some tests hang during make regtest on FreeBSD/amd64

[emaste]

Test stalled at:

sigsusp:         valgrind   -q ./sigsusp 

I did a kill -9 40899

 |           |         \--- 40899 emaste /usr/home/emaste/src/valgrind/./coregrind/valgrind --command-line-only=yes --memcheck:leak-check=no --tool=none -q ./sigsusp (none-amd64-freebsd)

Also

pth_cond_destroy_busy: valgrind   ./../../drd/tests/pth_cond_destroy_busy 
load: 0.39  cmd: helgrind-amd64-free 42230 [uwait] 95.66r 0.20u 0.12s 0% 42184k

[emaste]

And

pth_cond_destroy_busy: valgrind   ./pth_cond_destroy_busy 
load: 0.40  cmd: drd-amd64-freebsd 45676 [uwait] 260.04r 0.27u 0.10s 0% 28180k

[emaste]

And

trylock:         valgrind   ./trylock 
load: 1.55  cmd: drd-amd64-freebsd 47872 [uwrlck] 70.78r 0.20u 0.07s 0% 26512k

Then the test finished with == 626 tests, 208 stderr failures, 13 stdout failures, 10 stderrB failures, 6 stdoutB failures, 2 post failures ==

[markjdb]

I'll work through these.

[markjdb]

The pselect_alarm test also seems to spin forever.

[markjdb]

I've pushed fixes for a couple of hangs. The remaining hangs are caused by tests that assume specific behaviour from pthread functions in situations where the required behaviour is undefined. Basically, destroying a condition variable with a sleeper, and recursively write-locking a rwlock are the triggers. We're pretty inconsistent in how we deal with these situations; see FreeBSD bug 234805. I think we should just fix FreeBSD's libthr. With a couple of hacks to my local libthr I can get the test suite to run to completion without any hangs.

[plongstaff]

Sounds like a bug should be raised against the valgrind test if it assumes specific behaviour.

…

On Wed, Jan 9, 2019 at 4:45 PM Mark Johnston ***@***.***> wrote: I've pushed fixes for a couple of hangs. The remaining hangs are caused by tests that assume specific behaviour from pthread functions in situations where the required behaviour is undefined. Basically, destroying a condition variable with a sleeper, and recursively write-locking a rwlock are the triggers. We're pretty inconsistent in how we deal with these situations; see FreeBSD bug 234805. I think we should just fix FreeBSD's libthr. With a couple of hacks to my local libthr I can get the test suite to run to completion without any hangs. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#5 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AT2ddWuo38UgJ_Aqqc43GHMGs6Fcb1PQks5vBmL3gaJpZM4Z31Uq> .

[markjdb]

I submitted one: https://bugs.kde.org/show_bug.cgi?id=403212

I will try to fix the other issue (cond_destroy returning 0 when there are waiters) in FreeBSD libthr, as I think it's a bit of a bug.