Re: Cirrus-ci is lowering free CI cycles - what to do with cfbot, etc?

Hi,

On 2023-08-23 17:55:53 -0400, Tom Lane wrote:
> Andres Freund <andres(at)anarazel(dot)de> writes:
> > On 2023-08-23 17:02:51 -0400, Tom Lane wrote:
> >> ... unless we hit problems with, say, a different default port number or
> >> socket path compiled into one copy vs. the other? That seems like it's
> >> probably a "so don't do that" case, though.
>
> > If we were to find such a case, it seems we could just add whatever missing
> > parameter to the connection string? I think we would likely already hit such
> > problems though, the psql started by an installcheck pg_regress might use the
> > system libpq, I think?
>
> The trouble with that approach is that in "make installcheck", we
> don't really want to assume we know what the installed libpq's default
> connection parameters are. So we don't explicitly know where that
> libpq will connect.

Stepping back: I don't think installcheck matters for the concrete use of
libpq we're discussing - the only time we wait for server startup is the
non-installcheck case.

There are other potential uses for libpq in pg_regress though - I'd e.g. like
to have a "monitoring" session open, which we could use to detect that the
server crashed (by waiting for the FD to be become invalid). Where the
connection default issue could matter more?

I was wondering if we could create an unambiguous connection info, but that
seems like it'd be hard to do, without creating cross version hazards.

> As I said, we might be able to start treating installed-libpq-not-
> compatible-with-build as a "don't do it" case. Another idea is to try
> to ensure that pg_regress uses the same libpq that the psql-under-test
> does; but I'm not sure how to implement that.

I don't think that's likely to work, psql could use a libpq with a different
soversion. We could dlopen() libpq, etc, but that seems way too complicated.

What's the reason we don't force psql to come from the same build as
pg_regress?

Greetings,

Andres Freund