Importing `std::error::Error` causes lifetime error to appear in previously compiling code

[ginnyTheCat]

I tried this code:

use std::iter::successors;

use std::error::Error;

pub mod file {
    use std::{error::Error, fs::File};

    pub fn open_non_existent() -> Box<dyn Error> {
        let err = File::open("doesn't exist.jpg").unwrap_err();
        Box::new(err)
    }
}

fn main() {
    let err = file::open_non_existent();

    for part in successors(Some(&*err), |e| e.source()) {
        println!("{}", part);
    }
}

I expected to see this happen: It to compile, since it does with line 3 (use std::error::Error;) commented out

Instead, this happened: A lifetime error appears as long as e is not replaced with (*e)

error: lifetime may not live long enough
  --> src/main.rs:17:42
   |
17 |     for e in successors(Some(&*err), |e| e.source()) {
   |                                       -- ^^^^^^^^^^ returning this value requires that `'1` must outlive `'2`
   |                                       ||
   |                                       |return type of closure is Option<&'2 dyn std::error::Error>
   |                                       has type `&'1 &dyn std::error::Error`

error: could not compile `tmp` (bin "tmp") due to 1 previous error

Meta

This happens since version 1.51, where impl<'a, T: Error + ?Sized> Error for &'a T { was added, until nightly.

I'm unsure whether this is maybe even expected behaviour, since I also don't quite understand how source() could even be called without the import (maybe this has something to do with dyn?), since Error is not in the prelude as far as i can see.

rustc --version --verbose:

rustc 1.94.0 (4a4ef493e 2026-03-02)
binary: rustc
commit-hash: 4a4ef493e3a1488c6e321570238084b38948f6db
commit-date: 2026-03-02
host: x86_64-unknown-linux-gnu
release: 1.94.0
LLVM version: 21.1.8

[QuineDot]

The behavior is due to dyn: you can call its trait methods as if they were inherent methods.

When you import Error, method resolution finds the impl for references first instead.

The same code breakage was found, considered, and accepted in the PR. (Though, being a blanket implementation, it was a breaking change.)

[ginnyTheCat]

I kind of expected something like this. I thought about ways to improve/make the error message a bit more clear, but they all seem very difficult/intrusive to implement.

Thanks for answering to this, even though it was basicly the same code as in the PR.

[theemathas]

Seems similar to #101094

[xizheyin]

When we use std::error::Error;, method resolution includes trait method candidates as well, and it will match the impl for &'a T(&dyn Error). But without it, it may consider it as dyn Error(auto deref). This does not seem to be a bug from the compiler's perspective, but it certainly is one in terms of user experience.
We can use (*e).source to explicitly deref it.
https://play.rust-lang.org/?version=stable&mode=debug&edition=2024&gist=108d937824cec45cb2b13ed609216522