Re: suggestion: test error messages

[Bruno H. <br...@cl...>]

Joerg wrote:
> I wonder how many users use the french or German clisp locale.

Me at least, the German locale.

> It's full of bugs and incomplete / outdated.

This is normal, because clisp's translations are handled through the Free
Translation Project, and when you look at
   http://www.iro.umontreal.ca/translation/registry.cgi?domain=clisp
you see that no POT file has been submitted for the last 5 releases. In
other words, the translators are not even aware of clisp 2.36, 2.37, 2.38,
2.39, 2.40.

> I corrected some but this needs more time.

The maintainers (Sam, in this case) have the recommendation, from the
GNU gettext manual, to accept translation updates only through the
translator team for that particular language. So, if you want to fix
the French translations, you need to become a member of the French
translation team http://www.iro.umontreal.ca/translation/registry.cgi?team=fr
(write to Pierre Machard).

> This lead me to the following idea: the testsuite could, when ERROR is
> detected, independently try to print the error message (to a
> string-stream) and signal a distinct error when this fails.  This would
> give feedback on broken format strings.

What broken format strings? msgfmt should ensure that broken format strings
are not put into the .mo file, even if the translator made a mistake.

> I believe the testsuite already runs using the user's locale, so we'd
> get feedback on the various languages from the different users.
> 
> What I mean is that
> (if (equalp (ignore-errors (testcase)) expected) #)
> is different from
> (handler-case (testcase)
>   (:no-error ...)
>   (error (e)
>     (unless (with-output-to-string (ignore-errors (princ e s)))
>       (format log "Test additionaly uncovered unprintable error message"))))

Which bugs are uncovered by this test?

Bruno