"Heikki Linnakangas" <heikki@enterprisedb.com> writes:
> I'm not sure I have enough time to get all that done for 8.4, it's looking bad
> at the moment, so help would be much appreciated. If you don't agree with
> taking the visibility map approach, I would suggest working on the indexam API
> changes first, to allow returning index tuples from an index. I believe that
> part is the same regardless of how we check the visibility.
That part is the elephant in the room in all these discussions.
I wonder if we want to have a new plan node for the heap accesses so they can
be postponed up above a join or other quals. Even for tuples which are of
questionable visibility we ought to be able to check cheap quals first before
checking visibility (though we might need a new function property to indicate
it's safe to call extra times on data which doesn't really exist -- immutable
doesn't mean it might not throw errors, for example).
-- Gregory Stark EnterpriseDB http://www.enterprisedb.com Ask me about EnterpriseDB's PostGIS support!