Simon Riggs <simon(at)2ndquadrant(dot)com> writes:
> I believe this is safe.
I won't insult your intelligence by pointing out how I know that you
didn't even test the patch against hash or gist.
The major problem with the patch is that it's incapable of producing
correct tuple-count stats for partial indexes, which is really not
acceptable from a planning standpoint. What I'm currently fooling with
is skipping the bulkdelete scan only if the index isn't partial...
regards, tom lane