Regression in spacing of left margin in diagnostics

[dtolnay]

struct A;
//
//
//
//
//
//
//
fn main() {
    let _: () = A;
    let _: () = B;
}

$ cargo +nightly-2019-10-01 check

This is the correct output.

error[E0425]: cannot find value `B` in this scope
  --> src/main.rs:11:17
   |
11 |     let _: () = B;
   |                 ^ help: a unit struct with a similar name exists: `A`

error[E0308]: mismatched types
  --> src/main.rs:10:17
   |
10 |     let _: () = A;
   |                 ^ expected (), found struct `A`
   |
   = note: expected type `()`
              found type `A`

$ cargo +nightly-2019-10-02 check

Notice that one of the errors has a column of whitespace between the line number and the pipes, the next does not.

error[E0425]: cannot find value `B` in this scope
  --> src/main.rs:11:17
   |
11 |     let _: () = B;
   |                 ^ help: a unit struct with a similar name exists: `A`

error[E0308]: mismatched types
 --> src/main.rs:10:17
  |
10|     let _: () = A;
  |                 ^ expected (), found struct `A`
  |
  = note: expected type `()`
             found type `A`

Commit range: 22bc9e1...702b45e

[dtolnay]

Mentioning @AnthonyMikh @estebank who made changes in librustc_errors. Could this be from #64935?

[dtolnay]

It looks like this can cause the wrong line number to be printed. :(

If you bump the problematic line down to line 100, it prints as 10:

error[E0308]: mismatched types
 --> src/main.rs:100:17
  |
10|     let _: () = A;
  |                 ^ expected (), found struct `A`
  |
  = note: expected type `()`
             found type `A`

[estebank]

It seems like it might be.

[AnthonyMikh]

Mentioning @AnthonyMikh @estebank who made changes in librustc_errors. Could this be from #64935?

Indeed. Upon closer inspection I've found out that 7b4c5c6 actually changes the semantics of method: after applying this commit, if children is not empty, then max of them is not compared with primary, although primary could actually be greater.

The correct code should rather look like this:

    fn get_max_line_num(&mut self, span: &MultiSpan, children: &[SubDiagnostic]) -> usize {
        let primary = self.get_multispan_max_line_num(span);
        children.iter()
            .map(|sub| self.get_multispan_max_line_num(&sub.span))
            .max()
            .unwrap_or(0)
            .max(primary)
    }

Although it might be not the bug responsible for this issue, it definetely should be fixed. I'll make a PR with fix ASAP.

EDIT: Here it is

[estebank]

Subtle. That seems about right.

[AnthonyMikh]

For every problem there is a solution that is simple, neat—and wrong.