Follow up to discussion:
http://archives.postgresql.org/pgsql-hackers/2012-11/msg00817.php
I worked out a patch that replaces PD_ALL_VISIBLE with calls to
visibilitymap_test. It rips out a lot of complexity, with a net drop of
about 300 lines (not a lot, but some of that code is pretty complex).
The patch is quite rough, and I haven't yet added the code to keep the
VM buffer pins around longer, but I still don't see any major problems.
I'm fairly certain that scans will not be adversely affected, and I
think that inserts/updates/deletes should be fine as well.
The main worry I have with inserts/updates/deletes is that there will be
less spatial locality for allocating new buffers or for modifying
existing buffers. So keeping a pinned VM buffer might not help as much,
because it might need to pin a different one, anyway.
Adding to January commitfest, as it's past 11/15.
Regards,
Jeff Davis