Remove OuterAttributes for StructPatternEtCetera

[pushkine]

Rust in stable and nightly treats outer attributes above .. as whitespace.
Running cargo fmt even removes it.

Snippet:

struct Foo {
	a: u8
}

fn main() {
	let Foo { #[rest_attribute] .. } = Foo { a: 1 };
	let Foo { #[prop_attribute] a  } = Foo { a: 1 };
}

cargo run:
rest_attribute is ignored

error: cannot find attribute `prop_attribute` in this scope
 --> src\main.rs:7:14
  |
7 |     let Foo { #[prop_attribute] a  } = Foo { a: 1 };
  |                 ^^^^^^^^^^^^^^

cargo fmt:
rest_attribute is removed

struct Foo {
    a: u8,
}

fn main() {
    let Foo { .. } = Foo { a: 1 };
    let Foo {
        #[prop_attribute]
        a,
    } = Foo { a: 1 };
}

[ehuss]

Thanks for the PR! I don't think this is the correct change, as rustc does parse these attributes. This particular thing is a known issue tracked in rust-lang/rust#81282, where there is more discussion if this should be allowed or not.

[pushkine]

@ehuss
The fact that they are parsed and silently ignored is a bug, not a feature

Unless this is a "macro" situation, like visibility specifiers that are allowed but ignored in certain contexts?
In that case this should be mentioned, note on the other hand that syn does not parse it

https://astexplorer.net/#/gist/4a0c99b76353c971afe9576405234183/b8c362bf14d52215dd8410294069b10a2a305b18

[ehuss]

I don't think this is necessarily a macro situation. It's just a bug, and the lang team needs to make a call as to whether or not attributes should be allowed in that position. I personally would vote that it should be allowed to be consistent with other field patterns (and rust-lang/rust#81282 (comment) gives a motivating example).

[pushkine]

Is the argument that this is an agreed upon incoming feature that is already parsed by rustc?

I don't think that this is a feature that exists anywhere outside docs though

• The idea of such a feature was introduced by the docs 4+ years ago, there is no RFC supporting its implementation

• The compiler does not even define .. as a Node with attributes, there is a bug in rustc_parse skipping attributes silently

    /// A struct or struct variant pattern (e.g., `Variant {x, y, ..}`).
    /// The `bool` is `true` in the presence of a `..`.
    Struct(Option<QSelf>, Path, Vec<PatField>, bool),

https://github.com/rust-lang/rust/blob/21d613b1116eef86d613d5f7f10d8fa5f22e27b3/compiler/rustc_ast/src/ast.rs#L725-L727

[ehuss]

I'm not sure I understand your question. Attributes were added in rust-lang/rust#38814. There wasn't any discussion of attributes on the .. rest part. It parses it, but ignores it. I think that was an oversight, so the current behavior is a bug. Either rustc should honor the attributes, or it should reject them. That's a question for the lang team to answer in rust-lang/rust#81282.

[ehuss]

I'm going to close. Further followup should be done in rust-lang/rust#81282, and either rustc or the reference can be updated based on the outcome of that issue.