Error message usage of `_` is unclear

[mdinger]

Error messages use _ a lot but it's meaning is never explained. This could be improved.

Playpen

fn test(o: Option<i32>) {}

fn main() {
    let i = 7;
    test(i);
}

Error:

// What is `_` supposed to mean? Apparently integral value but it still seems meaningless...
... error ... expected `core::option::Option<i32>`, found `_`  ... found integral variable)
<anon>:5     test(i);
                  ^
error: aborting due to previous error

---

6/8/2017: updated error message:

error[E0308]: mismatched types
 --> <anon>:5:10
  |
5 |     test(i);
  |          ^ expected enum `std::option::Option`, found integral variable
  |
  = note: expected type `std::option::Option<i32>`
             found type `{integer}`

error: aborting due to previous error

[mdinger]

The book is getting a note which seems related: #22293

[steveklabnik]

Triage: no change.

[Mark-Simulacrum]

Nominating for discussion. What do we expect _ to mean? Ideally we'd get a decision on that and then see if we follow that throughout the code; but at least a consensus on the first would be good.

I believe the current state is that _ is an inference variable that we couldn't elaborate to anything more specific.

[nikomatsakis]

In the particular case of this example, we now print {integer}:

rustc 1.17.0 (56124baa9 2017-04-24)
error[E0308]: mismatched types
 --> <anon>:5:10
  |
5 |     test(i);
  |          ^ expected enum `std::option::Option`, found integral variable
  |
  = note: expected type `std::option::Option<i32>`
             found type `{integer}`

error: aborting due to previous error

[nikomatsakis]

Currently, _ is used to mean (roughly) "some type we've not inferred yet". Perhaps we could use this {} convention for that? e.g., {TBD}?

[nikomatsakis]

I will say that we've been back-and-forth on this point a lot. the primary advantage of _ here is that it can be copied-and-pasted into user's code and has roughly the same meaning ("infer this"). However, we have many types where that cannot be done (including {integer}), so I'm not sure how important of a principle that really is.

[mdinger]

FWIW, I think the it is definitely better than it used to be. I'm not sure what {} is though.

[nikomatsakis]

@mdinger it's not anything, that's the point. :) It's just a way for us to insert "meta" stuff.

[nikomatsakis]

I'm inclined to close this issue, absent a particular suggestion. The specific case in question has been improved and we don't know how to do much better. In any case, we have existing issues with more discussion, hence closing as a duplicate of #25633 and #40985.