Claudio Valderrama C. wrote:

>>
>>There's no point in retrofitting this stuff to old indexes. New indexes
>>sort duplicates by record number, which turns out to be important.
>
>
> I had the idea this record number sorting didn't happen for unique indices
> or did I misread some explanation in the code?
>

That doesn't matter for this algorithm because a unique index has only
one valid entry per key value for each transaction.


Regards,


Ann