`__module__` is not defined, seeming to contradict the Python Data Model.

Bug report

Bug description:

The __module__ variable might not be defined.

This is an update to the original bug post. After some investigation this doesn't seem like a bug in unittest. Here's a simple test case that shows a situation where __module__ is not defined:

test_str = """
class MyClass:
    pass

Derived = type("Derived", (MyClass,), {})
"""

test_ns = {}
exec(test_str, test_ns)
print("__module__ exists:", hasattr(test_ns["Derived"], "__module__"))
print("Module for drived:", test_ns["Derived"].__module__)

The original bug post is below:

https://github.com/python/cpython/blob/03fa2df92707b543c304a426732214002f81d671/Lib/unittest/loader.py#L220

The default test loader depends on __module__ being present. In rare cases it may not be. I encountered a crash when running unit tests using a Jupyter cell magic. The line should probably be:

            fullName = f'%s.%s.%s' % (
                testCaseClass.__module__ if hasattr(testCaseClass, "__module__") else "None", testCaseClass.__qualname__, attrname
            )

This problem has cropped up in bug reports for other projects. For example:

https://github.com/ray-project/ray/issues/4758
https://gitlab.orekit.org/orekit-labs/python-wrapper/-/issues/411

Thank you for the wonderful work!

CPython versions tested on:

3.12

Operating systems tested on:

Linux

Linked PRs

• gh-120957
• gh-120979

Start at Lib/unittest/loader.py#L220 and run the provided exec reproduction on Python 3.12. Check the default test loader’s handling of a class without module, then add or update a regression test in the unittest test suite. Done means the loader no longer crashes and the regression case passes.