Fix usage of byte2int with bytes

[Avasam]

Fixes#9145 . Restores functionality specifically with bytes.
Works around a mypy bug that thinks SupportsGetItem[int, <nothing>] is expected when passed bytes. May be python/mypy#14032

indexbytes seems unnafected

[JelleZijlstra]

Could you add a test case for this? Also worth reporting as a separate mypy bug I think.

[Avasam]

I was going to but I can't replicate it anymore O_o

fromtypingimportAny, Protocol, TypeVar, Union_T=TypeVar("_T")
_KT_contra=TypeVar("_KT_contra", contravariant=True)
_VT_co=TypeVar("_VT_co", covariant=True)
classSupportsGetItem(Protocol[_KT_contra, _VT_co]):
def__contains__(self, __x: Any) ->bool:
...
def__getitem__(self, __key: _KT_contra) ->_VT_co:
...
###deffoo(obj: SupportsGetItem[int, _T]) ->_T:
returnobj[0]
bar: int=foo(b"1")
###importsixsix.byte2int(b"1")
###defbyte2int(c: Union[bytes, int]) ->int:
ifisinstance(c, bytes):
returnsix.byte2int(c)
returncbar=byte2int(b"1")

mypy 0.990 (compiled: yes)
Python 3.9.13

[JelleZijlstra]

Interesting, different mypy version?

[Avasam]

Actually it differs whether I point to my local typeshed repo or not... Even if I revert the change to byte2int.

The other culprit could be SupportsGetItem. In my example above, I redefined it, and it's the same in my local typeshed repo (same as develop). But the typeshed stubs shipped with mypy still has def __contains__(self, __x: object) -> bool: ....

The change from object to any was actually important as part of #9117

Edit: Changing __x: Any back to __x: object in SupportsGetItem indeed reproduces the issue

Then would this PR make sense as a stop-gap until type-checkers update their shipped typeshed stubs?

[Avasam]

For the record, having to use Any instead of object or _VT_co sounds an awful lot like the comment in Mapping (which references discussion of python/typing#273 )

[AlexWaygood]

A better solution might be to just copy the new version of SupportsGetItem into the stubs for six, and then use that directly instead of importing it from _typeshed (with a TODO comment saying we should switch to the _typeshed version once mypy updates its vendored copy of typeshed and makes a new release). Then we wouldn't have to make the function an overloaded function.

[github-actions]

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

[AlexWaygood]

Thanks @Avasam!