On Sat, Dec 06, 2014 at 10:04:06PM +0100, Torsten Bögershausen wrote:
> I get this:
>
>
> expecting success:
> check_language "ko-KR, *;q=0.1" ko_KR.UTF-8 de_DE.UTF-8 ja_JP.UTF-8 en_US.UTF-8 &&
> check_language "de-DE, *;q=0.1" "" de_DE.UTF-8 ja_JP.UTF-8 en_US.UTF-8 &&
> check_language "ja-JP, *;q=0.1" "" "" ja_JP.UTF-8 en_US.UTF-8 &&
> check_language "en-US, *;q=0.1" "" "" "" en_US.UTF-8
>
> --- expect 2014-12-06 21:00:59.000000000 +0000
> +++ actual 2014-12-06 21:00:59.000000000 +0000
> @@ -1 +0,0 @@
> -Accept-Language: de-DE, *;q=0.1
> not ok 25 - git client sends Accept-Language based on LANGUAGE, LC_ALL, LC_MESSAGES and LANG
I can reproduce the same problem here (Debian unstable). I actually ran
into three issues (aside from needing to use Junio's SQUASH commit, to
avoid the "\r" bash-ism):
1. I couldn't build without including locale.h, for the
definition of setlocale() and the LC_MESSAGES constant (both used
in get_preferred_languages).
I'm not sure what portability issues there are with including it
unconditionally. Should this possibly be tied into gettext.c, which
already uses setlocale?
2. The call to setlocale(LC_MESSAGES, NULL) in get_preferred_languages
always returns "C" for me. This seems related to building with
NO_GETTEXT (which I typically do), as we never init setlocale
if NO_GETTEXT is set. This program demonstrates it:
#include <stdio.h>
#include <string.h>
#include <locale.h>
int main(int argc, char **argv)
{
if (argv[1] && !strcmp(argv[1], "init"))
setlocale(LC_MESSAGES, "");
printf("%s", setlocale(LC_MESSAGES, NULL));
return 0;
}
If I run it as "LANG=en_US.UTF-8 ./a.out", it prints "C". If I run
it as "LANG=en_US.UTF-8 ./a.out init", it prints "en_US.UTF-8". I
think we either need to start unconditionally calling setlocale()
as we do in git_setup_gettext, or we need to tie your feature to
using gettext.
This is what causes the failure of the de-DE test for me; building
without NO_GETTEXT makes it work. Note that this doesn't affect the
first test for ko-KR, because that test sets LANGUAGE, which we
read ourselves (so we never make a setlocale() call).
3. Even building with NO_GETTEXT, setlocale() does not want to
report ja_JP.UTF-8 for me, making the third test fail.
I think the issue is that I do not build the ja_JP locale on my
system. Running "dpkg-reconfigure locales" and asking it to build
ja_JP.UTF-8 makes the test pass. This is somewhat of a Debian-ism.
From "man locale-gen":
By default, the locale package which provides the base support
for localisation of libc-based programs does not contain usable
localisation files for every supported language. This limitation
has became necessary because of the substantial size of such
files and the large number of languages supported by libc. As a
result, Debian uses a special mechanism where we prepare the
actual localisation files on the target host and distribute only
the templates for them.
I suspect it is inherited by Debian derivatives like Ubuntu. But I
also don't know that we can count on other platforms having all of
the locales either (e.g., they may ship them as separate packages,
not all of which are installed).
So I'm not sure of an easy way around this. You want 4 separate
locales to thoroughly test, but you cannot rely on any particular
locale being present on the user's system.
Note that this is just a problem with the tests, probably not with
the feature itself. Presumably people setting LANG=ja_JP actually
have that locale on their system (though technically this feature
is about asking the _server_ to use that language, it seems like
you would do so because you were using that language locally, too).
-Peff
--
To unsubscribe from this list: send the line "unsubscribe git" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html