unused_parens lint suggestion also removes attributes on inner expression with feature(stmt_expr_attributes).

[zachs18]

I tried this code:

#![feature(stmt_expr_attributes)]
#![allow(unused_must_use)]

pub fn foo() -> impl Fn() {
    let _ = (#[inline] || println!("Hello!"));
    (#[inline] || println!("Hello!"));
    (#[inline] || println!("Hello!"))
}

I expected to see this happen: The unused_parens warning fires on all three lines of foo, and the suggestions should only suggest removing the parentheses, not the attribute also.

Instead, this happened: The unused_parens fires on the first and last lines of foo, and the suggestion also removes the #[inline].
No unused_parens warning fires on the second line (#[inline] || println!("Hello!")); (but this is probably a separate issue, unrelated to attributes; it happens even when the attribute is removed; maybe unused_must_use overrides it or something).

warning: unnecessary parentheses around assigned value
 --> a.rs:5:13
  |
5 |     let _ = (#[inline] || println!("Hello!"));
  |             ^^^^^^^^^^^                     ^
  |
  = note: `#[warn(unused_parens)]` on by default
help: remove these parentheses
  |
5 -     let _ = (#[inline] || println!("Hello!"));
5 +     let _ = || println!("Hello!");
  |

warning: unnecessary parentheses around block return value
 --> a.rs:7:5
  |
7 |     (#[inline] || println!("Hello!"))
  |     ^^^^^^^^^^^                     ^
  |
help: remove these parentheses
  |
7 -     (#[inline] || println!("Hello!"))
7 +     || println!("Hello!")
  |

warning: 2 warnings emitted

Meta

rustc --version --verbose:

rustc 1.82.0-nightly (0d634185d 2024-08-29)
binary: rustc
commit-hash: 0d634185dfddefe09047881175f35c65d68dcff1
commit-date: 2024-08-29
host: x86_64-unknown-linux-gnu
release: 1.82.0-nightly
LLVM version: 19.1.0

Requires use of #![feature(stmt_expr_attributes)], otherwise the #[inline] is not legal in that position (though note that the unused_parens warning and suggestion is the same on stable, there's just an error[E0658]: attributes on expressions are experimental first).

The issue is not specific to closures and #[inline], but that was the simplest example of a reasonable attribute-on-expression I thought of.

@rustbot label +A-diagnostics +requires-nightly

[surechen]

@rustbot claim