Clean up unused variables after fixes

[tylerlaprade]

x = None
try:
    x = func_that_might_throw()
except AttributeError:
    pass
return x

is getting simplified to

x = None  # Variable "x" is not accessed
with contextlib.suppress(AttributeError):
    return func_that_might_throw()

If this is the first time we've seen x, can we clean up the null assignment?

[zanieb]

Thanks for the report! Seems reasonable to me although it may be non-trivial.

[charliermarsh]

I believe this works as expected, assuming you enable the right set of rules: ruff check foo.py --select SIM --select RET --select F -n --fix.

We like to treat each fix as independent. Here, the SIM105 rule gives you:

import contextlib
def f():
    x = None
    with contextlib.suppress(AttributeError):
        x = func_that_might_throw()

    return x

Then the RET rule simplifies to:

import contextlib
def f():
    x = None
    with contextlib.suppress(AttributeError):
        return func_that_might_throw()

Then the F rule identifies the unused variable:

import contextlib
def f():
    with contextlib.suppress(AttributeError):
        return func_that_might_throw()

(This all happens in one ruff invocation, I'm just breaking it down for clarity.)

Perhaps you have F841 disabled?

[tylerlaprade]

Ahh, @charliermarsh is correct.

unfixable = ["B007", "ERA001", "F841", "RUF100"]

We disabled autofix on F841 because variables are often temporarily unused while rearranging a file. We don't want an errant ⌘ + S to accidentally delete useful code. In this case, though, there's no danger since it's only unused as a result of the other (safe) autofixes.

[zanieb]

@tylerlaprade I agree that seems like a fair use-case. We probably don't want to resolve this just for this specific rule though, instead we should work on a general solution that checks for unused variables introduced by an autofix.

Would you mind if I retitle this issue?

[tylerlaprade]

Yes, please do! I'm glad this could potentially turn out to be more generally applicable than just a single rule.

[zanieb]

In the meantime, for this use-case, I'd recommend leaving F841 enabled in your configuration and only disabling it in the flags passed when used by the IDE. Would that help?

[tylerlaprade]

Yes, that's a good suggestion, thank you Zanie!

[tylerlaprade]

Ah, no luck - that same flag is used for the recursive fixes, so ⌘ + S still has the same behavior.

[charliermarsh]

I think the suggestion was to leave you with a setup such that unused variables would be removed when running via the CLI, but not when running in the editor (which could happen "by accident").