F401 false positive (imported but unused)

[dllu]

from functools import reduce

import click

if click.confirm("do you want to disable reduce?", default=False):

    def reduce(f, d):  # noqa: F811
        return d


print(reduce(lambda x, y: x + y, [1, 2, 3, 4, 5]))

It is supposed to work like this:

snippets » python w0611.py
do you want to disable reduce? [y/N]:
15
snippets » python w0611.py
do you want to disable reduce? [y/N]: y
[1, 2, 3, 4, 5]

Here is what ruff outputs:

> ruff w0611.py
w0611.py:1:23: F401 `functools.reduce` imported but unused
Found 1 error(s).
1 potentially fixable with the --fix option.

If we apply the fix, then the code no longer works if the user presses "N" to not disable reduce.

> ruff --version
ruff 0.0.224

[alonme]

a simpler example without external dependencies

from math import sqrt

if True:
    def sqrt(f):
        print(f)

print(sqrt(5))

Regarding this specific issue:

pyflakes outputs

redefinition of unused 'sqrt' from line 1

pylint outputs

E0102: function already defined line 1 (function-redefined)

ruff outputs

F401 `math.sqrt` imported but unused
F811 Redefinition of unused `sqrt` from line 1

It does make sense to only output F811 in this case (ignoring if the conditional is True or False)

[charliermarsh]

Yeah -- this was an intentional deviation from Flake8, since it avoids a lot of false negatives, but perhaps the tradeoff here is wrong given that autofixing false positives is more dangerous.

[charliermarsh]

I'll likely change this today, to avoid these issues.

[charliermarsh]

Just to prove that this behavior isn't totally without merit, Ruff will flag this as unused (and remove it), unlike Flake8:

import os

def os():
    pass

But, I agree that it's probably wrong to do that right now given the false positives.

[alonme]

Just to prove that this behavior isn't totally without merit, Ruff will flag this as unused (and remove it), unlike Flake8:

import os

def os():
    pass

But, I agree that it's probably wrong to do that right now given the false positives.

just for completeness -
pylint will raise both function-redefined and unused-import

[charliermarsh]

👍 That makes sense. Pylint does more static analysis than we do right.

[charliermarsh]

The relevant PR that introduced this behavior is: #1173.

[MichaelTiemannOSC]

I came here with a problem in this file: https://github.com/MichaelTiemannOSC/pint-pandas/blob/ducks-unlimited/pint_pandas/testsuite/test_pandas_extensiontests.py

The problem is line 24. as_frame is a fixture I want to import. Later in my code (line 498) I seclare as_frame as a parameter, which is how I used the fixture imported above. I don't understand why this one example is the only thing across all my sources that trigger the problem, but it's blocking my CI/CD right now.

[charliermarsh]

Is this a pytest fixture issue? Similar to #3295 and #4046?

[MichaelTiemannOSC]

Yes, #4046 especially. Note: I'm not a maintainer of the project I'm contributing to, nor do I control the CI/CD. I'm just caught in the crossfire between how the developers factored the pytest code, what ruff thinks is OK or not, and when, if ever, the CI/CD will tell the maintainers they should bother looking at my code. I'm surprised that I cannot manually override this somehow.

[charliermarsh]

I think the error you're seeing is on line 498, not line 24:

foo.py:498:31: F811 Redefinition of unused `as_frame` from line 24

Flake8 gives the same result. You should be able to add a # noqa: F811 to line 498, or just rename that parameter?

[MichaelTiemannOSC]

Silly me! I didn't put the F811 where it belonged. Now fixed... Thanks!