"~const` is not allowed here" is a sub-optimal error message in one case

[leonardo-m]

I think this is a small diagnostic problem. This code compiles (with two warnings):

#![feature(const_fn_trait_bound)]
#![feature(const_trait_impl)]

trait HasBar {
    fn bar(&self);
}
impl const HasBar for usize {
    fn bar(&self) {}
}

struct Foo<T>(T)
where T: HasBar;

impl<T> Foo<T> where T: HasBar {
    const fn foo(&self)
    where T: ~const HasBar {
        self.0.bar();
    }
}

fn main() {}

(There in the impl foo I've used both HasBar and ~const HasBar, I'm not sure if this is necessary).

If you miss the 'const' from the impl foo:

#![feature(const_fn_trait_bound)]
#![feature(const_trait_impl)]

trait HasBar {
    fn bar(&self);
}
impl const HasBar for usize {
    fn bar(&self) {}
}

struct Foo<T>(T)
where T: HasBar;

impl<T> Foo<T> where T: HasBar {
    fn foo(&self)
    where T: ~const HasBar {
        self.0.bar();
    }
}

fn main() {}

It gives an error that doesn't suggest that a 'const' is missing in foo:

error: `~const` is not allowed here
  --> ...\test.rs:16:14
   |
16 |     where T: ~const HasBar {
   |              ^^^^^^^^^^^^^
   |
   = note: only allowed on bounds on traits' associated types and functions, const fns, const impls and its associated functions

[oli-obk]

hmm... so you're saying we should suggest to make the current function const?

[leonardo-m]

This is a minor diagnostic enhancement request. If you think this isn't worth doing, then we can close this down.

Currently the diagnostic says:
"only allowed on bounds on traits' associated types and functions, const fns, const impls and its associated functions"

I'd like a focused diagnostic that says something like:
"~const HasBar" isn't allowed as a bound for a non-const function."

I think less noisy error messages are better.

[oli-obk]

Oh, good point. Yea, that seems straight forward enough and def an improvement

[fee1-dead]

fixed in #103319.