Re: Procedural language definitions (was Re: 8.1 and syntax checking at create time)

elein(at)varlena(dot)com (elein) writes:
> On Wed, Aug 31, 2005 at 05:56:52PM -0400, Tom Lane wrote:
>> The basic idea is to create a shared catalog that contains "procedural
>> language templates". This catalog would essentially replace the
>> knowledge that's now hardwired in the createlang program.
>> ...
>> It's a shame that we didn't think about this before feature freeze,
>> as the recent changes to create PL support functions in pg_catalog
>> have made both pg_dump and createlang noticeably uglier than before.
>> We could have dispensed with those hacks. Oh well.

> This idea appears to me to be sound. It may be worth adding the
> feature during beta anyway to simplify the ugliness of pg_dump
> with createlang problems. The large number of weird configurations
> "out there" could use the beta testing of this release. I
> ran into this issue a lot with non-standard installations.

I was thinking the same thing, but it's a big change to put in during
beta.

We could trim back the size of the patch a good deal by not implementing
the ACL part just yet (ie, you'd still have to be superuser to create a
PL). However, we'd still need to force an initdb to add the new system
catalog, and I hate to do that to our long-suffering beta testers.

An even more trimmed-back version would not create a new system catalog
now, but would use a constant table of "known PLs" that's hardwired into
the CREATE LANGUAGE code. We could do that in a really localized
fashion, so it seems small enough for a post-beta change.

On the other hand: if we put that into beta2, and then get a related
bug report, we wouldn't be really sure if the reporter had a correct
PL definition or an incorrect one that he'd carried forward from beta1.
Forcing an initdb would let us be sure from the version what we were
dealing with.

Comments anyone?

regards, tom lane