Hello, On a gcc 4.7 tree, we are witnessing build failures on Aix 5.2 with libstdc++ failing on
<<
...pthread/libstdc++-v3/include/powerpc-ibm-aix5.2.0.0/bits/gthr-posix.h:789:69:
error: 'pthread_mutex_timedlock' was not declared in this scope
>>
The anomaly is visible from the libstdc++ configure log, where we have
configure:65567: checking whether it can be safely assumed that
mutex_timedlock is available
...
configure:65587: $? = 0
configure:65602: result: yes
then shortly after an error like the one quoted above.
The check does:
AC_TRY_COMPILE([#include <unistd.h>],
[
// In case of POSIX threads check _POSIX_TIMEOUTS.
#if (defined(_PTHREADS) \
&& (!defined(_POSIX_TIMEOUTS) || _POSIX_TIMEOUTS <= 0))
#error
#endif
], [ac_gthread_use_mutex_timedlock=1], [ac_gthread_use_mutex_timedlock=0])
Eventhough aix 5.2 is supposed to be posix compliant, _PTHREADS is not
defined and the test succeeds without even checking _POSIX_TIMEOUTS.
Still, on our aix 5.2 machine we find:
/usr/include/unistd.h:#define _POSIX_TIMEOUTS (-1)
While I understand that aix 5.2 is old now, it seems to me that
the attached patch might make sense in general.
The idea is simply to test only for the POSIX_TIMEOUTS conditions
so the test will fail as well if none is met and _PTHREADS is not
defined. It still seems legitimate for the test to fail in this
situation, as the aix 5.2 case actually illustrates.
OK to commit ?
Thanks in advance for your feedback,
With Kind Regards,
Olivier
libstdc++-v3/
* acinclude.m4 (check for mutex_timelock): Check straight
on _POSIX_TIMEOUTS, regardless of defined (_PTHREADS).
* configure: Regenerate.
timedlock.diff
Description: Binary data
