Mark desugaring of ..= as such when lowering

[samueltardieu]

When lowering, a..=b is transformed into RangeInclusive::new(a, b) without any desugaring information on the spans. It looks like a..=b requires Rust 1.27 (stabilization of RangeInclusive::new()), while a..=b was introduced in Rust 1.26.

Marking the desugaring as such in the HIR allows Clippy to filter it out and realize that this is not the original code coming from the user.

[rustbot]

r? @davidtwco

rustbot has assigned @davidtwco.
They will have a look at your PR within the next two weeks and either review your PR or reassign to another reviewer.

Use r? to explicitly pick a reviewer

[compiler-errors]

Why is this special-cased for RangeClosed? Shouldn't this also be applied to the open ranges?

[samueltardieu]

Why is this special-cased for RangeClosed? Shouldn't this also be applied to the open ranges?

Other kind of ranges build a struct instead of calling an initializer function, but this applies as well. I have updated the patch to do so.

I also made expr_field() private as it is not used anywhere, which let me change the span parameter into lowered_span as I want to use a span marked for desugaring. Would it be better to inline this function in lower_expr_range() instead, or is it ok like that?

[matthewjasper]

The compiler generally checks for hir::QPath::LangItem(...) to find range patterns (e.g. https://github.com/rust-lang/rust/blob/54cdc751df770517e70db0588573e32e6a7b9821/compiler/rustc_hir_typeck/src/method/suggest.rs#L2395-2407)

[samueltardieu]

The compiler generally checks for hir::QPath::LangItem(...) to find range patterns (e.g. https://github.com/rust-lang/rust/blob/54cdc751df770517e70db0588573e32e6a7b9821/compiler/rustc_hir_typeck/src/method/suggest.rs#L2395-2407)

We could specialize the MSRV check in Clippy to try and detect this desugaring, but we would have to also compare the source code strings to distinguish between a desugaring and a genuine user call to RangeInclusive::new(). It seemed simpler to use the dedicated mechanism present in the compiler for marking desugaring happening at lowering time.

[matthewjasper]

Writing RangeInclusive::new() wouldn't produce a QPath::LangItem

[samueltardieu]

Writing RangeInclusive::new() wouldn't produce a QPath::LangItem

Good point. While you're right on this, I can't understand if you're objecting to marking the desugaring as a desugaring and you suggest Clippy special-cases this particular desugaring MSRV (none other is), or if this was just a side remark.

[samueltardieu]

@matthewjasper While I would prefer to mark desugared hir as such, your comments made me realize that this is not necessary for what I want to achieve, as excluding all the QPath::LangItem from MSRV checking will give me the extra bit of information I need. Thanks!

I'll close this PR.