Re: pgsql: Add function to import operating system collations

Peter Eisentraut <peter(dot)eisentraut(at)2ndquadrant(dot)com> writes:
> What's fishy about this is that I can't reproduce it locally in a
> variety of VMs, and the buildfarm is not unanimous either.

Well, as I said, I get

$ locale -a | grep ^aa_ER
aa_ER
aa_ER.utf8
aa_ER(dot)utf8(at)saaho
aa_ER(at)saaho

What it looks like to me is that we see "aa_ER(dot)utf8(at)saaho", enter
that, strip it to "aa_ER(at)saaho" and enter that (if_not_exists,
which it doesn't), and then see "aa_ER(at)saaho" which we try to
enter and fail. IOW, the behavior is dependent on the order in
which "locale -a" returns the names, which I already mentioned
I do not think we should trust to be consistent.

The previous coding applied a sort so as not to depend on what
order "locale -a" had returned things in, and I think we need
to retain that. At the very least, all the normalized names
need to be saved up and entered in a second pass.

regards, tom lane