BytesIO.name and StringIO.name should not be defined

[srittau]

Please consider the following Python 3.6 code:

from io import BytesIO


class NamedBytesIO(BytesIO):

 def __init__(self, content: bytes, name: str) -> None:
 super().__init__(content)
 self.name = name

mypy 0.560 complains:

test.py:8: error: Property "name" defined in "NamedBytesIO" is read-only

This is due to the fact that in typeshed BytesIO derives from BinaryIO, which defines the name property. The actual BytesIO implementation does not have such an attribute:

>>> from io import BytesIO
>>> BytesIO(b"").name
Traceback (most recent call last):
 File "<stdin>", line 1, in <module>
AttributeError: '_io.BytesIO' object has no attribute 'name'

Interestingly, StringIO does not have this problem:

from io import StringIO


class NamedStringIO(StringIO):

 def __init__(self, content: str, name: str) -> None:
 super().__init__(content)
 self.name = name

This checks fine with mypy, since the StringIO stub includes a name field, which also does not match the implementation:

>>> from io import StringIO
>>> StringIO("").name
Traceback (most recent call last):
 File "<stdin>", line 1, in <module>
AttributeError: '_io.StringIO' object has no attribute 'name'

[gvanrossum]

Ow, this cuts deep. It seems class IO defined in types.py has a name property. This is one of the problems with poorly specified protocols like IO -- we guess at what's part of the spec and what isn't. We'll probably need another helper class to represent the kinds of files returned by open(), and tighten up the spec of types.IO.

[srittau]

A previous issue that discussed StringIO.name: #598.

[gvanrossum]

Rather than trying to fix the deep issues (which seem impossibly hard) let's just add

name = ... # type: str

to BytesIO so it matches the hack we have for StringIO.

[gvanrossum]

(Can you make a PR?)