Hi Mike,
> On Feb 9, 2017, at 1:31 AM, Dominik Vogt <v...@linux.vnet.ibm.com> wrote:
>>
>> Here's a case of the test failing now:
>>
>> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=79427
>>
>> Powerpc64 BE with glibc-2.17 (2.18 reported to work). I'd be
>> inclined to reply "upgrade Glibc to get rid of the FAIL" as that
>> is what the test is supposed to find after all. What do you
>> think?
>
> The config triplet seems to be wonderfully devoid of information. :-(
>
> They next way to do it would be to copy the <features.h> style test case
> smell out a power64 BE 2.17 system and then avoid such a system, if one
> wanted to make it prettier. Otherwise, just note in the PR this is a known
> bug in glibc 2.17 and then mark as WONTFIX. It'd leave it up to the target
> folks if they want to spend the energy on making it pretty or not. On
> darwin, we manage this, but having os releases mixed into the triplet name,
> and then we can always say darwin10* won't work. I just ran config.guess
> on my ubuntu box, and appears to the the same string is was a decade ago.
>
> The feature test would appear to be something like:
>
> #include <features.h>
> #if (__GLIBC__ < 1) || (__GLIBC__ == 2 && __GLIBC_MINOR__ < 18)
> die die die
> #endif
>
> if someone wants to make one.
I'd rather not go this route if it can be avoided: it's completely
contrary to the spirit of feature tests. If necessary, test for
__cxa_thread_atexit in libc using check_function_available, although
this seems only to apply to linux somehow.
Rainer
--
-----------------------------------------------------------------------------
Rainer Orth, Center for Biotechnology, Bielefeld University
