MagicMock, from unittest.mock, has a method reset_mock, which takes optional arguments return_value and side_effect, both with default values False.

In the body of reset_mock, reset_mock is again called on all the _mock_children of of the MagicMock object. However, here the arguments are not passed. This means that if you have a MagicMock object with children that are also mocked, and methods on these have been directly mocked, then it is not enough to call reset_mock on the parent object. A code example that demonstrates this follows below. Here, we could expect m to have been completely reset. But the final print statement shows that m.a() still returns 1.

from unittest.mock import MagicMock

m = MagicMock(a=MagicMock())
m.a.return_value = 1
m.reset_mock(return_value=True)
print(m.a())

Pertinent line in Github

I said MagicMock, but I meant Mock.

Thanks Vegard for the report. The docs say "Child mocks and the return value mock (if any) are reset as well." so it makes sense to me to reset the return_value and side_effect of children. But as mentioned in the PR it's a behaviour change that someone might be dependent on and can go on 3.9 . If we are not going ahead with the change then we can clarify the same in the docs and ask the user to reset mock recursively on their own so that it avoids confusion.

Oh, right! Thanks!

New changeset aef7dc8 by Chris Withers (Vegard Stikbakke) in branch 'master':
bpo-38932: Mock fully resets child objects on reset_mock(). (GH-17409)
aef7dc8