = (
        pFS->pCompress ? pFS->nBlocksize : (pFS->nBlocksize / pFS->nPagesize)
    );
    int iBlk = fsPageToBlock(pFS, iPg);
    int i;
    for(i=0; i<pRedir->n; i++){
      int iFrom = pRedir->a[i].iFrom;
      if( iFrom>iBlk ) break;
      if( iFrom==iBlk ){
        int iTo = pRedir->a[i].iTo;
        iReal = iPg - (LsmPgno)(iFrom - iTo) * nPagePerBlock;
        if( iTo==1 ){
          iReal += (fsFirstPageOnBlock(pFS, 1)-1);
        }
        break;
      }
    }
  }

  assert( iReal!=0 );
  return iReal;
}

/* Required by the circular fsBlockNext<->fsPageGet dependency. */
static int fsPageGet(FileSystem *, Segment *, LsmPgno, int, Page **, int *);

/*
** Parameter iBlock is a database file block. This function reads the value 
** stored in the blocks "next block" pointer and stores it in *piNext.
** LSM_OK is returned if everything is successful, or an LSM error code
** otherwise.
*/

  }
  return rc;
}

/*
** Return the page number of the last page on the same block as page iPg.
*/
LsmPgno fsLastPageOnPagesBlock(FileSystem *pFS, LsmPgno iPg){
  return fsLastPageOnBlock(pFS, fsPageToBlock(pFS, iPg));
}

/*
** Read nData bytes of data from offset iOff of the database file into
** buffer aData. If this means reading past the end of a block, follow
** the block pointer to the next block and continue reading.

** to the total number of free bytes before returning.
**
** If no error occurs, LSM_OK is returned. Otherwise, an lsm error code.
*/
static int fsPageGet(
  FileSystem *pFS,                /* File-system handle */
  Segment *pSeg,                  /* Block redirection to use (or NULL) */
  LsmPgno iPg,                    /* Page id */
  int noContent,                  /* True to not load content from disk */
  Page **ppPg,                    /* OUT: New page handle */
  int *pnSpace                    /* OUT: Bytes of free space */
){
  Page *p;
  int iHash;
  int rc = LSM_OK;

  /* In most cases iReal is the same as iPg. Except, if pSeg->pRedirect is 
  ** not NULL, and the block containing iPg has been redirected, then iReal
  ** is the page number after redirection.  */
  LsmPgno iReal = lsmFsRedirectPage(pFS, (pSeg ? pSeg->pRedirect : 0), iPg);

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

  /* Search the hash-table for the page */

/*
** Return true if the first or last page of segment pRun falls between iFirst
** and iLast, inclusive, and pRun is not equal to pIgnore.
*/
static int fsRunEndsBetween(
  Segment *pRun, 
  Segment *pIgnore, 
  LsmPgno iFirst, 
  LsmPgno iLast
){
  return (pRun!=pIgnore && (
        (pRun->iFirst>=iFirst && pRun->iFirst<=iLast)
     || (pRun->iLastPg>=iFirst && pRun->iLastPg<=iLast)
  ));
}

/*
** Return true if level pLevel contains a segment other than pIgnore for
** which the first or last page is between iFirst and iLast, inclusive.
*/
static int fsLevelEndsBetween(
  Level *pLevel, 
  Segment *pIgnore, 
  LsmPgno iFirst, 
  LsmPgno iLast
){
  int i;

  if( fsRunEndsBetween(&pLevel->lhs, pIgnore, iFirst, iLast) ){
    return 1;
  }
  for(i=0; i<pLevel->nRight; i++){

static int fsFreeBlock(
  FileSystem *pFS,                /* File system object */
  Snapshot *pSnapshot,            /* Worker snapshot */
  Segment *pIgnore,               /* Ignore this run when searching */
  int iBlk                        /* Block number of block to free */
){
  int rc = LSM_OK;                /* Return code */
  LsmPgno iFirst;                 /* First page on block iBlk */
  LsmPgno iLast;                  /* Last page on block iBlk */
  Level *pLevel;                  /* Used to iterate through levels */

  int iIn;                        /* Used to iterate through append points */
  int iOut = 0;                   /* Used to output append points */
  LsmPgno *aApp = pSnapshot->aiAppend;

  iFirst = fsFirstPageOnBlock(pFS, iBlk);
  iLast = fsLastPageOnBlock(pFS, iBlk);

  /* Check if any other run in the snapshot has a start or end page 
  ** within this block. If there is such a run, return early. */
  for(pLevel=lsmDbSnapshotLevel(pSnapshot); pLevel; pLevel=pLevel->pNext){

}

/*
** aPgno is an array containing nPgno page numbers. Return the smallest page
** number from the array that falls on block iBlk. Or, if none of the pages
** in aPgno[] fall on block iBlk, return 0.
*/
static LsmPgno firstOnBlock(
  FileSystem *pFS, 
  int iBlk, 
  LsmPgno *aPgno, 
  int nPgno
){
  LsmPgno iRet = 0;
  int i;
  for(i=0; i<nPgno; i++){
    LsmPgno iPg = aPgno[i];
    if( fsPageToBlock(pFS, iPg)==iBlk && (iRet==0 || iPg<iRet) ){
      iRet = iPg;
    }
  }
  return iRet;
}

#ifndef NDEBUG
/*
** Return true if page iPg, which is a part of segment p, lies on
** a redirected block. 
*/
static int fsPageRedirects(FileSystem *pFS, Segment *p, LsmPgno iPg){
  return (iPg!=0 && iPg!=lsmFsRedirectPage(pFS, p->pRedirect, iPg));
}

/*
** Return true if the second argument is not NULL and any of the first
** last or root pages lie on a redirected block. 
*/

** 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, 
  LsmPgno *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 += (int)(pRun->iFirst - fsFirstPageOnBlock(pFS, iBlk));

  while( rc==LSM_OK ){
    int iNext = 0;
    LsmPgno iFirst = firstOnBlock(pFS, iBlk, aPgno, nPgno);
    if( iFirst ){
      pRun->iFirst = iFirst;
      break;
    }
    rc = fsBlockNext(pFS, pRun, iBlk, &iNext);
    if( rc==LSM_OK ) rc = fsFreeBlock(pFS, pSnapshot, pRun, iBlk);
    pRun->nSize -= (int)(

**   *piNext = iPg + nByte;
**
** But take block overflow and redirection into account.
*/
static int fsNextPageOffset(
  FileSystem *pFS,                /* File system object */
  Segment *pSeg,                  /* Segment to move within */
  LsmPgno iPg,                    /* Offset of current page */
  int nByte,                      /* Size of current page including headers */
  LsmPgno *piNext                 /* OUT: Offset of next page. Or zero (EOF) */
){
  LsmPgno iNext;
  int rc;

  assert( pFS->pCompress );

  rc = fsAddOffset(pFS, pSeg, iPg, nByte-1, &iNext);
  if( pSeg && iNext==pSeg->iLastPg ){
    iNext = 0;

** LSM_OK is returned if no error occurs. Otherwise, an lsm error code.
** If any value other than LSM_OK is returned, then the final value of
** *piPrev is undefined.
*/
static int fsGetPageBefore(
  FileSystem *pFS, 
  Segment *pSeg, 
  LsmPgno iPg, 
  LsmPgno *piPrev
){
  u8 aSz[3];
  int rc;
  i64 iRead;

  assert( pFS->pCompress );

**
** Page references returned by this function should be released by the 
** caller using lsmFsPageRelease().
*/
int lsmFsDbPageNext(Segment *pRun, Page *pPg, int eDir, Page **ppNext){
  int rc = LSM_OK;
  FileSystem *pFS = pPg->pFS;
  LsmPgno iPg = pPg->iPg;

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

    do {
      if( eDir>0 ){

** already allocated block. If it is possible, the page number of the first
** page to use for the new segment is returned. Otherwise zero.
**
** If argument pLvl is not NULL, then this function will not attempt to
** start the new segment immediately following any segment that is part
** of the right-hand-side of pLvl.
*/
static LsmPgno findAppendPoint(FileSystem *pFS, Level *pLvl){
  int i;
  LsmPgno *aiAppend = pFS->pDb->pWorker->aiAppend;
  LsmPgno iRet = 0;

  for(i=LSM_APPLIST_SZ-1; iRet==0 && i>=0; i--){
    if( (iRet = aiAppend[i]) ){
      if( pLvl ){
        int iBlk = fsPageToBlock(pFS, iRet);
        int j;
        for(j=0; iRet && j<pLvl->nRight; j++){

  Snapshot *pSnapshot,
  Level *pLvl,
  int bDefer,
  Page **ppOut
){
  int rc = LSM_OK;
  Page *pPg = 0;
  LsmPgno iApp = 0;
  LsmPgno iNext = 0;
  Segment *p = &pLvl->lhs;
  LsmPgno iPrev = p->iLastPg;

  *ppOut = 0;
  assert( p->pRedirect==0 );

  if( pFS->pCompress || bDefer ){
    /* In compressed database mode the page is not assigned a page number
    ** or location in the database file at this point. This will be done

    ** Shift this extra block back to the free-block list. 
    **
    ** Otherwise, add the first free page in the last block used by the run
    ** to the lAppend list.
    */
    if( fsLastPageOnPagesBlock(pFS, p->iLastPg)!=p->iLastPg ){
      int i;
      LsmPgno *aiAppend = pFS->pDb->pWorker->aiAppend;
      for(i=0; i<LSM_APPLIST_SZ; i++){
        if( aiAppend[i]==0 ){
          aiAppend[i] = p->iLastPg+1;
          break;
        }
      }
    }else if( pFS->pCompress==0 ){

}

/*
** Obtain a reference to page number iPg.
**
** Return LSM_OK if successful, or an lsm error code if an error occurs.
*/
int lsmFsDbPageGet(FileSystem *pFS, Segment *pSeg, LsmPgno iPg, Page **ppPg){
  return fsPageGet(pFS, pSeg, iPg, 0, ppPg, 0);
}

/*
** Obtain a reference to the last page in the segment passed as the 
** second argument.
**
** Return LSM_OK if successful, or an lsm error code if an error occurs.
*/
int lsmFsDbPageLast(FileSystem *pFS, Segment *pSeg, Page **ppPg){
  int rc;
  LsmPgno iPg = pSeg->iLastPg;
  if( pFS->pCompress ){
    int nSpace;
    iPg++;
    do {
      nSpace = 0;
      rc = fsGetPageBefore(pFS, pSeg, iPg, &iPg);
      if( rc==LSM_OK ){

** number (*piPg) lies on block iFrom, then calculate the equivalent
** page on block iTo and set *piPg to this value before returning.
*/
static void fsMovePage(
  FileSystem *pFS,                /* File system object */
  int iTo,                        /* Destination block */
  int iFrom,                      /* Source block */
  LsmPgno *piPg                   /* IN/OUT: Page number */
){
  LsmPgno iPg = *piPg;
  if( iFrom==fsPageToBlock(pFS, iPg) ){
    const int nPagePerBlock = (
        pFS->pCompress ? pFS ->nBlocksize : (pFS->nBlocksize / pFS->nPagesize)
    );
    *piPg = iPg - (LsmPgno)(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 

/*
** Append raw data to a segment. Return the database file offset that the
** data is written to (this may be used as the page number if the data
** being appended is a new page record).
**
** This function is only used in compressed database mode.
*/
static LsmPgno fsAppendData(
  FileSystem *pFS,                /* File-system handle */
  Segment *pSeg,                  /* Segment to append to */
  const u8 *aData,                /* Buffer containing data to write */
  int nData,                      /* Size of buffer aData[] in bytes */
  int *pRc                        /* IN/OUT: Error code */
){
  LsmPgno iRet = 0;
  int rc = *pRc;
  assert( pFS->pCompress );
  if( rc==LSM_OK ){
    int nRem = 0;
    int nWrite = 0;
    LsmPgno iLastOnBlock;
    LsmPgno iApp = pSeg->iLastPg+1;

    /* If this is the first data written into the segment, find an append-point
    ** or allocate a new block.  */
    if( iApp==1 ){
      pSeg->iFirst = iApp = findAppendPoint(pFS, 0);
      if( iApp==0 ){
        int iBlk;

          assert( iApp==(fsPageToBlock(pFS, iApp)*pFS->nBlocksize)-4 );
          lsmPutU32(aPtr, iBlk);
          rc = lsmEnvWrite(pFS->pEnv, pFS->fdDb, iApp, aPtr, sizeof(aPtr));
        }

        /* Set the "prev" pointer on the new block */
        if( rc==LSM_OK ){
          LsmPgno iWrite;
          lsmPutU32(aPtr, fsPageToBlock(pFS, iApp));
          iWrite = fsFirstPageOnBlock(pFS, iBlk);
          rc = lsmEnvWrite(pFS->pEnv, pFS->fdDb, iWrite-4, aPtr, sizeof(aPtr));
          if( nRem>0 ) iApp = iWrite;
        }
      }else{
        /* The next block is already allocated. */

** LSM_OK is returned if successful, or an lsm error code otherwise. If
** any value other than LSM_OK is returned, then the final value of all
** output variables is undefined.
*/
static int fsAppendPage(
  FileSystem *pFS, 
  Segment *pSeg,
  LsmPgno *piNew,
  int *piPrev,
  int *piNext
){
  LsmPgno iPrev = pSeg->iLastPg;
  int rc;
  assert( iPrev!=0 );

  *piPrev = 0;
  *piNext = 0;

  if( fsIsLast(pFS, iPrev) ){

  }
  *pRc = rc;
}

/*
** If there exists a hash-table entry associated with page iPg, remove it.
*/
static void fsRemoveHashEntry(FileSystem *pFS, LsmPgno iPg){
  Page *p;
  int iHash = fsHashKey(pFS->nHash, iPg);

  for(p=pFS->apHash[iHash]; p && p->iPg!=iPg; p=p->pHashNext);

  if( p ){
    assert( p->nRef==0 || (p->flags & PAGE_FREE)==0 );

int lsmFsSortedPadding(
  FileSystem *pFS, 
  Snapshot *pSnapshot,
  Segment *pSeg
){
  int rc = LSM_OK;
  if( pFS->pCompress && pSeg->iFirst ){
    LsmPgno iLast2;
    LsmPgno iLast = pSeg->iLastPg;  /* Current last page of segment */
    int nPad;                       /* Bytes of padding required */
    u8 aSz[3];

    iLast2 = (1 + iLast/pFS->szSector) * pFS->szSector - 1;
    assert( fsPageToBlock(pFS, iLast)==fsPageToBlock(pFS, iLast2) );
    nPad = (int)(iLast2 - iLast);

int lsmFsSectorSize(FileSystem *pFS){
  return pFS->szSector;
}

/*
** Helper function for lsmInfoArrayStructure().
*/
static Segment *startsWith(Segment *pRun, LsmPgno iFirst){
  return (iFirst==pRun->iFirst) ? pRun : 0;
}

/*
** Return the segment that starts with page iFirst, if any. If no such segment
** can be found, return NULL.
*/
static Segment *findSegment(Snapshot *pWorker, LsmPgno iFirst){
  Level *pLvl;                    /* Used to iterate through db levels */
  Segment *pSeg = 0;              /* Pointer to segment to return */

  for(pLvl=lsmDbSnapshotLevel(pWorker); pLvl && pSeg==0; pLvl=pLvl->pNext){
    if( 0==(pSeg = startsWith(&pLvl->lhs, iFirst)) ){
      int i;
      for(i=0; i<pLvl->nRight; i++){

** eventually free the string using lsmFree().
**
** If an error occurs, *pzOut is set to NULL and an LSM error code returned.
*/
int lsmInfoArrayStructure(
  lsm_db *pDb, 
  int bBlock,                     /* True for block numbers only */
  LsmPgno iFirst,
  char **pzOut
){
  int rc = LSM_OK;
  Snapshot *pWorker;              /* Worker snapshot */
  Segment *pArray = 0;            /* Array to report on */
  int bUnlock = 0;

  }
  return rc;
}

int lsmFsSegmentContainsPg(
  FileSystem *pFS, 
  Segment *pSeg, 
  LsmPgno iPg, 
  int *pbRes
){
  Redirect *pRedir = pSeg->pRedirect;
  int rc = LSM_OK;
  int iBlk;
  int iLastBlk;
  int iPgBlock;                   /* Block containing page iPg */

** This function implements the lsm_info(LSM_INFO_ARRAY_PAGES) request.
** If successful, *pzOut is set to point to a nul-terminated string 
** containing the array structure and LSM_OK is returned. The caller should
** eventually free the string using lsmFree().
**
** If an error occurs, *pzOut is set to NULL and an LSM error code returned.
*/
int lsmInfoArrayPages(lsm_db *pDb, LsmPgno iFirst, char **pzOut){
  int rc = LSM_OK;
  Snapshot *pWorker;              /* Worker snapshot */
  Segment *pSeg = 0;              /* Array to report on */
  int bUnlock = 0;

  *pzOut = 0;
  if( iFirst==0 ) return LSM_ERROR;

#ifndef NDEBUG
/*
** Return true if pPg happens to be the last page in segment pSeg. Or false
** otherwise. This function is only invoked as part of assert() conditions.
*/
int lsmFsDbPageIsLast(Segment *pSeg, Page *pPg){
  if( pPg->pFS->pCompress ){
    LsmPgno iNext = 0;
    int rc;
    rc = fsNextPageOffset(pPg->pFS, pSeg, pPg->iPg, pPg->nCompress+6, &iNext);
    return (rc!=LSM_OK || iNext==0);
  }
  return (pPg->iPg==pSeg->iLastPg);
}
#endif

    case LSM_INFO_DB_STRUCTURE: {
      char **pzVal = va_arg(ap, char **);
      rc = lsmStructList(pDb, pzVal);
      break;
    }

    case LSM_INFO_ARRAY_STRUCTURE: {
      LsmPgno pgno = va_arg(ap, LsmPgno);
      char **pzVal = va_arg(ap, char **);
      rc = lsmInfoArrayStructure(pDb, 0, pgno, pzVal);
      break;
    }

    case LSM_INFO_ARRAY_PAGES: {
      LsmPgno pgno = va_arg(ap, LsmPgno);
      char **pzVal = va_arg(ap, char **);
      rc = lsmInfoArrayPages(pDb, pgno, pzVal);
      break;
    }

    case LSM_INFO_PAGE_HEX_DUMP:
    case LSM_INFO_PAGE_ASCII_DUMP: {
      LsmPgno pgno = va_arg(ap, LsmPgno);
      char **pzVal = va_arg(ap, char **);
      int bUnlock = 0;
      rc = infoGetWorker(pDb, 0, &bUnlock);
      if( rc==LSM_OK ){
        int bHex = (eParam==LSM_INFO_PAGE_HEX_DUMP);
        rc = lsmInfoPageDump(pDb, pgno, bHex, pzVal);
      }

#ifndef LSM_SEGMENTPTR_FREE_THRESHOLD
# define LSM_SEGMENTPTR_FREE_THRESHOLD 1024
#endif

typedef struct SegmentPtr SegmentPtr;
typedef struct LsmBlob LsmBlob;

struct LsmBlob {
  lsm_env *pEnv;
  void *pData;
  int nData;
  int nAlloc;
};

/*

  Level *pLevel;                /* Level object segment is part of */
  Segment *pSeg;                /* Segment to access */

  /* Current page. See segmentPtrLoadPage(). */
  Page *pPg;                    /* Current page */
  u16 flags;                    /* Copy of page flags field */
  int nCell;                    /* Number of cells on pPg */
  LsmPgno iPtr;                 /* Base cascade pointer */

  /* Current cell. See segmentPtrLoadCell() */
  int iCell;                    /* Current record within page pPg */
  int eType;                    /* Type of current record */
  LsmPgno iPgPtr;               /* Cascade pointer offset */
  void *pKey; int nKey;         /* Key associated with current record */
  void *pVal; int nVal;         /* Current record value (eType==WRITE only) */

  /* Blobs used to allocate buffers for pKey and pVal as required */
  LsmBlob blob1;
  LsmBlob blob2;
};

/*
** Used to iterate through the keys stored in a b-tree hierarchy from start
** to finish. Only First() and Next() operations are required.
**
**   btreeCursorNew()

  int iPg;                        /* Current entry in aPg[]. -1 -> EOF. */
  BtreePg *aPg;                   /* Pages from root to current location */

  /* Cache of current entry. pKey==0 for EOF. */
  void *pKey;
  int nKey;
  int eType;
  LsmPgno iPtr;

  /* Storage for key, if not local */
  LsmBlob blob;
};


/*
** A cursor used for merged searches or iterations through up to one
** Tree structure and any number of sorted files.
**

*/
struct MultiCursor {
  lsm_db *pDb;                    /* Connection that owns this cursor */
  MultiCursor *pNext;             /* Next cursor owned by connection pDb */
  int flags;                      /* Mask of CURSOR_XXX flags */

  int eType;                      /* Cache of current key type */
  LsmBlob key;                    /* Cache of current key (or NULL) */
  LsmBlob val;                    /* Cache of current value */

  /* All the component cursors: */
  TreeCursor *apTreeCsr[2];       /* Up to two tree cursors */
  int iFree;                      /* Next element of free-list (-ve for eof) */
  SegmentPtr *aPtr;               /* Array of segment pointers */
  int nPtr;                       /* Size of array aPtr[] */
  BtreeCursor *pBtCsr;            /* b-tree cursor (db writes only) */

  /* Comparison results */
  int nTree;                      /* Size of aTree[] array */
  int *aTree;                     /* Array of comparison results */

  /* Used by cursors flushing the in-memory tree only */
  void *pSystemVal;               /* Pointer to buffer to free */

  /* Used by worker cursors only */
  LsmPgno *pPrevMergePtr;
};

/*
** The following constants are used to assign integers to each component
** cursor of a multi-cursor.
*/
#define CURSOR_DATA_TREE0     0   /* Current tree cursor (apTreeCsr[0]) */

  lsm_db *pDb;                    /* Database handle */
  Level *pLevel;                  /* Worker snapshot Level being merged */
  MultiCursor *pCsr;              /* Cursor to read new segment contents from */
  int bFlush;                     /* True if this is an in-memory tree flush */
  Hierarchy hier;                 /* B-tree hierarchy under construction */
  Page *pPage;                    /* Current output page */
  int nWork;                      /* Number of calls to mergeWorkerNextPage() */
  LsmPgno *aGobble;               /* Gobble point for each input segment */

  LsmPgno iIndirect;
  struct SavedPgno {
    LsmPgno iPgno;
    int bStore;
  } aSave[2];
};

#ifdef LSM_DEBUG_EXPENSIVE
static int assertPointersOk(lsm_db *, Segment *, Segment *, int);
static int assertBtreeOk(lsm_db *, Segment *);

  aOut[3] = (u8)((nVal>>32) & 0xFF);
  aOut[4] = (u8)((nVal>>24) & 0xFF);
  aOut[5] = (u8)((nVal>>16) & 0xFF);
  aOut[6] = (u8)((nVal>> 8) & 0xFF);
  aOut[7] = (u8)((nVal    ) & 0xFF);
}

static int sortedBlobGrow(lsm_env *pEnv, LsmBlob *pBlob, int nData){
  assert( pBlob->pEnv==pEnv || (pBlob->pEnv==0 && pBlob->pData==0) );
  if( pBlob->nAlloc<nData ){
    pBlob->pData = lsmReallocOrFree(pEnv, pBlob->pData, nData);
    if( !pBlob->pData ) return LSM_NOMEM_BKPT;
    pBlob->nAlloc = nData;
    pBlob->pEnv = pEnv;
  }
  return LSM_OK;
}

static int sortedBlobSet(lsm_env *pEnv, LsmBlob *pBlob, void *pData, int nData){
  if( sortedBlobGrow(pEnv, pBlob, nData) ) return LSM_NOMEM;
  memcpy(pBlob->pData, pData, nData);
  pBlob->nData = nData;
  return LSM_OK;
}

#if 0
static int sortedBlobCopy(LsmBlob *pDest, LsmBlob *pSrc){
  return sortedBlobSet(pDest, pSrc->pData, pSrc->nData);
}
#endif

static void sortedBlobFree(LsmBlob *pBlob){
  assert( pBlob->pEnv || pBlob->pData==0 );
  if( pBlob->pData ) lsmFree(pBlob->pEnv, pBlob->pData);
  memset(pBlob, 0, sizeof(LsmBlob));
}

static int sortedReadData(
  Segment *pSeg,
  Page *pPg,
  int iOff,
  int nByte,
  void **ppData,
  LsmBlob *pBlob
){
  int rc = LSM_OK;
  int iEnd;
  int nData;
  int nCell;
  u8 *aData;

  return rc;
}

static int pageGetNRec(u8 *aData, int nData){
  return (int)lsmGetU16(&aData[SEGMENT_NRECORD_OFFSET(nData)]);
}

static LsmPgno pageGetPtr(u8 *aData, int nData){
  return (LsmPgno)lsmGetU64(&aData[SEGMENT_POINTER_OFFSET(nData)]);
}

static int pageGetFlags(u8 *aData, int nData){
  return (int)lsmGetU16(&aData[SEGMENT_FLAGS_OFFSET(nData)]);
}

static u8 *pageGetCell(u8 *aData, int nData, int iCell){

  return pageGetNRec(aData, nData);
}

/*
** Return the decoded (possibly relative) pointer value stored in cell 
** iCell from page aData/nData.
*/
static LsmPgno pageGetRecordPtr(u8 *aData, int nData, int iCell){
  LsmPgno iRet;                   /* Return value */
  u8 *aCell;                      /* Pointer to cell iCell */

  assert( iCell<pageGetNRec(aData, nData) && iCell>=0 );
  aCell = pageGetCell(aData, nData, iCell);
  lsmVarintGet64(&aCell[1], &iRet);
  return iRet;
}

static u8 *pageGetKey(
  Segment *pSeg,                  /* Segment pPg belongs to */
  Page *pPg,                      /* Page to read from */
  int iCell,                      /* Index of cell on page to read */
  int *piTopic,                   /* OUT: Topic associated with this key */
  int *pnKey,                     /* OUT: Size of key in bytes */
  LsmBlob *pBlob                  /* If required, use this for dynamic memory */
){
  u8 *pKey;
  int nDummy;
  int eType;
  u8 *aData;
  int nData;

static int pageGetKeyCopy(
  lsm_env *pEnv,                  /* Environment handle */
  Segment *pSeg,                  /* Segment pPg belongs to */
  Page *pPg,                      /* Page to read from */
  int iCell,                      /* Index of cell on page to read */
  int *piTopic,                   /* OUT: Topic associated with this key */
  LsmBlob *pBlob                  /* If required, use this for dynamic memory */
){
  int rc = LSM_OK;
  int nKey;
  u8 *aKey;

  aKey = pageGetKey(pSeg, pPg, iCell, piTopic, &nKey, pBlob);
  assert( (void *)aKey!=pBlob->pData || nKey==pBlob->nData );
  if( (void *)aKey!=pBlob->pData ){
    rc = sortedBlobSet(pEnv, pBlob, aKey, nKey);
  }

  return rc;
}

static LsmPgno pageGetBtreeRef(Page *pPg, int iKey){
  LsmPgno iRef;
  u8 *aData;
  int nData;
  u8 *aCell;

  aData = fsPageData(pPg, &nData);
  aCell = pageGetCell(aData, nData, iKey);
  assert( aCell[0]==0 );

#define GETVARINT64(a, i) (((i)=((u8*)(a))[0])<=240?1:lsmVarintGet64((a), &(i)))
#define GETVARINT32(a, i) (((i)=((u8*)(a))[0])<=240?1:lsmVarintGet32((a), &(i)))

static int pageGetBtreeKey(
  Segment *pSeg,                  /* Segment page pPg belongs to */
  Page *pPg,
  int iKey, 
  LsmPgno *piPtr, 
  int *piTopic, 
  void **ppKey,
  int *pnKey,
  LsmBlob *pBlob
){
  u8 *aData;
  int nData;
  u8 *aCell;
  int eType;

  aData = fsPageData(pPg, &nData);
  assert( SEGMENT_BTREE_FLAG & pageGetFlags(aData, nData) );
  assert( iKey>=0 && iKey<pageGetNRec(aData, nData) );

  aCell = pageGetCell(aData, nData, iKey);
  eType = *aCell++;
  aCell += GETVARINT64(aCell, *piPtr);

  if( eType==0 ){
    int rc;
    LsmPgno iRef;               /* Page number of referenced page */
    Page *pRef;
    aCell += GETVARINT64(aCell, iRef);
    rc = lsmFsDbPageGet(lsmPageFS(pPg), pSeg, iRef, &pRef);
    if( rc!=LSM_OK ) return rc;
    pageGetKeyCopy(lsmPageEnv(pPg), pSeg, pRef, 0, &eType, pBlob);
    lsmFsPageRelease(pRef);
    *ppKey = pBlob->pData;

static int btreeCursorLoadKey(BtreeCursor *pCsr){
  int rc = LSM_OK;
  if( pCsr->iPg<0 ){
    pCsr->pKey = 0;
    pCsr->nKey = 0;
    pCsr->eType = 0;
  }else{
    LsmPgno dummy;
    int iPg = pCsr->iPg;
    int iCell = pCsr->aPg[iPg].iCell;
    while( iCell<0 && (--iPg)>=0 ){
      iCell = pCsr->aPg[iPg].iCell-1;
    }
    if( iPg<0 || iCell<0 ) return LSM_CORRUPT_BKPT;

  assert( pCsr->iPg==pCsr->nDepth-1 );

  aData = fsPageData(pPg->pPage, &nData);
  nCell = pageGetNRec(aData, nData);
  assert( pPg->iCell<=nCell );
  pPg->iCell++;
  if( pPg->iCell==nCell ){
    LsmPgno iLoad;

    /* Up to parent. */
    lsmFsPageRelease(pPg->pPage);
    pPg->pPage = 0;
    pCsr->iPg--;
    while( pCsr->iPg>=0 ){
      pPg = &pCsr->aPg[pCsr->iPg];

  MergeInput *p
){
  int rc = LSM_OK;

  if( p->iPg ){
    lsm_env *pEnv = lsmFsEnv(pCsr->pFS);
    int iCell;                    /* Current cell number on leaf page */
    LsmPgno iLeaf;                /* Page number of current leaf page */
    int nDepth;                   /* Depth of b-tree structure */
    Segment *pSeg = pCsr->pSeg;

    /* Decode the MergeInput structure */
    iLeaf = p->iPg;
    nDepth = (p->iCell & 0x00FF);
    iCell = (p->iCell >> 8) - 1;

      pCsr->nDepth = nDepth;
      pCsr->aPg[pCsr->iPg].iCell = iCell;
      rc = lsmFsDbPageGet(pCsr->pFS, pSeg, iLeaf, pp);
    }

    /* Populate any other aPg[] array entries */
    if( rc==LSM_OK && nDepth>1 ){
      LsmBlob blob = {0,0,0};
      void *pSeek;
      int nSeek;
      int iTopicSeek;
      int iPg = 0;
      int iLoad = (int)pSeg->iRoot;
      Page *pPg = pCsr->aPg[nDepth-1].pPage;
 
      if( pageObjGetNRec(pPg)==0 ){
        /* This can happen when pPg is the right-most leaf in the b-tree.
        ** In this case, set the iTopicSeek/pSeek/nSeek key to a value
        ** greater than any real key.  */
        assert( iCell==-1 );
        iTopicSeek = 1000;
        pSeek = 0;
        nSeek = 0;
      }else{
        LsmPgno dummy;
        rc = pageGetBtreeKey(pSeg, pPg,
            0, &dummy, &iTopicSeek, &pSeek, &nSeek, &pCsr->blob
        );
      }

      do {
        Page *pPg2;

          iMax = iCell2-1;
          iMin = 0;

          while( iMax>=iMin ){
            int iTry = (iMin+iMax)/2;
            void *pKey; int nKey;         /* Key for cell iTry */
            int iTopic;                   /* Topic for key pKeyT/nKeyT */
            LsmPgno iPtr;                 /* Pointer for cell iTry */
            int res;                      /* (pSeek - pKeyT) */

            rc = pageGetBtreeKey(
                pSeg, pPg2, iTry, &iPtr, &iTopic, &pKey, &nKey, &blob
            );
            if( rc!=LSM_OK ) break;

      u8 *aData;
      int nData;

      pBtreePg = &pCsr->aPg[pCsr->iPg];
      aData = fsPageData(pBtreePg->pPage, &nData);
      pCsr->iPtr = btreeCursorPtr(aData, nData, pBtreePg->iCell+1);
      if( pBtreePg->iCell<0 ){
        LsmPgno dummy;
        int i;
        for(i=pCsr->iPg-1; i>=0; i--){
          if( pCsr->aPg[i].iCell>0 ) break;
        }
        assert( i>=0 );
        rc = pageGetBtreeKey(pSeg,
            pCsr->aPg[i].pPage, pCsr->aPg[i].iCell-1,

}

static int segmentPtrReadData(
  SegmentPtr *pPtr,
  int iOff,
  int nByte,
  void **ppData,
  LsmBlob *pBlob
){
  return sortedReadData(pPtr->pSeg, pPtr->pPg, iOff, nByte, ppData, pBlob);
}

static int segmentPtrNextPage(
  SegmentPtr *pPtr,              /* Load page into this SegmentPtr object */
  int eDir                       /* +1 for next(), -1 for prev() */

  pSeg = sortedSplitkeySegment(pLevel);
  if( rc==LSM_OK ){
    rc = lsmFsDbPageGet(pDb->pFS, pSeg, pMerge->splitkey.iPg, &pPg);
  }
  if( rc==LSM_OK ){
    int iTopic;
    LsmBlob blob = {0, 0, 0, 0};
    u8 *aData;
    int nData;
  
    aData = lsmFsPageData(pPg, &nData);
    if( pageGetFlags(aData, nData) & SEGMENT_BTREE_FLAG ){
      void *pKey;
      int nKey;
      LsmPgno dummy;
      rc = pageGetBtreeKey(pSeg,
          pPg, pMerge->splitkey.iCell, &dummy, &iTopic, &pKey, &nKey, &blob
      );
      if( rc==LSM_OK && blob.pData!=pKey ){
        rc = sortedBlobSet(pEnv, &blob, pKey, nKey);
      }
    }else{

*/
static int assertKeyLocation(
  MultiCursor *pCsr, 
  SegmentPtr *pPtr, 
  void *pKey, int nKey
){
  lsm_env *pEnv = lsmFsEnv(pCsr->pDb->pFS);
  LsmBlob blob = {0, 0, 0};
  int eDir;
  int iTopic = 0;                 /* TODO: Fix me */

  for(eDir=-1; eDir<=1; eDir+=2){
    Page *pTest = pPtr->pPg;

    lsmFsPageRef(pTest);

  return rc;
}

static int ptrFwdPointer(
  Page *pPage,
  int iCell,
  Segment *pSeg,
  LsmPgno *piPtr,
  int *pbFound
){
  Page *pPg = pPage;
  int iFirst = iCell;
  int rc = LSM_OK;

  do {

**   much better if the multi-cursor could do this lazily - only seek to the
**   level (N+1) page after the user has moved the cursor on level N passed
**   the big range-delete.
*/
static int segmentPtrFwdPointer(
  MultiCursor *pCsr,              /* Multi-cursor pPtr belongs to */
  SegmentPtr *pPtr,               /* Segment-pointer to extract FC ptr from */
  LsmPgno *piPtr                  /* OUT: FC pointer value */
){
  Level *pLvl = pPtr->pLevel;
  Level *pNext = pLvl->pNext;
  Page *pPg = pPtr->pPg;
  int rc;
  int bFound;
  LsmPgno iOut = 0;

  if( pPtr->pSeg==&pLvl->lhs || pPtr->pSeg==&pLvl->aRhs[pLvl->nRight-1] ){
    if( pNext==0 
        || (pNext->nRight==0 && pNext->lhs.iRoot)
        || (pNext->nRight!=0 && pNext->aRhs[0].iRoot)
      ){
      /* Do nothing. The pointer will not be used anyway. */

  int *pbStop
){
  int (*xCmp)(void *, int, void *, int) = pCsr->pDb->xCmp;
  int res = 0;                        /* Result of comparison operation */
  int rc = LSM_OK;
  int iMin;
  int iMax;
  LsmPgno iPtrOut = 0;

  /* If the current page contains an oversized entry, then there are no
  ** pointers to one or more of the subsequent pages in the sorted run.
  ** The following call ensures that the segment-ptr points to the correct 
  ** page in this case.  */
  rc = segmentPtrSearchOversized(pCsr, pPtr, iTopic, pKey, nKey);
  iPtrOut = pPtr->iPtr;

}

static int seekInBtree(
  MultiCursor *pCsr,              /* Multi-cursor object */
  Segment *pSeg,                  /* Seek within this segment */
  int iTopic,
  void *pKey, int nKey,           /* Key to seek to */
  LsmPgno *aPg,                   /* OUT: Page numbers */
  Page **ppPg                     /* OUT: Leaf (sorted-run) page reference */
){
  int i = 0;
  int rc;
  int iPg;
  Page *pPg = 0;
  LsmBlob blob = {0, 0, 0};

  iPg = (int)pSeg->iRoot;
  do {
    LsmPgno *piFirst = 0;
    if( aPg ){
      aPg[i++] = iPg;
      piFirst = &aPg[i];
    }

    rc = lsmFsDbPageGet(pCsr->pDb->pFS, pSeg, iPg, &pPg);
    assert( rc==LSM_OK || pPg==0 );

      iMin = 0;
      iMax = nRec-1;
      while( iMax>=iMin ){
        int iTry = (iMin+iMax)/2;
        void *pKeyT; int nKeyT;       /* Key for cell iTry */
        int iTopicT;                  /* Topic for key pKeyT/nKeyT */
        LsmPgno iPtr;                 /* Pointer associated with cell iTry */
        int res;                      /* (pKey - pKeyT) */

        rc = pageGetBtreeKey(
            pSeg, pPg, iTry, &iPtr, &iTopicT, &pKeyT, &nKeyT, &blob
        );
        if( rc!=LSM_OK ) break;
        if( piFirst && pKeyT==blob.pData ){

*/
static int seekInLevel(
  MultiCursor *pCsr,              /* Sorted cursor object to seek */
  SegmentPtr *aPtr,               /* Pointer to array of (nRhs+1) SPs */
  int eSeek,                      /* Search bias - see above */
  int iTopic,                     /* Key topic to search for */
  void *pKey, int nKey,           /* Key to search for */
  LsmPgno *piPgno,                /* IN/OUT: fraction cascade pointer (or 0) */
  int *pbStop                     /* OUT: See above */
){
  Level *pLvl = aPtr[0].pLevel;   /* Level to seek within */
  int rc = LSM_OK;                /* Return code */
  int iOut = 0;                   /* Pointer to return to caller */
  int res = -1;                   /* Result of xCmp(pKey, split) */
  int nRhs = pLvl->nRight;        /* Number of right-hand-side segments */

  void *pKey, int nKey, 
  int eSeek
){
  int eESeek = eSeek;             /* Effective eSeek parameter */
  int bStop = 0;                  /* Set to true to halt search operation */
  int rc = LSM_OK;                /* Return code */
  int iPtr = 0;                   /* Used to iterate through pCsr->aPtr[] */
  LsmPgno iPgno = 0;              /* FC pointer value */

  assert( pCsr->apTreeCsr[0]==0 || iTopic==0 );
  assert( pCsr->apTreeCsr[1]==0 || iTopic==0 );

  if( eESeek==LSM_SEEK_LEFAST ) eESeek = LSM_SEEK_LE;

  assert( eESeek==LSM_SEEK_EQ || eESeek==LSM_SEEK_LE || eESeek==LSM_SEEK_GE );

** differences are:
**
**   1. The record format is (usually, see below) as follows:
**
**         + Type byte (always SORTED_SEPARATOR or SORTED_SYSTEM_SEPARATOR),
**         + Absolute pointer value (varint),
**         + Number of bytes in key (varint),
**         + LsmBlob containing key data.
**
**   2. All pointer values are stored as absolute values (not offsets 
**      relative to the footer pointer value).
**
**   3. Each pointer that is part of a record points to a page that 
**      contains keys smaller than the records key (note: not "equal to or
**      smaller than - smaller than").

**
** See function seekInBtree() for the code that traverses b-tree pages.
*/

static int mergeWorkerBtreeWrite(
  MergeWorker *pMW,
  u8 eType,
  LsmPgno iPtr,
  LsmPgno iKeyPg,
  void *pKey,
  int nKey
){
  Hierarchy *p = &pMW->hier;
  lsm_db *pDb = pMW->pDb;         /* Database handle */
  int rc = LSM_OK;                /* Return Code */
  int iLevel;                     /* Level of b-tree hierachy to write to */

  return rc;
}

static int mergeWorkerBtreeIndirect(MergeWorker *pMW){
  int rc = LSM_OK;
  if( pMW->iIndirect ){
    LsmPgno iKeyPg = pMW->aSave[1].iPgno;
    rc = mergeWorkerBtreeWrite(pMW, 0, pMW->iIndirect, iKeyPg, 0, 0);
    pMW->iIndirect = 0;
  }
  return rc;
}

/*
** Append the database key (iTopic/pKey/nKey) to the b-tree under 
** construction. This key has not yet been written to a segment page.
** The pointer that will accompany the new key in the b-tree - that
** points to the completed segment page that contains keys smaller than
** (pKey/nKey) is currently stored in pMW->aSave[0].iPgno.
*/
static int mergeWorkerPushHierarchy(
  MergeWorker *pMW,               /* Merge worker object */
  int iTopic,                     /* Topic value for this key */
  void *pKey,                     /* Pointer to key buffer */
  int nKey                        /* Size of pKey buffer in bytes */
){
  int rc = LSM_OK;                /* Return Code */
  LsmPgno iPtr;                   /* Pointer value to accompany pKey/nKey */

  assert( pMW->aSave[0].bStore==0 );
  assert( pMW->aSave[1].bStore==0 );
  rc = mergeWorkerBtreeIndirect(pMW);

  /* Obtain the absolute pointer value to store along with the key in the
  ** page body. This pointer points to a page that contains keys that are

}

static int mergeWorkerFinishHierarchy(
  MergeWorker *pMW                /* Merge worker object */
){
  int i;                          /* Used to loop through apHier[] */
  int rc = LSM_OK;                /* Return code */
  LsmPgno iPtr;                   /* New right-hand-child pointer value */

  iPtr = pMW->aSave[0].iPgno;
  for(i=0; i<pMW->hier.nHier && rc==LSM_OK; i++){
    Page *pPg = pMW->hier.apHier[i];
    int nData;                    /* Size of aData[] in bytes */
    u8 *aData;                    /* Page data for pPg */

** zero records. The flags field is cleared. The page footer pointer field
** is set to iFPtr.
**
** If successful, LSM_OK is returned. Otherwise, an error code.
*/
static int mergeWorkerNextPage(
  MergeWorker *pMW,               /* Merge worker object to append page to */
  LsmPgno iFPtr                   /* Pointer value for footer of new page */
){
  int rc = LSM_OK;                /* Return code */
  Page *pNext = 0;                /* New page appended to run */
  lsm_db *pDb = pMW->pDb;         /* Database handle */

  rc = lsmFsSortedAppend(pDb->pFS, pDb->pWorker, pMW->pLevel, 0, &pNext);
  assert( rc || pMW->pLevel->lhs.iFirst>0 || pMW->pDb->compress.xCompress );

static int mergeWorkerStep(MergeWorker *pMW){
  lsm_db *pDb = pMW->pDb;       /* Database handle */
  MultiCursor *pCsr;            /* Cursor to read input data from */
  int rc = LSM_OK;              /* Return code */
  int eType;                    /* SORTED_SEPARATOR, WRITE or DELETE */
  void *pKey; int nKey;         /* Key */
  LsmPgno iPtr;
  int iVal;

  pCsr = pMW->pCsr;

  /* Pull the next record out of the source cursor. */
  lsmMCursorKey(pCsr, &pKey, &nKey);
  eType = pCsr->eType;

      multiCursorIgnoreDelete(pCsr);
    }
  }

  if( rc!=LSM_OK ){
    lsmMCursorClose(pCsr, 0);
  }else{
    LsmPgno iLeftPtr = 0;
    Merge merge;                  /* Merge object used to create new level */
    MergeWorker mergeworker;      /* MergeWorker object for the same purpose */

    memset(&merge, 0, sizeof(Merge));
    memset(&mergeworker, 0, sizeof(MergeWorker));

    pNew->pMerge = &merge;

  assert( pDb->pWorker );
  assert( pLevel->pMerge );
  assert( pLevel->nRight>0 );

  memset(pMW, 0, sizeof(MergeWorker));
  pMW->pDb = pDb;
  pMW->pLevel = pLevel;
  pMW->aGobble = lsmMallocZeroRc(pDb->pEnv, sizeof(LsmPgno)*pLevel->nRight,&rc);

  /* Create a multi-cursor to read the data to write to the new
  ** segment. The new segment contains:
  **
  **   1. Records from LHS of each of the nMerge levels being merged.
  **   2. Separators from either the last level being merged, or the
  **      separators attached to the LHS of the following level, or neither.

  lsm_db *pDb,                    /* Worker connection */
  MultiCursor *pCsr,              /* Multi-cursor being used for a merge */
  int iGobble                     /* pCsr->aPtr[] entry to operate on */
){
  int rc = LSM_OK;
  if( rtTopic(pCsr->eType)==0 ){
    Segment *pSeg = pCsr->aPtr[iGobble].pSeg;
    LsmPgno *aPg;
    int nPg;

    /* Seek from the root of the b-tree to the segment leaf that may contain
    ** a key equal to the one multi-cursor currently points to. Record the
    ** page number of each b-tree page and the leaf. The segment may be
    ** gobbled up to (but not including) the first of these page numbers.
    */
    assert( pSeg->iRoot>0 );
    aPg = lsmMallocZeroRc(pDb->pEnv, sizeof(LsmPgno)*32, &rc);
    if( rc==LSM_OK ){
      rc = seekInBtree(pCsr, pSeg, 
          rtTopic(pCsr->eType), pCsr->key.pData, pCsr->key.nData, aPg, 0
      ); 
    }

    if( rc==LSM_OK ){

/*
** Return a string representation of the segment passed as the only argument.
** Space for the returned string is allocated using lsmMalloc(), and should
** be freed by the caller using lsmFree().
*/
static char *segToString(lsm_env *pEnv, Segment *pSeg, int nMin){
  int nSize = pSeg->nSize;
  LsmPgno iRoot = pSeg->iRoot;
  LsmPgno iFirst = pSeg->iFirst;
  LsmPgno iLast = pSeg->iLastPg;
  char *z;

  char *z1;
  char *z2;
  int nPad;

  z1 = lsmMallocPrintf(pEnv, "%d.%d", iFirst, iLast);

    aBuf[0] = '\0';
  }

  return i;
}

void sortedDumpPage(lsm_db *pDb, Segment *pRun, Page *pPg, int bVals){
  LsmBlob blob = {0, 0, 0};       /* LsmBlob used for keys */
  LsmString s;
  int i;

  int nRec;
  int iPtr;
  int flags;
  u8 *aData;

    aCell = pageGetCell(aData, nData, i);
    eType = *aCell++;
    assert( (flags & SEGMENT_BTREE_FLAG) || eType!=0 );
    aCell += lsmVarintGet32(aCell, &iPgPtr);

    if( eType==0 ){
      LsmPgno iRef;               /* Page number of referenced page */
      aCell += lsmVarintGet64(aCell, &iRef);
      lsmFsDbPageGet(pDb->pFS, pRun, iRef, &pRef);
      aKey = pageGetKey(pRun, pRef, 0, &iTopic, &nKey, &blob);
    }else{
      aCell += lsmVarintGet32(aCell, &nKey);
      if( rtIsWrite(eType) ) aCell += lsmVarintGet32(aCell, &nVal);
      sortedReadData(0, pPg, (aCell-aData), nKey+nVal, (void **)&aKey, &blob);

  int bIndirect,                  /* True to follow indirect refs */
  Page *pPg,
  int iCell,
  int *peType,
  int *piPgPtr,
  u8 **paKey, int *pnKey,
  u8 **paVal, int *pnVal,
  LsmBlob *pBlob
){
  u8 *aData; int nData;           /* Page data */
  u8 *aKey; int nKey = 0;         /* Key */
  u8 *aVal = 0; int nVal = 0;     /* Value */
  int eType;
  int iPgPtr;
  Page *pRef = 0;                 /* Pointer to page iRef */
  u8 *aCell;

  aData = fsPageData(pPg, &nData);

  aCell = pageGetCell(aData, nData, iCell);
  eType = *aCell++;
  aCell += lsmVarintGet32(aCell, &iPgPtr);

  if( eType==0 ){
    int dummy;
    LsmPgno iRef;                 /* Page number of referenced page */
    aCell += lsmVarintGet64(aCell, &iRef);
    if( bIndirect ){
      lsmFsDbPageGet(pDb->pFS, pSeg, iRef, &pRef);
      pageGetKeyCopy(pDb->pEnv, pSeg, pRef, 0, &dummy, pBlob);
      aKey = (u8 *)pBlob->pData;
      nKey = pBlob->nData;
      lsmFsPageRelease(pRef);

#define INFO_PAGE_DUMP_DATA     0x01
#define INFO_PAGE_DUMP_VALUES   0x02
#define INFO_PAGE_DUMP_HEX      0x04
#define INFO_PAGE_DUMP_INDIRECT 0x08

static int infoPageDump(
  lsm_db *pDb,                    /* Database handle */
  LsmPgno iPg,                    /* Page number of page to dump */
  int flags,
  char **pzOut                    /* OUT: lsmMalloc'd string */
){
  int rc = LSM_OK;                /* Return code */
  Page *pPg = 0;                  /* Handle for page iPg */
  int i, j;                       /* Loop counters */
  const int perLine = 16;         /* Bytes per line in the raw hex dump */

  ** to pass a NULL in place of the segment pointer as the second argument
  ** to lsmFsDbPageGet() here.  */
  if( rc==LSM_OK ){
    rc = lsmFsDbPageGet(pDb->pFS, 0, iPg, &pPg);
  }

  if( rc==LSM_OK ){
    LsmBlob blob = {0, 0, 0, 0};
    int nKeyWidth = 0;
    LsmString str;
    int nRec;
    int iPtr;
    int flags2;
    int iCell;
    u8 *aData; int nData;         /* Page data and size thereof */

    if( bHex ) nKeyWidth = nKeyWidth * 2;

    for(iCell=0; iCell<nRec; iCell++){
      u8 *aKey; int nKey = 0;       /* Key */
      u8 *aVal; int nVal = 0;       /* Value */
      int iPgPtr;
      int eType;
      LsmPgno iAbsPtr;
      char zFlags[8];

      infoCellDump(pDb, pSeg, bIndirect, pPg, iCell, &eType, &iPgPtr,
          &aKey, &nKey, &aVal, &nVal, &blob
      );
      iAbsPtr = iPgPtr + ((flags2 & SEGMENT_BTREE_FLAG) ? 0 : iPtr);

  }

  return rc;
}

int lsmInfoPageDump(
  lsm_db *pDb,                    /* Database handle */
  LsmPgno iPg,                    /* Page number of page to dump */
  int bHex,                       /* True to output key/value in hex form */
  char **pzOut                    /* OUT: lsmMalloc'd string */
){
  int flags = INFO_PAGE_DUMP_DATA | INFO_PAGE_DUMP_VALUES;
  if( bHex ) flags |= INFO_PAGE_DUMP_HEX;
  return infoPageDump(pDb, iPg, flags, pzOut);
}

  iHdr = SEGMENT_EOF(nOrig, nEntry);
  memmove(&aData[iHdr + (nData-nOrig)], &aData[iHdr], nOrig-iHdr);
}

#ifdef LSM_DEBUG_EXPENSIVE
static void assertRunInOrder(lsm_db *pDb, Segment *pSeg){
  Page *pPg = 0;
  LsmBlob blob1 = {0, 0, 0, 0};
  LsmBlob blob2 = {0, 0, 0, 0};

  lsmFsDbPageGet(pDb->pFS, pSeg, pSeg->iFirst, &pPg);
  while( pPg ){
    u8 *aData; int nData;
    Page *pNext;

    aData = lsmFsPageData(pPg, &nData);

  Segment *pOne,                  /* Segment containing pointers */
  Segment *pTwo,                  /* Segment containing pointer targets */
  int bRhs                        /* True if pTwo may have been Gobble()d */
){
  int rc = LSM_OK;                /* Error code */
  SegmentPtr ptr1;                /* Iterates through pOne */
  SegmentPtr ptr2;                /* Iterates through pTwo */
  LsmPgno iPrev;

  assert( pOne && pTwo );

  memset(&ptr1, 0, sizeof(ptr1));
  memset(&ptr2, 0, sizeof(ptr1));
  ptr1.pSeg = pOne;
  ptr2.pSeg = pTwo;

  }

  if( rc==LSM_OK && ptr1.nCell>0 ){
    rc = segmentPtrLoadCell(&ptr1, 0);
  }
      
  while( rc==LSM_OK && ptr2.pPg ){
    LsmPgno iThis;

    /* Advance to the next page of segment pTwo that contains at least
    ** one cell. Break out of the loop if the iterator reaches EOF.  */
    do{
      rc = segmentPtrNextPage(&ptr2, 1);
      assert( rc==LSM_OK );
    }while( rc==LSM_OK && ptr2.pPg && ptr2.nCell==0 );

*/
static int assertBtreeOk(
  lsm_db *pDb,
  Segment *pSeg
){
  int rc = LSM_OK;                /* Return code */
  if( pSeg->iRoot ){
    LsmBlob blob = {0, 0, 0};     /* Buffer used to cache overflow keys */
    FileSystem *pFS = pDb->pFS;   /* File system to read from */
    Page *pPg = 0;                /* Main run page */
    BtreeCursor *pCsr = 0;        /* Btree cursor */

    rc = btreeCursorNew(pDb, pSeg, &pCsr);
    if( rc==LSM_OK ){
      rc = btreeCursorFirst(pCsr);