[pylint] - restrict iteration-over-set to only work on sets of literals (PLC0208)

[diceroll123]

Summary

Restricts iteration-over-set to only work on sets of literals. There are valid use cases for iterating over a set, for de-duplication purposes.

Fixes #11694

Test Plan

cargo test

[diceroll123]

I'm on the fence about still emitting a diagnostic without a fix even for "intentional" use of a set iteration, please advise. 😄

[github-actions]

ruff-ecosystem results

Linter (stable)

✅ ecosystem check detected no linter changes.

Linter (preview)

✅ ecosystem check detected no linter changes.

[MichaReiser]

Limiting the rule to only case where we can prove that the values are unique seems reasonable to me. Wdyt @AlexWaygood

But I suggest that we use ComparableExpr to ensure that the literals are unique.

[MichaReiser]

Technically, this also applies to literals. E.g. it's possible that someone wrote set(1, 2, 2) where fixing it to [1, 2, 2] would result in a semantic change.

That's why I'm inclined that we should instead use ComparableExpr to test if no two set entries are identical. See e.g.

ruff/crates/ruff_linter/src/rules/pyflakes/rules/repeated_keys.rs

Line 141 in aa0db33

let comparable_key = ComparableExpr::from(key);

[diceroll123]

Rule B033 catches duplicate set literals, so I've made it so this only emits a diagnostic if the set literals are unique.

The alternative is that we can still emit a diagnostic without a fix if the set literals are not unique. Let me know what you think! 😄

[MichaReiser]

LGTM, thank you. I'll let @AlexWaygood chime in to see if he agrees with the behavior change

[AlexWaygood]

I guess I'm not a massive fan of this rule in general :/

• Iteration over a list or tuple might be slightly more efficient in theory, but in practice it's a micro-optimisation that I would expect to very rarely make a difference
• Using a set rather than a tuple or list can make it clearer to readers of your code that the iteration order doesn't matter, and/or that you expect each element in the iterable to be unique
• Using a set rather than a tuple or list means you have deduplication of elements guaranteed, and means that Ruff will check your code with useful lints such as B033. With a tuple or list you could easily make an accidental typo such as (1, 1, 2, 3) when you meant (1, 2, 3).

However, all of that's pretty irrelevant to this PR, which I think pretty clearly makes the rule better!

[AlexWaygood]

Thank you!