gh-64490: Argument Clinic: Add support for **kwds

[AA-Turner]

This adds a scaffold of support, initially only working with strictly positional-only arguments. The FASTCALL calling convention is not yet supported.

I expect future PRs to relax these constraints.

Currently supported forms:

• f(**kwds)
• f(a, b, /, **kwds)
• f(*args, **kwds)
• f(a, b, /, *args, **kwds)

A

• Issue: Argument Clinic should understand *args and **kwargs parameters #64490

[vstinner]

Do you plan to convert some functions to Argument Clinic thanks for this change? Or is it just for completeness?

[AA-Turner]

Do you plan to convert some functions to Argument Clinic thanks for this change?

Yes, but I didn't want to make this PR bigger by including them.

A

[vstinner]

LGTM

[picnixz]

Should we test when someone uses **kwds: object or **kwds: int? Maybe we should enforce that **kwds is always annotated with "dict"? Also, I think more tests should be added for bad usage (for *args, the coverage is quite large, see test_clinic.py).

In addition, I would suggest using __clinic_kwds instead of kwds for the local variable when we have the (... *args, **kwargs) as someone could define kwds as being one of the positional arguments. This is what we do with args where we define __clinic_args if there are non-variadic parameters.

Those are just suggestions for fortifying the current interface though so they can be addressed later.

[serhiy-storchaka]

In general, looks right.

We need tests in Python that use functions defined in Modules/_testclinic.c. We need tests for different invalid combinations (var-keyword after keyword-or-positional or keyword-only parameters, different parameters after the var-keyword parameter, wrong converter for the var-keyword parameter, default value for the var-keyword parameter, the var-keyword parameter with groups, the var-keyword parameter with the defining_class parameter).

[erlend-aasland]

Thanks, looks good!

[serhiy-storchaka]

Thank you for addressing all my comments. It seems that the tests have not expose any bugs in code. Few more suggestions and nitpicks.

[serhiy-storchaka]

While we are here, could you also add a case with a var-positional parameter inside an optional group?

And I think that it would be worth to add a test for different kinds of parameters (keyword-or-positional, keyword-only, var-positional, var-keyword) after a valid optional group (if there are no such tests yet). This may be a separate test method. Some of these cases can be supported in principle in future. We want to know if we enabled them accidentally.