bad error message for missing static bounds on `&dyn Trait`

[sadan4]

Code

use std::error::Error;
fn is_caused_by<T: Error + 'static>(err: &dyn Error) -> bool {
    err.is::<T>()
        || err
            .source()
            .is_some_and(|cause| is_caused_by::<T>(cause))
}

Current output

error[E0521]: borrowed data escapes outside of function
 --> src/lib.rs:3:5
  |
2 | fn is_caused_by<T: StdError + 'static>(err: &dyn StdError) -> bool {
  |                                        ---  - let's call the lifetime of this reference `'1`
  |                                        |
  |                                        `err` is a reference that is only valid in the function body
3 |     err.is::<T>()
  |     ^^^^^^^^^^^^^
  |     |
  |     `err` escapes the function body here
  |     argument requires that `'1` must outlive `'static`

For more information about this error, try `rustc --explain E0521`.
error: could not compile `playground` (lib) due to 1 previous error

Desired output

the error message should mention the missing 'static bound on dyn Error or mention that the is method is implemented for dyn Error + 'static

Rationale and extra context

No response

Other cases

Rust Version

rustc 1.100.0-nightly (923c95cdf 2026-09-16)
binary: rustc
commit-hash: 923c95cdf5ba65cea505aa2ea829f578e1506ed8
commit-date: 2026-09-16
host: x86_64-unknown-linux-gnu
release: 1.100.0-nightly
LLVM version: 23.1.1

Anything else?

No response

[maxdexh]

Simpler:

use std::error::Error;
fn err_is<T: Error + 'static>(err: &dyn Error) -> bool {
    err.is::<T>()
}

So to me, the main issue here is that it is not even pointing out the correct lifetime. It is pointing at the lifetime of the reference, instead of the trait object lifetime. And it gives the typical "borrowed data escapes" message, which most of the time means "hey this reference doesn't live long enough" (though we do also use that message for Box<dyn Error + '_>). I would read the error message as "you need &'static dyn Error", which is misleading.

The lifetimes are only distinguished properly if the trait object lifetime is written explicitly, such as fn<'a, T: Error + 'static>(&(dyn Error + 'a)) -> bool. That will instead make the error message say that 'a must outlive 'static.

Anything else leads to it pointing to the lifetime of the reference.
A help message would be nice as well, but IMO the general case being misleading is the bigger issue here.

@rustbot label D-incorrect A-dyn-trait A-lifetimes T-types D-lack-of-suggestion

[fmease]

The "general case being misleading" is tracked in several issues already: https://github.com/rust-lang/rust/issues?q=is%3Aissue%20label%3AA-diagnostics%20label%3AA-lifetimes%20label%3AA-dyn-trait%20is%3Aopen. I don't have the time rn to go through the list to find the most relevant ones.

In any case, the basic issue is the fact that these kinds of diagnostics all fail to mention that the lifetime requirement arises from a default trait object lifetime (ImplicitObjectLifetimeDefault). PR #121274 tried to work towards fixing this but IIRC it was overly specific / narrow / targeted (by overindexing on user impls & 'static).

[estebank]

CC: #163115, it doesn't fix this, but adds some additional context.