Re: Remaining dependency on setlocale()

On Wed, 2025-11-12 at 19:41 +0100, Peter Eisentraut wrote:
> The impact area of LC_CTYPE is much larger and more complicated
> and possibly interacts with other settings and third-party libraries
> in
> ways that we don't understand yet and might not be able to change.
> That's why I'm more hesitant about it.

What do you think about making lc_ctype and/or lc_collate into GUCs
(like lc_messages), assuming we remove all known effects in the backend
first?

If we make the setting PGC_POSTMASTER, then it eliminates potential
problems with threading, because setlocale() would be called only once
before allowing connections. For platforms that support uselocale(), we
could allow it to be set more freely, for those who need it set to
different values in different backends.

It would also be easier to document, which would be nice. There could
be some confusion if various settings are inconsistent with each other,
but that's true currently. And we'd still enforce a ctype that's
consistent with the encoding, at least.

Regards,
    Jeff Davis