Peter Eisentraut <peter_e@gmx.net> writes:
> The other problem is that changing the language at runtime doesn't really
> work well, because there are obscure dependencies on LC_CTYPE and
> depending on the implementation of gettext there is some caching going on.
> So the real answer is actually, "It doesn't work, but the superuser is
> free to try."
Ugh. But LC_CTYPE isn't really going anywhere in our current setup, and
the feature would still be useful if we forced users to select their
message language at backend start time (via PGOPTIONS, for instance).
Can we get anywhere if we make those assumptions?
regards, tom lane