Heikki Linnakangas <heikki.linnakangas@enterprisedb.com> writes:
> ISTM we should not skip the analyze if run as a stand-alone ANALYZE. I'm
> not sure what problem the author of that comment envisioned with
> autovacuum, but resetting the live and dead tuple counters doesn't seem
> right to me.
I think that code probably originated at a time when palloc(0) threw an
error, and was meant to avoid any corner-case failures in the subsequent
logic. I can't get very excited about the case --- it seems to me the
*real* fix for this complaint is to give hstore some ANALYZE support ---
but if you've tested that it behaves sanely with no columns I have no
objection to committing it.
regards, tom lane
