Source code of Windows XP (NT5)
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
 
 
 
 
 
 

1427 lines
40 KiB

//+---------------------------------------------------------------------------
//
// Microsoft Windows
// Copyright (C) Microsoft Corporation, 1992 - 2000.
//
// File: tputget.cxx
//
// Contents: Classes to put, get and locate rows in the bigtable.
//
// Classes: CTableRowPutter, CTableRowGetter, CTableRowLocator
//
// Functions:
//
// History: 4-17-95 srikants Created
//
//----------------------------------------------------------------------------
#include "pch.cxx"
#pragma hdrstop
#include "tputget.hxx"
#include "tblwindo.hxx"
#include "tblbuket.hxx"
#include "winsplit.hxx"
#include "tabledbg.hxx"
#include "query.hxx"
#include "regtrans.hxx"
//+---------------------------------------------------------------------------
//
// Function: _LokShouldUseNext
//
// Synopsis: Tests whether the segment should be use for a putrow.
//
// Arguments: [icmp] -- compare result of current row to first of pSeg
// [pSeg] -- the segment in question
//
// Returns: TRUE if new row is in same category as the first row in
// [pSeg] and if the first row in [pSeg] is the first of its
// category.
//
// History: 11-7-95 dlee Created
//
//----------------------------------------------------------------------------
BOOL CTableRowPutter::_LokShouldUseNext(
int icmp,
CTableSegment * pSeg )
{
Win4Assert( _largeTable.IsCategorized() );
CTableWindow *pWin = (CTableWindow *) pSeg;
return ( ( ( (unsigned) (- icmp ) ) > _largeTable._cCategorizersTotal ) &&
( pWin->IsFirstRowFirstOfCategory() ) );
}
//+---------------------------------------------------------------------------
//
// Function: _IsHintGood
//
// Synopsis: Tests whether the segment passed in as a hint or one of
// its immediate neighbors can be used to insert the new row.
//
// Arguments: [pSegHint] - The input segment which is the hint.
//
// Returns: Pointer to either the hint or one of its neighbors if the
// row can be inserted into them. NULL if the hint is not good.
//
// History: 4-17-95 srikants Created
//
// Notes:
//
//----------------------------------------------------------------------------
CTableSegment * CTableRowPutter::_IsHintGood( CTableSegment * pSegHint )
{
if ( 0 == pSegHint )
return 0;
CTableSegment * pResult = 0;
Win4Assert( pSegHint->GetLowestKey().IsInitialized() );
int icmp = _comparator.Compare( _currRow, pSegHint->GetLowestKey() );
if ( icmp >= 0 )
{
//
// The current key is >= the lowest key in the hint. See if this is
// less than the key in the next segment.
//
if ( _segList.GetLast() != pSegHint )
{
CTableSegment * pNext = _segList.GetNext( pSegHint );
int icmpNext = _comparator.Compare( _currRow,
pNext->GetLowestKey() );
if ( icmpNext < 0 )
{
// If categorized AND
// new row in same category as first row of next window AND
// first row of next window is first of its category
// use the next window. Otherwise use the current window.
//
if ( ( _largeTable.IsCategorized() ) &&
( _LokShouldUseNext( icmpNext, pNext ) ) )
{
tbDebugOut(( DEB_ITRACE, "updated hint window for categorized putrow\n" ));
pResult = pNext;
}
else
{
//
// The current row is < the lowest key in the next segment.
// It is also not in the same category as the lowest key
// in the next segment.
// So, it is a good candidate.
//
pResult = pSegHint;
}
}
}
else
{
pResult = pSegHint;
}
}
else if ( _segList.GetFirst() == pSegHint )
{
//
// We are at the first segment and the current row is < the first
// segment. So, just use it.
//
pResult = pSegHint;
}
return pResult;
}
//+---------------------------------------------------------------------------
//
// Function: LokFindSegToInsert
//
// Synopsis: Finds the segment to insert the current row.
//
// Arguments: [pSegHint] - Segment used as a hint.
//
// Returns: The segment into which the row must be inserted.
//
// History: 4-17-95 srikants Created
//
//----------------------------------------------------------------------------
CTableSegment * CTableRowPutter::LokFindSegToInsert( CTableSegment * pSegHint )
{
//
// For the most common case, there will be only one segment. Just use
// that and don't setup _currRow.
//
if ( 1 == _segList.GetSegmentsCount() )
{
Win4Assert( 0 == pSegHint || _segList.GetFirst() == pSegHint );
return _segList.GetFirst();
}
//
// Initialize the sort key and use that to compare with various segments.
//
_currRow.MakeReady();
if ( _segList.IsEmpty() )
{
CTableWindow * pWindow = _largeTable._CreateNewWindow(
_largeTable._AllocSegId(),
_currRow,
_currRow);
_segListMgr.Push( pWindow );
tbDebugOut(( DEB_WINSPLIT,
" Adding Window 0x%X to an empty list\n", pWindow ));
return pWindow;
}
Win4Assert( 0 != _largeTable.GetSortSet() );
//
// See if the hint is usable.
//
CTableSegment * pSeg = _IsHintGood( pSegHint );
if ( 0 == pSeg )
{
//
// Either there was no hint or the row did not belong there.
//
pSeg = _segListMgr.GetSegmentArray().LookUp( _currRow );
if ( ( _largeTable.IsCategorized() ) &&
( _segList.GetLast() != pSeg ) )
{
// Pick a window so that the unique categorizer can make the
// assumption that:
// - an insert at the end of a window either belongs to the
// same category as the previous last row OR is a new
// category
// - an insert at the beginning of a window is the new first
// member of the category of the first row in the window
CTableSegment * pNext = _segList.GetNext( pSeg );
int icmpNext = _comparator.Compare( _currRow,
pNext->GetLowestKey() );
Win4Assert( icmpNext < 0 );
if ( _LokShouldUseNext( icmpNext, pNext ) )
{
tbDebugOut(( DEB_ITRACE, "updated window for categorized putrow\n" ));
pSeg = pNext;
}
}
// This assert will not be true when a row is the new first row
// of both a segment and a category.
if ( ! _largeTable.IsCategorized() )
{
Win4Assert( _IsHintGood(pSeg) == pSeg );
}
#if 0
CBackTableSegIter iter(_segList);
Win4Assert( !_segList.AtEnd(iter) );
do
{
CTableSegment & segment = *iter.GetSegment();
Win4Assert( segment.GetLowestKey().IsInitialized() );
if ( _comparator.Compare(_currRow, segment.GetLowestKey()) >= 0 )
{
//
// The current row is >= the lowest key in this segment. We
// can use it.
//
break;
}
else
{
_segList.BackUp(iter);
}
}
while ( !_segList.AtEnd(iter) );
if ( _segList.AtEnd(iter) )
{
//
// We have gone past the end of the list. Just use the
// first one.
//
pSeg = _segList.GetFirst();
Win4Assert( _comparator.Compare( _currRow,
pSeg->GetLowestKey() ) < 0 );
}
else
{
pSeg = iter.GetSegment();
Win4Assert( _comparator.Compare( _currRow,
pSeg->GetLowestKey() ) >= 0 );
}
#endif // 0
}
Win4Assert( 0 != pSeg );
Win4Assert( pSeg->GetLowestKey().IsInitialized() );
return pSeg;
}
//+---------------------------------------------------------------------------
//
// Function: LokSplitOrAddSegment
//
// Synopsis: When a segment gets too full, it needs to be either split
// or a new one added, if possible.
//
// A segment can be split if it is a window and the window
// contents permit a sorted split.
//
// If the segment is a window but either there is no sort set
// or all rows in it are equal, then we have to either append
// or insert a new segment before the current one.
//
// If the segment is bucket, then a new segment can either be
// inserted or appended only if the current row falls outside
// the range of the rows of the bucket.
//
// Arguments: [pSegToSplit] - The segment which we should check to split
// or insert/append to.
//
// Returns: The segment into which the row must be inserted.
//
// History: 4-17-95 srikants Created
//
//----------------------------------------------------------------------------
CTableSegment * CTableRowPutter::LokSplitOrAddSegment( CTableSegment * pSegToSplit )
{
Win4Assert( 0 != pSegToSplit );
Win4Assert( !_fNewWindowCreated );
_currRow.MakeReady();
CTableSegment * pSegment = pSegToSplit;
Win4Assert( pSegment->IsGettingFull() );
if ( pSegment->IsWindow() )
{
ULONG iSplit;
BOOL fSortedSplit = pSegment->IsSortedSplit(iSplit);
Win4Assert( fSortedSplit && "Unsorted window is not legal" );
CTableWindow * pWindow = (CTableWindow *) pSegment;
//
// This window will permit a sorted split.
//
pSegment = _largeTable._LokSplitWindow( &pWindow, iSplit );
pSegment = LokFindSegToInsert( pSegment );
_fNewWindowCreated = TRUE;
}
else
{
Win4Assert( pSegment->IsBucket() );
//
// The segment is a bucket. The row we have must be bigger than
// the smallest in the bucket.
//
Win4Assert( _comparator.Compare( _currRow,
pSegment->GetLowestKey() ) >= 0 );
//
// A new window can be added only if the current row is bigger than
// the biggest row in the bucket.
//
CTableBucket * pBucket = (CTableBucket *) pSegment;
if ( _comparator.Compare( _currRow, pBucket->GetHighestKey() ) > 0 )
{
//
// It is either an un-sorted set or a sorted set will all
// the rows in that window being equal.
//
CTableSegment * pNewSeg = _largeTable._CreateNewWindow(
_largeTable._AllocSegId(),
_currRow,
_currRow );
_fNewWindowCreated = TRUE;
_segListMgr.InsertAfter( pSegment, pNewSeg );
pSegment = pNewSeg;
}
}
return pSegment;
}
CTableRowGetter::CTableRowGetter( CLargeTable & largeTable,
CTableColumnSet const & outColumns,
CGetRowsParams & getParams,
CI_TBL_CHAPT chapt,
HWATCHREGION hRegion)
:
_largeTable(largeTable),
_segListMgr(largeTable._GetSegListMgr()),
_segList(_segListMgr.GetList()),
_outColumns(outColumns),
_getParams(getParams),
_hRegion(hRegion),
_pBucketToExpand(0),
_widLastTransferred(widInvalid),
_chapt(chapt),
_cRowsToTransfer(_getParams.RowsToTransfer())
{
}
void CTableRowGetter::SetRowsToTransfer( ULONG cRowsToTransfer )
{
Win4Assert( cRowsToTransfer <= _getParams.RowsToTransfer() );
_cRowsToTransfer = cRowsToTransfer;
}
//+---------------------------------------------------------------------------
//
// Function: _GetRowsFromWindow
//
// Synopsis:
//
// Arguments: [iter] -
//
// Returns:
//
// Modifies:
//
// History: 7-13-95 srikants Split from GetRowsAtSegment to make it
// more modular
//
// Notes:
//
//----------------------------------------------------------------------------
BOOL
CTableRowGetter::_GetRowsFromWindow( CDoubleTableSegIter & iter )
{
Win4Assert( iter.GetSegment()->IsWindow() );
CTableWindow * pWindow = iter.GetWindow();
//
// Real work done here!
//
_status = pWindow->GetRows( _hRegion,
_widStart,
_chapt,
_cRowsToTransfer,
_outColumns,
_getParams,
_widLastTransferred );
if ( _getParams.GetFwdFetch() )
_segList.Advance(iter);
else
_segList.BackUp(iter);
BOOL fContinue = TRUE;
if ( _status != DB_S_ENDOFROWSET && _status != S_OK )
{
fContinue = FALSE;
if (_fAsync && _hRegion != 0 && _status == DB_S_BLOCKLIMITEDROWS)
{
//
// NEWFEATURE - Finish extending the region
//
}
}
else if ( _segList.AtEnd(iter) ||
0 == _cRowsToTransfer )
{
//
// Add the last workid retrieved to the MRU list of
// wids.
//
ULONG endOffset;
if ( pWindow->RowOffset( _widLastTransferred, endOffset ) )
{
_segListMgr.SetInUseByClient( _widLastTransferred,
pWindow );
}
//
// If transfer is complete, don't return DB_S_ENDOFROWSET
//
if ( 0 == _cRowsToTransfer )
_status = S_OK;
fContinue = FALSE;
}
else
{
//
// Retrieve rows from the next/prev segment.
//
if ( _getParams.GetFwdFetch() )
_widStart = WORKID_TBLFIRST;
else
_widStart = WORKID_TBLLAST;
}
return fContinue;
}
//+---------------------------------------------------------------------------
//
// Function: _ProcessBucket
//
// Synopsis:
//
// Arguments: [iter] -
// [xBktsToExpand] -
//
// Returns:
//
// Modifies:
//
// History: 7-13-95 srikants Split from GetRowsAtSegment to make it
// more modular
//
// Notes:
//
//----------------------------------------------------------------------------
BOOL
CTableRowGetter::_ProcessBucket( CDoubleTableSegIter & iter,
XPtr<CTableBucket> & xBktToExplode )
{
CTableSegment * pSegment = iter.GetSegment();
Win4Assert( pSegment->IsBucket() );
Win4Assert( 0 == xBktToExplode.GetPointer() );
BOOL fContinue = TRUE;
if (!_fAsync)
{
//
// Current segment is a bucket. It must be expanded before
// we can continue to retrieve rows.
//
Win4Assert( pSegment->IsBucket() );
CTableBucket * pBktToExplode =
_largeTable._LokReplaceWithEmptyWindow( iter );
xBktToExplode.Set( pBktToExplode );
fContinue = FALSE;
_status = DB_S_BLOCKLIMITEDROWS;
}
else
{
//
// How did we end up with buckets in the fetch region. All buckets
// in the fetch region should have been converted into windows and
// scheduled for expansion. These windows cannot be converted into
// buckets because they have watch regions on them.
//
Win4Assert( !"Found Buckets in the Fetch Region" );
//
// Just skip them ??
//
_segList.Advance( iter );
fContinue = !_segList.AtEnd(iter);
}
return fContinue;
}
//+---------------------------------------------------------------------------
//
// Member: CTableRowGetter::LokGetRowsAtSegment
//
// Synopsis: Retrieves rows starting at the specified segment. It will
// continue fetching until either all the requested rows are
// returned or we are past the end of the list of segments or
// we hit a bucket.
//
// Arguments: [pStartSeg] - The starting segment to use for retrieveing
// rows.
// [widStart] - The starting workid in that segment.
// [fAsync] - Explode buckets asynchronously
// [xBktToExplode] - Safe pointer for the bucket to be exploded
// (if any)
//
// Returns: Status of the operation.
//
// History: 4-18-95 SrikantS Created (Moved from CLargeTable)
// 6-28-95 BartoszM Added watch region support
// 7-13-95 SrikantS Enhanced for watch region support
//
// Notes: The watch region handle has been verified!
//
//----------------------------------------------------------------------------
SCODE CTableRowGetter::LokGetRowsAtSegment( CTableSegment * pStartSeg,
WORKID widStart,
BOOL fAsync,
XPtr<CTableBucket> & xBktToExplode
)
{
_InitForGetRows( widStart, fAsync );
if ( 0 == pStartSeg )
{
return DB_S_ENDOFROWSET;
}
else
{
Win4Assert( !_segList.IsEmpty() );
}
//
// Skip over the nodes upto the segment passed.
//
for ( CFwdTableSegIter iter( _segList ); !_segList.AtEnd(iter);
_segList.Advance(iter) )
{
if ( iter.GetSegment() == pStartSeg )
{
break;
}
}
Win4Assert( !_segList.AtEnd(iter) );
Win4Assert( iter.GetSegment() == pStartSeg );
//
// Do the GetRows to a window, crossing segment boundaries as needed.
//
//
// Add the starting wid to the "MRU" list of wid/segment pairs.
//
if ( !IsSpecialWid(widStart) && iter.GetSegment()->IsWindow() )
{
#if DBG==1
CTableWindow * pTemp = iter.GetWindow();
ULONG startOffset;
BOOL fFound = pTemp->RowOffset( widStart, startOffset );
Win4Assert( fFound );
#endif // DBG==1
_segListMgr.SetInUseByClient( widStart, iter.GetSegment() );
}
BOOL fContinue = TRUE;
do
{
CTableSegment * pSegment = iter.GetSegment();
if ( pSegment->IsWindow() )
{
fContinue = _GetRowsFromWindow( iter );
}
else
{
fContinue = _ProcessBucket( iter, xBktToExplode );
}
}
while (fContinue);
return _status;
}
//+---------------------------------------------------------------------------
//
// Member: CTableRowLocator::LokLocate
//
// Synopsis: Locates the starting bookmark in the list of segments.
//
// Arguments: [hRegion] -
// [fAsync] -
// [iter] -
// [transformer] -
//
// Returns: Status of the operation.
//
// History: 4-18-95 srikants Created
// 6-29-95 BartoszM Rewrote for notifications
// Notes:
//
//----------------------------------------------------------------------------
SCODE CTableRowLocator::LokLocate( HWATCHREGION hRegion,
BOOL fAsync,
CDoubleTableSegIter& iter,
CRegionTransformer& transformer )
{
SCODE status = S_OK;
if ( _segList.IsEmpty() )
{
if ( IsSpecialWid( _widStart ) )
{
_widFound = _widStart;
return DB_S_ENDOFROWSET;
}
else
{
QUIETTHROW (CException( DB_E_BADSTARTPOSITION ));
}
}
CTableSegment* pSegmentWatch = 0;
if ( fAsync )
{
CWatchRegion *pRegion = transformer.Region();
if (pRegion)
{
pSegmentWatch = pRegion->Segment();
Win4Assert( 0 == pSegmentWatch || pSegmentWatch->IsWindow() );
}
}
//
// Do not modify the _widStart and _iRowOffset member variables here
// as they are "input" values and must not be modified here. The client
// of this class can modify them if needed.
//
WORKID widStart = _widStart;
LONG iRowOffset = _iRowOffset;
Win4Assert( WORKID_TBLBEFOREFIRST != widStart &&
WORKID_TBLAFTERLAST != widStart );
//
// Locate either the segment in which the fetch bookmark is present
// or the segment which is the beginning of the watch region -
// whichever comes first in the segment list.
//
if (widStart != WORKID_TBLFIRST)
{
if (widStart == WORKID_TBLLAST)
{
//
// PERFFIX - can we optimize by initializing the iterator with the
// pSegmentWatch if it is not 0.
//
while ( !_segList.IsLast(iter) && iter.GetSegment() != pSegmentWatch)
{
_segList.Advance(iter);
}
}
else
{
while ( !_segList.AtEnd(iter) && iter.GetSegment() != pSegmentWatch)
{
if ( iter.GetSegment()->IsRowInSegment(widStart) )
{
tbDebugOut(( DEB_REGTRANS, "found wid %#x in segment %#x\n",
widStart, iter.GetSegment() ));
break;
}
_segList.Advance(iter);
}
}
}
if (_segList.AtEnd(iter))
{
THROW (CException ( DB_E_BADBOOKMARK));
}
TBL_OFF obRow; // dummy
ULONG iRowInSeg = 0; // position of bookmark in current segment
//
// The iterator is positioned either at the beginning of the watch region
// or at the segment that contains widStart
//
if (iter.GetSegment() == pSegmentWatch)
{
//
// We are in the watch region
//
Win4Assert (0 != pSegmentWatch && pSegmentWatch->IsWindow() );
CTableWindow* pWindow = iter.GetWindow();
transformer.SetWatchPos( pWindow->GetWatchStart(hRegion) );
if ( pWindow->FindBookMark( widStart, obRow, iRowInSeg ))
{
//
// both the watch and the starting wid are in the same window
//
transformer.SetFetchBmkPos (iRowInSeg);
}
else
{
DBROWCOUNT iFetch = 0;
// skip segments to get to the starting bookmark
do
{
iFetch += iter.GetSegment()->RowCount();
_segList.Advance(iter);
if (_segList.AtEnd(iter))
{
//
// Bookmarks do become invalid.
// For instance it could have
// gone to an expanding bucket.
//
THROW (CException ( DB_E_BADBOOKMARK));
}
}
while ( !iter.GetSegment()->IsRowInSegment(widStart));
//
// Is the starting bookmark in a bucket?
//
if (!iter.GetSegment()->IsWindow())
{
//
// We hit a bucket. The bookmark became invisible
// to the user. In the asynchronous world bookmarks
// are not valid forever, buddy!
//
THROW (CException ( DB_E_BADBOOKMARK));
}
else
{
//
// The starting bookmark is in a window.
//
pWindow = iter.GetWindow();
pWindow->FindBookMark ( widStart, obRow, iRowInSeg );
iFetch += iRowInSeg;
}
transformer.SetFetchBmkPos ( iFetch );
}
}
else if ( 0 != hRegion )
{
//
// Found widStart but not the watch region.
// We now have to search for the watch region in the subsequent
// segments.
//
if (!iter.GetSegment()->IsWindow())
{
//
// We hit a bucket. The bookmark became invisible
// to the user. In the asynchronous world bookmarks
// are not valid forever, buddy!
//
THROW (CException ( DB_E_BADBOOKMARK));
}
else
{
CTableWindow* pWindow = iter.GetWindow ();
ULONG iOffset, iRowInSeg;
pWindow->FindBookMark ( widStart, obRow, iRowInSeg );
//
// iFetch is now the offset from the beginning of the bookmark
// segment.
//
transformer.SetFetchBmkPos ( iRowInSeg );
if ( 0 != pSegmentWatch )
{
DBROWCOUNT iWatch = 0;
// Search for the beginning of watch region
CDoubleTableSegIter iter2 (iter);
do
{
iWatch += iter2.GetSegment()->RowCount();
_segList.Advance (iter2);
Win4Assert (!_segList.AtEnd(iter2));
} while (iter2.GetSegment() != pSegmentWatch);
pWindow = iter2.GetWindow();
iWatch += pWindow->GetWatchStart (hRegion);
transformer.SetWatchPos (iWatch);
}
}
}
// CLEANCODE: Can we make this function void since it's throwing exceptions?
return S_OK;
}
//+---------------------------------------------------------------------------
//
// Member: CTableRowLocator::LokRelocate
//
// Synopsis: Finds the start of the fetch region by counting the offset
// from the start bookmark.
//
// Arguments: [fAsync] -
// [iter] -
// [transformer] -
// [xBktToExplode] -
// [xBktToConvert] -
//
// Returns:
//
// Modifies:
//
// History: 7-25-95 srikants Created
//
// Notes:
//
//----------------------------------------------------------------------------
void CTableRowLocator::LokRelocate( BOOL fAsync,
CDoubleTableSegIter& iter,
CRegionTransformer& transformer,
XPtr<CTableBucket>& xBktToExplode,
CDynStack<CTableBucket>& xBktToConvert )
{
#if CIDBG==1
tbDebugOut(( DEB_REGTRANS, "Before Relocating\n" ));
transformer.DumpState();
#endif // CIDBG
//
// 1. We have located the "anchor" bookmark. The iterator
// is positioned at that segment.
//
// 2. If there was a watch region, we have located the watch region
//
//
// Now, we have to locate the segment which
// is at the particular offset from the "anchor" bookmark.
//
if ( iter.GetSegment()->IsBucket() )
{
Win4Assert( !fAsync &&
"Must be synchronous when a bookmark is in a bucket" );
CTableBucket * pBucket =
_largeTable._LokReplaceWithEmptyWindow( iter );
xBktToExplode.Set( pBucket );
return;
}
//
// The first segment in the iterator is a window.
//
WORKID widStart = _widStart;
DBCOUNTITEM cRowsInSeg = iter.GetSegment()->RowCount();
DBROWCOUNT cRowsResidual = transformer.GetFetchOffsetFromAnchor();
_cRowsBeyond = 0;
Win4Assert( WORKID_TBLBEFOREFIRST != widStart &&
WORKID_TBLAFTERLAST != widStart );
BOOL fGoingForward = cRowsResidual >= 0;
if ( !fGoingForward )
{
//
// Make the residual row count always positive.
//
cRowsResidual = -cRowsResidual;
}
//
// Do the computations for the first segment.
// The residual row count will always be WRT the beginning of a segment
// if going forward and will be WRT to the end of a segment if going
// backward. It will always be positive.
//
ULONG iRowStart;
if ( WORKID_TBLLAST == widStart )
{
iRowStart = (ULONG) (cRowsInSeg-1);
}
else if ( WORKID_TBLFIRST == widStart )
{
iRowStart = 0;
}
else
{
CTableWindow * pStartWindow = iter.GetWindow();
BOOL fFound = pStartWindow->RowOffset( widStart, iRowStart );
Win4Assert( fFound && "Must Find Starting BookMark" );
}
if ( fGoingForward )
{
// WRT to the beginning of the current segment
cRowsResidual += iRowStart;
}
else
{
// WRT to the end of the current segment
cRowsResidual += (cRowsInSeg-iRowStart)-1;
}
//
// Look for the row in this segment.
//
while ( !IsOffsetFound(iter, cRowsResidual) )
{
cRowsResidual -= cRowsInSeg;
Win4Assert( cRowsResidual >= 0 );
if ( fGoingForward )
{
_segList.Advance(iter);
}
else
{
_segList.BackUp(iter);
}
if ( _segList.AtEnd(iter) )
{
break;
}
cRowsInSeg = iter.GetSegment()->RowCount();
if ( iter.GetSegment()->IsBucket() &&
fAsync &&
transformer.IsContiguous() )
{
//
// We are doing asynchronous get rows in a contiguous
// watch region. So, we should schedule the buckets for
// later expansion and NOT count the rows in them for the
// offset computation.
//
CTableBucket * pBktToExpand =
_largeTable._LokReplaceWithEmptyWindow( iter );
xBktToConvert.Push( pBktToExpand );
//
// We should not count the rows in a bucket in async case.
//
cRowsInSeg = 0;
}
}
if ( _segList.AtEnd( iter ) )
{
if ( fGoingForward )
{
_cRowsBeyond = -cRowsResidual;
if ( 0 == _cRowsBeyond )
{
//
// Should be the first row after the last segment.
//
_cRowsBeyond = 1;
}
}
else
{
_cRowsBeyond = -cRowsResidual;
if ( 0 == _cRowsBeyond )
{
//
// Is the first row before this segment.
//
_cRowsBeyond = -1;
}
}
_widFound = widInvalid;
}
else
{
if ( !iter.GetSegment()->IsWindow() )
{
_RelocateThruBuckets( fAsync, iter, transformer, xBktToExplode );
}
else
{
CTableWindow * pWindow = iter.GetWindow();
Win4Assert( pWindow->RowCount() > (ULONG) cRowsResidual );
if ( fGoingForward )
{
_widFound = pWindow->GetBookMarkAt( (ULONG) cRowsResidual );
}
else
{
_widFound = pWindow->GetBookMarkAt( (ULONG) (
pWindow->RowCount() - cRowsResidual - 1 ) );
}
}
}
#if CIDBG==1
tbDebugOut(( DEB_REGTRANS, "After Relocating\n" ));
transformer.DumpState();
#endif // CIDBG
return;
}
//+---------------------------------------------------------------------------
//
// Member: CTableRowLocator::_RelocateThruBuckets
//
// Synopsis: Does the processing needed after locating the segment in
// which the requested row is contained. If the current segment
// is a bucket, the following actions must be taken:
//
// 1. If it is synchronous fetch, must EXPLODE the bucket.
//
// 2. O/W, it must a non-contiguous fetch. In this case, we
// must locate the closest window.
//
// If the fetch is forward,
// (+ve count of rows), we must locate the first window in
// the forward direction and position at the beginning of the
// window.
//
// If the fetch is backward (-ve count of rows), we must
// locate the first window in the backward direction and
// position at the end of the window.
//
// Arguments: [fAsync] -
// [iter] -
// [transformer] -
// [xBktToExplode] -
//
// History: 7-25-95 srikants Created
//
// Notes:
//
//----------------------------------------------------------------------------
void
CTableRowLocator::_RelocateThruBuckets( BOOL fAsync,
CDoubleTableSegIter& iter,
CRegionTransformer& transformer,
XPtr<CTableBucket>& xBktToExplode
)
{
//
// If the iterator is not at the end of the list, then the current
// segment MUST contain the row we are looking for.
//
Win4Assert( !iter.GetSegment()->IsWindow() );
if (!fAsync)
{
//
// We are in synchronous mode and the iterator is positioned at
// a bucket. Must explode the bucket.
//
CTableBucket * pBucket = _largeTable._LokReplaceWithEmptyWindow( iter );
xBktToExplode.Set( pBucket );
return;
}
//
// If the fetch is contiguous, we must be either at a window or past the
// end of the list. Both cases are eliminated above this.
//
Win4Assert( !transformer.IsContiguous() );
//
// Search for the nearest window - either in the positive
// or negative direction depending upon the count of the
// rows.
//
BOOL fForward = transformer.GetFetchCount() >= 0 ;
while ( !_segList.AtEnd(iter) )
{
CTableSegment * pSeg = iter.GetSegment();
if ( pSeg->IsBucket() ||
0 == pSeg->RowCount() )
{
if ( fForward )
{
_segList.Advance(iter);
}
else
{
_segList.BackUp(iter);
}
}
else
{
Win4Assert( pSeg->IsWindow() );
break;
}
}
if ( !_segList.AtEnd(iter) )
{
if ( fForward )
{
_widFound = iter.GetWindow()->GetBookMarkAt(0);
}
else
{
_widFound = iter.GetWindow()->GetBookMarkAt( (ULONG) (iter.GetSegment()->RowCount()-1) );
}
}
return;
}
//+---------------------------------------------------------------------------
//
// Member: CTableRowLocator::LokSimulateFetch
//
// Synopsis: Simulates fetching rows from windows and schedules the buckets
// in the path to be converted to windows. This allows setting
// the watch region correctly over all the required segments.
//
// Arguments: [iter] - The iterator is positioned at the segment to
// start fetching from.
// [transformer] - The transformer is initialized with the old
// and new fetch/watch arguments.
// [xBktToConvert] - On output, will have buckets to be converted
// into windows. These buckets were in the fetch region and will be
// replace with empty windows here.
//
// History: 7-26-95 srikants Created
//
// Notes:
//
//----------------------------------------------------------------------------
void CTableRowLocator::LokSimulateFetch( CDoubleTableSegIter& iter,
CRegionTransformer& transformer,
CDynStack<CTableBucket>& xBktToConvert
)
{
#if CIDBG==1
tbDebugOut(( DEB_REGTRANS, " Before SimulateFetch\n" ));
transformer.DumpState();
#endif // CIDBG
Win4Assert( iter.GetSegment()->IsWindow() );
Win4Assert( widInvalid != _widFound );
// Win4Assert( IsRowFound(iter) );
//
// If it is asynchronous fetch, we do not have categorization and so
// we don't have to worry about checking for going beyond chapters.
//
BOOL fForward = transformer.GetFetchCount() >= 0;
DBROWCOUNT cRowsToRetrieve = 0; // tracks the number of rows still to retrieve.
if ( fForward )
cRowsToRetrieve = transformer.GetFetchCount();
else
cRowsToRetrieve = -transformer.GetFetchCount();
CTableWindow * pFirstWindow = iter.GetWindow();
ULONG iRowOffset;
TBL_OFF dummy;
BOOL fFound = pFirstWindow->FindBookMark( _widFound, dummy, iRowOffset );
Win4Assert( fFound );
//
// For determining the lowest fetch position.
//
ULONG iOffsetInFirstWindow = iRowOffset;
//
// Compute the number of rows that can be retrieved from current window.
//
ULONG cRowsFromCurrWindow = 0;
if ( fForward )
cRowsFromCurrWindow = (ULONG) (pFirstWindow->RowCount()-iRowOffset);
else
cRowsFromCurrWindow = iRowOffset+1;
if ( cRowsToRetrieve > (long) cRowsFromCurrWindow )
{
//
// We still have more rows to be retrieved.
//
cRowsToRetrieve -= cRowsFromCurrWindow;
}
else
{
//
// This window can give us all the rows to be fetched.
//
if ( !fForward )
{
Win4Assert( iRowOffset+1 >= (ULONG) cRowsToRetrieve );
iOffsetInFirstWindow = (ULONG) ( iRowOffset+1-cRowsToRetrieve );
}
cRowsToRetrieve = 0;
}
while ( cRowsToRetrieve > 0 )
{
Win4Assert( !_segList.AtEnd(iter) );
if ( fForward )
{
_segList.Advance( iter );
}
else
{
_segList.BackUp(iter);
}
if ( _segList.AtEnd(iter) )
{
break;
}
if ( iter.GetSegment()->IsBucket() )
{
CTableBucket * pBucket =
_largeTable._LokReplaceWithEmptyWindow(iter);
xBktToConvert.Push(pBucket);
}
else
{
CTableWindow * pWindow = iter.GetWindow();
cRowsFromCurrWindow = (ULONG) pWindow->RowCount();
if ( cRowsToRetrieve > (long) cRowsFromCurrWindow )
{
cRowsToRetrieve -= cRowsFromCurrWindow;
}
else
{
if ( !fForward )
{
iOffsetInFirstWindow = (ULONG) (
pWindow->RowCount()-cRowsToRetrieve );
pFirstWindow = pWindow;
}
cRowsToRetrieve = 0;
}
}
}
if ( _segList.AtEnd(iter) && !fForward )
{
//
// we have gone beyond the beginning of the table but still
// cannot satisfy the row count to be retrieved. Is it okay to assume
// that the first window and the first row are the lowest fetch points
//
Win4Assert( _segList.GetFirst() &&
_segList.GetFirst()->IsWindow() );
pFirstWindow = (CTableWindow *) _segList.GetFirst();
iOffsetInFirstWindow = 0;
}
Win4Assert( 0 != pFirstWindow );
Win4Assert( iOffsetInFirstWindow < pFirstWindow->RowCount() );
transformer.SetLowFetchPos( pFirstWindow, iOffsetInFirstWindow );
if ( transformer.GetFetchOffsetFromOrigin() < 0 )
{
transformer.MoveOrigin( transformer.GetFetchOffsetFromOrigin() -
iOffsetInFirstWindow );
}
#if CIDBG==1
tbDebugOut(( DEB_REGTRANS, " SimulateFetch Done\n" ));
#endif // CIDBG==1
}
void
CTableRowLocator::SeekAndSetFetchBmk( WORKID wid , CDoubleTableSegIter & iter )
{
_widFound = widInvalid;
if ( _segList.IsEmpty() )
return;
CTableWindow * pWindow;
ULONG iRowOffset;
if ( WORKID_TBLFIRST == wid )
{
//
// We want the first bookmark in the table.
//
_segList.SetToBeginning( iter );
pWindow = iter.GetWindow();
iRowOffset = 0;
}
else
{
Win4Assert( WORKID_TBLLAST == wid );
//
// We want the last bookmark in the table.
//
_segList.SetToEnd(iter);
pWindow = iter.GetWindow();
if ( pWindow->RowCount() > 0 )
iRowOffset = (ULONG) (pWindow->RowCount()-1);
}
if ( pWindow->RowCount() > 0 )
{
_widFound = pWindow->GetBookMarkAt( iRowOffset );
}
}