Re: [bug-gettext] lang-* tests

[Daiki Ueno]

Bruno Haible <address@hidden> writes:

> These tests are not pointless. They serve 3 purposes:

Thanks Bruno, those are all good points.

>   1) They are integration tests. They test the internationalization
>      from the beginning (a source file with marked strings) to the
>      end (the receipt of a localized message at runtime).
>      In many of the languages / language runtime systems such tests
>      with gettext are not part of the language package. Therefore they
>      have to be here.
>
>   2) They serve as a test for the documentation. If a test fails on
>      some platform, with some version of PHP, Vala, or whatever, then
>      - even if it's not the fault of the xgettext and msgfmt tools -
>      you know that you have to update the documentation section of
>      the gettext manual about the particular language.
>
>   3) They serve as a reference: what options need to be passed to
>      xgettext and msgfmt in order to make these tools useful for
>      the particular language. This info ought to be contained in
>      the documentation, but sometimes the documentation is too
>      terse or may be misnuderstood.

Considering those and the fact that they could fail independently with
gettext, doesn't it make more sense to actually install them?
https://wiki.gnome.org/Initiatives/GnomeGoals/InstalledTests

> Yes, the lang-* tests are portability hassles. Yes, when they fail,
> most of the time the cause is not inside the gettext package.

So, I was wondering if a failure of the lang-* tests should be
considered as a failure of the standard package building procedure
(i.e., make && make check).

We could provide a way to skip them, in a similar way that coreutils
skips expensive tests by default:
http://git.savannah.gnu.org/cgit/coreutils.git/tree/init.cfg#n362

Regards,
-- 
Daiki Ueno