Collin Funk wrote: > $ time ./gltests/test-pthread-once2 > Starting test_once ... OK > > real 8m6.167s > user 56m55.342s > sys 7m44.460s > > $ time ./gltests/test-call_once2 > Starting test_once ...Alarm clock > > real 10m4.626s > user 71m5.161s > sys 9m31.302s
So, there is definitely a problem. I would concentrate on test-pthread-once2, since pthread_once is a more common API than call_once. > When using '#define ENABLE_DEBUGGING 1': > > So I guess the printf'ing causes there to be less contention and > therefore the test passes before the alarm? Yes, it looks like this. I've seen this effect too, under other circumstances, in the past. Hope you can debug it. I can't — when I give 8 CPUs to my NetBSD VM, I can't log in to the console any more because every keystroke is repeated 3 or 4 times. Bruno
