test_posix: test_sched_setaffinity() fails on CentOS9

[vstinner]

Bug report

Example of AMD64 CentOS9 3.x failure: https://buildbot.python.org/all/#/builders/838/builds/5082

======================================================================
FAIL: test_sched_setaffinity (test.test_posix.PosixTester.test_sched_setaffinity)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/home/buildbot/buildarea/3.x.cstratak-CentOS9-ppc64le/build/Lib/test/test_posix.py", line 1343, in test_sched_setaffinity
    self.assertRaises(OSError, posix.sched_setaffinity, 0, [])
    ~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
AssertionError: OSError not raised by sched_setaffinity

Linked PRs

• gh-117061: Fix test_posix.test_sched_setaffinity() on RHEL9 #117126
• [3.12] gh-117061: Fix test_posix.test_sched_setaffinity() on RHEL9 (GH-117126) #117137
• [3.11] gh-117061: Fix test_posix.test_sched_setaffinity() on RHEL9 (GH-117126) #117138

[serhiy-storchaka]

The test calls sched_setaffinity() with a clear CPU affinity mask.

From the man page:

EINVAL

The affinity bit mask mask contains no processors that are currently physically on the system and permitted to the process according to any restrictions that may be imposed by the "cpuset" mechanism described in cpuset(7).

The operation should fail with EINVAL since the clear mask obviously does not contain any processors. Was something changed in a new kernel?

[vstinner]

IMO we should remove the test self.assertRaises(OSError, posix.sched_setaffinity, 0, []). We have to test that Python pass the right values to the syscall, but not test all cases of the syscall. I don't know why Centos9 kernel behaves differently than my Fedora 39 kernel.

[serhiy-storchaka]

Well, I think that we should test that it does not raise other exceptions (TypeError, SystemError, etc).

[vstinner]

Ok, I will propose a change to expect either: OSError or <no exception>.