[Gluster-infra] [Bug 1757804] Code coverage - call frequency is wrong

bugzilla at redhat.com bugzilla at redhat.com
Thu Oct 3 10:21:12 UTC 2019


--- Comment #3 from M. Scherer <mscherer at redhat.com> ---
We are using the RHEL 7 version (gcov), so yeah, that's indeed old. We could
switch to Fedora builders, but that would imply that the test suite is running
on Fedora and compile on it, even with a newer GCC (as this did create issue on
the past). We can also push that on Centos 8 once we are ready for it, I didn't
had time to look yet.

After (painfully) reading the assembly, I think the problem is that
LOCK_DESTROY can take 2 paths, one using pthread_spin_destroy, another using
pthread_mutex_destroy, and this could mess with the counter somehow. Now, why
does it happen, I suspect there is somehow a bug. 

My build (on RHEL 7) do have HAVE_SPINLOCK defined (in config.log), and yet,
the compiled code for fd-lk.c seems to act as if LOCKING_IMPL was undefined
(that's in variable defined in ./libglusterfs/src/locking.c). I am not sure if
that's by design or not, it could be, but having different behavior for the
same macro strike me as unusual.

I guess someone who have a idea of the locking code on gluster should take a
look, cause we are at the limits of my skills
(still no answer from compiler folks)

You are receiving this mail because:
You are on the CC list for the bug.

More information about the Gluster-infra mailing list