I am not sure of the exact cause about this failure but `./python.exe -m test -m test_gc` fails though `./python.exe -m test -v test_gc` passes. This test was recently added with 175421b58cc97a2555e474f479f30a6c5d2250b0 and issue36016.

./python.exe -m test -m test_gc
== CPython 3.8.0a2+ (heads/master:f684d83d86, Mar 1 2019, 10:39:16) [Clang 7.0.2 (clang-700.1.81)]
== macOS-10.10.4-x86_64-i386-64bit little-endian
== cwd: /Users/karthikeyansingaravelan/stuff/python/cpython/build/test_python_21045
== CPU count: 4
== encodings: locale=UTF-8, FS=utf-8
Run tests sequentially
0:00:00 load avg: 1.85 [  1/420] test_grammar
[snip]
0:00:36 load avg: 2.17 [156/420] test_gc -- test_future5 run no tests
test test_gc failed -- Traceback (most recent call last):
  File "/Users/karthikeyansingaravelan/stuff/python/cpython/Lib/test/test_gc.py", line 775, in test_get_objects
    self.assertNotIn(l, gc.get_objects(generation=2))
AssertionError: [[...]] unexpectedly found in <list object at 0x112b5d668>
0:00:48 load avg: 2.14 [157/420/1] test_gdb -- test_gc failed

It seems that some collection from some other test is happening between the three calls to get_objects:

self.assertIn(l, gc.get_objects(generation=0))
self.assertNotIn(l, gc.get_objects(generation=1))
self.assertNotIn(l, gc.get_objects(generation=2))

The easiest solution is deactivating the gc at the beginning of the test and reactivating it afterwards, as the test is relying on manual collection. In this way, external collections cannot affect the test. I will prepare a PR.

This may be more complicated that it seems as these two statements are true at the same time:

l in gc.get_objects(generation=0)
True
p l in gc.get_objects(generation=2)
True

But this only happens when running the test suite as ./python.exe -m test -m test_gc

Ok, this is happening because there is a unittest.mock._ANY in the second generation.

If I understand this correctly any combination that imports mock._ANY affects test_gc like below combination that uses mock._ANY causes test_gc to fail ?

./python.exe -m unittest unittest.test.testmock test.test_gc
[snip output]
----------------------------------------------------------------------
Ran 404 tests in 8.551s

FAILED (failures=1, skipped=1)

Yup, I was actually using:

./python.exe -m test  test_asyncio test_gc -m test_gc 

when I found out thst the core cause was mock._ANY :)

New changeset d60a79a1015aa26ff7ee3166820ec43558911b60 by Pablo Galindo in branch 'master':
bpo-36155: Check for identity on test_gc.test_get_objects (GH-12116)
https://github.com/python/cpython/commit/d60a79a1015aa26ff7ee3166820ec43558911b60

Ok, this one has been fun :)

Thanks for finding this one @xtreak!

No problem. Thanks for fix :)

I stumbled upon it due to a typo where I used -m instead of -v in python -m test -m test_gc instead of python -m test -v test_gc . Any suggestion on how you debugged it was mock ?

Well, unless there was a bug on the gc, the only way the list l could be on both lists is if in one of them there was something that was saying that is equal to it. To confirm I checked what was equal to l in the second generation and I saw it was mock.ANY 

:)

Thanks for the explanation :)