On Tue, 29 Dec 2009, Mindaugas Kavaliauskas wrote:
Hi,
> >It's defined indirectly by dynamic descend flag which causes
> >that seek last should work like in Harbour native RDDs and
> >CL53/COMIX. In such implementation seek last is fully revertible
> >by switching to descending order ( ordDescend(,,!ordDescend()) )
> >and forward seek. Otherwise we will have some anomalies.
> Ok, if understand the idea of seeklast now, then it means "try doing
> seek backward (in reversed index order, like using DESCEND)", but
> not "return the last record (having maximal recno value) of the
> records having the same key value".
Too much complicated for me ;-)
I rather suggest you to think about functionality. This code:
dbSeek( dtFirst, .T., .F. )
while field->DATE <= dtLast .and. !EOF()
[...]
dbSkip( 1 )
enddo
should scan the same record set as this code:
dbSeek( dtLast, .T., .T. )
while field->DATE >= dtFirst .and. !BOF()
[...]
dbSkip( -1 )
enddo
and this also precisely defines which record should be chosen
when the exact key does not exists and for soft seek last.
> But ADS still have bug in the first test. If softseek positions at
> EOF seeklast does not help to stop on last record. Here is the test
> indicating both correct (in some cases) and wrong (in another cases)
> behavior of ADS:
Yes, and your patch is correct though probably we can remove
AdsAtEOF() checking and always call:
u32RetVal = AdsSkip( pArea->hOrdCurrent, -1 );
I also suggest to store in u32RetVal AdsSeek() and AdsSeekLast() results
and move:
if( u32RetVal != AE_SUCCESS )
{
commonError( pArea, EG_CORRUPTION, ( HB_ERRCODE ) u32RetVal, 0, NULL,
EF_CANDEFAULT, NULL );
return HB_FAILURE;
}
to always generate RTE also when AdsSeek*() fails.
Please only check the AdsSeek*() results for empty ADT and DBF tables.
best regards,
Przemek
_______________________________________________
Harbour mailing list (attachment size limit: 40KB)
[email protected]
http://lists.harbour-project.org/mailman/listinfo/harbour
