"# type: ignore[no-redef]" causing "Module 'x' has no attribute 'y'" errors.

[jonathanslenders]

This is probably a bug in mypy.

It can be reproduced by creating a directory a with only two files in there: an __init__.py and b.py

 $ cat a/__init__.py
from .b import X

and

 $ cat a/b.py
# type: ignore[no-redef]
class X:
    pass
__all__ = ['X']

Mypy gives the following output:

$ mypy a/
a/__init__.py:1: error: Module 'a.b' has no attribute 'X'
Found 1 error in 1 file (checked 2 source files)

Remove the type: ignore[no-redef] and mypy succeeds.

I would not expect that ignoring something (even leaving type: ignore in there makes mypy fail), would break the imports.

This is an issue for prompt_toolkit: prompt-toolkit/python-prompt-toolkit#1008

[ilevkivskyi]

Hm, this is interesting. This is an artefact of how top-level # type: ignore is implemented. I assume you are aware that per PEP 484 a # type: ignore on line of its own should ignore all errors until the end of file. This is implemented in mypy by marking code after the # type: ignore as unreachable.

So this is essentially a duplicate of #7839 (there is also a workaround mentioned).

[jonathanslenders]

Thanks for the quick reply @ilevkivskyi ! I really appreciate it.

So # mypy: ignore-errors will ignore all typing errors?
Actually, I really want these files to be type checked, except for one single no-redef warning.

The code does register key bindings callbacks like this in the snippet below, and I don't like to think of hundreds of different names for functions that are never called by that name.

@handle_key('control-c')
def _():
   ...

@handle_key('control-a')
   ...

If I could configure mypy to ignore no-redef when functions are named _, that would be fine too.

Thanks again. I'll see when a fix arrives. :-)

[ilevkivskyi]

Actually, I really want these files to be type checked, except for one single no-redef warning.

Yes, support for selectively silencing error codes is planned in near future (see #7599)