cpu_affinity() segfaults on Linux on CentOS / manylinux

[giampaolo]

See: #824 (comment)

psutil.tests.test_process.TestProcess.test_cmdline ... ok
psutil.tests.test_process.TestProcess.test_cpu_affinity ... ok
psutil.tests.test_process.TestProcess.test_cpu_affinity_errs ... config.sh: line 17:    55 Segmentation fault      (core dumped) PSUTIL_TESTING=1 python -Wa ../psutil/psutil/tests/__main__.py

[giampaolo]

Note: as far as I understand the segfault occurs when using the wheel built on CentOS 5. That means a different implementation of psutil_proc_cpu_affinity_get is used because CPU_ALLOC is not defined:

psutil/psutil/_psutil_linux.c

Line 350 in fbece8e

/*

[webknjaz]

FTR I've faced a similar issue when building manylinux1 wheel for pygit2 (libgit2/pygit2#793 (comment)): tests segfaulted under the same manylinux1 container. But the wheels themselves worked under other envs + they worked after changing directory on my dev machine and crashed when I tried using those from within project dir probably picking up something from a current working directory rather than virtualenv...

[giampaolo]

To further elaborate on this, and as a note to self:

• the change was introduced because of undefined symbol: CPU_ALLOC #536
• and merged here: Fix psutil_proc_cpu_affinity_get to work without CPU_ALLOC #537
• the alternate method should be tested/fixed on CentOS

[webknjaz]

@giampaolo is this still a problem with wheels shipped as v0.28.0?