The expectation is that the __module__ attribute for a patched function should persist after patching.

Minimal test case is attached. Simply run pytest in a venv with the files.

Output:
def test_zxc():
with mock.patch.object(mymodule, 'asd', side_effect=mymodule.asd, autospec=True) as spy_asd:

      assert spy_asd.\_\_module__ == 'mymodule'

E AssertionError: assert None == 'mymodule'
E + where None = <function asd at 0x7fe4cd6fd620>.__module__

test_mymodule.py:8: AssertionError

Originally reported at pytest-dev/pytest-mock#146 before it was determined this was a unittest.mock issue.

Happens on both Python 2.7 and 3.7. Probably not really tied to a specific Python version and more of mock library issue.

My local venv:
Python 3.7.2
pytest 4.4.1

I guess the problem is that to patch 'bar' in foo.bar the signature of bar is taken and then it's evaluated using exec [1] where the function body has _check_sig that checks for the signature during function call. _check_sig has attributes of func copied. Since it's evaluated using exec the returned funcopy variable might not have __module__ set as per the execution context. One possible approach would be to _copy_func_details(func, funcopy) so that the __module__ and other attributes are copied.

This produces a test failure where in case of lambdas the __name__ is <lambda> which is not a valid identifier and hence funcopy is used. Then __name__ is set as '<lambda>' by my patch. Hence, 'funcopy' is replaced by '<lambda>' causing below test failure. This could be resolved by setting __name__ again to the expected value. This was newly added for test coverage with adbf178.

My main worry is that it's used in create_autospec hence do we really want to copy __module__ and others to the patched function for places with create_autospec also with functions supplying autospec=True. This is a user expectation but I would like to know maintainers opinion as well of this change.

Reproducer

# cat /tmp/foo.py

def bar(a: int):
    pass

# /tmp/bar.py
from unittest import mock

import foo

with mock.patch.object(foo, 'bar', autospec=True) as mocked_attribute:
    assert mocked_attribute.__module__ == 'foo'

# Python 3.7 __module__ is None
➜  cpython git:(master) ✗ python3.7 /tmp/bar.py
Traceback (most recent call last):
  File "/tmp/bar.py", line 6, in <module>
    assert mocked_attribute.__module__ == 'foo', f'__module__ is {mocked_attribute.__module__}'
AssertionError: __module__ is None

# With patch

➜ cpython git:(master) ✗ ./python.exe /tmp/bar.py # No failure

# Patch

diff --git a/Lib/unittest/mock.py b/Lib/unittest/mock.py
index 1e8057d5f5..a7006fcf7d 100644
--- a/Lib/unittest/mock.py
+++ b/Lib/unittest/mock.py
@@ -165,6 +165,8 @@ def _set_signature(mock, original, instance=False):
     exec (src, context)
     funcopy = context[name]
     _setup_func(funcopy, mock, sig)
+    _copy_func_details(func, funcopy)
+    funcopy.__name__ = name
     return funcopy

# Test failure without funcopy.__name__ = name in my patch where I don't restore the 'funcopy' for lambdas. But this can be fixed.

======================================================================
FAIL: test_spec_function_no_name (unittest.test.testmock.testhelpers.SpecSignatureTest)
----------------------------------------------------------------------

Traceback (most recent call last):
  File "/Users/karthikeyansingaravelan/stuff/python/cpython/Lib/unittest/test/testmock/testhelpers.py", line 946, in test_spec_function_no_name
    self.assertEqual(mock.__name__, 'funcopy')
AssertionError: '<lambda>' != 'funcopy'
- <lambda>
+ funcopy

Ran 386 tests in 2.911s

[1]