[pyflakes] Mark the fix as unsafe when it creates a docstring (F541)

[baltasarblanco]

Summary

The fix is now unsafe in docstring positions. The reason is that removing the f turns the f-string into a string literal, and this changes what the program exposes as documentation.

It does not touch the checker or any other rule. The scope is wider than what was reported in the issue. Listing the positions to make the scope clear:

• the first statement of a module body
• the first statement of a class body
• the first statement of a function body
• after a simple assignment (a = 1) at module level or in a class body
• after an annotated assignment (b: int = 2), likewise

I say documentation and not runtime because attribute docstrings don't touch __doc__.

c = d = 3 stays safe because this mirrors the checker's own criterion.

Fixes #18807

Test Plan

2828 passed, 0 failed. Tested with:

• new snapshot with 11 diagnostics: 6 with the unsafe note, 5 without it
• --fix without the flag on the unsafe case and the file didn't change; with --unsafe-fixes it did. The safe case was fixed without the flag
• the F541.py snapshot didn't move
• cargo clippy --workspace --all-targets --all-features -- -D warnings clean
• prek run --all-files: 14 hooks

[astral-sh-bot]

ruff-ecosystem results

Linter (stable)

✅ ecosystem check detected no linter changes.

Linter (preview)

✅ ecosystem check detected no linter changes.

[baltasarblanco]

Changes from the first push:

The helper was extracted to Checker::in_docstring_position in checkers/ast/mod.rs. The helpers F541 and ISC003 were using go away, since they now use this one. Neither rule uses &Expr because it only needs a TextRange.

convert_f_string_to_regular_string returns an Edit because Fix::applicable_edit now happens there, next to where the applicability is computed.

I applied the suggestions @ntBre asked for (## Fix safety is now a single line).

Most importantly: ISC003 inherited attribute docstrings. Its fixture had an x = 1 followed by a concatenation, commented as "Not a docstring position: not the first statement in the module body", and it's unsafe now. I fixed that comment and its snapshot changes on that particular line.

I also think it's worth mentioning that F541_docstring.py and its snapshot were replaced by resources/mdtest/pyflakes/f-string-missing-placeholders.md, with a new case (previous statement that is not an assignment).

Tests came out at 2828, 38 mdtests, clippy clean.

[astral-sh-bot]

We support using AI (i.e., LLMs) while coding. However, AI should not be used to communicate with maintainers.

Please read and follow our AI policy when submitting issues and pull requests.

[baltasarblanco]

I’ll write shorter and more direct from now on.

[baltasarblanco]

Hi @ntBre, the changes from the last review are done and the branch is up to date too!

[ntBre]

Thanks! Codex noticed a small issue with weird code like:

class C:
    if condition:
        x = 1
        f"Attribute documentation"

So I pushed a commit addressing it. This otherwise looked good to me.

[baltasarblanco]

Thanks! Codex noticed a small issue with weird code like:

class C:
    if condition:
        x = 1
        f"Attribute documentation"

So I pushed a commit addressing it. This otherwise looked good to me.

Thanks for the review. Im very focused on improving my code for this project!