Re: "could not adopt C locale" failure at startup on Windows

On Wed, Jun 17, 2015 at 01:43:55PM -0400, Tom Lane wrote:
> Noah Misch <noah(at)leadboat(dot)com> writes:
> > On Mon, Jun 15, 2015 at 12:37:43PM -0400, Tom Lane wrote:
> >> It's mere chance that the order of calls to pg_perm_setlocale() is
> >> such that the code works now; and it's not hard to imagine future
> >> changes in gettext, or reordering of our own code, that would break it.
>
> > pg_bind_textdomain_codeset() looks at one piece of the locale environment,
> > namely setlocale(LC_CTYPE, NULL), so the order of pg_perm_setlocale() calls
> > does not affect it.
>
> Well, my point is that that is a larger assumption about the behavior of
> pg_bind_textdomain_codeset() than I think we ought to make, even if it
> happens to be true today.

Perhaps it's just me, but I can envision changes of similar plausibility that
break under each approach and not the other. Without a way to distinguish on
that basis, I'm left shrugging about a proposal to switch. For that matter,
if pg_bind_textdomain_codeset() starts to act on other facets of the locale,
that's liable to be a bug independent of our choice here. However locale
facets conflict, we expect LC_CTYPE to control the message codeset.

> > There's nothing acutely bad about the alternative you
> > identify here, but it's no better equipped to forestall mistakes. Moving the
> > call later would slightly expand the window during which translated messages
> > are garbled.
>
> I'm not exactly sure that they wouldn't be garbled anyway during the
> interval where we're setting these things up. Until DatabaseEncoding,
> ClientEncoding, and gettext's internal notions are all in sync, we
> are at risk of that type of issue no matter what.

True; the window exists and is small enough both ways. This is merely one
more reason to fix the bug without fixing what ain't broke.