SQLite

/*
** 2001 September 15
**
** The author disclaims copyright to this source code.  In place of
** a legal notice, here is a blessing:
**
**    May you do good and not evil.
**    May you find forgiveness for yourself and forgive others.
**    May you share freely, never taking more than you give.
**
*************************************************************************
** $Id: btree.c,v 1.69 2002/08/11 20:10:47 drh Exp $
**
** This file implements a external (disk-based) database using BTrees.
** For a detailed discussion of BTrees, refer to
**
**     Donald E. Knuth, THE ART OF COMPUTER PROGRAMMING, Volume 3:
**     "Sorting And Searching", pages 473-480. Addison-Wesley
**     Publishing Company, Reading, Massachusetts.

** 1, not 0.)  Thus a minimum database contains 2 pages.
*/
#include "sqliteInt.h"
#include "pager.h"
#include "btree.h"
#include <assert.h>

/*
** Macros used for byteswapping.  B is a pointer to the Btree
** structure.  This is needed to access the Btree.needSwab boolean
** in order to tell if byte swapping is needed or not.
** X is an unsigned integer.  SWAB16 byte swaps a 16-bit integer.
** SWAB32 byteswaps a 32-bit integer.
*/
#define SWAB16(B,X)   ((B)->needSwab? swab16(X) : (X))
#define SWAB32(B,X)   ((B)->needSwab? swab32(X) : (X))
#define SWAB_ADD(B,X,A) \
   if((B)->needSwab){ X=swab32(swab32(X)+A); }else{ X += (A); }

/*
** The following global variable - available only if SQLITE_TEST is
** defined - is used to determine whether new databases are created in
** native byte order or in non-native byte order.  Non-native byte order
** databases are created for testing purposes only.  Under normal operation,
** only native byte-order databases should be created, but we should be
** able to read or write existing databases regardless of the byteorder.
*/
#ifdef SQLITE_TEST
int btree_native_byte_order = 1;
#endif

/*
** Forward declarations of structures used only in this file.
*/
typedef struct PageOne PageOne;
typedef struct MemPage MemPage;
typedef struct PageHdr PageHdr;
typedef struct Cell Cell;

/*
** The key and data size are split into a lower 16-bit segment and an
** upper 8-bit segment in order to pack them together into a smaller
** space.  The following macros reassembly a key or data size back
** into an integer.
*/
#define NKEY(b,h)  (SWAB16(b,h.nKey) + h.nKeyHi*65536)
#define NDATA(b,h) (SWAB16(b,h.nData) + h.nDataHi*65536)

/*
** The minimum size of a complete Cell.  The Cell must contain a header
** and at least 4 bytes of payload.
*/
#define MIN_CELL_SIZE  (sizeof(CellHdr)+4)

struct Btree {
  Pager *pPager;        /* The page cache */
  BtCursor *pCursor;    /* A list of all open cursors */
  PageOne *page1;       /* First page of the database */
  u8 inTrans;           /* True if a transaction is in progress */
  u8 inCkpt;            /* True if there is a checkpoint on the transaction */
  u8 readOnly;          /* True if the underlying file is readonly */
  u8 needSwab;          /* Need to byte-swapping */
  Hash locks;           /* Key: root page number.  Data: lock count */
};
typedef Btree Bt;

/*
** A cursor is a pointer to a particular entry in the BTree.
** The entry is identified by its MemPage and the index in
** MemPage.apCell[] of the entry.
*/
struct BtCursor {
  Btree *pBt;               /* The Btree to which this cursor belongs */
  BtCursor *pNext, *pPrev;  /* Forms a linked list of all cursors */
  Pgno pgnoRoot;            /* The root page of this tree */
  MemPage *pPage;           /* Page that contains the entry */
  int idx;                  /* Index of the entry in pPage->apCell[] */
  u8 wrFlag;                /* True if writable */
  u8 bSkipNext;             /* sqliteBtreeNext() is no-op if true */
  u8 iMatch;                /* compare result from last sqliteBtreeMoveto() */
};

/*
** Routines for byte swapping.
*/
u16 swab16(u16 x){
  return ((x & 0xff)<<8) | ((x>>8)&0xff);
}
u32 swab32(u32 x){
  return ((x & 0xff)<<24) | ((x & 0xff00)<<8) |
         ((x>>8) & 0xff00) | ((x>>24)&0xff);
}

/*
** Compute the total number of bytes that a Cell needs on the main
** database page.  The number returned includes the Cell header,
** local payload storage, and the pointer to overflow pages (if
** applicable).  Additional space allocated on overflow pages
** is NOT included in the value returned from this routine.
*/
static int cellSize(Btree *pBt, Cell *pCell){
  int n = NKEY(pBt, pCell->h) + NDATA(pBt, pCell->h);
  if( n>MX_LOCAL_PAYLOAD ){
    n = MX_LOCAL_PAYLOAD + sizeof(Pgno);
  }else{
    n = ROUNDUP(n);
  }
  n += sizeof(CellHdr);
  return n;
}

/*
** Defragment the page given.  All Cells are moved to the
** beginning of the page and all free space is collected 
** into one big FreeBlk at the end of the page.
*/
static void defragmentPage(Btree *pBt, MemPage *pPage){
  int pc, i, n;
  FreeBlk *pFBlk;
  char newPage[SQLITE_PAGE_SIZE];

  assert( sqlitepager_iswriteable(pPage) );
  assert( pPage->isInit );
  pc = sizeof(PageHdr);
  pPage->u.hdr.firstCell = SWAB16(pBt, pc);
  memcpy(newPage, pPage->u.aDisk, pc);
  for(i=0; i<pPage->nCell; i++){
    Cell *pCell = pPage->apCell[i];

    /* This routine should never be called on an overfull page.  The
    ** following asserts verify that constraint. */
    assert( Addr(pCell) > Addr(pPage) );
    assert( Addr(pCell) < Addr(pPage) + SQLITE_PAGE_SIZE );

    n = cellSize(pBt, pCell);
    pCell->h.iNext = SWAB16(pBt, pc + n);
    memcpy(&newPage[pc], pCell, n);
    pPage->apCell[i] = (Cell*)&pPage->u.aDisk[pc];
    pc += n;
  }
  assert( pPage->nFree==SQLITE_PAGE_SIZE-pc );
  memcpy(pPage->u.aDisk, newPage, pc);
  if( pPage->nCell>0 ){
    pPage->apCell[pPage->nCell-1]->h.iNext = 0;
  }
  pFBlk = (FreeBlk*)&pPage->u.aDisk[pc];
  pFBlk->iSize = SWAB16(pBt, SQLITE_PAGE_SIZE - pc);
  pFBlk->iNext = 0;
  pPage->u.hdr.firstFree = SWAB16(pBt, pc);
  memset(&pFBlk[1], 0, SQLITE_PAGE_SIZE - pc - sizeof(FreeBlk));
}

/*
** Allocate nByte bytes of space on a page.  nByte must be a 
** multiple of 4.
**
** Return the index into pPage->u.aDisk[] of the first byte of
** the new allocation. Or return 0 if there is not enough free
** space on the page to satisfy the allocation request.
**
** If the page contains nBytes of free space but does not contain
** nBytes of contiguous free space, then this routine automatically
** calls defragementPage() to consolidate all free space before 
** allocating the new chunk.
*/
static int allocateSpace(Btree *pBt, MemPage *pPage, int nByte){
  FreeBlk *p;
  u16 *pIdx;
  int start;
  int cnt = 0;
  int iSize;

  assert( sqlitepager_iswriteable(pPage) );
  assert( nByte==ROUNDUP(nByte) );
  assert( pPage->isInit );
  if( pPage->nFree<nByte || pPage->isOverfull ) return 0;
  pIdx = &pPage->u.hdr.firstFree;
  p = (FreeBlk*)&pPage->u.aDisk[SWAB16(pBt, *pIdx)];
  while( (iSize = SWAB16(pBt, p->iSize))<nByte ){
    assert( cnt++ < SQLITE_PAGE_SIZE/4 );
    if( p->iNext==0 ){
      defragmentPage(pBt, pPage);
      pIdx = &pPage->u.hdr.firstFree;
    }else{
      pIdx = &p->iNext;
    }
    p = (FreeBlk*)&pPage->u.aDisk[SWAB16(pBt, *pIdx)];
  }
  if( iSize==nByte ){
    start = SWAB16(pBt, *pIdx);
    *pIdx = p->iNext;
  }else{
    FreeBlk *pNew;
    start = SWAB16(pBt, *pIdx);
    pNew = (FreeBlk*)&pPage->u.aDisk[start + nByte];
    pNew->iNext = p->iNext;
    pNew->iSize = SWAB16(pBt, iSize - nByte);
    *pIdx = SWAB16(pBt, start + nByte);
  }
  pPage->nFree -= nByte;
  return start;
}

/*
** Return a section of the MemPage.u.aDisk[] to the freelist.
** The first byte of the new free block is pPage->u.aDisk[start]
** and the size of the block is "size" bytes.  Size must be
** a multiple of 4.
**
** Most of the effort here is involved in coalesing adjacent
** free blocks into a single big free block.
*/
static void freeSpace(Btree *pBt, MemPage *pPage, int start, int size){
  int end = start + size;
  u16 *pIdx, idx;
  FreeBlk *pFBlk;
  FreeBlk *pNew;
  FreeBlk *pNext;
  int iSize;

  assert( sqlitepager_iswriteable(pPage) );
  assert( size == ROUNDUP(size) );
  assert( start == ROUNDUP(start) );
  assert( pPage->isInit );
  pIdx = &pPage->u.hdr.firstFree;
  idx = SWAB16(pBt, *pIdx);
  while( idx!=0 && idx<start ){
    pFBlk = (FreeBlk*)&pPage->u.aDisk[idx];
    iSize = SWAB16(pBt, pFBlk->iSize);
    if( idx + iSize == start ){
      pFBlk->iSize = SWAB16(pBt, iSize + size);
      if( idx + iSize + size == SWAB16(pBt, pFBlk->iNext) ){
        pNext = (FreeBlk*)&pPage->u.aDisk[idx + iSize + size];
        if( pBt->needSwab ){
          pFBlk->iSize = swab16(swab16(pNext->iSize)+iSize+size);
        }else{
          pFBlk->iSize += pNext->iSize;
        }
        pFBlk->iNext = pNext->iNext;
      }
      pPage->nFree += size;
      return;
    }
    pIdx = &pFBlk->iNext;
    idx = SWAB16(pBt, *pIdx);
  }
  pNew = (FreeBlk*)&pPage->u.aDisk[start];
  if( idx != end ){
    pNew->iSize = SWAB16(pBt, size);
    pNew->iNext = SWAB16(pBt, idx);
  }else{
    pNext = (FreeBlk*)&pPage->u.aDisk[idx];
    pNew->iSize = SWAB16(pBt, size + SWAB16(pBt, pNext->iSize));
    pNew->iNext = pNext->iNext;
  }
  *pIdx = SWAB16(pBt, start);
  pPage->nFree += size;
}

/*
** Initialize the auxiliary information for a disk block.
**
** The pParent parameter must be a pointer to the MemPage which
** is the parent of the page being initialized.  The root of the
** BTree (usually page 2) has no parent and so for that page, 
** pParent==NULL.
**
** Return SQLITE_OK on success.  If we see that the page does
** not contained a well-formed database page, then return 
** SQLITE_CORRUPT.  Note that a return of SQLITE_OK does not
** guarantee that the page is well-formed.  It only shows that
** we failed to detect any corruption.
*/
static int initPage(Bt *pBt, MemPage *pPage, Pgno pgnoThis, MemPage *pParent){
  int idx;           /* An index into pPage->u.aDisk[] */
  Cell *pCell;       /* A pointer to a Cell in pPage->u.aDisk[] */
  FreeBlk *pFBlk;    /* A pointer to a free block in pPage->u.aDisk[] */
  int sz;            /* The size of a Cell in bytes */
  int freeSpace;     /* Amount of free space on the page */

  if( pPage->pParent ){
    assert( pPage->pParent==pParent );
    return SQLITE_OK;
  }
  if( pParent ){
    pPage->pParent = pParent;
    sqlitepager_ref(pParent);
  }
  if( pPage->isInit ) return SQLITE_OK;
  pPage->isInit = 1;
  pPage->nCell = 0;
  freeSpace = USABLE_SPACE;
  idx = SWAB16(pBt, pPage->u.hdr.firstCell);
  while( idx!=0 ){
    if( idx>SQLITE_PAGE_SIZE-MIN_CELL_SIZE ) goto page_format_error;
    if( idx<sizeof(PageHdr) ) goto page_format_error;
    if( idx!=ROUNDUP(idx) ) goto page_format_error;
    pCell = (Cell*)&pPage->u.aDisk[idx];
    sz = cellSize(pBt, pCell);
    if( idx+sz > SQLITE_PAGE_SIZE ) goto page_format_error;
    freeSpace -= sz;
    pPage->apCell[pPage->nCell++] = pCell;
    idx = SWAB16(pBt, pCell->h.iNext);
  }
  pPage->nFree = 0;
  idx = SWAB16(pBt, pPage->u.hdr.firstFree);
  while( idx!=0 ){
    int iNext;
    if( idx>SQLITE_PAGE_SIZE-sizeof(FreeBlk) ) goto page_format_error;
    if( idx<sizeof(PageHdr) ) goto page_format_error;
    pFBlk = (FreeBlk*)&pPage->u.aDisk[idx];
    pPage->nFree += SWAB16(pBt, pFBlk->iSize);
    iNext = SWAB16(pBt, pFBlk->iNext);
    if( iNext>0 && iNext <= idx ) goto page_format_error;
    idx = iNext;
  }
  if( pPage->nCell==0 && pPage->nFree==0 ){
    /* As a special case, an uninitialized root page appears to be
    ** an empty database */
    return SQLITE_OK;
  }
  if( pPage->nFree!=freeSpace ) goto page_format_error;
  return SQLITE_OK;

page_format_error:
  return SQLITE_CORRUPT;
}

/*
** Set up a raw page so that it looks like a database page holding
** no entries.
*/
static void zeroPage(Btree *pBt, MemPage *pPage){
  PageHdr *pHdr;
  FreeBlk *pFBlk;
  assert( sqlitepager_iswriteable(pPage) );
  memset(pPage, 0, SQLITE_PAGE_SIZE);
  pHdr = &pPage->u.hdr;
  pHdr->firstCell = 0;
  pHdr->firstFree = SWAB16(pBt, sizeof(*pHdr));
  pFBlk = (FreeBlk*)&pHdr[1];
  pFBlk->iNext = 0;
  pPage->nFree = SQLITE_PAGE_SIZE - sizeof(*pHdr);
  pFBlk->iSize = SWAB16(pBt, pPage->nFree);
  pPage->nCell = 0;
  pPage->isOverfull = 0;
}

/*
** This routine is called when the reference count for a page
** reaches zero.  We need to unref the pParent pointer when that

  if( rc!=SQLITE_OK ) return rc;

  /* Do some checking to help insure the file we opened really is
  ** a valid database file. 
  */
  if( sqlitepager_pagecount(pBt->pPager)>0 ){
    PageOne *pP1 = pBt->page1;
    if( strcmp(pP1->zMagic,zMagicHeader)!=0 ||
          (pP1->iMagic!=MAGIC && swab32(pP1->iMagic)!=MAGIC) ){
      rc = SQLITE_CORRUPT;
      goto page1_init_failed;
    }
    pBt->needSwab = pP1->iMagic!=MAGIC;
  }
  return rc;

page1_init_failed:
  sqlitepager_unref(pBt->page1);
  pBt->page1 = 0;
  return rc;

  if( rc ) return rc;
  rc = sqlitepager_write(pRoot);
  if( rc ){
    sqlitepager_unref(pRoot);
    return rc;
  }
  strcpy(pP1->zMagic, zMagicHeader);
#ifdef SQLITE_TEST
  if( btree_native_byte_order ){
    pP1->iMagic = MAGIC;
    pBt->needSwab = 0;
  }else{
    pP1->iMagic = swab32(MAGIC);
    pBt->needSwab = 1;
  }
#else
  pP1->iMagic = MAGIC;
  pBt->needSwab = 0;
#endif
  zeroPage(pBt, pRoot);
  sqlitepager_unref(pRoot);
  return SQLITE_OK;
}

/*
** Attempt to start a new transaction.
**

    goto create_cursor_exception;
  }
  pCur->pgnoRoot = (Pgno)iTable;
  rc = sqlitepager_get(pBt->pPager, pCur->pgnoRoot, (void**)&pCur->pPage);
  if( rc!=SQLITE_OK ){
    goto create_cursor_exception;
  }
  rc = initPage(pBt, pCur->pPage, pCur->pgnoRoot, 0);
  if( rc!=SQLITE_OK ){
    goto create_cursor_exception;
  }
  nLock = (ptr)sqliteHashFind(&pBt->locks, 0, iTable);
  if( nLock<0 || (nLock>0 && wrFlag) ){
    rc = SQLITE_LOCKED;
    goto create_cursor_exception;

  MemPage *pPage;

  pPage = pCur->pPage;
  if( pPage==0 || pCur->idx >= pPage->nCell ){
    *pSize = 0;
  }else{
    pCell = pPage->apCell[pCur->idx];
    *pSize = NKEY(pCur->pBt, pCell->h);
  }
  return SQLITE_OK;
}

/*
** Read payload information from the entry that the pCur cursor is
** pointing to.  Begin reading the payload at "offset" and read
** a total of "amt" bytes.  Put the result in zBuf.
**
** This routine does not make a distinction between key and data.
** It just reads bytes from the payload area.
*/
static int getPayload(BtCursor *pCur, int offset, int amt, char *zBuf){
  char *aPayload;
  Pgno nextPage;
  int rc;
  Btree *pBt = pCur->pBt;
  assert( pCur!=0 && pCur->pPage!=0 );
  assert( pCur->idx>=0 && pCur->idx<pCur->pPage->nCell );
  aPayload = pCur->pPage->apCell[pCur->idx]->aPayload;
  if( offset<MX_LOCAL_PAYLOAD ){
    int a = amt;
    if( a+offset>MX_LOCAL_PAYLOAD ){
      a = MX_LOCAL_PAYLOAD - offset;
    }
    memcpy(zBuf, &aPayload[offset], a);
    if( a==amt ){
      return SQLITE_OK;
    }
    offset = 0;
    zBuf += a;
    amt -= a;
  }else{
    offset -= MX_LOCAL_PAYLOAD;
  }
  if( amt>0 ){
    nextPage = SWAB32(pBt, pCur->pPage->apCell[pCur->idx]->ovfl);
  }
  while( amt>0 && nextPage ){
    OverflowPage *pOvfl;
    rc = sqlitepager_get(pBt->pPager, nextPage, (void**)&pOvfl);
    if( rc!=0 ){
      return rc;
    }
    nextPage = SWAB32(pBt, pOvfl->iNext);
    if( offset<OVERFLOW_SIZE ){
      int a = amt;
      if( a + offset > OVERFLOW_SIZE ){
        a = OVERFLOW_SIZE - offset;
      }
      memcpy(zBuf, &pOvfl->aPayload[offset], a);
      offset = 0;

  if( amt==0 ) return 0;
  pPage = pCur->pPage;
  if( pPage==0 ) return 0;
  if( pCur->idx >= pPage->nCell ){
    return 0;
  }
  pCell = pPage->apCell[pCur->idx];
  if( amt+offset > NKEY(pCur->pBt, pCell->h) ){
    amt = NKEY(pCur->pBt, pCell->h) - offset;
    if( amt<=0 ){
      return 0;
    }
  }
  getPayload(pCur, offset, amt, zBuf);
  return amt;
}

  MemPage *pPage;

  pPage = pCur->pPage;
  if( pPage==0 || pCur->idx >= pPage->nCell ){
    *pSize = 0;
  }else{
    pCell = pPage->apCell[pCur->idx];
    *pSize = NDATA(pCur->pBt, pCell->h);
  }
  return SQLITE_OK;
}

/*
** Read part of the data associated with cursor pCur.  A maximum
** of "amt" bytes will be transfered into zBuf[].  The transfer
** begins at "offset".  The number of bytes actually read is
** returned.  The amount returned will be smaller than the
** amount requested if there are not enough bytes in the data
** to satisfy the request.
*/
int sqliteBtreeData(BtCursor *pCur, int offset, int amt, char *zBuf){
  Cell *pCell;
  MemPage *pPage;
  int nData;

  if( amt<0 ) return 0;
  if( offset<0 ) return 0;
  if( amt==0 ) return 0;
  pPage = pCur->pPage;
  if( pPage==0 || pCur->idx >= pPage->nCell ){
    return 0;
  }
  pCell = pPage->apCell[pCur->idx];
  nData = NDATA(pCur->pBt, pCell->h);
  if( amt+offset > nData ){
    amt = nData - offset;
    if( amt<=0 ){
      return 0;
    }
  }
  getPayload(pCur, offset + NKEY(pCur->pBt, pCell->h), amt, zBuf);
  return amt;
}

/*
** Compare an external key against the key on the entry that pCur points to.
**
** The external key is pKey and is nKey bytes long.  The last nIgnore bytes

  int nKey,             /* Number of bytes in pKey */
  int nIgnore,          /* Ignore this many bytes at the end of pCur */
  int *pResult          /* Write the result here */
){
  Pgno nextPage;
  int n, c, rc, nLocal;
  Cell *pCell;
  Btree *pBt = pCur->pBt;
  const char *zKey  = (const char*)pKey;

  assert( pCur->pPage );
  assert( pCur->idx>=0 && pCur->idx<pCur->pPage->nCell );
  pCell = pCur->pPage->apCell[pCur->idx];
  nLocal = NKEY(pBt, pCell->h) - nIgnore;
  if( nLocal<0 ) nLocal = 0;
  n = nKey<nLocal ? nKey : nLocal;
  if( n>MX_LOCAL_PAYLOAD ){
    n = MX_LOCAL_PAYLOAD;
  }
  c = memcmp(pCell->aPayload, zKey, n);
  if( c!=0 ){
    *pResult = c;
    return SQLITE_OK;
  }
  zKey += n;
  nKey -= n;
  nLocal -= n;
  nextPage = SWAB32(pBt, pCell->ovfl);
  while( nKey>0 && nLocal>0 ){
    OverflowPage *pOvfl;
    if( nextPage==0 ){
      return SQLITE_CORRUPT;
    }
    rc = sqlitepager_get(pBt->pPager, nextPage, (void**)&pOvfl);
    if( rc ){
      return rc;
    }
    nextPage = SWAB32(pBt, pOvfl->iNext);
    n = nKey<nLocal ? nKey : nLocal;
    if( n>OVERFLOW_SIZE ){
      n = OVERFLOW_SIZE;
    }
    c = memcmp(pOvfl->aPayload, zKey, n);
    sqlitepager_unref(pOvfl);
    if( c!=0 ){

/*
** Move the cursor down to a new child page.
*/
static int moveToChild(BtCursor *pCur, int newPgno){
  int rc;
  MemPage *pNewPage;
  Btree *pBt = pCur->pBt;

  rc = sqlitepager_get(pBt->pPager, newPgno, (void**)&pNewPage);
  if( rc ) return rc;
  rc = initPage(pBt, pNewPage, newPgno, pCur->pPage);
  if( rc ) return rc;
  sqlitepager_unref(pCur->pPage);
  pCur->pPage = pNewPage;
  pCur->idx = 0;
  return SQLITE_OK;
}

  pParent = pCur->pPage->pParent;
  if( pParent==0 ) return SQLITE_INTERNAL;
  oldPgno = sqlitepager_pagenumber(pCur->pPage);
  sqlitepager_ref(pParent);
  sqlitepager_unref(pCur->pPage);
  pCur->pPage = pParent;
  pCur->idx = pParent->nCell;
  oldPgno = SWAB32(pCur->pBt, oldPgno);
  for(i=0; i<pParent->nCell; i++){
    if( pParent->apCell[i]->h.leftChild==oldPgno ){
      pCur->idx = i;
      break;
    }
  }
  return SQLITE_OK;
}

/*
** Move the cursor to the root page
*/
static int moveToRoot(BtCursor *pCur){
  MemPage *pNew;
  int rc;
  Btree *pBt = pCur->pBt;

  rc = sqlitepager_get(pBt->pPager, pCur->pgnoRoot, (void**)&pNew);
  if( rc ) return rc;
  rc = initPage(pBt, pNew, pCur->pgnoRoot, 0);
  if( rc ) return rc;
  sqlitepager_unref(pCur->pPage);
  pCur->pPage = pNew;
  pCur->idx = 0;
  return SQLITE_OK;
}

/*
** Move the cursor down to the left-most leaf entry beneath the
** entry to which it is currently pointing.
*/
static int moveToLeftmost(BtCursor *pCur){
  Pgno pgno;
  int rc;

  while( (pgno = pCur->pPage->apCell[pCur->idx]->h.leftChild)!=0 ){
    rc = moveToChild(pCur, SWAB32(pCur->pBt, pgno));
    if( rc ) return rc;
  }
  return SQLITE_OK;
}

/* Move the cursor to the first entry in the table.  Return SQLITE_OK
** on success.  Set *pRes to 0 if the cursor actually points to something

  assert( pCur->pPage->isInit );
  if( pCur->pPage->nCell==0 ){
    *pRes = 1;
    return SQLITE_OK;
  }
  *pRes = 0;
  while( (pgno = pCur->pPage->u.hdr.rightChild)!=0 ){
    rc = moveToChild(pCur, SWAB32(pCur->pBt, pgno));
    if( rc ) return rc;
  }
  pCur->idx = pCur->pPage->nCell-1;
  pCur->bSkipNext = 0;
  return rc;
}

      chldPg = pPage->apCell[lwr]->h.leftChild;
    }
    if( chldPg==0 ){
      pCur->iMatch = c;
      if( pRes ) *pRes = c;
      return SQLITE_OK;
    }
    rc = moveToChild(pCur, SWAB32(pCur->pBt, chldPg));
    if( rc ) return rc;
  }
  /* NOT REACHED */
}

/*
** Advance the cursor to the next entry in the database.  If

    pCur->bSkipNext = 0;
    if( pRes ) *pRes = 0;
    return SQLITE_OK;
  }
  pCur->idx++;
  if( pCur->idx>=pCur->pPage->nCell ){
    if( pCur->pPage->u.hdr.rightChild ){
      rc = moveToChild(pCur, SWAB32(pCur->pBt, pCur->pPage->u.hdr.rightChild));
      if( rc ) return rc;
      rc = moveToLeftmost(pCur);
      if( rc ) return rc;
      if( pRes ) *pRes = 0;
      return SQLITE_OK;
    }
    do{

  int rc;
  if( pPage1->freeList ){
    OverflowPage *pOvfl;
    FreelistInfo *pInfo;

    rc = sqlitepager_write(pPage1);
    if( rc ) return rc;
    SWAB_ADD(pBt, pPage1->nFree, -1);
    rc = sqlitepager_get(pBt->pPager, SWAB32(pBt, pPage1->freeList),
                        (void**)&pOvfl);
    if( rc ) return rc;
    rc = sqlitepager_write(pOvfl);
    if( rc ){
      sqlitepager_unref(pOvfl);
      return rc;
    }
    pInfo = (FreelistInfo*)pOvfl->aPayload;
    if( pInfo->nFree==0 ){
      *pPgno = SWAB32(pBt, pPage1->freeList);
      pPage1->freeList = pOvfl->iNext;
      *ppPage = (MemPage*)pOvfl;
    }else{
      int closest, n;
      n = SWAB32(pBt, pInfo->nFree);
      if( n>1 && nearby>0 ){
        int i, dist;
        closest = 0;
        dist = SWAB32(pBt, pInfo->aFree[0]) - nearby;
        if( dist<0 ) dist = -dist;
        for(i=1; i<n; i++){
          int d2 = SWAB32(pBt, pInfo->aFree[i]) - nearby;
          if( d2<0 ) d2 = -d2;
          if( d2<dist ) closest = i;
        }
      }else{
        closest = 0;
      }
      SWAB_ADD(pBt, pInfo->nFree, -1);
      *pPgno = SWAB32(pBt, pInfo->aFree[closest]);
      pInfo->aFree[closest] = pInfo->aFree[n-1];
      rc = sqlitepager_get(pBt->pPager, *pPgno, (void**)ppPage);
      sqlitepager_unref(pOvfl);
      if( rc==SQLITE_OK ){
        sqlitepager_dont_rollback(*ppPage);
        rc = sqlitepager_write(*ppPage);
      }
    }

    sqlitepager_unref(pMemPage->pParent);
    pMemPage->pParent = 0;
  }
  rc = sqlitepager_write(pPage1);
  if( rc ){
    return rc;
  }
  SWAB_ADD(pBt, pPage1->nFree, 1);
  if( pPage1->nFree!=0 && pPage1->freeList!=0 ){
    OverflowPage *pFreeIdx;
    rc = sqlitepager_get(pBt->pPager, SWAB32(pBt, pPage1->freeList),
                        (void**)&pFreeIdx);
    if( rc==SQLITE_OK ){
      FreelistInfo *pInfo = (FreelistInfo*)pFreeIdx->aPayload;
      int n = SWAB32(pBt, pInfo->nFree);
      if( n<(sizeof(pInfo->aFree)/sizeof(pInfo->aFree[0])) ){
        rc = sqlitepager_write(pFreeIdx);
        if( rc==SQLITE_OK ){
          pInfo->aFree[n] = SWAB32(pBt, pgno);
          SWAB_ADD(pBt, pInfo->nFree, 1);
          sqlitepager_unref(pFreeIdx);
          sqlitepager_dont_write(pBt->pPager, pgno);
          return rc;
        }
      }
      sqlitepager_unref(pFreeIdx);
    }
  }
  if( pOvfl==0 ){
    assert( pgno>0 );
    rc = sqlitepager_get(pBt->pPager, pgno, (void**)&pOvfl);
    if( rc ) return rc;
    needUnref = 1;
  }
  rc = sqlitepager_write(pOvfl);
  if( rc ){
    if( needUnref ) sqlitepager_unref(pOvfl);
    return rc;
  }
  pOvfl->iNext = pPage1->freeList;
  pPage1->freeList = SWAB32(pBt, pgno);
  memset(pOvfl->aPayload, 0, OVERFLOW_SIZE);
  if( needUnref ) rc = sqlitepager_unref(pOvfl);
  return rc;
}

/*
** Erase all the data out of a cell.  This involves returning overflow
** pages back the freelist.
*/
static int clearCell(Btree *pBt, Cell *pCell){
  Pager *pPager = pBt->pPager;
  OverflowPage *pOvfl;
  Pgno ovfl, nextOvfl;
  int rc;

  if( NKEY(pBt, pCell->h) + NDATA(pBt, pCell->h) <= MX_LOCAL_PAYLOAD ){
    return SQLITE_OK;
  }
  ovfl = SWAB32(pBt, pCell->ovfl);
  pCell->ovfl = 0;
  while( ovfl ){
    rc = sqlitepager_get(pPager, ovfl, (void**)&pOvfl);
    if( rc ) return rc;
    nextOvfl = SWAB32(pBt, pOvfl->iNext);
    rc = freePage(pBt, pOvfl, ovfl);
    if( rc ) return rc;
    sqlitepager_unref(pOvfl);
    ovfl = nextOvfl;
  }
  return SQLITE_OK;
}

  int n, rc;
  int nPayload;
  const char *pPayload;
  char *pSpace;
  Pgno nearby = 0;

  pCell->h.leftChild = 0;
  pCell->h.nKey = SWAB16(pBt, nKey & 0xffff);
  pCell->h.nKeyHi = nKey >> 16;
  pCell->h.nData = SWAB16(pBt, nData & 0xffff);
  pCell->h.nDataHi = nData >> 16;
  pCell->h.iNext = 0;

  pNext = &pCell->ovfl;
  pSpace = pCell->aPayload;
  spaceLeft = MX_LOCAL_PAYLOAD;
  pPayload = pKey;

        nearby = *pNext;
      }
      if( pPrior ) sqlitepager_unref(pPrior);
      if( rc ){
        clearCell(pBt, pCell);
        return rc;
      }
      if( pBt->needSwab ) *pNext = swab32(*pNext);
      pPrior = pOvfl;
      spaceLeft = OVERFLOW_SIZE;
      pSpace = pOvfl->aPayload;
      pNext = &pOvfl->iNext;
    }
    n = nPayload;
    if( n>spaceLeft ) n = spaceLeft;

** Reparent all children of the given page to be the given page.
** In other words, for every child of pPage, invoke reparentPage()
** to make sure that each child knows that pPage is its parent.
**
** This routine gets called after you memcpy() one page into
** another.
*/
static void reparentChildPages(Btree *pBt, MemPage *pPage){
  int i;
  Pager *pPager = pBt->pPager;
  for(i=0; i<pPage->nCell; i++){
    reparentPage(pPager, SWAB32(pBt, pPage->apCell[i]->h.leftChild), pPage);
  }
  reparentPage(pPager, SWAB32(pBt, pPage->u.hdr.rightChild), pPage);
}

/*
** Remove the i-th cell from pPage.  This routine effects pPage only.
** The cell content is not freed or deallocated.  It is assumed that
** the cell content has been copied someplace else.  This routine just
** removes the reference to the cell from pPage.
**
** "sz" must be the number of bytes in the cell.
**
** Do not bother maintaining the integrity of the linked list of Cells.
** Only the pPage->apCell[] array is important.  The relinkCellList() 
** routine will be called soon after this routine in order to rebuild 
** the linked list.
*/
static void dropCell(Btree *pBt, MemPage *pPage, int idx, int sz){
  int j;
  assert( idx>=0 && idx<pPage->nCell );
  assert( sz==cellSize(pBt, pPage->apCell[idx]) );
  assert( sqlitepager_iswriteable(pPage) );
  freeSpace(pBt, pPage, Addr(pPage->apCell[idx]) - Addr(pPage), sz);
  for(j=idx; j<pPage->nCell-1; j++){
    pPage->apCell[j] = pPage->apCell[j+1];
  }
  pPage->nCell--;
}

/*
** Insert a new cell on pPage at cell index "i".  pCell points to the
** content of the cell.
**
** If the cell content will fit on the page, then put it there.  If it
** will not fit, then just make pPage->apCell[i] point to the content
** and set pPage->isOverfull.  
**
** Do not bother maintaining the integrity of the linked list of Cells.
** Only the pPage->apCell[] array is important.  The relinkCellList() 
** routine will be called soon after this routine in order to rebuild 
** the linked list.
*/
static void insertCell(Btree *pBt, MemPage *pPage, int i, Cell *pCell, int sz){
  int idx, j;
  assert( i>=0 && i<=pPage->nCell );
  assert( sz==cellSize(pBt, pCell) );
  assert( sqlitepager_iswriteable(pPage) );
  idx = allocateSpace(pBt, pPage, sz);
  for(j=pPage->nCell; j>i; j--){
    pPage->apCell[j] = pPage->apCell[j-1];
  }
  pPage->nCell++;
  if( idx<=0 ){
    pPage->isOverfull = 1;
    pPage->apCell[i] = pCell;
  }else{
    memcpy(&pPage->u.aDisk[idx], pCell, sz);
    pPage->apCell[i] = (Cell*)&pPage->u.aDisk[idx];
  }
}

/*
** Rebuild the linked list of cells on a page so that the cells
** occur in the order specified by the pPage->apCell[] array.  
** Invoke this routine once to repair damage after one or more
** invocations of either insertCell() or dropCell().
*/
static void relinkCellList(Btree *pBt, MemPage *pPage){
  int i;
  u16 *pIdx;
  assert( sqlitepager_iswriteable(pPage) );
  pIdx = &pPage->u.hdr.firstCell;
  for(i=0; i<pPage->nCell; i++){
    int idx = Addr(pPage->apCell[i]) - Addr(pPage);
    assert( idx>0 && idx<SQLITE_PAGE_SIZE );
    *pIdx = SWAB16(pBt, idx);
    pIdx = &pPage->apCell[i]->h.iNext;
  }
  *pIdx = 0;
}

/*
** Make a copy of the contents of pFrom into pTo.  The pFrom->apCell[]

  int iCur;                    /* apCell[iCur] is the cell of the cursor */
  MemPage *pOldCurPage;        /* The cursor originally points to this page */
  int totalSize;               /* Total bytes for all cells */
  int subtotal;                /* Subtotal of bytes in cells on one page */
  int cntNew[4];               /* Index in apCell[] of cell after i-th page */
  int szNew[4];                /* Combined size of cells place on i-th page */
  MemPage *extraUnref = 0;     /* A page that needs to be unref-ed */
  Pgno pgno, swabPgno;         /* Page number */
  Cell *apCell[MX_CELL*3+5];   /* All cells from pages being balanceed */
  int szCell[MX_CELL*3+5];     /* Local size of all cells */
  Cell aTemp[2];               /* Temporary holding area for apDiv[] */
  MemPage aOld[3];             /* Temporary copies of pPage and its siblings */

  /* 
  ** Return without doing any work if pPage is neither overfull nor
  ** underfull.
  */
  assert( sqlitepager_iswriteable(pPage) );
  if( !pPage->isOverfull && pPage->nFree<SQLITE_PAGE_SIZE/2 
        && pPage->nCell>=2){
    relinkCellList(pBt, pPage);
    return SQLITE_OK;
  }

  /*
  ** Find the parent of the page to be balanceed.
  ** If there is no parent, it means this page is the root page and
  ** special rules apply.
  */
  pParent = pPage->pParent;
  if( pParent==0 ){
    Pgno pgnoChild;
    MemPage *pChild;
    assert( pPage->isInit );
    if( pPage->nCell==0 ){
      if( pPage->u.hdr.rightChild ){
        /*
        ** The root page is empty.  Copy the one child page
        ** into the root page and return.  This reduces the depth
        ** of the BTree by one.
        */
        pgnoChild = SWAB32(pBt, pPage->u.hdr.rightChild);
        rc = sqlitepager_get(pBt->pPager, pgnoChild, (void**)&pChild);
        if( rc ) return rc;
        memcpy(pPage, pChild, SQLITE_PAGE_SIZE);
        pPage->isInit = 0;
        rc = initPage(pBt, pPage, sqlitepager_pagenumber(pPage), 0);
        assert( rc==SQLITE_OK );
        reparentChildPages(pBt, pPage);
        if( pCur && pCur->pPage==pChild ){
          sqlitepager_unref(pChild);
          pCur->pPage = pPage;
          sqlitepager_ref(pPage);
        }
        freePage(pBt, pChild, pgnoChild);
        sqlitepager_unref(pChild);
      }else{
        relinkCellList(pBt, pPage);
      }
      return SQLITE_OK;
    }
    if( !pPage->isOverfull ){
      /* It is OK for the root page to be less than half full.
      */
      relinkCellList(pBt, pPage);
      return SQLITE_OK;
    }
    /*
    ** If we get to here, it means the root page is overfull.
    ** When this happens, Create a new child page and copy the
    ** contents of the root into the child.  Then make the root
    ** page an empty page with rightChild pointing to the new

    pChild->isOverfull = 1;
    if( pCur && pCur->pPage==pPage ){
      sqlitepager_unref(pPage);
      pCur->pPage = pChild;
    }else{
      extraUnref = pChild;
    }
    zeroPage(pBt, pPage);
    pPage->u.hdr.rightChild = SWAB32(pBt, pgnoChild);
    pParent = pPage;
    pPage = pChild;
  }
  rc = sqlitepager_write(pParent);
  if( rc ) return rc;
  assert( pParent->isInit );
  
  /*
  ** Find the Cell in the parent page whose h.leftChild points back
  ** to pPage.  The "idx" variable is the index of that cell.  If pPage
  ** is the rightmost child of pParent then set idx to pParent->nCell 
  */
  idx = -1;
  pgno = sqlitepager_pagenumber(pPage);
  swabPgno = SWAB32(pBt, pgno);
  for(i=0; i<pParent->nCell; i++){
    if( pParent->apCell[i]->h.leftChild==swabPgno ){
      idx = i;
      break;
    }
  }
  if( idx<0 && pParent->u.hdr.rightChild==swabPgno ){
    idx = pParent->nCell;
  }
  if( idx<0 ){
    return SQLITE_CORRUPT;
  }

  /*

  if( nxDiv<0 ) nxDiv = 0;
  nDiv = 0;
  for(i=0, k=nxDiv; i<3; i++, k++){
    if( k<pParent->nCell ){
      idxDiv[i] = k;
      apDiv[i] = pParent->apCell[k];
      nDiv++;
      pgnoOld[i] = SWAB32(pBt, apDiv[i]->h.leftChild);
    }else if( k==pParent->nCell ){
      pgnoOld[i] = SWAB32(pBt, pParent->u.hdr.rightChild);
    }else{
      break;
    }
    rc = sqlitepager_get(pBt->pPager, pgnoOld[i], (void**)&apOld[i]);
    if( rc ) goto balance_cleanup;
    rc = initPage(pBt, apOld[i], pgnoOld[i], pParent);
    if( rc ) goto balance_cleanup;
    nOld++;
  }

  /*
  ** Set iCur to be the index in apCell[] of the cell that the cursor
  ** is pointing to.  We will need this later on in order to keep the

  ** into aTemp[] and remove the the divider Cells from pParent.
  */
  nCell = 0;
  for(i=0; i<nOld; i++){
    MemPage *pOld = &aOld[i];
    for(j=0; j<pOld->nCell; j++){
      apCell[nCell] = pOld->apCell[j];
      szCell[nCell] = cellSize(pBt, apCell[nCell]);
      nCell++;
    }
    if( i<nOld-1 ){
      szCell[nCell] = cellSize(pBt, apDiv[i]);
      memcpy(&aTemp[i], apDiv[i], szCell[nCell]);
      apCell[nCell] = &aTemp[i];
      dropCell(pBt, pParent, nxDiv, szCell[nCell]);
      assert( SWAB32(pBt, apCell[nCell]->h.leftChild)==pgnoOld[i] );
      apCell[nCell]->h.leftChild = pOld->u.hdr.rightChild;
      nCell++;
    }
  }

  /*
  ** Figure out the number of pages needed to hold all nCell cells.

      apOld[i] = 0;
      sqlitepager_write(apNew[i]);
    }else{
      rc = allocatePage(pBt, &apNew[i], &pgnoNew[i], pgnoNew[i-1]);
      if( rc ) goto balance_cleanup;
    }
    nNew++;
    zeroPage(pBt, apNew[i]);
    apNew[i]->isInit = 1;
  }

  /* Free any old pages that were not reused as new pages.
  */
  while( i<nOld ){
    rc = freePage(pBt, apOld[i], pgnoOld[i]);

  */
  j = 0;
  for(i=0; i<nNew; i++){
    MemPage *pNew = apNew[i];
    while( j<cntNew[i] ){
      assert( pNew->nFree>=szCell[j] );
      if( pCur && iCur==j ){ pCur->pPage = pNew; pCur->idx = pNew->nCell; }
      insertCell(pBt, pNew, pNew->nCell, apCell[j], szCell[j]);
      j++;
    }
    assert( pNew->nCell>0 );
    assert( !pNew->isOverfull );
    relinkCellList(pBt, pNew);
    if( i<nNew-1 && j<nCell ){
      pNew->u.hdr.rightChild = apCell[j]->h.leftChild;
      apCell[j]->h.leftChild = SWAB32(pBt, pgnoNew[i]);
      if( pCur && iCur==j ){ pCur->pPage = pParent; pCur->idx = nxDiv; }
      insertCell(pBt, pParent, nxDiv, apCell[j], szCell[j]);
      j++;
      nxDiv++;
    }
  }
  assert( j==nCell );
  apNew[nNew-1]->u.hdr.rightChild = aOld[nOld-1].u.hdr.rightChild;
  if( nxDiv==pParent->nCell ){
    pParent->u.hdr.rightChild = SWAB32(pBt, pgnoNew[nNew-1]);
  }else{
    pParent->apCell[nxDiv]->h.leftChild = SWAB32(pBt, pgnoNew[nNew-1]);
  }
  if( pCur ){
    if( j<=iCur && pCur->pPage==pParent && pCur->idx>idxDiv[nOld-1] ){
      assert( pCur->pPage==pOldCurPage );
      pCur->idx += nNew - nOld;
    }else{
      assert( pOldCurPage!=0 );
      sqlitepager_ref(pCur->pPage);
      sqlitepager_unref(pOldCurPage);
    }
  }

  /*
  ** Reparent children of all cells.
  */
  for(i=0; i<nNew; i++){
    reparentChildPages(pBt, apNew[i]);
  }
  reparentChildPages(pBt, pParent);

  /*
  ** balance the parent page.
  */
  rc = balance(pBt, pParent, pCur);

  /*

  if( rc ) return rc;
  pPage = pCur->pPage;
  assert( pPage->isInit );
  rc = sqlitepager_write(pPage);
  if( rc ) return rc;
  rc = fillInCell(pBt, &newCell, pKey, nKey, pData, nData);
  if( rc ) return rc;
  szNew = cellSize(pBt, &newCell);
  if( loc==0 ){
    newCell.h.leftChild = pPage->apCell[pCur->idx]->h.leftChild;
    rc = clearCell(pBt, pPage->apCell[pCur->idx]);
    if( rc ) return rc;
    dropCell(pBt, pPage, pCur->idx, cellSize(pBt, pPage->apCell[pCur->idx]));
  }else if( loc<0 && pPage->nCell>0 ){
    assert( pPage->u.hdr.rightChild==0 );  /* Must be a leaf page */
    pCur->idx++;
  }else{
    assert( pPage->u.hdr.rightChild==0 );  /* Must be a leaf page */
  }
  insertCell(pBt, pPage, pCur->idx, &newCell, szNew);
  rc = balance(pCur->pBt, pPage, pCur);
  /* sqliteBtreePageDump(pCur->pBt, pCur->pgnoRoot, 1); */
  /* fflush(stdout); */
  return rc;
}

/*

** pointing to the first entry after the deleted entry.
*/
int sqliteBtreeDelete(BtCursor *pCur){
  MemPage *pPage = pCur->pPage;
  Cell *pCell;
  int rc;
  Pgno pgnoChild;
  Btree *pBt = pCur->pBt;

  assert( pPage->isInit );
  if( pCur->pPage==0 ){
    return SQLITE_ABORT;  /* A rollback destroyed this cursor */
  }
  if( !pCur->pBt->inTrans ){
    return SQLITE_ERROR;  /* Must start a transaction first */
  }
  if( pCur->idx >= pPage->nCell ){
    return SQLITE_ERROR;  /* The cursor is not pointing to anything */
  }
  if( !pCur->wrFlag ){
    return SQLITE_PERM;   /* Did not open this cursor for writing */
  }
  rc = sqlitepager_write(pPage);
  if( rc ) return rc;
  pCell = pPage->apCell[pCur->idx];
  pgnoChild = SWAB32(pBt, pCell->h.leftChild);
  clearCell(pBt, pCell);
  if( pgnoChild ){
    /*
    ** The entry we are about to delete is not a leaf so if we do not
    ** do something we will leave a hole on an internal page.
    ** We have to fill the hole by moving in a cell from a leaf.  The
    ** next Cell after the one to be deleted is guaranteed to exist and
    ** to be a leaf so we can use it.
    */
    BtCursor leafCur;
    Cell *pNext;
    int szNext;
    getTempCursor(pCur, &leafCur);
    rc = sqliteBtreeNext(&leafCur, 0);
    if( rc!=SQLITE_OK ){
      return SQLITE_CORRUPT;
    }
    rc = sqlitepager_write(leafCur.pPage);
    if( rc ) return rc;
    dropCell(pBt, pPage, pCur->idx, cellSize(pBt, pCell));
    pNext = leafCur.pPage->apCell[leafCur.idx];
    szNext = cellSize(pBt, pNext);
    pNext->h.leftChild = SWAB32(pBt, pgnoChild);
    insertCell(pBt, pPage, pCur->idx, pNext, szNext);
    rc = balance(pBt, pPage, pCur);
    if( rc ) return rc;
    pCur->bSkipNext = 1;
    dropCell(pBt, leafCur.pPage, leafCur.idx, szNext);
    rc = balance(pBt, leafCur.pPage, pCur);
    releaseTempCursor(&leafCur);
  }else{
    dropCell(pBt, pPage, pCur->idx, cellSize(pBt, pCell));
    if( pCur->idx>=pPage->nCell ){
      pCur->idx = pPage->nCell-1;
      if( pCur->idx<0 ){ 
        pCur->idx = 0;
        pCur->bSkipNext = 1;
      }else{
        pCur->bSkipNext = 0;
      }
    }else{
      pCur->bSkipNext = 1;
    }
    rc = balance(pBt, pPage, pCur);
  }
  return rc;
}

/*
** Create a new BTree table.  Write into *piTable the page
** number for the root page of the new table.

  }
  if( pBt->readOnly ){
    return SQLITE_READONLY;
  }
  rc = allocatePage(pBt, &pRoot, &pgnoRoot, 0);
  if( rc ) return rc;
  assert( sqlitepager_iswriteable(pRoot) );
  zeroPage(pBt, pRoot);
  sqlitepager_unref(pRoot);
  *piTable = (int)pgnoRoot;
  return SQLITE_OK;
}

/*
** Create a new BTree index.  Write into *piTable the page

  Cell *pCell;
  int idx;

  rc = sqlitepager_get(pBt->pPager, pgno, (void**)&pPage);
  if( rc ) return rc;
  rc = sqlitepager_write(pPage);
  if( rc ) return rc;
  rc = initPage(pBt, pPage, pgno, 0);
  if( rc ) return rc;
  idx = SWAB16(pBt, pPage->u.hdr.firstCell);
  while( idx>0 ){
    pCell = (Cell*)&pPage->u.aDisk[idx];
    idx = SWAB16(pBt, pCell->h.iNext);
    if( pCell->h.leftChild ){
      rc = clearDatabasePage(pBt, SWAB32(pBt, pCell->h.leftChild), 1);
      if( rc ) return rc;
    }
    rc = clearCell(pBt, pCell);
    if( rc ) return rc;
  }
  if( pPage->u.hdr.rightChild ){
    rc = clearDatabasePage(pBt, SWAB32(pBt, pPage->u.hdr.rightChild), 1);
    if( rc ) return rc;
  }
  if( freePageFlag ){
    rc = freePage(pBt, pPage, pgno);
  }else{
    zeroPage(pBt, pPage);
  }
  sqlitepager_unref(pPage);
  return rc;
}

/*
** Delete all information from a single table in the database.

  rc = sqlitepager_get(pBt->pPager, (Pgno)iTable, (void**)&pPage);
  if( rc ) return rc;
  rc = sqliteBtreeClearTable(pBt, iTable);
  if( rc ) return rc;
  if( iTable>2 ){
    rc = freePage(pBt, pPage, iTable);
  }else{
    zeroPage(pBt, pPage);
  }
  sqlitepager_unref(pPage);
  return rc;  
}

/*
** Read the meta-information out of a database file.
*/
int sqliteBtreeGetMeta(Btree *pBt, int *aMeta){
  PageOne *pP1;
  int rc;
  int i;

  rc = sqlitepager_get(pBt->pPager, 1, (void**)&pP1);
  if( rc ) return rc;
  aMeta[0] = SWAB32(pBt, pP1->nFree);
  for(i=0; i<sizeof(pP1->aMeta)/sizeof(pP1->aMeta[0]); i++){
    aMeta[i+1] = SWAB32(pBt, pP1->aMeta[i]);
  }
  sqlitepager_unref(pP1);
  return SQLITE_OK;
}

/*
** Write meta-information back into the database.
*/
int sqliteBtreeUpdateMeta(Btree *pBt, int *aMeta){
  PageOne *pP1;
  int rc, i;
  if( !pBt->inTrans ){
    return SQLITE_ERROR;  /* Must start a transaction first */
  }
  if( pBt->readOnly ){
    return SQLITE_READONLY;
  }
  pP1 = pBt->page1;
  rc = sqlitepager_write(pP1);
  if( rc ) return rc;   
  for(i=0; i<sizeof(pP1->aMeta)/sizeof(pP1->aMeta[0]); i++){
    pP1->aMeta[i] = SWAB32(pBt, aMeta[i+1]);
  }
  return SQLITE_OK;
}

/******************************************************************************
** The complete implementation of the BTree subsystem is above this line.
** All the code the follows is for testing and troubleshooting the BTree
** subsystem.  None of the code that follows is used during normal operation.

  unsigned char payload[20];
  rc = sqlitepager_get(pBt->pPager, (Pgno)pgno, (void**)&pPage);
  if( rc ){
    return rc;
  }
  if( recursive ) printf("PAGE %d:\n", pgno);
  i = 0;
  idx = SWAB16(pBt, pPage->u.hdr.firstCell);
  while( idx>0 && idx<=SQLITE_PAGE_SIZE-MIN_CELL_SIZE ){
    Cell *pCell = (Cell*)&pPage->u.aDisk[idx];
    int sz = cellSize(pBt, pCell);
    sprintf(range,"%d..%d", idx, idx+sz-1);
    sz = NKEY(pBt, pCell->h) + NDATA(pBt, pCell->h);
    if( sz>sizeof(payload)-1 ) sz = sizeof(payload)-1;
    memcpy(payload, pCell->aPayload, sz);
    for(j=0; j<sz; j++){
      if( payload[j]<0x20 || payload[j]>0x7f ) payload[j] = '.';
    }
    payload[sz] = 0;
    printf(
      "cell %2d: i=%-10s chld=%-4d nk=%-4d nd=%-4d payload=%s\n",
      i, range, (int)pCell->h.leftChild, 
      NKEY(pBt, pCell->h), NDATA(pBt, pCell->h),
      payload
    );
    if( pPage->isInit && pPage->apCell[i]!=pCell ){
      printf("**** apCell[%d] does not match on prior entry ****\n", i);
    }
    i++;
    idx = SWAB16(pBt, pCell->h.iNext);
  }
  if( idx!=0 ){
    printf("ERROR: next cell index out of range: %d\n", idx);
  }
  printf("right_child: %d\n", SWAB32(pBt, pPage->u.hdr.rightChild));
  nFree = 0;
  i = 0;
  idx = SWAB16(pBt, pPage->u.hdr.firstFree);
  while( idx>0 && idx<SQLITE_PAGE_SIZE ){
    FreeBlk *p = (FreeBlk*)&pPage->u.aDisk[idx];
    sprintf(range,"%d..%d", idx, idx+p->iSize-1);
    nFree += SWAB16(pBt, p->iSize);
    printf("freeblock %2d: i=%-10s size=%-4d total=%d\n",
       i, range, SWAB16(pBt, p->iSize), nFree);
    idx = SWAB16(pBt, p->iNext);
    i++;
  }
  if( idx!=0 ){
    printf("ERROR: next freeblock index out of range: %d\n", idx);
  }
  if( recursive && pPage->u.hdr.rightChild!=0 ){
    idx = SWAB16(pBt, pPage->u.hdr.firstCell);
    while( idx>0 && idx<SQLITE_PAGE_SIZE-MIN_CELL_SIZE ){
      Cell *pCell = (Cell*)&pPage->u.aDisk[idx];
      sqliteBtreePageDump(pBt, SWAB32(pBt, pCell->h.leftChild), 1);
      idx = SWAB16(pBt, pCell->h.iNext);
    }
    sqliteBtreePageDump(pBt, SWAB32(pBt, pPage->u.hdr.rightChild), 1);
  }
  sqlitepager_unref(pPage);
  return SQLITE_OK;
}
#endif

#ifdef SQLITE_TEST

**   aResult[7] =  Page number of the right child for the whole page
**
** This routine is used for testing and debugging only.
*/
int sqliteBtreeCursorDump(BtCursor *pCur, int *aResult){
  int cnt, idx;
  MemPage *pPage = pCur->pPage;
  Btree *pBt = pCur->pBt;
  aResult[0] = sqlitepager_pagenumber(pPage);
  aResult[1] = pCur->idx;
  aResult[2] = pPage->nCell;
  if( pCur->idx>=0 && pCur->idx<pPage->nCell ){
    aResult[3] = cellSize(pBt, pPage->apCell[pCur->idx]);
    aResult[6] = SWAB32(pBt, pPage->apCell[pCur->idx]->h.leftChild);
  }else{
    aResult[3] = 0;
    aResult[6] = 0;
  }
  aResult[4] = pPage->nFree;
  cnt = 0;
  idx = SWAB16(pBt, pPage->u.hdr.firstFree);
  while( idx>0 && idx<SQLITE_PAGE_SIZE ){
    cnt++;
    idx = SWAB16(pBt, ((FreeBlk*)&pPage->u.aDisk[idx])->iNext);
  }
  aResult[5] = cnt;
  aResult[7] = SWAB32(pBt, pPage->u.hdr.rightChild);
  return SQLITE_OK;
}
#endif

#ifdef SQLITE_TEST
/*
** Return the pager associated with a BTree.  This routine is used for

    if( sqlitepager_get(pCheck->pPager, (Pgno)iPage, (void**)&pOvfl) ){
      sprintf(zMsg, "failed to get page %d", iPage);
      checkAppendMsg(pCheck, zContext, zMsg);
      break;
    }
    if( isFreeList ){
      FreelistInfo *pInfo = (FreelistInfo*)pOvfl->aPayload;
      int n = SWAB32(pCheck->pBt, pInfo->nFree);
      for(i=0; i<n; i++){
        checkRef(pCheck, SWAB32(pCheck->pBt, pInfo->aFree[i]), zMsg);
      }
      N -= n;
    }
    iPage = SWAB32(pCheck->pBt, pOvfl->iNext);
    sqlitepager_unref(pOvfl);
  }
}

/*
** Return negative if zKey1<zKey2.
** Return zero if zKey1==zKey2.

  int nUpper            /* Number of characters in zUpperBound */
){
  MemPage *pPage;
  int i, rc, depth, d2, pgno;
  char *zKey1, *zKey2;
  int nKey1, nKey2;
  BtCursor cur;
  Btree *pBt;
  char zMsg[100];
  char zContext[100];
  char hit[SQLITE_PAGE_SIZE];

  /* Check that the page exists
  */
  cur.pBt = pBt = pCheck->pBt;
  if( iPage==0 ) return 0;
  if( checkRef(pCheck, iPage, zParentContext) ) return 0;
  sprintf(zContext, "On tree page %d: ", iPage);
  if( (rc = sqlitepager_get(pCheck->pPager, (Pgno)iPage, (void**)&pPage))!=0 ){
    sprintf(zMsg, "unable to get the page. error code=%d", rc);
    checkAppendMsg(pCheck, zContext, zMsg);
    return 0;
  }
  if( (rc = initPage(pBt, pPage, (Pgno)iPage, pParent))!=0 ){
    sprintf(zMsg, "initPage() returns error code %d", rc);
    checkAppendMsg(pCheck, zContext, zMsg);
    sqlitepager_unref(pPage);
    return 0;
  }

  /* Check out all the cells.
  */
  depth = 0;
  if( zLowerBound ){
    zKey1 = sqliteMalloc( nLower+1 );
    memcpy(zKey1, zLowerBound, nLower);
    zKey1[nLower] = 0;
  }else{
    zKey1 = 0;
  }
  nKey1 = nLower;
  cur.pPage = pPage;

  for(i=0; i<pPage->nCell; i++){
    Cell *pCell = pPage->apCell[i];
    int sz;

    /* Check payload overflow pages
    */
    nKey2 = NKEY(pBt, pCell->h);
    sz = nKey2 + NDATA(pBt, pCell->h);
    sprintf(zContext, "On page %d cell %d: ", iPage, i);
    if( sz>MX_LOCAL_PAYLOAD ){
      int nPage = (sz - MX_LOCAL_PAYLOAD + OVERFLOW_SIZE - 1)/OVERFLOW_SIZE;
      checkList(pCheck, 0, SWAB32(pBt, pCell->ovfl), nPage, zContext);
    }

    /* Check that keys are in the right order
    */
    cur.idx = i;
    zKey2 = sqliteMalloc( nKey2+1 );
    getPayload(&cur, 0, nKey2, zKey2);
    if( zKey1 && keyCompare(zKey1, nKey1, zKey2, nKey2)>=0 ){
      checkAppendMsg(pCheck, zContext, "Key is out of order");
    }

    /* Check sanity of left child page.
    */
    pgno = SWAB32(pBt, pCell->h.leftChild);
    d2 = checkTreePage(pCheck, pgno, pPage, zContext, zKey1,nKey1,zKey2,nKey2);
    if( i>0 && d2!=depth ){
      checkAppendMsg(pCheck, zContext, "Child page depth differs");
    }
    depth = d2;
    sqliteFree(zKey1);
    zKey1 = zKey2;
    nKey1 = nKey2;
  }
  pgno = SWAB32(pBt, pPage->u.hdr.rightChild);
  sprintf(zContext, "On page %d at right child: ", iPage);
  checkTreePage(pCheck, pgno, pPage, zContext, zKey1,nKey1,zUpperBound,nUpper);
  sqliteFree(zKey1);
 
  /* Check for complete coverage of the page
  */
  memset(hit, 0, sizeof(hit));
  memset(hit, 1, sizeof(PageHdr));
  for(i=SWAB16(pBt, pPage->u.hdr.firstCell); i>0 && i<SQLITE_PAGE_SIZE; ){
    Cell *pCell = (Cell*)&pPage->u.aDisk[i];
    int j;
    for(j=i+cellSize(pBt, pCell)-1; j>=i; j--) hit[j]++;
    i = SWAB16(pBt, pCell->h.iNext);
  }
  for(i=SWAB16(pBt,pPage->u.hdr.firstFree); i>0 && i<SQLITE_PAGE_SIZE; ){
    FreeBlk *pFBlk = (FreeBlk*)&pPage->u.aDisk[i];
    int j;
    for(j=i+SWAB16(pBt,pFBlk->iSize)-1; j>=i; j--) hit[j]++;
    i = SWAB16(pBt,pFBlk->iNext);
  }
  for(i=0; i<SQLITE_PAGE_SIZE; i++){
    if( hit[i]==0 ){
      sprintf(zMsg, "Unused space at byte %d of page %d", i, iPage);
      checkAppendMsg(pCheck, zMsg, 0);
      break;
    }else if( hit[i]>1 ){

  sCheck.anRef = sqliteMalloc( (sCheck.nPage+1)*sizeof(sCheck.anRef[0]) );
  sCheck.anRef[1] = 1;
  for(i=2; i<=sCheck.nPage; i++){ sCheck.anRef[i] = 0; }
  sCheck.zErrMsg = 0;

  /* Check the integrity of the freelist
  */
  checkList(&sCheck, 1, SWAB32(pBt, pBt->page1->freeList),
            SWAB32(pBt, pBt->page1->nFree), "Main freelist: ");

  /* Check all the tables.
  */
  for(i=0; i<nRoot; i++){
    if( aRoot[i]==0 ) continue;
    checkTreePage(&sCheck, aRoot[i], 0, "List of tree roots: ", 0,0,0,0);
  }

**    May you share freely, never taking more than you give.
**
*************************************************************************
** This header file defines the interface that the sqlite B-Tree file
** subsystem.  See comments in the source code for a detailed description
** of what each interface routine does.
**
** @(#) $Id: btree.h,v 1.25 2002/08/11 20:10:48 drh Exp $
*/
#ifndef _BTREE_H_
#define _BTREE_H_

typedef struct Btree Btree;
typedef struct BtCursor BtCursor;

char *sqliteBtreeIntegrityCheck(Btree*, int*, int);

#ifdef SQLITE_TEST
int sqliteBtreePageDump(Btree*, int, int);
int sqliteBtreeCursorDump(BtCursor*, int*);
struct Pager *sqliteBtreePager(Btree*);
int btree_native_byte_order;
#endif

#endif /* _BTREE_H_ */

**
*************************************************************************
** This file contains SQLite's grammar for SQL.  Process this file
** using the lemon parser generator to generate C code that runs
** the parser.  Lemon will also generate a header file containing
** numeric codes for all of the tokens.
**
** @(#) $Id: parse.y,v 1.80 2002/08/11 20:10:48 drh Exp $
*/
%token_prefix TK_
%token_type {Token}
%default_type {Token}
%extra_argument {Parse *pParse}
%syntax_error {
  sqliteSetString(&pParse->zErrMsg,"syntax error",0);

carg ::= DEFAULT PLUS FLOAT(X).      {sqliteAddDefaultValue(pParse,&X,0);}
carg ::= DEFAULT MINUS FLOAT(X).     {sqliteAddDefaultValue(pParse,&X,1);}
carg ::= DEFAULT NULL. 

// In addition to the type name, we also care about the primary key and
// UNIQUE constraints.
//
ccons ::= NULL onconf.
ccons ::= NOT NULL onconf(R).               {sqliteAddNotNull(pParse, R);}
ccons ::= PRIMARY KEY sortorder onconf(R).  {sqliteAddPrimaryKey(pParse,0,R);}
ccons ::= UNIQUE onconf(R).            {sqliteCreateIndex(pParse,0,0,0,R,0,0);}
ccons ::= CHECK LP expr RP onconf.
ccons ::= references.
ccons ::= defer_subclause.
ccons ::= COLLATE id(C).  {

**    May you share freely, never taking more than you give.
**
*************************************************************************
** Code for testing the btree.c module in SQLite.  This code
** is not included in the SQLite library.  It is used for automated
** testing of the SQLite library.
**
** $Id: test3.c,v 1.18 2002/08/11 20:10:48 drh Exp $
*/
#include "sqliteInt.h"
#include "pager.h"
#include "btree.h"
#include "tcl.h"
#include <stdlib.h>
#include <string.h>

  aRoot = malloc( sizeof(int)*(argc-2) );
  for(i=0; i<argc-2; i++){
    if( Tcl_GetInt(interp, argv[i+2], &aRoot[i]) ) return TCL_ERROR;
  }
  zResult = sqliteBtreeIntegrityCheck(pBt, aRoot, nRoot);
  if( zResult ){
    Tcl_AppendResult(interp, zResult, 0);
    sqliteFree(zResult); 
  }
  return TCL_OK;
}

/*
** Usage:   btree_cursor ID TABLENUM WRITEABLE
**

  Tcl_CreateCommand(interp, "btree_key", btree_key, 0, 0);
  Tcl_CreateCommand(interp, "btree_data", btree_data, 0, 0);
  Tcl_CreateCommand(interp, "btree_payload_size", btree_payload_size, 0, 0);
  Tcl_CreateCommand(interp, "btree_first", btree_first, 0, 0);
  Tcl_CreateCommand(interp, "btree_cursor_dump", btree_cursor_dump, 0, 0);
  Tcl_CreateCommand(interp, "btree_integrity_check", btree_integrity_check,0,0);
  Tcl_LinkVar(interp, "pager_refinfo_enable", (char*)&pager_refinfo_enable,
     TCL_LINK_INT);
  Tcl_LinkVar(interp, "btree_native_byte_order",(char*)&btree_native_byte_order,
     TCL_LINK_INT);
  return TCL_OK;
}

# 2001 September 15
#
# The author disclaims copyright to this source code.  In place of
# a legal notice, here is a blessing:
#
#    May you do good and not evil.
#    May you find forgiveness for yourself and forgive others.
#    May you share freely, never taking more than you give.
#
#***********************************************************************
# This file runs all tests.
#
# $Id: all.test,v 1.16 2002/08/11 20:10:49 drh Exp $

set testdir [file dirname $argv0]
source $testdir/tester.tcl
rename finish_test really_finish_test
proc finish_test {} {memleak_check}

if {[file exists ./sqlite_test_count]} {
  set COUNT [exec cat ./sqlite_test_count]
} else {
  set COUNT 4
}

# LeakList will hold a list of the number of unfreed mallocs after
# each round of the test.  This number should be constant.  If it
# grows, it may mean there is a memory leak in the library.
#
set LeakList {}

set EXCLUDE {
  all.test
  quick.test
  malloc.test
  misuse.test
}
#  btree2.test

for {set Counter 0} {$Counter<$COUNT && $nErr==0} {incr Counter} {
  set btree_native_byte_order [expr {($Counter>>1)&0x1}]
  if {$Counter%2} {
    set ::SETUP_SQL {PRAGMA default_synchronous=off;}
  } else {
    catch {unset ::SETUP_SQL}
  }
  foreach testfile [lsort -dictionary [glob $testdir/*.test]] {
    set tail [file tail $testfile]

#    May you share freely, never taking more than you give.
#
#***********************************************************************
# This file implements regression tests for SQLite library.  The
# focus of this file is testing the ability of the library to detect
# past or future file format version numbers and respond appropriately.
#
# $Id: version.test,v 1.5 2002/08/11 20:10:49 drh Exp $

set testdir [file dirname $argv0]
source $testdir/tester.tcl

# Current file format version
set VX 3

} {1 2 3 4 5 6 7 8}

# Make sure the version number is set correctly
#
do_test version-1.2 {
  db close
  set ::bt [btree_open test.db]
  btree_begin_transaction $::bt
  set ::meta [btree_get_meta $::bt]
  btree_rollback $::bt
  lindex $::meta 2
} $VX

# Increase the file_format number by one.  Verify that the
# file will refuse to open.
#
do_test version-1.3 {

#
# Run this script to generated a faq.html output file
#
set rcsid {$Id: faq.tcl,v 1.14 2002/08/11 20:10:49 drh Exp $}

puts {<html>
<head>
  <title>SQLite Frequently Asked Questions</title>
</head>
<body bgcolor="white">
<h1 align="center">Frequently Asked Questions</h1>

  </ol>
}
        
faq {
  My linux box is not able to read an SQLite database that was created
  on my SparcStation.
} {
  <p>You need to upgrade your SQLite library to version 2.6.3 or later.</p>

  <p>The x86 processor on your linux box is little-endian (meaning that
  the least significant byte of integers comes first) but the Sparc is
  big-endian (the most significant bytes comes first).  SQLite databases
  created on a little-endian architecture cannot be on a big-endian
  machine by version 2.6.2 or earlier of SQLite.  Beginning with
  version 2.6.3, SQLite should be able to read and write database files




  regardless of byte order of the machine on which the file was created.</p>









}

faq {
  Can multiple applications or multiple instances of the same
  application access a single database file at the same time?
} {
  <p>Multiple processes can have the same database open at the same

#
# Run this TCL script to generate HTML for the index.html file.
#
set rcsid {$Id: index.tcl,v 1.61 2002/08/11 20:10:49 drh Exp $}

puts {<html>
<head><title>SQLite: An SQL Database Engine In A C Library</title></head>
<body bgcolor=white>
<h1 align=center>SQLite: An SQL Database Engine In A C Library</h1>
<p align=center>}
puts "This page was last modified on [lrange $rcsid 3 4] UTC<br>"

}

puts {<h2>Features</h2>

<p><ul>
<li>Implements most of SQL92.</li>
<li>A complete database (with multiple tables and indices) is
    stored in a single byte-order independent disk file.</li>
<li>Atomic commit and rollback protect data integrity.</li>
<li>Small memory footprint: less than 20K lines of C code.</li>
<li><a href="speed.html">Four times faster</a> than PostgreSQL.
    Twice as fast as SQLite 1.0.</li>
<li>Very simple 
<a href="c_interface.html">C/C++ interface</a> requires the use of only
three functions and one opaque structure.</li>

#
# Run this Tcl script to generate the sqlite.html file.
#
set rcsid {$Id: opcode.tcl,v 1.10 2002/08/11 20:10:49 drh Exp $}

puts {<html>
<head>
  <title>SQLite Virtual Machine Opcodes</title>
</head>
<body bgcolor=white>
<h1 align=center>

Execution continues until (1) a Halt instruction is seen, or 
(2) the program counter becomes one greater than the address of
last instruction, or (3) there is an execution error.
When the virtual machine halts, all memory
that it allocated is released and all database cursors it may
have had open are closed.  If the execution stopped due to an
error, any pending transactions are terminated and changes made
to the database are rolled back.</p>

<p>The virtual machine also contains an operand stack of unlimited
depth.  Many of the opcodes use operands from the stack.  See the
individual opcode descriptions for details.</p>

<p>The virtual machine can have zero or more cursors.  Each cursor
is a pointer into a single table or index within the database.