SQLite

/*
** 2004 April 6
**
** 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 1.232 2005/01/14 13:50:12 danielk1977 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.

** in exchange for a larger degradation in INSERT and UPDATE performance.
** The value of NN appears to give the best results overall.
*/
#define NN 1             /* Number of neighbors on either side of pPage */
#define NB (NN*2+1)      /* Total pages involved in the balance */

/* Forward reference */
static int balance(MemPage*, int);

static int balance_quick(MemPage *pPage, MemPage *pParent){
  int rc;
  MemPage *pNew;
  Pgno pgnoNew;
  u8 *pCell;
  int szCell;
  CellInfo info;

  u8 parentCell[64];              /* How big should this be? */
  int parentIdx = pParent->nCell;
  int parentSize;

  /* Allocate a new page. Insert the overflow cell from pPage
  ** into it. Then remove the overflow cell from pPage.
  */
  rc = allocatePage(pPage->pBt, &pNew, &pgnoNew, 0, 0);
  if( rc!=SQLITE_OK ){
    return rc;
  }
  pCell = pPage->aOvfl[0].pCell;
  szCell = cellSizePtr(pPage, pCell);
  zeroPage(pNew, pPage->aData[0]);
  assemblePage(pNew, 1, &pCell, &szCell);
  pPage->nOverflow = 0;

  /* pPage is currently the right-child of pParent. Change this
  ** so that the right-child is the new page allocated above and
  ** pPage is the next-to-right child. Then balance() the parent
  ** page, in case it is now overfull.
  */
  parseCellPtr(pPage, findCell(pPage, pPage->nCell-1), &info);
  rc = fillInCell(pParent, parentCell, 0, info.nKey, 0, 0, &parentSize);
  if( rc!=SQLITE_OK ){
    return SQLITE_OK;
  }
  assert( parentSize<64 );
  rc = insertCell(pParent, parentIdx, parentCell, parentSize, 0, 4);
  if( rc!=SQLITE_OK ){
    return SQLITE_OK;
  }
  put4byte(findOverflowCell(pParent,parentIdx), pPage->pgno);
  put4byte(&pParent->aData[pParent->hdrOffset+8], pgnoNew);

  pNew->pParent = pParent;
  sqlite3pager_ref(pParent->aData);

  releasePage(pNew);
  return balance(pParent, 0);
}

/*
** This routine redistributes Cells on pPage and up to NN*2 siblings
** of pPage so that all pages have about the same amount of free space.
** Usually NN siblings on either side of pPage is used in the balancing,
** though more siblings might come from one side if pPage is the first
** or last child of its parent.  If pPage has fewer than 2*NN siblings

  assert( pPage->isInit );
  assert( sqlite3pager_iswriteable(pPage->aData) );
  pBt = pPage->pBt;
  pParent = pPage->pParent;
  sqlite3pager_write(pParent->aData);
  assert( pParent );
  TRACE(("BALANCE: begin page %d child of %d\n", pPage->pgno, pParent->pgno));

#ifdef SQLITE_BALANCE_QUICK
  if( pPage->leaf &&
      pPage->intKey &&
      pPage->leafData &&
      pPage->nOverflow==1 &&
      pPage->aOvfl[0].idx==pPage->nCell &&
      get4byte(&pParent->aData[pParent->hdrOffset+8])==pPage->pgno
  ){
    return balance_quick(pPage, pParent);
  }
#endif

  /*
  ** Allocate space for memory structures
  */
  mxCellPerPage = MX_CELL(pBt);
  apCell = sqliteMallocRaw( 
       (mxCellPerPage+2)*NB*(sizeof(u8*)+sizeof(int))

  ** Balance the parent page.  Note that the current page (pPage) might
  ** have been added to the freelist is it might no longer be initialized.
  ** But the parent page will always be initialized.
  */
  assert( pParent->isInit );
  /* assert( pPage->isInit ); // No! pPage might have been added to freelist */
  /* pageIntegrity(pPage);    // No! pPage might have been added to freelist */ 
  rc = balance(pParent, 0);
  
  /*
  ** Cleanup before returning.
  */
balance_cleanup:
  sqliteFree(apCell);
  for(i=0; i<nOld; i++){

  return rc;
}

/*
** Decide if the page pPage needs to be balanced.  If balancing is
** required, call the appropriate balancing routine.
*/
static int balance(MemPage *pPage, int insert){
  int rc = SQLITE_OK;
  if( pPage->pParent==0 ){
    if( pPage->nOverflow>0 ){
      rc = balance_deeper(pPage);
    }
    if( rc==SQLITE_OK && pPage->nCell==0 ){
      rc = balance_shallower(pPage);
    }
  }else{
    if( pPage->nOverflow>0 || 
        (!insert && pPage->nFree>pPage->pBt->usableSize*2/3) ){
      rc = balance_nonroot(pPage);
    }
  }
  return rc;
}

/*

    pCur->idx++;
    pCur->info.nSize = 0;
  }else{
    assert( pPage->leaf );
  }
  rc = insertCell(pPage, pCur->idx, newCell, szNew, 0, 0);
  if( rc!=SQLITE_OK ) goto end_insert;
  rc = balance(pPage, 1);
  /* sqlite3BtreePageDump(pCur->pBt, pCur->pgnoRoot, 1); */
  /* fflush(stdout); */
  if( rc==SQLITE_OK ){
    moveToRoot(pCur);
  }
end_insert:
  sqliteFree(newCell);

    szNext = cellSizePtr(leafCur.pPage, pNext);
    assert( MX_CELL_SIZE(pBt)>=szNext+4 );
    tempCell = sqliteMallocRaw( MX_CELL_SIZE(pBt) );
    if( tempCell==0 ) return SQLITE_NOMEM;
    rc = insertCell(pPage, pCur->idx, pNext-4, szNext+4, tempCell, 0);
    if( rc!=SQLITE_OK ) return rc;
    put4byte(findOverflowCell(pPage, pCur->idx), pgnoChild);
    rc = balance(pPage, 0);
    sqliteFree(tempCell);
    if( rc ) return rc;
    dropCell(leafCur.pPage, leafCur.idx, szNext);
    rc = balance(leafCur.pPage, 0);
    releaseTempCursor(&leafCur);
  }else{
    TRACE(("DELETE: table=%d delete from leaf %d\n",
       pCur->pgnoRoot, pPage->pgno));
    dropCell(pPage, pCur->idx, cellSizePtr(pPage, pCell));
    rc = balance(pPage, 0);
  }
  moveToRoot(pCur);
  return rc;
}

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

  sCheck.pPager = pBt->pPager;
  sCheck.nPage = sqlite3pager_pagecount(sCheck.pPager);
  if( sCheck.nPage==0 ){
    unlockBtreeIfUnused(pBt);
    return 0;
  }
  sCheck.anRef = sqliteMallocRaw( (sCheck.nPage+1)*sizeof(sCheck.anRef[0]) );
  if( !sCheck.anRef ){
    unlockBtreeIfUnused(pBt);
    return sqlite3MPrintf("Unable to malloc %d bytes", 
        (sCheck.nPage+1)*sizeof(sCheck.anRef[0]));
  }
  for(i=0; i<=sCheck.nPage; i++){ sCheck.anRef[i] = 0; }
  i = PENDING_BYTE_PAGE(pBt);
  if( i<=sCheck.nPage ){
    sCheck.anRef[i] = 1;
  }
  sCheck.zErrMsg = 0;

# This file implements regression tests for SQLite library.  The
# focus of this file is testing for correct handling of I/O errors
# such as writes failing because the disk is full.
# 
# The tests in this file use special facilities that are only
# available in the SQLite test fixture.
#
# $Id: ioerr.test,v 1.13 2005/01/14 13:50:13 danielk1977 Exp $

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

set ::AV [execsql {pragma auto_vacuum}]

# Usage: do_ioerr_test <test number> <options...>

  CREATE TABLE t1(a,b,c);
  CREATE TABLE test2.t2(a,b,c);
  COMMIT;
} -exclude [expr [execsql {pragma auto_vacuum}] ? 8 : 0]

# Test IO errors when replaying two hot journals from a 2-file 
# transaction. This test only runs on UNIX.
if {$tcl_platform(platform)=="unix" && [file exists ./crashtest]} {
  do_ioerr_test 6 -tclprep {
    set rc [crashsql 2 test2.db-journal {
      ATTACH 'test2.db' as aux;
      PRAGMA cache_size = 10;
      BEGIN;
      CREATE TABLE aux.t2(a, b, c);
      CREATE TABLE t1(a, b, c);