`impl Trait for (use<Self>) {}` unexpected token, cfgd out

[matthiaskrgr]

The following code

//@ check-pass

#![cfg(false)]

trait Foo {}
impl Foo for (use<Self>) {}

fn main(x: impl AsRef<str>) {
    let _: () = foo(Lt::<'static>::None);
}

Compiled successfully ( with --crate-type=lib --cap-lints=warn ):

However rasur was found to have errors (-e 2015):

error: found `use` but expected type
 --> ./2025_12_29__22_59_36/52989afeed5c5d1e752743e9c7a1231e5c7c9f28.rs:6:15
  |
6 | impl Foo for (use<Self>) {}
  |               ^^^ unexpected token

[matthiaskrgr]

I guess rustc actually ignores everything because of the #![cfg(false)] but rasur still tries to do its best?

[fmease]

use<Self> is actually a (bare) trait object type that contains a precise-capturing list (cf: dyn use<Self>). Syntactically (only), it's a valid type, so rustc is right. Even in modern editions, bare trait object types continue to exist (sadly).

rasur doesn't accept it yet since I haven't actually implemented bare trait object types, like at all (it's the first bullet point in issue #6). I have a local branch I'm working on as we speak that'll impl all of that.

[matthiaskrgr]

yeh I saw #6, right now bare trait object types probably are around 80-90% of my findings

[fmease]

I guess rustc actually ignores everything because of the #![cfg(false)]

Somewhat. You probably know parts of this already. #[cfg(false)] / #![cfg(false)] is a way to determine if a snippet is syntactically valid Rust since it suppresses any further semantic analysis steps.

When test-running rasur against r-l/r's test suite I actually use rustc -Zparse-crate-root-only as my fundamental measure for checking parity with rustc. It's not perfect since rustc doesn't reject "emoji identifiers" under this mode (I'm not kidding) but it's way better than trying to inject #![cfg(false)] into source files I want to check (I'd need to account for shebang and/or frontmatter and it may falsify certain results actually, -Zcrate-attrs='cfg(false)' properly injects the attr, so my concern is moot). -Zparse-crate-root-only also suppresses feature gate errors but that's actually perfect for me since I don't want to replicate feature gates (for now?) and they'd skew the results when comparing exit codes.