fix: add missing overload for sub

[trim21]

add missing overload

class _Regexp(Generic[AnyStr]):
    @overload
    def sub(self, repl: Callable[[_Match[AnyStr]], AnyStr], text: AnyStr, count: int = 0) -> AnyStr: ...

[ddn0]

Context: The point of bytes vs AnyStr in the stubs is to convey the fact that if the input is a str the output will be a Match[str]. If we used AnyStr, then input were str then the output could be either Match[bytes] or Match[str] which loses some typing constraints.

Anyway, happy to just take the missing overload without the AnyStr change if you could scope down this PR to just that.

[trim21]

Looks like I misunderstood how re2 handle inputs and you misunderstood how AnyStr works here.

I find that methods on re2.compile(str(...)).match/findall/sub/... can handle bytes input, so it dones't need to be AnyStr. For python stdlib re, pattern and string must have same type but it's not true for re2, re2 can run bytes pattren on string or string pattern on bytes, which means re.findall("1", b"0 1 2") doesn't work but re2.findall("1", b"0 1 2") works.

And AnyStr is a TypeVar and in class _Regexp(Generic[AnyStr]) it will be narrowed base on the type of re2.compile(), but not narrowed at function/method level by the input string.

This make most methods in _Regexp have wrong typing, I'll send a seprated PR to fix them later

[ddn0]

Thanks for the investigation!

I based the stubs on the idea that re2 would be used as a drop-in for re so I made the constraints stronger than re2 itself supports, but I don't have an issue with making things match the re2 implementation more.

The main issue with that though is since this repo isn't directly related to the re2 one it is easy for the stubs to drift from the implementation in a way that wouldn't be helpful for stub users, but we can discuss that on the next PR if you want.

[ddn0]

Looking at the in progress stubs produced by the re2 maintainers, google/re2#496, it looks like they are more in line with your original AnyStr change.

[trim21]

yes, which mean using typing from re is fine.

Maybe we can just lock version to google-re2==1.1.* so downstream users will fail to resolve deps if re2 release 1.2, and we can check typing then.

[trim21]

Looking at the in progress stubs produced by the re2 maintainers, google/re2#496, it looks like they are more in line with your original AnyStr change.

can you point out where the file is? I didn't find it in that issue

[trim21]

OK, I find it at https://github.com/google/re2/compare/main...cosmicexplorer:type-stubs?expand=1

But it's in-complete, as I explained before.

checkout #5 please, I have lint error fixed in #5 and I don't want to fix it again here.