Support for pexpect.spawn(..., logfile=sys.stdout)

[PokkaKiyo]

We can set the logfile for pexpect's spawn using sys.stdout, but the type hint shows an error.

[srittau]

Thanks! Looking at the source code, it might be better to just define a custom protocol, which we can use for both SpawnBase and spawn:

class _Logfile(Protocol):
    def write(self, __s: str) -> object:...
    def flush(self) -> object: ...

Source:

https://github.com/pexpect/pexpect/blob/c53497663c1a25603ccf8ba7daca03b319387344/pexpect/spawnbase.py#L128-L130

[AlexWaygood]

Thanks! Looking at the source code, it might be better to just define a custom protocol, which we can use for both SpawnBase and spawn:

class _Logfile(Protocol):
    def write(self, __s: str) -> object:...
    def flush(self) -> object: ...

We already define a protocol like that elsewhere in typeshed. Maybe we should add it to stdlib/_typeshed/__init__.pyi?:

typeshed/stdlib/builtins.pyi

Lines 1640 to 1641 in 6c5d1d3

	class _SupportsWriteAndFlush(SupportsWrite[_T_contra], Protocol[_T_contra]):
	def flush(self) -> None: ...

EDIT: Actually, no, because we wouldn't be able to use that in our pexpect stubs until we've had a mypy release that includes the updates to the _typeshed directory in its vendored stubs. So we should just copy that protocol into our pexpect stubs, like @srittau suggests.

[srittau]

We already define a protocol like that elsewhere in typeshed. Maybe we should add it to stdlib/_typeshed/init.pyi?

While I would be ok with that, I am not a huge fan of adding all sometimes-used permutations of I/O methods to _typeshed. I'd rather just switch to intersection types (python/typing#213) at some point in the far future. That said, moving the protocol to _typeshed could help use find places where we could use intersection types in the future.

(Also, flush() should return object, not None.)

[github-actions]

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

[srittau]

Thanks!