Bug? Inconsistent type inference between open() and Path().open()

[BuongiornoTexas]

After spending quite a while one this, I'm not sure if this is an intended mode of operation, a mypy issue or a typeshed issue. If I have this in the wrong place or if it is not an issue, please close and I will re-raise in typeshed if appropriate.

I have run into an issue with inconsistent type-inferences for open() and pathlib.Path().open().

#!/usr/bin/env python3

from pathlib import Path
fp = open('README.rst', 'rt')
print(type(fp))
fp1 = Path('README.rst').open('rt')
print(type(fp1))

yields:

<class '_io.TextIOWrapper'>
<class '_io.TextIOWrapper'>

However, running mypy reveal_type on fp and fp1 yields:

Revealed type is 'typing.TextIO'
Revealed type is 'typing.IO[Any]'

I would expect both of these to reveal typing.TextIO. I found this issue when using a routine that switched TextIO output between sys.stdout (TextIO) and a file opened with Path().open() (IO).

This is on python 3.7.2 and mypy 0.700 with arguments: --follow-imports=silent, --show-column-numbers, --disallow-untyped-defs.

[gvanrossum]

Right now we have a plugin that special-cases builtins.open() but not Path.open(). This should be solved in typeshed by making the definition of Path.open() overloaded on various Literal types. So I am closing this as "not a mypy bug" but I recommend that you submit a PR to typeshed to add this.

[BuongiornoTexas]

Thanks. Will do.

I figured with an issue number of 6666, this was always going to be short lived :-).

[gvanrossum]

Well, there's always https://en.wikipedia.org/wiki/Chinese_numerology#Six

[ilevkivskyi]

Superseded by python/typeshed#2911