`&-ptr` should be less used in compiler errors

[barosl]

When comparing mismatched types, today's rustc prints &-ptr when one of the types is a reference. For example,

 expected `&i32`,
    found `f64`
(expected &-ptr,
    found f64) [E0308]

expected `core::result::Result<_, &str>`,
    found `core::result::Result<i32, collections::string::String>`
(expected &-ptr,
    found struct `collections::string::String`) [E0308]

There might be reasons for this, but it's inconvenient in many cases. If the compiler is giving attention to the &-difference of the two types, it should print &-ptr only when the two compared types could be deduced from one type. (e.g. &i32 and i32) But in general, if those two types are fairly distinct, just printing the full type name would be much more handy.

[arielb1]

We currently just print the non-matching type constructors - I mean, should we also do this between Box<T> and U? Between Option<T> and U?

[Manishearth]

I think we should go one level deeper if we find a sigil-based type constructor (currently, & and * and [])

No, no need to do this with box, because that is a textual type ctor.

[steveklabnik]

I believe this changed recently, and it seems so:

error[E0308]: mismatched types
 --> foo.rs:2:19
  |
2 |     let y: &i32 = 5.0f64;
  |                   ^^^^^^ expected &i32, found f64
  |
  = note: expected type `&i32`
  = note:    found type `f64`

Closing!