Re: postgres.h included from relcache.h - but removing it breaks pg_upgrade

On Tue, Sep 14, 2021 at 6:53 AM Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
> Andres Freund <andres(at)anarazel(dot)de> writes:
> > On 2021-09-13 22:40:19 -0400, Tom Lane wrote:
> >> As for the fix ... what in the world is pg_upgrade doing including
> >> relcache.h? It seems like there's a more fundamental problem here:
> >> either relcache.h is declaring something that needs to be elsewhere,
> >> or pg_upgrade is doing something it should not.
>
> > We could split visibilitymap.h into two, or we could forward-declare Relation
> > and not include relcache...
>
> Without having looked at the details, I think using a forward-declare
> to avoid including relcache.h in visibilitymap.h might be a reasonably
> non-painful fix.

I like that idea, but I didn't find an appropriate existing header for
forward-declaration of Relation. relation.h isn't suitable, because
it includes primnodes.h. A separate header for just
forward-definition of Relation seems too much.

> TOH, in the long run it might be worth the effort
> to split visibilitymap.h to separate useful file-contents knowledge
> from backend function declarations.

I've drafted a patch splitting visibilitymap_maros.h from
visibilitymap.h. What do you think?

------
Regards,
Alexander Korotkov