Regression: this `#[deprecated]` annotation has no effect

[kpcyrd]

Code

I tried this code:

struct X {}

#[deprecated(since="0.2.2", note="Use the `AsRef` or `AsMut` implementation instead")]
impl ::std::ops::Index<::std::ops::RangeFull> for X {
    type Output = [u8];
    fn index(&self, _index: ::std::ops::RangeFull) -> &[u8] {
        todo!();
    }
}

fn main() {
    println!("Hello, world!");
}

This code currently compiles on stable rust, but stops working on nightly and beta. The example code is a reduced case from sodiumoxide which is currently broken on nightly and beta.

% cargo check      
    Checking regress v0.1.0 (/<redacted>)
    Finished dev [unoptimized + debuginfo] target(s) in 0.62s

% cargo +beta check
    Checking regress v0.1.0 (/<redacted>)
error: this `#[deprecated]` annotation has no effect
 --> src/main.rs:3:1
  |
3 | #[deprecated(since="0.2.2", note="Use the `AsRef` or `AsMut` implementation instead")]
  | ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ help: remove the unnecessary deprecation attribute
  |
  = note: `#[deny(useless_deprecated)]` on by default

error: aborting due to previous error

error: could not compile `regress`

To learn more, run the command again with --verbose.

Version it worked on

It most recently worked on: 1.48.0

Version with regression

rustc --version --verbose:

rustc 1.49.0-beta.1 (21dea46d8 2020-11-18)
binary: rustc
commit-hash: 21dea46d8347c8b4117c5567949703f0fbb51649
commit-date: 2020-11-18
host: x86_64-unknown-linux-gnu
release: 1.49.0-beta.1

[jonas-schievink]

I believe this is intended behavior added in #78626. It shouldn't cause any issues when using sodiumoxide as a dependency, since all lints are silenced in that case.

cc @m-ou-se

[petrochenkov]

If #[deprecated] is displayed by rustdoc (like #[stable]/#[unstable]), then #78626 is a regression.
Even if stability attributes are not checked on impls now (but they may be checked in the future), they are still useful as documentation.

[jyn514]

This should be a warning, not an error I think.

[jyn514]

@petrochenkov deprecated is not shown by rustdoc on impls:

So the attribute does have no effect.

[camelid]

Should this be closed as intended behavior then?

[kpcyrd]

Since this is silenced for dependencies and we've already fixed this on our master branch this issue is resolved for me (thanks for the help!). I'm not closing it myself in case you still want to discuss this until the 1.49.0 release, but feel free to do so.

[m-ou-se]

@kpcyrd Thanks for the report and the update.

As mentioned above, this is "working as intended" in the sense that this was expected outcome of #78626. That does not mean it's necessarily the best behaviour. Maybe a warning would make more sense here, depending on how much trouble this change causes.

this issue is resolved for me

Okay, closing this for now then.

Please re-open the issue or leave a comment if you (anybody) think something needs to be done here.