homepage

I get an error when running test_wait4 with trunk on AIX:
test_wait (__main__.Wait4Test) ... FAIL
======================================================================
FAIL: test_wait (__main__.Wait4Test)
----------------------------------------------------------------------
Traceback (most recent call last):
 File "/san_cis/home/cis/.buildbot/python-aix6/3.x.phenix.xlc/build/Lib/test/fork_wait.py", line 72, in test_wait
 self.wait_impl(cpid)
 File "./Lib/test/test_wait4.py", line 23, in wait_impl
 self.assertEqual(spid, cpid)
AssertionError: 0 != 1486954
----------------------------------------------------------------------
Ran 1 test in 12.030s
FAILED (failures=1)
Traceback (most recent call last):
 File "./Lib/test/test_wait4.py", line 32, in <module>
 test_main()
 File "./Lib/test/test_wait4.py", line 28, in test_main
 run_unittest(Wait4Test)
 File "/san_cis/home/cis/.buildbot/python-aix6/3.x.phenix.xlc/build/Lib/test/support.py", line 1145, in run_unittest
 _run_suite(suite)
 File "/san_cis/home/cis/.buildbot/python-aix6/3.x.phenix.xlc/build/Lib/test/support.py", line 1128, in _run_suite
 raise TestFailed(err)
test.support.TestFailed: Traceback (most recent call last):
 File "/san_cis/home/cis/.buildbot/python-aix6/3.x.phenix.xlc/build/Lib/test/fork_wait.py", line 72, in test_wait
 self.wait_impl(cpid)
 File "./Lib/test/test_wait4.py", line 23, in wait_impl
 self.assertEqual(spid, cpid)
AssertionError: 0 != 1486954
Thanks in advance

This issue already existed on Python 2.5.2 with AIX 5.2:
http://www.mail-archive.com/python-list@python.org/msg192219.html
The documentation for WNOHANG says:
http://docs.python.org/library/os.html#os.WNOHANG
"""
The option for waitpid() to return immediately if no child process status is available immediately. The function returns (0, 0) in this case.
"""
It seems wait4 always returns 0 on AIX when WNOHANG is specified.
Removing WNOHANG will make the test succeed.
waitpid does not have the same limitation.
I suppose this is a bug of AIX, though there is not even a man page to describe wait4 on this platform.
Here is a proposition for a patch that will workaround this bug...
Index: Lib/test/test_wait4.py
===================================================================
--- Lib/test/test_wait4.py (revision 88430)
+++ Lib/test/test_wait4.py (working copy)
@@ -3,6 +3,7 @@
 
 import os
 import time
+import sys
 from test.fork_wait import ForkWait
 from test.support import run_unittest, reap_children, get_attribute
 
@@ -13,10 +14,14 @@
 
 class Wait4Test(ForkWait):
 def wait_impl(self, cpid):
+ option = os.WNOHANG
+ if sys.platform.startswith('aix'):
+ # wait4 is broken on AIX and will always return 0 with WNOHANG
+ option = 0
 for i in range(10):
 # wait4() shouldn't hang, but some of the buildbots seem to hang
 # in the forking tests. This is an attempt to fix the problem.
- spid, status, rusage = os.wait4(cpid, os.WNOHANG)
+ spid, status, rusage = os.wait4(cpid, option)
 if spid == cpid:
 break
 time.sleep(1.0)

If test_wait3 and test_fork1 pass, then yes, it's probably an issue with AIX's wait4.
See http://fixunix.com/aix/84872-sigchld-recursion.html:
"""
Replace the wait4() call with a waitpid() call...
....like this:
for(n=0;waitpid(-1, &status, WNOHANG) > 0; n++) ;
Or, compile the existing code with the BSD library:
cc -o demo demo.c -D_BSD -lbsd
Both will work...
The current problem is that child process is not "seen" by the wait4()
call,
so that when "signal" is rearmed, it immediately goes (recursively)
into the
child_handler() function.
"""
So it seems that under AIX, posix_wait4 should be compiled with -D_BSD -lbsd.
Could you try this ?
If this doesn't do the trick, then avoiding passing WNOHANG could be the second option.

I had seen that post you mentioned and already tested the -lbsd without success.
wait4 is not even present in libbsd.
phenix:~$ nm /usr/lib/libbsd.a | grep wait
phenix:~$ 
Maybe it was present on older versions of the system. But I couldn't find any documentation mentioning wait4 and -lbsd anywhere.
Actually wait4 is never mentioned in the IBM documentation concerning AIX.
wait4 without WNOHANG works fine. waitpid works fine even with WNOHANG.
I don't know which workaround is the better.
I will also try to report this bug to IBM so that a future version of AIX could work correctly.

> wait4 without WNOHANG works fine. waitpid works fine even with WNOHANG.
> I don't know which workaround is the better.
As far as the test is concerned, it's of course better to use wait4
without WNOHANG in a test names test_wait4 (especially since waitpid
is tested elsewhere)...

Yes, for the test, as I put in msg128727, it works fine by removing WNOHANG.
However I should put a note in the AIX-NOTES file to explain that wait4 is broken with WNOHANG on AIX and suggest the 2 workarounds.

The patch in msg128727 is correct for AIX and should be applied.

New changeset b3ea1b5a1617 by Antoine Pitrou in branch '3.3':
Issue #11185: Fix test_wait4 under AIX. Patch by Sébastien Sablé.
http://hg.python.org/cpython/rev/b3ea1b5a1617
New changeset 8055521e372f by Antoine Pitrou in branch 'default':
Issue #11185: Fix test_wait4 under AIX. Patch by Sébastien Sablé.
http://hg.python.org/cpython/rev/8055521e372f 

New changeset e3fd5fc5dc47 by Antoine Pitrou in branch '2.7':
Issue #11185: Fix test_wait4 under AIX. Patch by Sébastien Sablé.
http://hg.python.org/cpython/rev/e3fd5fc5dc47 

Thank you. This should be fixed now. Please reopen if not.