Fn traits with array args no longer work with const_generics enabled

[Manishearth]

#![feature(const_generics)]

fn main() {
    takes_closure_of_array_3(|[x, y, z]| {
        println!("args: x={}, y={}, z={}", x, y, z);
    });
}

fn takes_closure_of_array_3<F>(f: F)
where
    F: Fn([i32; 3]),
{
    f([1, 2, 3]);
}

(playpen)

Gives the error:

   Compiling playground v0.0.1 (/playground)
warning: the feature `const_generics` is incomplete and may cause the compiler to crash
 --> src/main.rs:1:12
  |
1 | #![feature(const_generics)]
  |            ^^^^^^^^^^^^^^
  |
  = note: `#[warn(incomplete_features)]` on by default

error[E0391]: cycle detected when const-evaluating + checking `takes_closure_of_array_3::{{constant}}#0`
  --> src/main.rs:11:17
   |
11 |     F: Fn([i32; 3]),
   |                 ^
   |
note: ...which requires const-evaluating + checking `takes_closure_of_array_3::{{constant}}#0`...
  --> src/main.rs:11:17
   |
11 |     F: Fn([i32; 3]),
   |                 ^
......................
9  | / fn takes_closure_of_array_3<F>(f: F)
10 | | where
11 | |     F: Fn([i32; 3]),
12 | | {
13 | |     f([1, 2, 3]);
14 | | }
   | |_^

This compiles fine with const_generics not enabled, and is a regression

cc @Centril @oli-obk

[Manishearth]

(discovered by @jcoglan)

[Centril]

cc @nikomatsakis @eddyb @varkor

Simplified:

#![feature(const_generics)]

fn main() {
    takes_closure_of_array(|_| {});
}

fn takes_closure_of_array<F>(f: F)
where
    F: FnOnce([i32; 0]),
{
    f([]);
}

[jonas-schievink]

Isn't this expected fallout?

[varkor]

This is an expected consequence of #66883.

[Manishearth]

Seems like #60471 is what's supposed to eventually fix this?

[pnkfelix]

triage: P-medium (nightly-only regression on feature that is known to have bugs)

[eddyb]

@varkor @oli-obk I think we've discussed this before, but what do you think of special-casing literals, to unblock at least some common usecases like this?

[oli-obk]

I can't reproduce on master, this may have gotten fixed? (Note that the playground is severely outdated, a fix is underway for that though)

[varkor]

@eddyb pointed out that #68118 most likely fixed this.