On Tue, Jul 16, 2019 at 10:49:39AM +1200, Thomas Munro wrote:
> On Tue, Jul 16, 2019 at 1:49 AM Robert Haas <robertmhaas@gmail.com> wrote:
> > [long form -1]
> >
> > But how about just using a magic database OID?
>
> This patch was just an experiment based on discussion here:
>
>
https://www.postgresql.org/message-id/flat/CA%2BhUKG%2BDE0mmiBZMtZyvwWtgv1sZCniSVhXYsXkvJ_Wo%2B83vvw%40mail.gmail.com
>
> I learned some things. The main one is that you don't just need space
> the buffer tag (which has plenty of spare bits) but also in WAL block
> references, and that does seem to be a strike against the idea. I
> don't want lack of agreement here to hold up other work. So here's
> what I propose:
>
> I'll go and commit the simple refactoring bits of this work, which
> just move some stuff belonging to md.c out of smgr.c (see attached).
> I'll go back to using a magic database OID for the undo log patch set
> for now. We could always reconsider the SMGR discriminator later.
> For now I'm not going to consider this question a blocker for the
> later undo code when it's eventually ready for commit.
Agree that we should move on at this point. The magic OIDs do not block
us from moving to this model later if needed.
--
Shawn Debnath
Amazon Web Services (AWS)