On Thu, Mar 05, 2020 at 05:46:44PM -0500, Tom Lane wrote:
>I wrote:
>> If not, we probably should bite the bullet and go for #1, since
>> I have little doubt that we'll need that someday anyway.
>> The trick will be to keep down the cache invalidation overhead...
>
>Here's a version that does it like that. I'm less worried about the
>overhead than I was before, because I realized that we already had
>a syscache callback for pg_type there. And it was being pretty
>stupid about which entries it reset, too, so this version might
>actually net out as less overhead (in some workloads anyway).
>
>For ease of review I just added the new TCFLAGS value out of
>sequence, but I'd be inclined to renumber the bits back into
>sequence before committing.
>
LGTM. If I had to nitpick, I'd say that the example in docs should be
ALTER TYPE mytype SET (
SEND = mytypesend,
RECEIVE = mytyperecv
);
i.e. with uppercase SEND/RECEIVE, because that's how we spell it in
other examples in CREATE TYPE etc.
regards
--
Tomas Vondra http://www.2ndQuadrant.com
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services