This is related to the new AsyncMock[0] class in Python 3.8b1. A simple reproducer is:

from unittest import mock

mock_obj = mock.MagicMock()
mock_obj.mock_func = mock.MagicMock(spec=lambda x: x)

with mock.patch.object(mock_obj, "mock_func") as nested:
    print(type(nested))


Instead of a MagicMock (the behavior in Python 3.7) in Python 3.8b1 this results in an AsyncMock.

[0]https://github.com/python/cpython/pull/9296

Following up from xtreak's proposal (https://github.com/python/cpython/pull/9296) I think checking if __code__ is actually a CodeType is a good idea. It's simple and doesn't change any other functionality in an unwanted way.

I will wait for a couple of days for suggestions and will raise a PR to check for __code__ to be a CodeType. Thanks.

I created PR to ensure the __code__ object is checked to be a CodeType and converted the report to a unittest. I also found a similar function _is_async_func [0] which also seems to perform similar check but is used only in create_autospec. creating an autospec function out of MagicMock with a function spec is not possible so though the change could be made it's not testable. Also changing _is_async_func to _is_async_obj in create_autospec shows no test case failure. Can this be removed to use only _is_async_obj? Is there a difference in their usage due to isawaitable check present in _is_async_obj that needs a test?

# create_autospec with MagicMock(spec=lambda x: x)

$ cpython git:(bpo37251) ./python.exe
Python 3.9.0a0 (heads/master:7a68f8c28b, Jun 15 2019, 21:00:05)
[Clang 7.0.2 (clang-700.1.81)] on darwin
Type "help", "copyright", "credits" or "license" for more information.
>>> from unittest.mock import *
>>> create_autospec(MagicMock())
<MagicMock spec='MagicMock' id='4370353280'>
>>> create_autospec(MagicMock(spec=lambda x: x))
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/Users/karthikeyansingaravelan/stuff/python/cpython/Lib/unittest/mock.py", line 2547, in create_autospec
    mock = Klass(parent=_parent, _new_parent=_parent, _new_name=_new_name,
  File "/Users/karthikeyansingaravelan/stuff/python/cpython/Lib/unittest/mock.py", line 2066, in __init__
    super().__init__(*args, **kwargs)
  File "/Users/karthikeyansingaravelan/stuff/python/cpython/Lib/unittest/mock.py", line 1996, in __init__
    _safe_super(AsyncMagicMixin, self).__init__(*args, **kw)
  File "/Users/karthikeyansingaravelan/stuff/python/cpython/Lib/unittest/mock.py", line 1007, in __init__
    _safe_super(CallableMixin, self).__init__(
  File "/Users/karthikeyansingaravelan/stuff/python/cpython/Lib/unittest/mock.py", line 442, in __init__
    self._mock_add_spec(spec, spec_set, _spec_as_instance, _eat_self)
  File "/Users/karthikeyansingaravelan/stuff/python/cpython/Lib/unittest/mock.py", line 503, in _mock_add_spec
    res = _get_signature_object(spec,
  File "/Users/karthikeyansingaravelan/stuff/python/cpython/Lib/unittest/mock.py", line 99, in _get_signature_object
    return func, inspect.signature(sig_func)
  File "/Users/karthikeyansingaravelan/stuff/python/cpython/Lib/inspect.py", line 3093, in signature
    return Signature.from_callable(obj, follow_wrapped=follow_wrapped)
  File "/Users/karthikeyansingaravelan/stuff/python/cpython/Lib/inspect.py", line 2842, in from_callable
    return _signature_from_callable(obj, sigcls=cls,
  File "/Users/karthikeyansingaravelan/stuff/python/cpython/Lib/inspect.py", line 2292, in _signature_from_callable
    return _signature_from_function(sigcls, obj,
  File "/Users/karthikeyansingaravelan/stuff/python/cpython/Lib/inspect.py", line 2175, in _signature_from_function
    parameters.append(Parameter(name, annotation=annotation,
  File "/Users/karthikeyansingaravelan/stuff/python/cpython/Lib/inspect.py", line 2495, in __init__
    raise TypeError(msg)
TypeError: name must be a str, not a MagicMock


[0] https://github.com/python/cpython/blob/3a1d50e7e573efb577714146bed5c03b9c95f466/Lib/unittest/mock.py#L55

Thanks for the patch!

To answer your question, I do not think we can remove _is_async_func in favor of _is_async_obj, _is_async_obj will evaluate to True in cases where _is_async_func would not. 

For example:

>>> class NewCoroutine(Awaitable):
...     def __await__():
...         pass
...
>>> c = NewCoroutine()
>>> import inspect
>>> inspect.isawaitable(c)
True
>>> inspect.iscoroutinefunction(c)
False


BUT I think removing the `if getattr(obj, '__code__', None)` from `_is_async_obj` actually makes this work correctly. It is possible for a coroutine object to not have a __code__, but I don't think it is possible for a coroutine function to be missing a __code__. 

Before removing the __code__ check:

>>> from unittest.mock import _is_async_func, _is_async_obj
>>> import asyncio
>>> _is_async_obj(asyncio.sleep(1))
<stdin>:1: RuntimeWarning: coroutine 'sleep' was never awaited
RuntimeWarning: Enable tracemalloc to get the object allocation traceback
False
>>> _is_async_func(asyncio.sleep(1))
False

_is_async_obj evaluates to False when it should be True

After removing it:

>>> from unittest.mock import _is_async_func, _is_async_obj
>>> import asyncio
>>> _is_async_obj(asyncio.sleep(1))
<stdin>:1: RuntimeWarning: coroutine 'sleep' was never awaited
RuntimeWarning: Enable tracemalloc to get the object allocation traceback
True
>>> _is_async_func(asyncio.sleep(1))
False

It correctly evaluates to True

All tests pass as well. What do you think?

> BUT I think removing the `if getattr(obj, '__code__', None)` from `_is_async_obj` actually makes this work correctly. It is possible for a coroutine object to not have a __code__, but I don't think it is possible for a coroutine function to be missing a __code__. 

Sorry, I am little confused here. If the code attribute check is removed then my test case in PR fails since obj.__code__ that is passed through iscoroutinefunction returns True. Maybe something along the lines of below that if there is a __code__ attribute then always check it's of CodeType. So that my test passes with MagicMock.__code__ detected.

If I understand the awaitable examples correctly, mocking the obj which is an Awaitable should be returning an AsyncMock. But obj doesn't contain __code__ and hence check for inspect.isawaitable is never done causing _is_async_obj(obj) to return False and subsequently it's patched with MagicMock.


from collections.abc import Awaitable
from unittest.mock import patch

class NewCoroutine(Awaitable):
    def __await__():
        pass

obj = NewCoroutine()

with patch(f"{__name__}.obj") as m:
    print(m)

$ ./python.exe ../backups/bpo37251_awaitable.py
<MagicMock name='obj' id='4552158896'>

On removing the __code__ attribute check my test case of MagicMock with __code__ passes through iscoroutinefunction. Perhaps an acceptable tradeoff would be to check for __code__ and if present to be a CodeType or else to resume normal check like below. This way an AsyncMock is returned. Also there is no test failure. I have less understanding on asyncio terminologies over coroutine and awaitables so feel free to correct me if I am wrong. I guess it would be also helpful to have good number of tests for different asyncio object cases so that this could also be documented.

$ ./python.exe ../backups/bpo37251_awaitable.py
<AsyncMock name='obj' id='4363294672'>

# _is_async_obj to check for __code__ to be CodeType only if present.

def _is_async_obj(obj):
    code = getattr(obj, '__code__', None)
    if code and not isinstance(code, CodeType):
        return False
    return asyncio.iscoroutinefunction(obj) or inspect.isawaitable(obj)

# Also verified asyncio.sleep() to return True for _is_async_obj with above definition

>>> from unittest.mock import _is_async_func, _is_async_obj
>>> import asyncio
>>> _is_async_obj(asyncio.sleep(1))
<stdin>:1: RuntimeWarning: coroutine 'sleep' was never awaited
RuntimeWarning: Enable tracemalloc to get the object allocation traceback
True

Yes, sorry I wasn't clear, I was thinking about the functions and testing without your PR. I think removing the __code__ object (or working around it) is the correct way to go, but just removing it wouldn't solve this particular problem.

"If I understand the awaitable examples correctly, mocking the obj which is an Awaitable should be returning an AsyncMock. But obj doesn't contain __code__ and hence check for inspect.isawaitable is never done causing _is_async_obj(obj) to return False and subsequently it's patched with MagicMock."

Exactly! This is why I think technically removing the __code__ check is correct. Probably removing the __code__ attribute for any AsyncMock that is mocking an async object and not an async function is best, but I don't know how I would do that. 

I may also be misunderstanding some asyncio concepts, that is just what I observed :)


What if instead of checking for the __code__ object at all we check if there it is a Mock object (excluding AsyncMock):
    

    def _is_async_obj(obj):
        sync_mocks = [MagicMock, Mock, PropertyMock, NonCallableMock, NonCallableMagicMock]
        if (any(isinstance(obj, sync_mock) for sync_mock in sync_mocks)
                and not isinstance(obj, AsyncMock)):
            return False
        return asyncio.iscoroutinefunction(obj) or inspect.isawaitable(obj)

I made a new branch with the changes I am suggesting here to try to be more clear: https://github.com/lisroach/cpython/tree/issue37251

What do you think?

New changeset f1a297acb60b88917712450ebd3cfa707e6efd6b by Lisa Roach in branch 'master':
bpo-37251: Removes __code__ check from _is_async_obj. (GH-15830)
https://github.com/python/cpython/commit/f1a297acb60b88917712450ebd3cfa707e6efd6b

New changeset c3008dd480d645678409273eecbfd24bbc9669d7 by Miss Islington (bot) in branch '3.8':
bpo-37251: Removes __code__ check from _is_async_obj. (GH-15830)
https://github.com/python/cpython/commit/c3008dd480d645678409273eecbfd24bbc9669d7

I don't think this was actually fixed for the create_autospec case. create_autospec still uses the only is_async_func check to enable use of AsyncMock and that still does a __code__ check.

There was a test submitted to check this case but the test itself was bugged and discovered in the process of implementing https://bugs.python.org/issue43478.