Stubtest: option to ignore "runtime does not have argument" when implementation accepts *args/**kwargs

[twoertwein]

Feature

Pandas uses many decorators. Some of them change the function signature, e.g., @deprecate_nonkeyword_arguments(...) exposes a function that accepts *args, **kwargs instead of the original function. It would be nice to have an option to suppress "runtime does not have argument"-warnings when the implementation accepts *args/**kwargs.

Pitch

Add an option to deal with decorators or args/kwargs.

@hauntsaninja

[hauntsaninja]

I assume you're referring to errors like the following?

pandas.DataFrame.ffill is inconsistent, runtime argument "inplace" is not keyword-only
pandas.DataFrame.ffill is inconsistent, runtime argument "limit" is not keyword-only
pandas.DataFrame.ffill is inconsistent, runtime argument "downcast" is not keyword-only
pandas.DataFrame.ffill is inconsistent, runtime does not have argument "downcast"
pandas.DataFrame.ffill is inconsistent, runtime does not have argument "inplace"
pandas.DataFrame.ffill is inconsistent, runtime does not have argument "limit"

stubtest does actually handle decorators and args / kwargs mostly well. The issue is that pandas.DataFrame.ffill does not look like it takes *args, **kwargs. Because deprecate_nonkeyword_arguments uses functools.wraps, inspect.signature sees a more complete signature:

In [4]: inspect.signature(pandas.DataFrame.ffill)
Out[4]: <Signature (self: 'DataFrame', axis: 'None | Axis' = None, inplace: 'bool' = False, limit: 'None | int' = None, downcast=None) -> 'DataFrame | None'>

stubtest's output^ is confusing, but it basically is reporting errors twice, once when checking whether positional args match and once when checking whether keyword-only args match:

pandas.DataFrame.ffill is inconsistent, runtime argument "downcast" is not keyword-only
pandas.DataFrame.ffill is inconsistent, runtime does not have [keyword-only] argument "downcast"

If I were to fix the duplication, you'd still get the following (because inspect.signature sees the full signature because of functools.wraps):

pandas.DataFrame.ffill is inconsistent, runtime argument "inplace" is not keyword-only
pandas.DataFrame.ffill is inconsistent, runtime argument "limit" is not keyword-only
pandas.DataFrame.ffill is inconsistent, runtime argument "downcast" is not keyword-only

[twoertwein]

Thank you, that makes sense that it errors because of the arguments not being keyword-only! I focused only on the many "runtime does not have argument" warnings which did not seem very intuitive.

[hauntsaninja]

#13307 will fix the duplicate error, so will just say:

pandas.DataFrame.ffill is inconsistent, runtime argument "inplace" is not keyword-only
pandas.DataFrame.ffill is inconsistent, runtime argument "limit" is not keyword-only
pandas.DataFrame.ffill is inconsistent, runtime argument "downcast" is not keyword-only