The following test fails on FreeBSD buildbot:

    def test_finalize_with_trace(self):
        # Issue1733757
        # Avoid a deadlock when sys.settrace steps into threading._shutdown
        assert_python_ok("-c", """if 1:
            import sys, threading

            # A deadlock-killer, to prevent the
            # testsuite to hang forever
            def killer():
                import os, time
                time.sleep(2)
                print('program blocked; aborting')
                os._exit(2)
            t = threading.Thread(target=killer)
            t.daemon = True
            t.start()

            # This is the trace function
            def func(frame, event, arg):
                threading.current_thread()
                return func

            sys.settrace(func)
            """)

I ran it manually on my FreeBSD 9.1 VM, I get the following error. I don't know if it's the same error than the buildbot.
----
(...)
# clear builtins._
(...)
# restore sys.stderr
# cleanup __main__
# cleanup[1] _sysconfigdata
(...)
# cleanup[1] threading
Exception ignored in: <function WeakSet.__init__.<locals>._remove at 0x801b11058>
Traceback (most recent call last):
  File "/usr/home/haypo/prog/python/default/Lib/_weakrefset.py", line 38, in _remove
  File "x.py", line 17, in func
AttributeError: 'NoneType' object has no attribute 'current_thread'
# cleanup[1] _weakrefset
(...)
# cleanup[3] _codecs
PyThreadState_Clear: warning: thread still has a frame
----

The weakref is probably threading._dangling.

IMO Py_Finalize() should first clear the trace function. Tracing Python exception while Python is dying (exiting) is insane. I'm surprised that it works :-)

I'm not sure if this issue is/was related, but it seems the commit addressing #18178 has taken care of the test_finalize_runnning_thread failure.

I note that your description specifies test_finalize_"with_trace", perhaps suggesting your reproduction case may be something slightly different?

FWIW, I don't think I've seen the 'with_trace' test fail on my buildbots (but correct me if I'm wrong)

After the commit mentioned above, the FreeBSD 10 buildbot is now green on all branches.

It may be related to #18408 (changeset 5bd9db528aed) and #18664.

I don't see this issue anymore on FreeBSD buildbots. It was probably fixed. The issue #19466 changed also the Python shutdown procedure.