Improve @overload detection

[asottile]

Resolves#434

[Zac-HD]

Hurray!

This fixes another bug too: in the current released version, using a decorator named overload causes an internal error (because pyflakes thinks that it's typing.overload, and tries to access a nonexistent scope name). We first noticed this a while ago in Hypothesis' build, but it's reproducible by running pyflakes on the following file:

overload=lambdaf: f# or `def overload(arg): return arg@overloaddeffunc(arg): passdeffunc(arg): return ...

Happily I've confirmed that this change fixes pyflakes for both the minimal case above (which might be worth adding as another test case for this PR?) and for Hypothesis - thanks @asottile 😍

[asottile]

@Zac-HD 🎉 though it was probably #423 that fixed that one ~o/

[Zac-HD]

Yeah, probably... still fixed though and still fixed by you 😁

[asottile]

heh only fair -- I broke it in the first place /o\

[kelseyfrancis]

Thanks for fixing this! Could you all please cut a pypi release that includes this fix?

[omry]

When is this expected to make it to flake8?

[asottile]

no current plan, but you can read my thoughts here or as a temporary workaround install from git+https://github.com/pycqa/pyflakes@revision_here

[omry]

Looks like this issue still happens with master on python 3.8.

[asottile]

doesn't happen for me? try uninstalling pyflakes before installing from git (I suspect you're not using the actual git install)

[omry]

You are right, I had a system wide flake8 installed and bash already had it cached as flake8.

[omry]

@ asottile, unfortunately depending on a git version in setup.py prevents publishing a package.
would you be able to make a dev/pre release?

[asottile]

it's a bit unusual to depend on pyflakes in setup.py -- could you elaborate on the usecase?

(I also don't have pypi permission, I believe @bitglue manages that)

[omry]

Sure:
I am depending on flake for linting my code here.

Because of this overload bug, I am also depending on pyflake to get this bugfix.
This dependency (as well as the one with isort which is due to a similar problem) are preventing me from publishing the package as is (and I need to hack those out before publishing).

[asottile]

ah I see -- those dependencies aren't for your consumers so there isn't really a need to put them in setup.py -- maybe put them in your nox file instead?

[omry]

developers are also consumers.
I am using "pip install -e .[dev]" to setup the development environment.
I have git commit hooks that requires the proper version, It's not just a CI thing.

my noxfile is also settings up in the same way.

[asottile]

yeah I notice you're using pre-commit -- I'm the author (you've got it set up in a less-than-supported manner -- first you don't need .pre-commit-hooks.yaml that's for providing hooks, next you're using the escape hatch local + system hooks instead of versioning directly in .pre-commit-config.yaml) -- but that's a separate discussion entirely :)

when I said consumers above, I meant people that would install your package -- I understand you still want to have a developer workflow, I'm just suggesting that while setup.py does work for that, it isn't really the intended mechanism (whereas something more for your developer workflow such as nox / tox / poetry or even requirements-dev.txt is more properly suited for that)

[omry]

Thanks for the pointers! didn't realize you are the author of pre-commit hooks.
About that, I was following the instructions from Black, I was a bit confused about the purpose of the second file.
Looking now, looks like Black got better instructions. I update and clean up the mess.

[omry]

I fixed the pre-commit config for black to not use the local escape hatch, but I am not sure how to do it for flake8 given the need to install a specific revision of pyflakes.

[asottile]

something like this:

- repo: https://gitlab.com/pycqa/flake8rev: 3.7.9hooks:
- id: flake8additional_dependencies: [-e, git+https://github.com/pycqa/pyflakes@...]

[omry]

Awesome, thanks!