Fails mypy --strict because of reexports

[AMDmi3]

Long story short

With python 3.7, fresh mypy 0.711:

import aiohttp
aiohttp.ClientSession

% mypy --strict test.py
test.py:3: error: Module has no attribute "ClientSession"

This happens because mypy --strict expects imported stuff reexported as in

from .client import BaseConnector as BaseConnector

while aiohttp/__init.py__ just imports stuff:

from .client import (BaseConnector, ...)

Related: python/mypy#7067

Expected behaviour

Clean mypy --strict, as these errors break checking code which uses aiohttp

[webknjaz]

Feel free to send a PR :)

[AMDmi3]

Would it be OK stylewise to just convert to a bunch of from .foo import bar as bar lines?

[webknjaz]

That's a question to @asvetlov. I don't mind personally.

[asvetlov]

Would it be OK stylewise to just convert to a bunch of from .foo import bar as bar lines?

Yes. Whatever works for you and isort tool :)

Please make sure that both __init__.py and web.py are processed.
[mypy] section from setup.cfg worth to be upgraded to support the most strict mode.
The problem is that mypy doesn't support strict=true in config files but requires decomposition of this mode into a bunch of separate flags. mypy is expanding, new options are added to the tool but we didn't change our config file yet.

[AMDmi3]

Pull requests are ready. I'm not touching other mypy stuff yet as more work is required there, and these changes seem to be sufficient to unbreak consumer code.

[AMDmi3]

Ping?

[webknjaz]

I'm deferring the merge to @asvetlov

[asvetlov]

Fixed by #3872

[webknjaz]

should we also make our CIs use strict mode?

[asvetlov]

Make sense.
mypy changes a set of concrete options implied by --strict.
Maybe we have to use mypy --strict explicitly instead of a bunch of detailed options in a config file.

[webknjaz]

I don't know why our CI didn't catch it then

[AMDmi3]

Mypy version for CI is fixed at rather old 0.670 for aiohttp, while the latest version is 0.720 - that's why many issues are not caught. Bumping it would require fixing a handful of them first (I've tried, should be not too hard for someone more familiar with the code).

Regarding strict mode, it would be nice to enable it too, but note that it can't be set via setup.cfg (it's just not supported by mypy), so it's only possible to list all options which --strict enables explicitly. Here's what I use in setup.cfg:

[mypy]
#strict = True  # not supported
warn_unused_configs = True
disallow_subclassing_any = True
disallow_any_generics = True
disallow_untyped_calls = True
disallow_untyped_defs = True
disallow_incomplete_defs = True
check_untyped_defs = True
disallow_untyped_decorators = True
no_implicit_optional = True
warn_redundant_casts = True
warn_unused_ignores = True
warn_return_any = True
implicit_reexport = False

I also set warn_unreachable = True, it's new in 0.720 and not included into --strict, but I find it quite useful for it revealed a bunch of problems in my code. There are some false positives which can be fixed by adding explicit optional types. Example:

foo = None
...
foo = 1
if bar && foo:  # "right hand of && is always false"
   ...  # "dead code"

Fixed like this:

foo: Optional[int] = None
...
foo = 1
if bar && foo:
   ...

[webknjaz]

@AMDmi3 I'd be glad to accept a series of PRs gradually fixing those mypy issues along with adding options and modifying mypy invocations in the test runner scripts.

[AMDmi3]

Ok, I'll try to figure something out.

[AMDmi3]

See #3927