Re: Proposal: SET ROLE hook

Joe Conway <mail(at)joeconway(dot)com> writes:
> I still don't see any point in trying to pass data back from the hooks
> as the extension can maintain that state just fine, although it looks
> like it would be pretty trivial to do using a new void* member added to
> role_auth_extra. Tom (or anyone else), any comment on that?

The point of packaging GUC-related state into a blob that guc.c
knows about is that then the right things will happen when guc.c handles
something like a SET LOCAL, GUC reversion at subtransaction rollback,
SET clauses attached to functions, yadda yadda. Are you sure your
extension can, or wants to, track all those possibilities for itself?

I remember thinking that we probably would need to extend role_auth_extra
to make this work, so I have no objection if you're finding that that's
actually the case. Might need to think about how more than one hook
could include state into the blob.

(Note: I've not actually read this version of your patch, although
I could make time for that next week sometime.)

regards, tom lane