&[T, ..N] casts to *T instead of *[T, ..N] in structs

[mahkoh]

Consider the following code:

struct X {
    y: [u8, ..2],
}

fn main() {
    let x1 = X { y: [0, 0] };

    let p1: *u8 = &x1.y as *_;
    let t1: *[u8, ..2] = &x1.y as *_;
    let h1: *[u8, ..2] = &x1.y as *[u8, ..2];
}

The second cast is an error. The other casts work.

Consider the following code:

fn main() {
    let mut x1 = X { y: [0, 0] };

    let p1: *mut u8 = &mut x1.y as *mut _;
    let t1: *mut [u8, ..2] = &mut x1.y as *mut _;
    let h1: *mut [u8, ..2] = &mut x1.y as *mut [u8, ..2];
}

The first cast is an error. The other casts work.

[huonw]

There's a comment seeming to imply that &[T, ..n] as *T is explicitly ok, so it's weird that it doesn't work with *mut.

Nominating.

[jdm]

I'm pretty sure this was caused by #14856.

[brson]

1.0, P-backcompat-lang

[nikomatsakis]

This is related to how we do coercions. Right now we must do this "eagerly", I'd prefer to leverage a lazy approach like trait matching does. I'm going to try and write this up somewhere, perhaps an RFC.