[ER] A simple case of missed type inference

[leonardo-m]

A type inference problem:

#![feature(generic_arg_infer)]
fn main() {
    let foo = |x| x + x; // OK
    let _a = foo(10_u8);

    let bar = |[a,b,c,d,e,f,g,h,i]: [_; _]| [c,f,i,b,e,h,a,d,g]; // OK
    let _b = bar([0_u8; 9]);

    let spam = |[a,b,c,d,e,f,g,h,i]| [c,f,i,b,e,h,a,d,g]; // error[E0282]: type annotations needed
    let _c = spam([0_u8; 9]);
}

In the second case the added : [_; _] adds very little information, at best it just states it's an array. I think Rustc could perform this little step and accept the "spam" case too.

[scottmcm]

Possible dup of #76342 where it was pointed out that references are also possible: the spam closure could also validly type as &[T; 9] -> [&T; 9]. It's always unclear to me when closures can care about things after themselves, to pick up call here that would disambiguate the rest of the way.

[leonardo-m]

related:

let test = |[[a,b,c], [d,e,f], [g,h,i]]: [[_; _]; _]| [[g,d,a], [h,e,b], [i,f,c]];

[jyn514]

I'm going to close this as a duplicate of #76342; I think they have the same cause, that [a,b,c,d,e,f,g,h,i] could be inferred as either an array or a slice.