On Tue, Feb 17, 2015 at 10:07 PM, Paul Eggert <egg...@cs.ucla.edu> wrote:
> Kevin Cernekee wrote:
>>>>
>>>> test-localename is calling strdup() on a NULL pointer returned by
>>>> >>gl_locale_name_thread(), and crashing. From reading
>>>> >>gl_locale_name_thread() it seems like NULL is a valid result. Maybe
>>>> >> we
>>>> >>need to change the test case to "return 77" on systems that are
>>>> >> unaware
>>>> >>of locales?
>>>
>>> >
>>> >
>>> >Yes, that sounds right. Can you propose a patch along those lines?
>>
>> OK. Should I just check for NULL, or is there a better test that
>> would indicate that the locale support is completely absent?
>
>
> I haven't read that part of the code, but it sounds like a NULL test would
> be good enough.
What I found is that for Android API levels < 21, gnulib correctly
detects that locale support is absent and the test becomes a no-op.
For API level >= 21, the only missing piece was obtaining an ASCII
name for the current thread's locale ("C" or "C.UTF-8"), which my
patch accomplished by checking MB_CUR_MAX. So now we don't get a NULL
pointer, and the test passes on Android.
We might still want to add a NULL check in the test program, rather
than just segfaulting.
