homepage

test_shutil fails with the following error under some AIX 5.3 machine:
======================================================================
FAIL: test_on_error (__main__.TestShutil)
----------------------------------------------------------------------
Traceback (most recent call last):
 File "Lib/test/test_shutil.py", line 35, in test_on_error
 shutil.rmtree(TESTFN, onerror=self.check_args_to_onerror)
 File "/qakplus/qa_test/SandBoxes/Antoine/Python-2.5.2/Lib/shutil.py",
line 161, in rmtree
 onerror(os.listdir, path, sys.exc_info())
 File "Lib/test/test_shutil.py", line 49, in check_args_to_onerror
 self.assertEqual(func, os.remove)
AssertionError: <built-in function listdir> != <built-in function remove>
The problem seems to be that the system is stricter with permissions
(please note I don't know AIX at all, and the filesystems are unknown
brands of network filesystems :-)), and @test in that test case is set
to chmod 0400:
$ chmod 400 @test
$ ./python -c "import os; print os.listdir('@test')"
Traceback (most recent call last):
 File "<string>", line 1, in <module>
OSError: [Errno 13] Permission denied: '@test'
$ chmod 100 @test
$ ./python -c "import os; print os.listdir('@test')"
Traceback (most recent call last):
 File "<string>", line 1, in <module>
OSError: [Errno 13] Permission denied: '@test'
$ chmod 500 @test
$ ./python -c "import os; print os.listdir('@test')"
['a']
The error was witnessed with Python 2.5.2 but the test case doesn't seem
to have changed in trunk so it should be the same there too.

Here is a working patch.
I've also added a try/finally block, because otherwise the following
tests in regrtest.py couldn't run properly (@test was unremovable
because of wrong permissions).

Is there any particular reason to assert for failed function at all?
This test seems to be for 'onerror' function and the test would be valid
even without asserting whether the failed API is 'remove' or 'listdir'
etc. Isn't it?

Apparently it dates back to this commit:
http://hg.pitrou.net/public/cpython/trunk/rev/56254b99fb78
Perhaps someone should ask the original author. But indeed I'm not sure
why it's testing such implementation details. The only requirement about
the first argument passed to the onerror callback is that it's one of
(os.listdir, os.remove, os.rmdir).

For info: I have no problem with test_shutil on AIX 6.1 with python 2.6.6 or python 3.2alpha2.
There may be no problem anymore.

Ok, closing. The original install was on a rather strange setup.