question: how to type compatibility-only args / kwargs?

[MarcoGorelli]

I'm trying to use pyright --verifytypes to check which public methods are untyped, and a lot of the results come from kwargs and args.

There seem to be 3 cases for them:

1. they're undocumented, and just silently exist to allow for numpy-compatible code. For example: Series.std
2. they get passed down to a third-party function. For example: most plotting funtions
3. when they get passed to other pandas functions which have known args/kwargs. For example, the kwargs in Series.pct_change.

How should kwargs and args get typed in these cases? I'm thinking:
1., 2.: use Any
3.: make a Protocol with only the args/kwargs which can be accepted

I think if we can get these typed (even if just with Any) then pyright's verifytypes should become a lot more useful to identify missing parts of the public API surface

repo where I'm developing this type-coverage-measurement functionality: https://github.com/MarcoGorelli/measuring-pandas-type-completeness (note: it's currently quite patchy, but I've already found it useful to decide what to focus on)

[Dr-Irv]

We've used *args: Any, **kwargs: Any, in various places, just not consistently. I'm fine if you want to introduce that.

In the cases where we know that the possible kwargs are limited (e.g., Series.pct_change(), it would be best to just replace kwargs with the possible keywords that could be used. But as a first pass, we could just let it be **kwargs: Any