SIM118 Use key in vd instead of key in vd.keys() assumes __iter__ is over keys()

[taldcroft]

Example code:

class ValuesDict:
   """Dict-like container that iterates over the values."""
    def __init__(self, value):
        self.value = value

    def __iter__(self):
        yield from self.values()

    def keys(self):
        return self.value.keys()

    def values(self):
        return self.value.values()


d = {"a": 1, "b": 2, "c": 3}
vd = ValuesDict(d)

for key in vd.keys():
    print(key)

This is the original expected output:

$ python row.py
a
b
c

Applying fixes for SIM118:

$ ruff --select=SIM118 --fix row.py
Found 1 error(s) (1 fixed, 0 remaining).

$ git diff
diff --git a/row.py b/row.py
index 1e7536d..7751a3b 100644
--- a/row.py
+++ b/row.py
@@ -18,5 +18,5 @@ class ValuesDict:
 d = {"a": 1, "b": 2, "c": 3}
 vd = ValuesDict(d)
 
-for key in vd.keys():
+for key in vd:
     print(key)

This changes the code behavior:

$ python row.py                                 
1
2
3

[not-my-profile]

Yes I think we currently have many lints that assume a conventional implementation of some __dunder__ methods. I agree that we should categorize the applicability of such rules, I just opened #1997 for that :)

[charliermarsh]

Yeah this is probably impossible to avoid completely right now. But we should categorize fixes as described in the issue and consider requiring opt-in for those that could be "dangerous".

[taldcroft]

Thanks, sounds like a plan. This one certainly caught me by surprise.