Trait object coercion regression

[sfackler]

This code has stopped compiling some time in the last week:

trait Foo {}

impl Foo for i32 {}

fn test(_: &[&Foo]) {}

fn main() {
    test([&1i32]);
}

test.rs:8:10: 8:17 error: mismatched types: expected `&[&Foo]`, found `[&i32, .. 1]` (expected &-ptr, found vector)
test.rs:8     test([&1i32]);
                   ^~~~~~~
error: aborting due to previous error

[sfackler]

cc @luqmana

[luqmana]

I'm guessing this is fallout from DST? @nick29581

[huonw]

Works with test(&[&1i32]); (i.e. a pointer to a [&i32, .. 1] coercing to a &[&i32]), which is the long term solution, I believe.

[nrc]

Apparently we used to coerce [T, ..n] to &[T]. Since Rust is meant to never coerce a value type to a pointer type, I regard that as a bug that has been fixed. &[T, ..n] will still coerce to &[T]. And we even seem to do the nested coercion to trait objects (I'm not sure if we should....).