Tom Lane wrote:
> Bruce Momjian <bruce@momjian.us> writes:
> > Are we done with the sort interrupt issue mentioned in the subject line,
> > and the issue outlined below?
>
> I'm inclined not to apply the proposed patch (adding
> CHECK_FOR_INTERRUPTS) because of the risk of memory leakage inside
> qsort. OTOH you could argue that there's an unfixable risk of memory
> leakage there anyway, because it's always possible that the invoked
> datatype comparison routine exits with elog(ERROR) for some reason,
> or even contains a CHECK_FOR_INTERRUPTS call itself. Comments?
OK, we do check somewhere during sorting, I assume. I can't imagine a
single qsort() call taking all that long because we do them in batches
anyway.
> As for the question of whether we should try to detoast sort keys before
> sorting, I'd suggest adding that to TODO. Investigating whether this
> would be a good idea will take more time than we have for 8.2, so it's
> gonna have to wait for a future cycle.
Added to TODO:
* Consider detoasting keys before sorting
--
Bruce Momjian bruce@momjian.us
EnterpriseDB http://www.enterprisedb.com
+ If your life is a hard drive, Christ can be your backup. +