Improve error messages for unexpected keyword arguments in overloaded functions

[KevinRK29]

This PR improves error messages when calling overloaded functions with unexpected keyword arguments, making it easier to identify and fix typos.

[JukkaL]

Don't report the line number here. It could be useful to report it, but we generally use a note, since the function could be in a different file so line number by itself isn't sufficient.

[KevinRK29]

That's a good point, I'll exclude the line number from the error messaging

[JukkaL]

Additional test ideas:

• Test multiple invalid keyword arguments
• Test both invalid keyword argument and incompatible positional argument
• Test both valid an invalid keyword arguments in the same call

[JukkaL]

Can you investigate the mypy_primer diff? Were there no errors reported in the past, or do we now generate multiple errors for some calls, and are the new errors valid?

[KevinRK29]

@JukkaL so looking into this:

prefect (https://github.com/PrefectHQ/prefect)
+ src/prefect/futures.py:224: error: Unexpected keyword argument "_sync" for overloaded function "result" of "State"  [call-overload]
+ src/prefect/utilities/engine.py:764: error: Unexpected keyword argument "_sync" for overloaded function "result" of "State"  [call-overload]
+ src/prefect/task_engine.py:529: error: Unexpected keyword argument "_sync" for overloaded function "result" of "State"  [call-overload]

this seems like it's because none of the function result overloads have the parameter _sync. Apparently the parameter is injected with a decorator @async_dispatch. So it works at runtime but fails during type checking. So before this PR, it would have already had a "No overload variant of 'result' matches argument types..." error but now we throw an additional unexpected keyword argument error as well. However they type ignore the line of code (1, 2) with reportCallIssue (except this).

[KevinRK29]

This is the same trend as above with the one in the scipy repo where the __call__ overloaded function takes in axes (plural), but it allows the usage of the parameter axis during runtime.

And for the third one, the np.arange where it passes in a start keyword argument when the python function doesn't define it. But this function is written in C, so it has it's own argument parsing which allows the start keyword.

So there were errors reported in the past for all of these instances but they were of the generic message like No overload variant of "result" of "State" matches argument types .... But now we add this specific error messaging alongside it, which seems valid to me.

[JukkaL]

But now we add this specific error messaging alongside it, which seems valid to me.

Yeah, that is a reasonable change.

[JukkaL]

Additional test case ideas:

• Test a call where there are multiple positional and keyword arguments, and everything else is valid, but one of the keyword arguments is a misspelling (and not the first one).
• Test a call where there are multiple keyword arguments, and two of them are misspelling of different target keyword argument names, but otherwise the call is fine.
• A test case like obj.f(fobar=1), where we have a misspelling in a method call.

[github-actions]

Diff from mypy_primer, showing the effect of this PR on open source code:

prefect (https://github.com/PrefectHQ/prefect)
+ src/prefect/futures.py:224: error: Unexpected keyword argument "_sync" for overloaded function "result" of "State"  [call-overload]
+ src/prefect/utilities/engine.py:765: error: Unexpected keyword argument "_sync" for overloaded function "result" of "State"  [call-overload]
+ src/prefect/task_engine.py:530: error: Unexpected keyword argument "_sync" for overloaded function "result" of "State"  [call-overload]