Bug#311259: error and omission in documentation of LANGUAGE
On Mon, Jun 06, 2005 at 08:24:34PM -0400, Daniel Jacobowitz wrote:
[...]
> > > They remain as C.
> >
> > Yes, LANGUAGE is only checked to select message catalogs, nothing else.
>
> This is also not documented. And it's surprising - to me anyway. The
> documentation presents it as an extension to LANG.
Basically the LANGUAGE variable is a GNU extension to allow displaying
messages in several languages, so that untranslated messages for your
favorite language can be displayed with other languages than English.
This is not a replacement for LANG; your environment must be setup
correctly to display translated messages (which means that LC_MESSAGES
and LC_CTYPE must be set and have compatible values).
E.g. debian-installer sets
LANG=nn_NO
LANGUAGE=nn_NO:nn:no_NO:no:nb_NO:nb:da:sv:en_GB:en
when Norwegian Nynorsk is selected, because Norwegian translators
believed that most Norwegian users will prefer Danish or Swedish over
English. This is fine because all these locales have the same
ISO-8859-1 encoding.
Now compare
$ env - LANG=ja_JP LANGUAGE=de_DE cat -h
cat: Ung«ältige Option -- h
,,cat --help¡È gibt weitere Informationen.
$ env - LANG=fr_FR LANGUAGE=de_DE cat -h
cat: Ungültige Option -- h
,,cat --help" gibt weitere Informationen.
The latter works as expected, but the former does not because of encoding
mismatch. In order to avoid such problems, LANGUAGE variable should almost
be set under UTF-8 locales only. You can check that the some command
with LANG=ja_JP.UTF-8 works fine.
When LC_MESSAGES and LC_CTYPE are correctly set, LANGUAGE can contain a
list of locale components, and gettext will check in turn all these message
catalogs and return the first translation of this msgid. In fact, the
initial example could be shortened to
LANGUAGE=nn_NO:no_NO:nb_NO:da:sv:en_GB
because message catalogs are also searched in xx after xx_XX.
Denis
Reply to: