E0094 needs better underline

[sophiajt]

Currently, E0094 looks like:

error[E0094]: intrinsic has wrong number of type parameters: found 2, expected 1
  --> src/test/compile-fail/E0094.rs:13:5
   |
13 |     fn size_of<T, U>() -> usize; //~ ERROR E0094
   |     ^^^^^^^^^^^^^^^^^^^^^^^^^^^^ expected 1 type parameter

A better span would underline the type parameters themselves.

error[E0094]: intrinsic has wrong number of type parameters: found 2, expected 1
  --> src/test/compile-fail/E0094.rs:13:5
   |
13 |     fn size_of<T, U>() -> usize; //~ ERROR E0094
   |                ^^^^ expected 1 type parameter

[KiChjang]

So we're looking at an Item of the ItemFn variant. More specifically, it's the Generics struct that resides in an ItemFn. This should be pretty straight-forward since Generics contains a span method which we can call to get us the span.

[kyrias]

Tried that, but getting

error[E0094]: intrinsic has wrong number of type parameters: found 2, expected 1
  --> src/test/compile-fail/E0094.rs:13:5
   |
13 |     fn size_of<T, U>() -> usize; //~ ERROR E0094
   |     ^^^^^^^^^^------^^^^^^^^^^^^
   |               |
   |               expected 1 type parameter

instead, which is closer, but still not quite. Going to look more at it tomorrow though.

[sophiajt]

@kyrias - almost there, just need to also update the span at the start of the error message as well. When both are updated, it will will turn the new underline into the primary span (the ^^^ underline)

[kyrias]

Ah, didn't think about that span! (And I just realized it's almost 01:00.)

To get there I have to match on rustc::hir::ForeignItem_ which can be either ForeignItemFn or ForeignItemStatic though, and if I add the original error to the second match branch I of course get a warning that E0094 has already been defined, so I'm not sure how to solve that correctly. Can it.node be ForeignItemStatic at that point? Haven't had time to look at it right now, have to sleep so I don't end up too late tomorrow, but should have time during the day tomorrow.

Edit: I could of course set a variable with either the span from the generics or the it.span, but I don't think that would be appropriate, since they end up printing different error messages for the same error code.

[KiChjang]

You don't need to overthink this:

let span = match it.node {
    ForeignItem_::ForeignItemFn(_, ref generics) => generics.span().unwrap_or(it.span),
    _ => it.span
};

And then use span as the first argument to span_label.

[KiChjang]

Essentially, it.span here is being treated as a fallback, and I think that's fine. They do not "print different error messages" as you put it, only that the spans would be different, and it's no big deal.

[kyrias]

Yeah, I'm not sure what I was thinking. That serves me right for doing things while going to bed is 3 hours overdue. Currently compiling, will submit a PR in a while when all tests are done.