Unused type params for trait impls cause weird errors about Sized

[ben0x539]

trait Tr {}
impl<T> Tr for () {}
fn main() {}

foo.rs:2:1: 2:21 error: the trait `core::kinds::Sized` is not implemented for the type `<generic #0>`
foo.rs:2 impl<T> Tr for () {}
         ^~~~~~~~~~~~~~~~~~~~
foo.rs:2:1: 2:21 note: the trait `core::kinds::Sized` must be implemented because it is required by `Tr`
foo.rs:2 impl<T> Tr for () {}
         ^~~~~~~~~~~~~~~~~~~~

[japaric]

Reproducible on:

rustc 0.12.0-pre (5d653c17a 2014-09-26 11:44:01 +0000)

cc @nick29581

[nrc]

This has a slightly different error message now (int or () give the same results):

foo.rs:3:1: 3:22 error: unable to infer enough type information to locate the impl of the trait `core::kinds::Sized` for the type `<generic #0>`; type annotations required
foo.rs:3 impl<T> Tr for int {}
         ^~~~~~~~~~~~~~~~~~~~~
foo.rs:3:1: 3:22 note: the trait `core::kinds::Sized` must be implemented because it is required by `Tr`
foo.rs:3 impl<T> Tr for int {}
         ^~~~~~~~~~~~~~~~~~~~~

I'm not actually sure if this is wrong. We do require the Sized bound on T and there is not enough type information to infer it. Just because T is not used, doesn't mean its bounds don't have to be satisfied by the concrete type (int, here). It would be nicer to say T rather than <generic #0> in the error message, but otherwise I don't think this is a bug.

Propose closing this issue. cc @nikomatsakis (since this is a trait reform thing)

[nikomatsakis]

Sorry I've neglected this. I'm trying to power through some the multidispatch changes etc and haven't forgotten about this issue. I'll try to make time to dig a bit deeper soon.

[nikomatsakis]

This is presently the expected behavior. I do agree it is suboptimal. Perhaps we should simply improve the error message to place more focus on the insufficient information, annotations required part and less on the specific trait in question.