False positive for PLR1706

[UnknownPlatypus]

Here is a minimal example for the issue I had with PLR1706 and the autofix using ruff v0.1.7.

-print(True and False or 2) # This print '2'
+print(False if True else 2) # This print 'False'

I used this command:

echo "print(True and False or 2)" > t.py
ruff check --select PLR1706 --isolated --preview --fix --unsafe-fixes t.py

Generally speaking, A and B or C is transformed into B if A else C but this is not equivalent when A is truphy and B is falsy because the first one returns C and the second returns B.

Let me know if I'm missing something.

[charliermarsh]

Does look like a bug, thanks for filing.

[charliermarsh]

It turns out that this rule relies on type inference. If B is false-y, we shouldn't be triggering it.

[charliermarsh]

Specifically, you have to be able to resolve B to a specific value.

Even x >= y and x or y is not safe to convert when x and y are both integers -- it fails for x, y = 0, -1. (Pylint correctly does not flag this example, but I'm more pointing out that it's a very limited rule.)

We may want to deprecate.

[UnknownPlatypus]

Ye I feel like with the current infrastructure it might be hard to get this right.
This rule is targeting a very limited use case anyway since it's supposed to be a pattern for python <2.5.

I would probably deprecate it too, and remove the autofix altogether (it only introduced behavior changes when I tried it myself so I'd rather let people do manual changes if they really want the rule).

Anyway thanks for the quick reply and it's a pleasure to use ruff.

[charliermarsh]

👍 Makes sense. I'll mark it in the 0.2.0 issue. Thanks for the kind words, I appreciate it.

[zanieb]

See #9691 which removes this rule