Re: Allow file inclusion in pg_hba and pg_ident files

Hi,

On Thu, Jun 02, 2022 at 10:08:15AM +0900, Michael Paquier wrote:
> On Thu, May 26, 2022 at 03:26:57PM +0800, Julien Rouhaud wrote:
>
> > After a bit more digging, I think that this comes from the fact that there's no
> > "official" name for this file. Even the documentation just says "the
> > pg_hba.conf file" [1]. So using pg_hba.conf can either means explicitly
> > $PGDATA/pg_hba.conf or the instance's HBA file in general, whatever its
> > location.
> >
> > I think it would be good to improve this, including in the doc, but I'm
> > assuming it's entirely for HEAD only, including the error messages?
>
> Yes, that would be a set of changes only for HEAD, once 16~ opens for
> business. FWIW, the acronym "HBA" is defined as "Host-Based
> Authentication", so we could use that as a base for the description of
> the file, using simply HBA in the follow-up paragraphs for simplicity,
> telling that pg_hba.conf is the default.

Ok.

> > If so, should I also change the doc to replace "pg_hba.conf" with something
> > else when it's not referring to the file default name?
> >
> > I'm thinking of using "HBA file" to replace pg_hba.conf, and using
> > "authentication file" when it can be either the "HBA file" and the "User Name
> > Maps file", would that be ok?
>
> Using "HBA file" in the docs is fine by me, knowing that the acronym
> is already defined. The modified parts of the docs should perhaps
> mention once something like "Host-Based Authentication file (or HBA
> file)" for clarity. For the error message, I think that we tend to
> avoid those acronyms, don't we?

I don't have an extensive knowledge of all the user facing error messages, but
after a quick grep I see multiple usage of OID, PID, GIN and other defined
acronyms. I also see multiple occurrences of "only heap AM is supported",
while AM isn't even a defined acronym.

It doesn't seem that my proposal would be inconsistent with existing messages
and will help to reduce the message length, but if you prefer to keep the full
name I'm fine with it. Those should be very rare and specialized errors
anyway.

While on the bikeshedding part, are you ok with the proposed keywords (include
and include_dir), behaving exactly like for postgresql.conf, and to also add
include_if_exists, so that we have the exact same possibilities with
postgresql.conf, pg_hba.conf and pg_ident.conf?