Re: includedir_internal headers are not self-contained

Robert Haas <robertmhaas(at)gmail(dot)com> writes:
> On Mon, Apr 28, 2014 at 12:14 PM, Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
>> As far as pg_xlogdump goes, I agree that symbolic fork names are probably
>> nice, but I think the case for printing db/ts/rel OIDs as a pathname is a
>> whole lot weaker --- to my taste, that's actually an anti-feature.

> I might be missing something, but, why?

It's more verbose, it's not actually any more information, and in many
cases it's actively misleading, because what's printed is NOT the real
file name --- it omits segment numbers for instance. As a particularly
egregious example, in xact_desc_commit() we print a pathname including
MAIN_FORKNUM, which is a flat out lie to the reader, because what will
actually get deleted is all forks.

So yeah, it's an anti-feature.

BTW, for the same reasons I'm less than impressed with the uses of
relpath in error messages in, eg, reorderbuffer.c:

relation = RelationIdGetRelation(reloid);

if (relation == NULL)
elog(ERROR, "could open relation descriptor %s",
relpathperm(change->data.tp.relnode, MAIN_FORKNUM));

Printing anything other than the relation OID here is irrelevant,
misleading, and inconsistent with our practice everywhere else.
Let's not even mention the missing "not" in the message text.

regards, tom lane