Re: Oid registry

Antonin Houska <antonin(dot)houska(at)gmail(dot)com> writes:
> I'm also implementing an extension where direct access to non-fixed OIDs
> (i.e. no catalog cache lookup by name) would be very helpful. I spent some
> time thinking about a workaround that makes OID registry unnecessary.
> How about the following?

> 1. Add a new varlena column to pg_proc catalog table, say
> 'ext_types',containing
> C array of OIDs.

> 2. Let each extension declare requirements like the following in its
> configuration
> files:

> "I expect <some type's name> type at 0-th position of 'ext_types' array."
> "I expect <other type's name> type at 1-st position of 'ext_types' array."
> etc.

I think this just begs the question: how do you specify <some type> and
how do you know that whatever was found is what you want?

Beyond that, nothing in what you said can't be done today by a function
that does type name lookups and caches the results internally. And I'd
just as soon not burden the function-call infrastructure with more
overhead to support something only a small fraction of functions would
need.

Another point is that server-internal features don't help client-side
code, which is actually where most of the pain is AFAICT. We aren't
providing any infrastructure that helps clients interpret PQftype()
values for non-core types.

regards, tom lane