Andreas Fromm <Andreas.Fromm@physik.uni-erlangen.de> writes:
> Tryed it out, but the initdb still fails:
> The database cluster will be initialized with locales:
> COLLATE: de_DE CTYPE: de_DE MESSAGES: C
> MONETARY: de_DE NUMERIC: de_DE TIME: de_DE
> initializing pg_shadow... FATAL: XX000: failed to initialize
> lc_messages to ""
> LOCATION: InitializeGUCOptions, guc.c:1871
Hm. After looking at the code a little bit, I think I was mistaken to
suppose there was something wrong with LC_MESSAGES in particular.
It looks like this is simply the first place that tries to set a locale
setting and checks the return value from setlocale(). So the most
likely theory is that *all* setlocale calls are failing, ie, there's
something broken about your system's locale configuration.
I don't know enough about locale stuff to know how to fix that;
you might try Tomka Gergely's advice to start with.
> In /etc/locale.gen there is nothing set, but shouldn't I be able to
> create a UTF-8-aware DB on a machine that doesn't have localisation
> support at all?
This is unrelated to your character-encoding desires. You have a broken
system library that Postgres depends on. It would still depend on it no
matter what -E value you ask for.
regards, tom lane