@larsoner I've written a simple regression test, but I'm unsure if this fits in the existing testing pipeline. Please don't hesitate to indicate any issue or possible improvement, I don't mind rewritting these to better fit the existing tests.

On another note, I've just noticed that a similar bug still exists for @property decorated methods. The fix proposed here does not work, as inspect.getsourcelines raises a TypeError when passed a property object, leading Validator.source_file_def_line to return None. This may be related to #486 and #494.

I think we may be able to fix this by passing the fget attribute of property objects, which is the getter function, to inspect.getsourcelines. I think numpydoc only parses the getter anyway, so that should not be a problem. Shall I fix this in this PR or open a new one?

EDIT: I've tested the proposed fix and it works correctly when fixing Validator.source_file_name also. I think this might require some new tests though.

Failure looks related, can you check?

I think we may be able to fix this by passing the fget attribute of property objects, which is the getter function, to inspect.getsourcelines. I think numpydoc only parses the getter anyway, so that should not be a problem. Shall I fix this in this PR or open a new one?

Probably better as a follow-up PR if we can merge this one quickly, which I think we can. But the current changes are small so if you think it makes sense to combine them then I can live with that, too.

Ok this is a weird one. Seems like inspect.getsourcelines does not catch decorators from a decorated class in Python 3.8. It does for functions though. I'm not sure why this is the case. I'll modify the tests to take it into account.

This was indeed a bug in CPython: python/cpython#60060. The fix was only implemented starting with 3.9 if I understand it correctly. Tests should be passing for 3.8 now.

Feel free to merge if everything is OK and I'll start the follow up PR. 🙂

Marking for merge-when-green, thanks in advance @sdiebolt !