On Wed, 27.01.10 08:42, Daniel Mack (daniel at caiaq.de) wrote: > As the subject says, tests/once-test.c fails to build on OS X because > pthread_barrier_* is unimplemented on this OS. *sigh* > > What puzzles me is that this test covers functions which aren't used > anywhere else in the PA sources. What's the idea of testing these > functions then? The code that is tested is the pa_once stuff. I added this test because I was tracking down a race and suspected the problem to be in pa_once. To increase the chance of triggering the race I used the barrier stuff so that the many different threads try their best at executing the same code at the same time. Now in the end this turned out not to be fruitful, the bug was elsewhere and outside of PA and the pa_once stuff is apparently rock solid. Since that however is also some valuable information I kept the test so that for all future we can be sure that pa_once is free of any doubt. ;-) > If the test should stay as it is, we need to build it conditionally for > !OS_IS_DARWIN. I can send a patch if you like. Yes, that should safe. Given that the pa_once code is identical for MacOS and Linux it should be safe having this test on Linux only. Lennart -- Lennart Poettering Red Hat, Inc. lennart [at] poettering [dot] net http://0pointer.net/lennart/ GnuPG 0x1A015CC4
