Michael Paquier <michael@paquier.xyz> writes:
> I have noticed this thread and 4e54d23 as a result this morning. If
> you want to spread this style more, wouldn't it be better to do that
> in all the places of pg_upgrade where we store paths to files? I can
> see six code paths with log_opts.basedir that could do the same, as of
> the attached. The hardcoded file names have various lengths, and some
> of them are quite long making the generated paths more exposed to
> being cut in the middle.
Well, I just fixed the ones in make_outputdirs because it seemed weird
that that part of the function was not doing something the earlier parts
did. I didn't look around for more trouble.
I think that pg_fatal'ing on the grounds of path-too-long once we've
already started the upgrade isn't all that great. Really we want to
fail on that early on --- so coding make_outputdirs like this is
fine, but maybe we need a different plan for files made later.
regards, tom lane