According to mypy_primer, this change has no effect on the checked open source code. 🤖🎉

This change breaks a lot of code that assumes the list doesn't contain Nones. In most cases, it can't contain Nones, because the regex used doesn't contain optional groups.

We have some options:

• Change the return type to List[Optional[str]] as in this PR. This breaks a lot of code, and fixing it requires annoying assert thing is not None checks or similar.
• Change the return type to List[Any]. Might hide bugs that type checkers could otherwise find.
• Do nothing. This seems like the best option to me.

What do you think?

This can also be implemented more precisely in a type checker plugin that could look at the regex to see if there are optional groups.

I guess so.

I was just startled when mypy assured me that split returns list[str] but my code gave me Nones. However annoying it might be in the former case, the latter case is misleading... so I don't know what's right and what's not.

There's not a good way to express this in stubs. If you want, you can create a feature request to mypy.

See #3902 for discussion about match.group(). I think the best solution would be to write at least a mypy plugin that overrides the stubs and then fix the stubs to match the implementation.

Another option is to use List[Union[AnyStr, Any]]. This may be a bit non-intuitive, but it means that the value can be str/bytes or something else. This provides most of the type checking precision we had before, but it's also more correct, as the Any type indicates that other values are possible. It also wouldn't break existing code, and it would be better for mypyc and other tools that do strict runtime checking based on types.

Type checkers can still optionally special case this based on the regular expression string and provide more precise types when possible.

Yeah that sounds nice.

According to mypy_primer, this change has no effect on the checked open source code. 🤖🎉