Lib/threading.py gets into infinite recursion while testing (or using)
it with verbose set. This didn't happen before because Thread.__started
didn't use to be an Event, but now it is.

Typical situation:
_DummyThread calls _Thread__started.set, which calls
__started.__cond.notify_all(), which calls __cond.notify(). If there are
no waiters, it calls _Verbose._note, which calls currentThread, but if
there is no current thread _DummyThread is called, which calls
_Thread__started.set ...

Could someone with knowledge of threading provide a unit test for this, assuming that you want or need to force threading into infinite recursion.  Then we can test a patch that only moves two lines.

The patch looks ok.
A more robust solution would be to modify _note() so that it never creates a DummyThread, and instead uses the raw thread id if the current thread doesn't have a corresponding Thread object.

I spoke too fast. With the patch, test_3_join_in_forked_from_thread in test_threading still fails with _VERBOSE manually set to True.

Here is a patch with the alternative approach suggested above.
There still is the aforementioned test_threading issue when run standalone.

I can confirm that the patch fixes the recursion problem if threading._VERBOSE is set to true, but the test Antoine mentioned hangs when the test suite is run.

_VERBOSE is an internal, undocumented facility, so perhaps the priority on this isn't really "high".

On the other hand, Antoine's patch takes things from non-functional to at least partially functional, so perhaps it is worth applying as is, pending someone figuring out where the test hang is coming from.

I looked in to this briefly, but I have no clue how to trigger this in a unit test, since it seems to happen when regrtest imports logging which imports threading, and appears to my threading-ignorant eyes to be tied to conditions that only exist at initial module import.

Ok, I committed the patch in r87341 (3.2), r87342 (3.1) and r87343 (2.7).

The hanging test still hangs for me with _VERBOSE set to True on py3k trunk.

sorry, i saw the patches committed and thought that had fixed it.  oops.  still more to be done.

The threading._VERBOSE attribute was removed in issue 13550 (see also changeset http://hg.python.org/cpython/rev/8ec51b2e57c2).