regression: AsRef inference breakage

[Mark-Simulacrum]

I'm not quickly finding the PR which introduced the new impls, but it looks like some AsRef impls were added in 1.51, and those are causing breakage as determined through beta crater. Usually we don't do anything here, but would be good to at least ping authors.

• https://crater-reports.s3.amazonaws.com/beta-1.51-1/beta-2021-02-14/gh/phideg.neptungen/log.txt
• https://crater-reports.s3.amazonaws.com/beta-1.51-1/beta-2021-02-14/gh/ttarikbnr.mars_discovery/log.txt
• https://crater-reports.s3.amazonaws.com/beta-1.51-1/beta-2021-02-14/reg/c_string-0.7.1-obsolete/log.txt
• https://crater-reports.s3.amazonaws.com/beta-1.51-1/beta-2021-02-14/reg/localghost-0.2.0/log.txt
• https://crater-reports.s3.amazonaws.com/beta-1.51-1/beta-2021-02-14/reg/sv-0.2.1/log.txt

[apiraino]

I've tried bisecting this issue (if it helps):

searched nightlies: from nightly-2021-01-01 to nightly-2021-03-03
regressed nightly: nightly-2021-01-26
searched commits: from 1d0d76f to f4eb5d9
regressed commit: d3163e9

bisected with cargo-bisect-rustc v0.6.0

Host triple: x86_64-unknown-linux-gnu
Reproduce with:

cargo bisect-rustc --start=2021-01-01 

somewhere in this rollup #81355

[m-ou-se]

somewhere in this rollup #81355

Hm, none of the PRs in that rollup should affect AsRef for String, which is what all these regressions are about. There's only one trait implementation added in that rollup: impl Error for &(impl Error).

[JohnTitor]

This shows the regression (playground):

use std::io;

pub fn foo() -> io::Result<i32> {
    Err(io::Error::new(
        io::ErrorKind::Other,
        format!("Error!").as_ref(),
    ))
}

fn main() {}

And yes, confirmed that #75180 is the cause.

[LeSeulArtichaut]

It looks like those were expected regressions (see #75180 (comment)).

[JohnTitor]

Triage: Close this issue as an acceptable regression, we mentioned it in #75180 (comment) and consider it acceptable through FCP.