Possible pthread_cond_timedwait() regression on kfreebsd-i386
since recently, I get some test failure in apr on kfreebsd-i386 that seem
to indicate a bug in the kernel or libc. The failing test is part of
testlock.c and basically does a pthread_cond_timedwait() with 5 second
timeout and measures that it returns no later than 5.1 seconds after the
call. The failure happens most of the time but not always. In apr 1.4.6-3,
I have added some debug output that prints how late after the 5 seconds
the call to pthread_cond_timedwait() returns.
On fischer.d.o (which is qemu-instance, however) I got greatly varying
differences from around -2 to +1.5 seconds. The latest build of apr
1.4.6-3 on field.d.o (which is a real server) gave 1 second offset (i.e.
the call returned after 6 seconds). On all linux and the kfreebsd-amd64
buildds, the offset is less than 1 millisecond. On hurd, it's 11ms.
Therefore I think the ~ 1000 ms on kfreebsd-i386 constitute a bug. As apr
used to build reliably on kfreebsd-i386 until around march, I think this
may be a regression.
Further help to debug this (or a hint against which package I should file
a bug) is wellcome.
PS: Please cc me, I am not subscribed to this list.