Rule `A003` intended behaviour

[billf11]

Question

Hi Ruff-Team,

I am wondering what the intended behavior of the rule A003 is as the output of ruff, by running it installed with uv tool install ruff, on the following example code is rather confusing:

#!/usr/bin/env python3


class Testing:
    def __init__(self) -> None:
        self._format: str | None = None

    @property
    def format(self) -> str | None:
        return self._format

    @format.setter
    def format(self, value: str | None) -> None:
        self._format = value


if __name__ == "__main__":
    test = Testing()
    test.format = "example"
    print(f"Format is set to: {test.format}")  # noqa: T201

ruff output:

$ ruff version && ruff check test.py 
ruff 0.15.0 (ce5f7b612 2026-02-03)
A003 Python builtin is shadowed by method `format` from line 9
  --> test.py:12:6
   |
10 |         return self._format
11 |
12 |     @format.setter
   |      ^^^^^^
13 |     def format(self, value: str | None) -> None:
14 |         self._format = value
   |

A003 Python builtin is shadowed by method `format` from line 13
  --> test.py:12:6
   |
10 |         return self._format
11 |
12 |     @format.setter
   |      ^^^^^^
13 |     def format(self, value: str | None) -> None:
14 |         self._format = value
   |

Found 2 errors.

Thus, I am wondering why is it indicating two errors on line 12 in the setter decorator with reference to line 9 and 13 and not at in their corresponding line.

Thanks!

Version

ruff 0.15.0 (ce5f7b6 2026-02-03)

[ntBre]

Ruff's old diagnostic system was only capable of recording one range for a diagnostic, so we had to pick between emitting the diagnostic on the usage of the shadowed name and the place where it was shadowed. We opted for the former and included the location of the shadowing in the message. However, I think we could improve this now that our diagnostics support additional ranges! As it turns out, I previously mentioned A003 in #17203 (comment), which tracks some of our diagnostic improvements.

I think these diagnostics would probably look more sensible if the main ranges were on lines 9 and 13 and each pointed to the reference on line 12 as a secondary annotation, but we'd have to make that change in preview because it changes the noqa range. However, it also seems more intuitive to me for the noqa to go on the method doing the shadowing than on the reference to it anyway.

[ntBre]

On second thought, I guess I got a bit fixated on the diagnostic, but maybe there's an argument that @format.setter shouldn't count as a shadowed reference anyway since it's pretty clear that this is intentionally referring to the format property.

[billf11]

On further thought I agree and also started to wonder why the rule is even applied here, as it is clearly no shadowing of any built-in function. Otherwise stuff like def [built-in](self, *args, **kwargs): should trigger A003 by itself.

[dscorbett]

#7806 explains why A003 only applies to references, not definitions.