On Fri, Jul 19, 2002 at 05:15:43PM -0400, Tom Lane wrote:
> nconway@klamath.dyndns.org (Neil Conway) writes:
> > On Fri, Jul 19, 2002 at 01:21:10PM -0400, Tom Lane wrote:
> >> I'm very unthrilled with this approach to faking up a composite type
> >> for pg_show_locks to return.
>
> > As am I, and I agree that the proper long-term answer is some new
> > infrastructure for adding builtin SRFs. However, I don't think that's
> > a really good reason for rejecting the patch --
>
> I'm not wanting to reject the patch; I'm wanting to restructure it as
> additions to lmgr.c plus a contrib module that includes the API function
> and perhaps some sample views. The contrib module's install script
> could avoid these pesky problems because it can just CREATE a dummy
> table or view and then CREATE the function. Once we have a better
> answer about declaring built-in SRFs, we can migrate the code into the
> core.
Personally, that doesn't strike me as a lot cleaner than just putting
the code into the core in the first place. Since the changes to adapt
the SRF to a new composite type scheme would be trivial (less than
20 lines of changes, probably less), I'd personally vote to include it
in the core, and put up with a little bit of ugliness in initdb until
we get a proper solution.
I've attached a revised patch which incorporates Tom's suggestions, as
well as including a few more code cleanups/fixes, and doesn't remove
the USER_LOCKS code.
Cheers,
Neil
--
Neil Conway <neilconway@rogers.com>
PGP Key ID: DB3C29FC