Coercion to an unspecified but easily inferrable raw pointer (`as *const _`) has regressed

[chris-morgan]

A couple of days ago it was all working fine, but in a fresh build (rustc 1.0.0-dev (593db005d 2015-03-25) (built 2015-03-25)) trouble is happening:

#![allow(dead_code)]

// Note that it’s got to be *const T; T compiles fine!
fn gimme_a_raw_pointer<T>(_: *const T) { }
fn gimme_a_raw_pointer_to_unit(_: *const ()) { }
fn gimme_a_raw_pointer_to_a_u8(_: *const u8) { }

fn main() {
    // This builds fine:
    //let pointer: *const _ = &();
    // But add an `as *const _` (the type ascription on `let pointer` is optional and meaningless):
    let pointer = &() as *const _;

    // Clearly `pointer` must be of type `*const ()`.

    // The real failure case was a “simple” `println!("{:?}", pointer)`.
    // I have extracted the essence of the failure here:
    gimme_a_raw_pointer(pointer);
    // Uncomment either of these two and it will suddenly figure out that that _ has got to be ().
    //gimme_a_raw_pointer_to_unit(pointer);
    //gimme_a_raw_pointer_to_a_u8(pointer);
}

<anon>:18:5: 18:24 error: unable to infer enough type information about `_`; type annotations required [E0282]
<anon>:18     gimme_a_raw_pointer(pointer);
              ^~~~~~~~~~~~~~~~~~~
error: aborting due to previous error

[Ms2ger]

#23630?

[Marwes]

Smaller test case of what I believe to be the same bug.

fn test<T>(t: T) { }

fn main() {
    let t = test as fn (i32);
    t(0i32);
}

I don't have any knowledge of the inner workings of rustc but it seems that as doesn't unify the expression to the type at all. leading to the expr to still have an unknown type (fn (T) rather than fn (i32)).

[brson]

Both examples seem to work. Just add them to the test suite.

[jneem]

I can do that