Backward spans are displayed strangely

[dtolnay]

macro_rules! m {
    ($a:tt $b:tt) => {
        $b $a;
    }
}

fn main() {
    m!(S struct);
}

This is a regression in 1.18.0-beta but unclear whether it affects anything in practice. We encountered this in #rust-libs discussing #40939.

$ rustc +stable --version
rustc 1.17.0 (56124baa9 2017-04-24)
$ cargo +stable build
   Compiling testing v0.1.0
warning: struct is never used: `S`
 --> src/main.rs:8:10
  |
8 |     m!(S struct);
  |     -----^------- in this macro invocation
  |
  = note: #[warn(dead_code)] on by default

    Finished dev [unoptimized + debuginfo] target(s) in 0.34 secs

$ rustc +beta --version
rustc 1.18.0-beta.2 (f4e8e81c3 2017-05-11)
$ cargo +beta build
   Compiling testing v0.1.0
warning: struct is never used: `S`
 --> src/main.rs:8:10
  |
3 | |         $b $a;
  | |______________^
...
8 |       m!(S struct);
  |  _____-----^-------
  | |     |
  | |     in this macro invocation
  |
  = note: #[warn(dead_code)] on by default

    Finished dev [unoptimized + debuginfo] target(s) in 0.45 secs

[dtolnay]

cc @estebank who touched this code recently.

[estebank]

Oh, beautiful... I'm seeing many problems with this output, beyond the obvious. I feel the proper output should either be:

warning: struct is never used: `S`
 --> src/main.rs:8:10
  |
3 |         $b $a;
  |         ^^^^^^ unused struct
...
8 |     m!(S struct);
  |     ------------- in this macro invocation
  |
  = note: #[warn(dead_code)] on by default

or

warning: struct is never used: `S`
 --> src/main.rs:8:10
  |
8 |     m!(S struct);
  |     ---^^^^^^^^-- in this macro invocation
  |        |
  |        unused struct
  = note: #[warn(dead_code)] on by default

I can make (or mentor) a small patch to either swap end and start if they are reversed or only point at the last end for that case, but I'd prefer to properly fix the spans generated in macro invocations in a better way regardless.

[dtolnay]

This made it into 1.18.0.

[brson]

Help wanted. @estebank can mentor.

[estebank]

@brson should this be backported to 1.19? I don't feel there's a need to rush it.

[Mark-Simulacrum]

I'm going to suggest that we don't backport this; it's not high-impact enough.

[estebank]

That's my thinking as well.