In debug mode, on my Ubuntu Linux box, this fails on the last iteration:

$ ./python Lib/test/regrtest.py -R:20 test_ctypes
test_ctypes
beginning 25 repetitions
1234567890123456789012345
........................test test_ctypes failed -- Traceback (most
recent call last):
  File
"/usr/local/google/home/guido/python/py3kd/Lib/ctypes/test/test_values.py",
line 18, in test_an_integer
    self.failUnlessEqual(x*2, ctdll.get_an_integer())
AssertionError: 2818572288 != -1476395008

1 test failed:
    test_ctypes
[104230 refs]
$ 

Also, when using -R:19, I get some random leaks reported of 33 or -33
references leaked.  Does this test use threads?  It seems to have a race
condition.

Thanks!  (I agree with Eric Smith that this is mysterious for the
innocent bystander.)

Also, what about the -33/+33 leaks?  I suppose these are harmless, but
it would be better if the test reliably didn't report leaks...

Guido van Rossum schrieb:
> Thanks!  (I agree with Eric Smith that this is mysterious for the
> innocent bystander.)

I have added a comment.

> Also, what about the -33/+33 leaks?  I suppose these are harmless, but
> it would be better if the test reliably didn't report leaks...

I am using my own definition of leaking:  a test is leaking if references
grow without bounds ;-).

Ok, I found that the test in charge is Lib/ctypes/test/test_loading.py,
LoaderTest.test_find.  This test uses the ctypes.util.find_library function,
which contains several calls to os.popen(...).read().

The minimal code reporting the -33/+33 leaks that I found is simply this:

   os.popen('ls').read()

Is there a problem in "os.popen(...)", or do I something wrong here?

> The minimal code reporting the -33/+33 leaks that I found is simply this:
>
>    os.popen('ls').read()
>
> Is there a problem in "os.popen(...)", or do I something wrong here?

Try this instead:

 p = os.popen('ls')
 try:
   x = p.read()
 finally:
   p.close()

IOW close the pipe explicitly. That should make the code execute in a
more reproducible fashion. I've fixed this in ctypes/util.py now:
r59477.

> more reproducible fashion. I've fixed this in ctypes/util.py now:
> r59477.

Cool!  And the 'leak' is gone.