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 {
  int iFirst;                     /* First page of this run */
  int iLast;                      /* Last page of this run */
  Pgno iRoot;                     /* Root page number (if any) */
  int nSize;                      /* Size of this run in pages */
};

/*
** iSplitTopic/pSplitKey/nSplitKey:
**   If nRight>0, this buffer contains a copy of the largest key that has
**   already been written to the left-hand-side of the level.
*/
struct Level {
  Segment lhs;                    /* Left-hand (main) segment */
  int iAge;                       /* Number of times data has been written */
  int nRight;                     /* Size of apRight[] array */
  Segment *aRhs;                  /* Old segments being merged into this */
  int iSplitTopic;                /* Split key topic (if nRight>0) */
  void *pSplitKey;                /* Pointer to split-key (if nRight>0) */
  int nSplitKey;                  /* Number of bytes in split-key */


  Merge *pMerge;                  /* Merge operation currently underway */
  Level *pNext;                   /* Next level in tree */
};

/*
** A structure describing an ongoing merge. There is an instance of this
** structure for every Level currently undergoing a merge in the worker
................................................................................
int lsmCheckpointDeserialize(lsm_db *, int, u32 *, Snapshot **);

int lsmCheckpointLoadWorker(lsm_db *pDb);
int lsmCheckpointStore(lsm_db *pDb, int);

int lsmCheckpointLoad(lsm_db *pDb, int *);
int lsmCheckpointLoadOk(lsm_db *pDb, int);


i64 lsmCheckpointId(u32 *, int);
u32 lsmCheckpointNWrite(u32 *, int);
i64 lsmCheckpointLogOffset(u32 *);
int lsmCheckpointPgsz(u32 *);
int lsmCheckpointBlksz(u32 *);
void lsmCheckpointLogoffset(u32 *aCkpt, DbLog *pLog);

  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 {
  Pgno iFirst;                     /* First page of this run */
  Pgno iLast;                      /* Last page of this run */
  Pgno iRoot;                     /* Root page number (if any) */
  int nSize;                      /* Size of this run in pages */
};

/*
** iSplitTopic/pSplitKey/nSplitKey:
**   If nRight>0, this buffer contains a copy of the largest key that has
**   already been written to the left-hand-side of the level.
*/
struct Level {
  Segment lhs;                    /* Left-hand (main) segment */

  int nRight;                     /* Size of apRight[] array */
  Segment *aRhs;                  /* Old segments being merged into this */
  int iSplitTopic;                /* Split key topic (if nRight>0) */
  void *pSplitKey;                /* Pointer to split-key (if nRight>0) */
  int nSplitKey;                  /* Number of bytes in split-key */

  int iAge;                       /* Number of times data has been written */
  Merge *pMerge;                  /* Merge operation currently underway */
  Level *pNext;                   /* Next level in tree */
};

/*
** A structure describing an ongoing merge. There is an instance of this
** structure for every Level currently undergoing a merge in the worker
................................................................................
int lsmCheckpointDeserialize(lsm_db *, int, u32 *, Snapshot **);

int lsmCheckpointLoadWorker(lsm_db *pDb);
int lsmCheckpointStore(lsm_db *pDb, int);

int lsmCheckpointLoad(lsm_db *pDb, int *);
int lsmCheckpointLoadOk(lsm_db *pDb, int);
int lsmCheckpointClientCacheOk(lsm_db *);

i64 lsmCheckpointId(u32 *, int);
u32 lsmCheckpointNWrite(u32 *, int);
i64 lsmCheckpointLogOffset(u32 *);
int lsmCheckpointPgsz(u32 *);
int lsmCheckpointBlksz(u32 *);
void lsmCheckpointLogoffset(u32 *aCkpt, DbLog *pLog);

**     2. Checksum value 2.
**
** In the above, a segment record is:
**
**     1. First page of array,
**     2. Last page of array,
**     3. Root page of array (or 0),
**     4. Size of array in pages,
*/

/*
** LARGE NUMBERS OF LEVEL RECORDS:
**
** A limit on the number of rhs segments that may be present in the database
** file. Defining this limit ensures that all level records fit within
................................................................................

int lsmCheckpointLoadOk(lsm_db *pDb, int iSnap){
  u32 *aShm;
  assert( iSnap==1 || iSnap==2 );
  aShm = (iSnap==1) ? pDb->pShmhdr->aSnap1 : pDb->pShmhdr->aSnap2;
  return (lsmCheckpointId(pDb->aSnapshot, 0)==lsmCheckpointId(aShm, 0) );
}









int lsmCheckpointLoadWorker(lsm_db *pDb){
  int rc;
  ShmHeader *pShm = pDb->pShmhdr;
  int nInt1;
  int nInt2;

**     2. Checksum value 2.
**
** In the above, a segment record is:
**
**     1. First page of array,
**     2. Last page of array,
**     3. Root page of array (or 0),
**     4. Size of array in pages.
*/

/*
** LARGE NUMBERS OF LEVEL RECORDS:
**
** A limit on the number of rhs segments that may be present in the database
** file. Defining this limit ensures that all level records fit within
................................................................................

int lsmCheckpointLoadOk(lsm_db *pDb, int iSnap){
  u32 *aShm;
  assert( iSnap==1 || iSnap==2 );
  aShm = (iSnap==1) ? pDb->pShmhdr->aSnap1 : pDb->pShmhdr->aSnap2;
  return (lsmCheckpointId(pDb->aSnapshot, 0)==lsmCheckpointId(aShm, 0) );
}

int lsmCheckpointClientCacheOk(lsm_db *pDb){
  return ( pDb->pClient 
        && pDb->pClient->iId==lsmCheckpointId(pDb->aSnapshot, 0)
        && pDb->pClient->iId==lsmCheckpointId(pDb->pShmhdr->aSnap1, 0)
        && pDb->pClient->iId==lsmCheckpointId(pDb->pShmhdr->aSnap2, 0)
  );
}

int lsmCheckpointLoadWorker(lsm_db *pDb){
  int rc;
  ShmHeader *pShm = pDb->pShmhdr;
  int nInt1;
  int nInt2;

*/
static void assert_db_state(lsm_db *pDb){

  /* If there is at least one cursor or a write transaction open, the database
  ** handle must be holding a pointer to a client snapshot. And the reverse 
  ** - if there are no open cursors and no write transactions then there must 
  ** not be a client snapshot.  */
  assert( (pDb->pCsr!=0 || pDb->nTransOpen>0)==(pDb->pClient!=0) );

  assert( pDb->nTransOpen>=0 );
}
#else
# define assert_db_state(x) 
#endif

................................................................................
int lsm_close(lsm_db *pDb){
  int rc = LSM_OK;
  if( pDb ){
    assert_db_state(pDb);
    if( pDb->pCsr || pDb->nTransOpen ){
      rc = LSM_MISUSE_BKPT;
    }else{


      lsmDbDatabaseRelease(pDb);
      lsmFsClose(pDb->pFS);
      lsmFree(pDb->pEnv, pDb->aTrans);
      lsmFree(pDb->pEnv, pDb->apShm);
      lsmFree(pDb->pEnv, pDb);
    }
  }

*/
static void assert_db_state(lsm_db *pDb){

  /* If there is at least one cursor or a write transaction open, the database
  ** handle must be holding a pointer to a client snapshot. And the reverse 
  ** - if there are no open cursors and no write transactions then there must 
  ** not be a client snapshot.  */
  assert( (pDb->pCsr!=0 || pDb->nTransOpen>0)==(pDb->iReader>=0) );

  assert( pDb->nTransOpen>=0 );
}
#else
# define assert_db_state(x) 
#endif

................................................................................
int lsm_close(lsm_db *pDb){
  int rc = LSM_OK;
  if( pDb ){
    assert_db_state(pDb);
    if( pDb->pCsr || pDb->nTransOpen ){
      rc = LSM_MISUSE_BKPT;
    }else{
      lsmFreeSnapshot(pDb->pEnv, pDb->pClient);
      pDb->pClient = 0;
      lsmDbDatabaseRelease(pDb);
      lsmFsClose(pDb->pFS);
      lsmFree(pDb->pEnv, pDb->aTrans);
      lsmFree(pDb->pEnv, pDb->apShm);
      lsmFree(pDb->pEnv, pDb);
    }
  }

*/
int lsmBeginReadTrans(lsm_db *pDb){
  const int MAX_READLOCK_ATTEMPTS = 10;
  int rc = LSM_OK;                /* Return code */
  int iAttempt = 0;

  assert( pDb->pWorker==0 );
  assert( (pDb->pClient!=0)==(pDb->iReader>=0) );

  while( rc==LSM_OK && pDb->pClient==0 && (iAttempt++)<MAX_READLOCK_ATTEMPTS ){
    int iTreehdr = 0;
    int iSnap = 0;
    assert( pDb->pCsr==0 && pDb->nTransOpen==0 );

    /* Load the in-memory tree header. */
    rc = lsmTreeLoadHeader(pDb, &iTreehdr);

    /* Load the database snapshot */
    if( rc==LSM_OK ){



      rc = lsmCheckpointLoad(pDb, &iSnap);



    }

    /* Take a read-lock on the tree and snapshot just loaded. Then check
    ** that the shared-memory still contains the same values. If so, proceed.
    ** Otherwise, relinquish the read-lock and retry the whole procedure
    ** (starting with loading the in-memory tree header).  */
    if( rc==LSM_OK ){
................................................................................
        if( lsmTreeLoadHeaderOk(pDb, iTreehdr)
         && lsmCheckpointLoadOk(pDb, iSnap)
        ){
          /* Read lock has been successfully obtained. Deserialize the 
          ** checkpoint just loaded. TODO: This will be removed after 
          ** lsm_sorted.c is changed to work directly from the serialized
          ** version of the snapshot.  */

          rc = lsmCheckpointDeserialize(pDb, 0, pDb->aSnapshot, &pDb->pClient);

          assert( (rc==LSM_OK)==(pDb->pClient!=0) );
          assert( pDb->iReader>=0 );
        }else{
          rc = lsmReleaseReadlock(pDb);
        }
      }
      if( rc==LSM_BUSY ){
................................................................................
      (int)pDb->treehdr.root.iTransId,
      (int)pDb->treehdr.iOldShmid
  );
  fflush(stdout);
}
#endif
  }
  if( pDb->pClient==0 && rc==LSM_OK ) rc = LSM_BUSY;





  return rc;
}

/*
** Close the currently open read transaction.
*/
void lsmFinishReadTrans(lsm_db *pDb){
................................................................................
  /* Worker connections should not be closing read transactions. And
  ** read transactions should only be closed after all cursors and write
  ** transactions have been closed. Finally pClient should be non-NULL
  ** only iff pDb->iReader>=0.  */
  assert( pDb->pWorker==0 );
  assert( pDb->pCsr==0 && pDb->nTransOpen==0 );


  if( pClient ){
    lsmFreeSnapshot(pDb->pEnv, pDb->pClient);
    pDb->pClient = 0;
  }

  if( pDb->iReader>=0 ) lsmReleaseReadlock(pDb);
  assert( (pDb->pClient!=0)==(pDb->iReader>=0) );
}

/*
** Open a write transaction.
*/
int lsmBeginWriteTrans(lsm_db *pDb){
  int rc;                         /* Return code */
................................................................................

/*
** Release the read-lock currently held by connection db.
*/
int lsmReleaseReadlock(lsm_db *db){
  int rc = LSM_OK;
  if( db->iReader>=0 ){
    assert( db->pClient==0 );
    rc = lsmShmLock(db, LSM_LOCK_READER(db->iReader), LSM_LOCK_UNLOCK, 0);
    db->iReader = -1;
  }
  return rc;
}

/*

*/
int lsmBeginReadTrans(lsm_db *pDb){
  const int MAX_READLOCK_ATTEMPTS = 10;
  int rc = LSM_OK;                /* Return code */
  int iAttempt = 0;

  assert( pDb->pWorker==0 );


  while( rc==LSM_OK && pDb->iReader<0 && (iAttempt++)<MAX_READLOCK_ATTEMPTS ){
    int iTreehdr = 0;
    int iSnap = 0;
    assert( pDb->pCsr==0 && pDb->nTransOpen==0 );

    /* Load the in-memory tree header. */
    rc = lsmTreeLoadHeader(pDb, &iTreehdr);

    /* Load the database snapshot */
    if( rc==LSM_OK ){
      if( lsmCheckpointClientCacheOk(pDb)==0 ){
        lsmFreeSnapshot(pDb->pEnv, pDb->pClient);
        pDb->pClient = 0;
        rc = lsmCheckpointLoad(pDb, &iSnap);
      }else{
        iSnap = 1;
      }
    }

    /* Take a read-lock on the tree and snapshot just loaded. Then check
    ** that the shared-memory still contains the same values. If so, proceed.
    ** Otherwise, relinquish the read-lock and retry the whole procedure
    ** (starting with loading the in-memory tree header).  */
    if( rc==LSM_OK ){
................................................................................
        if( lsmTreeLoadHeaderOk(pDb, iTreehdr)
         && lsmCheckpointLoadOk(pDb, iSnap)
        ){
          /* Read lock has been successfully obtained. Deserialize the 
          ** checkpoint just loaded. TODO: This will be removed after 
          ** lsm_sorted.c is changed to work directly from the serialized
          ** version of the snapshot.  */
          if( pDb->pClient==0 ){
            rc = lsmCheckpointDeserialize(pDb, 0, pDb->aSnapshot,&pDb->pClient);
          }
          assert( (rc==LSM_OK)==(pDb->pClient!=0) );
          assert( pDb->iReader>=0 );
        }else{
          rc = lsmReleaseReadlock(pDb);
        }
      }
      if( rc==LSM_BUSY ){
................................................................................
      (int)pDb->treehdr.root.iTransId,
      (int)pDb->treehdr.iOldShmid
  );
  fflush(stdout);
}
#endif
  }


  if( rc!=LSM_OK ){
    lsmReleaseReadlock(pDb);
  }
  if( pDb->pClient==0 && rc==LSM_OK ) rc = LSM_BUSY;
  return rc;
}

/*
** Close the currently open read transaction.
*/
void lsmFinishReadTrans(lsm_db *pDb){
................................................................................
  /* Worker connections should not be closing read transactions. And
  ** read transactions should only be closed after all cursors and write
  ** transactions have been closed. Finally pClient should be non-NULL
  ** only iff pDb->iReader>=0.  */
  assert( pDb->pWorker==0 );
  assert( pDb->pCsr==0 && pDb->nTransOpen==0 );

#if 0
  if( pClient ){
    lsmFreeSnapshot(pDb->pEnv, pDb->pClient);
    pDb->pClient = 0;
  }
#endif
  if( pDb->iReader>=0 ) lsmReleaseReadlock(pDb);

}

/*
** Open a write transaction.
*/
int lsmBeginWriteTrans(lsm_db *pDb){
  int rc;                         /* Return code */
................................................................................

/*
** Release the read-lock currently held by connection db.
*/
int lsmReleaseReadlock(lsm_db *db){
  int rc = LSM_OK;
  if( db->iReader>=0 ){

    rc = lsmShmLock(db, LSM_LOCK_READER(db->iReader), LSM_LOCK_UNLOCK, 0);
    db->iReader = -1;
  }
  return rc;
}

/*

  int flags;                      /* Mask of CURSOR_XXX flags */
  int (*xCmp)(void *, int, void *, int);         /* Compare function */
  int eType;                      /* Cache of current key type */
  Blob key;                       /* Cache of current key (or NULL) */
  Blob val;                       /* Cache of current value */


  TreeCursor *apTreeCsr[2];       /* One or two tree cursors */
  int nSegCsr;                    /* Size of aSegCsr[] array */
  LevelCursor *aSegCsr;           /* Array of cursors open on sorted files */
  int nTree;
  int *aTree;
  BtreeCursor *pBtCsr;

  int flags;                      /* Mask of CURSOR_XXX flags */
  int (*xCmp)(void *, int, void *, int);         /* Compare function */
  int eType;                      /* Cache of current key type */
  Blob key;                       /* Cache of current key (or NULL) */
  Blob val;                       /* Cache of current value */

  /* All the component cursors: */
  TreeCursor *apTreeCsr[2];       /* One or two tree cursors */
  int nSegCsr;                    /* Size of aSegCsr[] array */
  LevelCursor *aSegCsr;           /* Array of cursors open on sorted files */
  int nTree;
  int *aTree;
  BtreeCursor *pBtCsr;