I have not thought this through completely yet, but I do think that these are made slightly harder by a few factors. I think Tim is right about removing the | None when those things are non-optional.

I am, however, not sure that that will have the desired result, especially once it goes through boilerplate for the purposes of pyplot.py. Currently boilerplate does not fully contemplate overloads for the functions it generates. A quick rereading of boilerplate I believe we only check the first overload (and even if we let one particular loop continue instead of early returning, not sure it would to the right thing). It could (and maybe should) do so, but that is a larger and more complicated change. In particular, I think if you remove the | None, which is the correct signature in each individual case, the logic currently used by boilerplate may make the pyplot version break by not including the None in the type hint (but I think it will still set the default...)

One option would be to move this method out of the autogenerated portion of pyplot and just include the overloads (and subsequent implementation) in the handwritten portion. We do this for a few things already. This is perhaps a reasonable short term action (though I would have to look again at the precise actions that would need to be taken there).

One thing that complicates the analysis at least it is the inclusion of **kwargs, which means that the signatures overlap more than you want to think of them logically, which could cause additional problems for type checkers (though I think this is a minor concern)

I initially removed the | None options, as you can see from the first commit to this PR.

However, in order to keep the original behaviors of the function without introducing any regression errors, I added the | None options back. When I tested my changes with the following test cases:

1. using axline with xy2 as a parameter
2. using axline with slope as a parameter
3. using axline with both xy2 and slope as parameters
4. using axline without xy2 or slope in the parameters

Test cases 1 and 2 worked as expected and successfully drew a line, test cases 3 and 4 threw errors with a message that says only one of xy2 or slope must be presented, which were working as expected as well. Therefore confirming the current changes did not introduce any regression errors.

It does make sense to remove the | None options, but it might involve more changes as the current implementation of axline specifies the default values for xy2 and slope to be None as well. Additionally, I am not sure if I should refactor any code because of overloading the function signature (i.e the actual function implementation)