Unhelpful error message and cycles with const_evaluatable_checked

[WorldSEnder]

Given the following code: playground link

#![feature(const_generics, const_evaluatable_checked)]

trait FiniteGroup : Sized {
    const group_size: usize;
    const elems: [Self; Self::group_size];
}

impl FiniteGroup for () {
    const group_size: usize = 1;
    const elems: [Self; Self::group_size] = [()];
}

Currently the compiler complains

error: unconstrained generic constant
 --> src/lib.rs:5:5
  |
5 |     const elems: [Self; Self::group_size];
  |     ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  |
  = help: try adding a `where` bound using this expression: `where [(); Self::group_size]:`

This is problematic:

• it is not clear where to put the suggested where bound. The parser errors when put behind elems: or before the final ;. Maybe I'm missing something obvious?
• the suggested expression is incomplete, and most likely missing a : Sized

Possible related to #80976, #82934 and #82509 but additionally, the bound is required on the trait's own associated constant, not in an impl, a struct or construction related to the trait.

Note: I believe correct code would look like

trait FiniteGroup : Sized where [Self; Self::group_size]: Sized {
    const group_size: usize;
    const elems: [Self; Self::group_size];
}

but this leads to

cycle detected when building an abstract representation for FiniteGroup::{constant#0}

note: ...which requires processing MIR for `FiniteGroup::{constant#0}`...
note: ...which requires unsafety-checking `FiniteGroup::{constant#0}`...
note: ...which requires building MIR for `FiniteGroup::{constant#0}`...
note: ...which requires type-checking `FiniteGroup::{constant#0}`...
note: ...which again requires building an abstract representation for FiniteGroup::{constant#0}, completing the cycle

[BoxyUwU]

the suggested expression is incomplete, and most likely missing a : Sized

it's actually not where T: is a valid where clause 😅 (not to say this isnt confusing)

[WorldSEnder]

Maybe I did after all incorrectly understand what the compiler was complaining about, and I'm more confused! I thought it was complaining that [Self; Self::group_size] can not be used like that as the type for a const item, since that would have to be Sized.

[lcnr]

the cycle error with the where bound is tracked in #79356 (comment)

and the unclear + confusing error messages are well known. While I don't know whether we have an explicit tracking issue for them, we do keep them in mind and will deal with them as part of rust-lang/project-const-generics#41.

decided during triage (https://rust-lang.zulipchat.com/#narrow/stream/260443-project-const-generics/topic/.60F-generic_const_exprs.60.20triage/near/287342437) to close this issue as its content is already tracked elsewhere. Still thank you for opening this issue 👍