In Lib/unittest/test/testmock/testmock.py, there are two test_attribute_deletion. One of them is overshadowed (not executed) by the other.

    def test_attribute_deletion(self):
        # this behaviour isn't *useful*, but at least it's now tested...
        for Klass in Mock, MagicMock, NonCallableMagicMock, NonCallableMock:
            m = Klass()
            original = m.foo
            m.foo = 3
            del m.foo
            self.assertEqual(m.foo, original)

            new = m.foo = Mock()
            del m.foo
            self.assertEqual(m.foo, new)

    def test_attribute_deletion(self):
        for mock in Mock(), MagicMock():
            self.assertTrue(hasattr(mock, 'm'))

            del mock.m
            self.assertFalse(hasattr(mock, 'm'))

            del mock.f
            self.assertFalse(hasattr(mock, 'f'))
            self.assertRaises(AttributeError, getattr, mock, 'f')

They are testing the same thing but with different expectations. The first one is invalid and should be removed. The patch altered a bit of the second test to incorporate more mock classes.

Good catch - thanks!

Although I'm not certain the first test is "invalid". It's testing a different case than the second test. So if the first test passes it should be renamed rather than removed. If it *fails* then I'd like to look at the behaviour and specify (test) that.

The first test fails. I already tested it.

I think they test the same thing, what happens to the attribute of mock instance if you delete it.

Well, they actually test slightly different scenarios - they're *not* just duplicates of each other.

Okay, I found the difference.

The second test case is to test the case on which we delete the attribute of the mock instance before being referenced by any variable. The we make sure the attribute is gone for good.

The first test case is to test the case on which we delete the attribute of the mock instance after being referenced by any variable (or maybe just needed to be used by __getattr__), then we make sure the attribute is resetted to the original value, not deleted.

I oversight this because I never thought that people could rely on this behaviour.

The duplicate_code_names.py script in issue 16079 did find that there
are duplicate test_attribute_deletion tests in testmock.py (see
http://bugs.python.org/file27376/std_lib_duplicates.txt).

On balance I think that this test failing is *correct* (after deleting an attribute the old mock should *not* come back). I'd rather adjust the test to assert this than just remove it though.

Michael, attached the patch to accommodate your request. Yeah, I think this is the best way because I have seen no possible solutions to make both tests work without altering them.

Xavier, that's cool. I didn't know such tool exists. Usually I found the duplicate function/method by using flake plugin in Vim.

Can we have a formal review on the latest patch please.

This has been fixed in https://hg.python.org/cpython/rev/c8c11082bd0c.