The fix for issue 10326 landing on Python 2.7 trunk has interfered with a hack I wrote in Bazaar to break the reference cycles the private `unittest.TestCase._type_equality_funcs` member creates.

As the change to make pickling work is nearly enough to avoid the test and the dict referencing each other, it would be good to remove that breakage and the need for the hack in the first place.

Downstream bug:

<https://bugs.launchpad.net/ubuntu/+source/bzr/+bug/809048>

It isn't clear to me exactly what fix you are suggesting for Python 2.7?

Michael: See attached patch, that should merge up to Python 3 without too much pain as well.

New changeset 02a591131360 by Benjamin Peterson in branch '2.7':
this can be done without a custom dict (also fixes #12544)
http://hg.python.org/cpython/rev/02a591131360

New changeset 842f5ed06255 by Benjamin Peterson in branch '3.2':
this can be done without a custom dict (also fixes #12544)
http://hg.python.org/cpython/rev/842f5ed06255

New changeset 47a36d2d2b44 by Benjamin Peterson in branch 'default':
merge 3.2 (#12544)
http://hg.python.org/cpython/rev/47a36d2d2b44

...typo in the first patch, this one should be okay. But... already landed without the testcase?

What’s the typo?

About the test case: Benjamin probably judged that testing this internal detail was not worth it, as long as the code was readable and working.

Actually, what happened was I saw the bug report, didn't see your patch, and started working on my own patch. I see we came to roughly the same conclusion, though. :) About the test, I'm not sure testcases having no cyclic references is part of the guarantee of the API. Perhaps Michael can chime in on that.

I wouldn't object to having a cyclic reference test for TestCase, although if it ever becomes a problem for the development of unittest we may have to disable it again. I think Benjamin said he would like to modify the test from Martin's patch.

Done.

2011/7/15 Michael Foord <report@bugs.python.org>:
>
> Michael Foord <michael@voidspace.org.uk> added the comment:
>
> I wouldn't object to having a cyclic reference test for TestCase, although if it ever becomes a problem for the development of unittest we may have to disable it again. I think Benjamin said he would like to modify the test from Martin's patch.
>
> ----------
>
> _______________________________________
> Python tracker <report@bugs.python.org>
> <http://bugs.python.org/issue12544>
> _______________________________________
>

FTR, some assertions about GC are embedded in the tests, for example test_set.