TYP: DataFrame.(dot, __matmul__)

[arw2019]

• closes #xxxx
• tests added / passed
• passes black pandas
• passes git diff upstream/master -u -- "*.py" | flake8 --diff
• whatsnew entry

xref https://github.com/pandas-dev/pandas/pull/38416/files#r549438292

[arw2019]

cc @simonjayhawkins This currently fails with mypy for reasons I haven't figured out:

pandas/core/frame.py:1162: error: Overloaded function signatures 1 and 2 overlap with incompatible return types  [misc]
pandas/core/frame.py:1280: error: Overloaded function signatures 1 and 2 overlap with incompatible return types  [misc]

but I believe what I have is the correct logic for these methods:

• if other is Series we return Series
• if other is a DataFrame, np.ndarray, ExtensionArray or Index we return a DataFrame
• otherwise we throw

[jreback]

why is the return Optional?

[arw2019]

It's a (likely misguided) attempt to please mypy. There's no return statement after the raise at the end so technically returns None, although ofc we never get there

(I'm honestly not sure why what I have, without the optional return, isn't working - I think it should be right)

[simonjayhawkins]

cc @simonjayhawkins This currently fails with mypy for reasons I haven't figured out:

from experience, error: Overloaded function signatures 1 and 2 overlap with incompatible return types occur since np.ndarray resolves to Any.

[simonjayhawkins]

Thanks @arw2019 for the PR. when adding overloads also add the types to the actual method.

The overloads are used by mypy when establishing the return type when checking functions with calls to this method. The typing on the actual method is used to check this method.

[arw2019]

cc @simonjayhawkins This currently fails with mypy for reasons I haven't figured out:

from experience, error: Overloaded function signatures 1 and 2 overlap with incompatible return types occur since np.ndarray resolves to Any.

Right. Is it right there's no great way to fix that before NumPy 1.2?

Currently I added a type: ignore which can be removed once we're checking against NumPy 1.2 (and I believe mypy will force that because the type: ignore will become unused).

That said, I know type: ignore use is an anti-pattern, so alternatively I can just delete the overloads and create an issue in the tracker to be taken up once we've upgraded NumPy.

Thanks @arw2019 for the PR. when adding overloads also add the types to the actual method.

The overloads are used by mypy when establishing the return type when checking functions with calls to this method. The typing on the actual method is used to check this method.

Thanks, that was the problem (in addition to the ndarray issue)

[jreback]

why is this ignore?

[jreback]

i see your comment, can you make this the actual type error as a comment

[arw2019]

Done

[jreback]

thanks @arw2019

[simonjayhawkins]

@arw2019 just FYI

error: Overloaded function implementation cannot satisfy signature 2 due to inconsistencies in how they use type variables [misc]

when we add numpy types.

the problem is the Series in AnyArrayLike could be a Series subclass.

I would like to change AnyArrayLike to a Union instead of a TypeVar to avoid many issues like this

good news though. the ignore above is not needed.