PLR1736 yields invalid suggestion and unsafe fix that changes semantics.

[Skylion007]

l = list(range(1, 10))
b = []
for i, a in enumerate(l):
    if a < 5:
        a = -a
    b.append(l[i])
assert b == l, "b and l should match"

but it gets flagged and transform by the fixes into

l = list(range(1, 10))
b = []
for i, a in enumerate(l):
    if a < 5:
        a = -a
    b.append(a)
assert b == l, "b and l should match"

which is wrong and now fails the assert. Seems like the refactor may have missed this test case. I ran this test on the PR before it was merged a few weeks agi and it ran fine so it might have something to do with either last minute changes to the PR or changes @charliermarsh made to the PR after it was merged (#8932). This is a contrived example with an unused, but it's simplified from a real failure case in torch/optim/adamw.py in the pytorch repo.

Current ruff version is a build done on the latest master as of today 2023/12/2.

[Skylion007]

I flagged a similar issue here: #7999 (comment) in the PR, but that one appeared to be fixed.

[charliermarsh]

Thanks, will fix. Just confirming, this isn’t released yet, right?

[Skylion007]

Correct, and it's still a preview rule.

[charliermarsh]

Awesome, thanks for reporting from main!