>>>>> NB: The same benchmark revealed that CREATE TABLE (or maybe it's CREATE
>>>>> INDEX) leaks about 2k of memory.
> Hmm. So the space *is* leaked. Grumble. Another TODO list item...
> thanks for following up on it.
OK, I think this bug is swatted --- it's not relcache really, just a
couple of generic memory leaks that happened to occur during relcache
entry construction (worst possible time). If you care to try the
repaired code, see our CVS server, or grab a nightly snapshot dated
later than this message.
regards, tom lane