Type errors in macros have poor spans

[Mark-Simulacrum]

For example, in this code, it's hard to tell which part of the inputs to the macro has the wrong type. This is obviously a reduced example, but I've seen the same problem in real code.

I suspect this might be a duplicate issue, but couldn't find anything...

fn main() {
    assert_eq!(10u64, 10usize);
}

provides:

rustc 1.16.0-nightly (4ecc85beb 2016-12-28)
error[E0308]: mismatched types
 --> <anon>:2:5
  |
2 |     assert_eq!(10u64, 10usize);
  |     ^^^^^^^^^^^^^^^^^^^^^^^^^^^ expected u64, found usize
  |
  = note: this error originates in a macro outside of the current crate

error: aborting due to previous error

[KalitaAlexey]

Is anyone working on it?

[estebank]

On nightly using -Z external-macro-backtrace this looks like this:

error[E0308]: mismatched types
 --> <assert_eq macros>:5:22
  |
5 | if ! ( * left_val == * right_val ) {
  |                      ^^^^^^^^^^^ expected u64, found usize
file2.rs:2:5: 2:32 note: in this expansion of assert_eq! (defined in <assert_eq macros>)

error: aborting due to previous error

and once #46605 lands it'll look like this:

error[E0308]: mismatched types
 --> <assert_eq macros>:5:22
  |
5 | if ! ( * left_val == * right_val ) {
  |                      ^^^^^^^^^^^ expected u64, found usize
 ::: file.rs
  |
2 |     assert_eq!(10u64, 10usize);
  |     --------------------------- in this expansion of `assert_eq!`

error: aborting due to previous error

[estebank]

We should probably leave this ticket open for us to stabilize external-macro-backtrace.

This reminds me, -Zteach should probably imply a simplified version of -Zexternal-macro-backtrace.

[estebank]

I think the "only" work left to do here is to add a way to annotate macro inner spans as coming from a macro argument. This annotation has to be very viral in order to work in this case. Looking at the current output for external-macro-backtrace:

error[E0308]: mismatched types
  --> <assert_eq macros>:5:22
   |
1  | / ( $ left : expr , $ right : expr ) => (
2  | | {
3  | | match ( & $ left , & $ right ) {
4  | | ( left_val , right_val ) => {
5  | | if ! ( * left_val == * right_val ) {
   | |                      ^^^^^^^^^^^ expected u64, found usize
...  |
20 | |  right: `{:?}`: {}"# ,
21 | | left_val , right_val , format_args ! ( $ ( $ arg ) + ) ) } } } } ) ;
   | |____________________________________________________________________- in this expansion of `assert_eq!`
   |
  ::: test.rs:2:5
   |
2  |       assert_eq!(10u64, 10usize);
   |       --------------------------- in this macro invocation

the following spans should have the following internal_macro_backtrace:

1  | ( $ left : expr , $ right : expr ) => (
   |     ^^^^ 1          ^^^^^ 2
2  | {
3  | match ( & $ left , & $ right ) {
   |         --------  ---------- 4 through 2
   |         |
   |         3 through 1
4  | ( left_val , right_val ) => {
   |   --------  ---------- 6 through 4
   |   |
   |   5 through 3
5  | if ! ( * left_val == * right_val ) {
   |        ^^^^^^^^^^    ^^^^^^^^^^^ 8 though 6
   |        |
   |        7 through 5

Once we have a way to travel back to an individual argument, we could see that the error comes from 8, which is tied all the way to 2, which corresponds to the second match argument in the macro call and emit the following without the backtrace:

error[E0308]: mismatched types
 --> <anon>:2:5
  |
2 |     assert_eq!(10u64, 10usize);
  |                       ^^^^^^^ expected u64, found usize
  |
  = note: this error originates in a macro outside of the current crate

or

error[E0308]: mismatched types
 --> <anon>:2:5
  |
2 |     assert_eq!(10u64, 10usize);
  |     ------------------^^^^^^^- in this macro call
  |                       |
  |                       expected u64, found usize
  |
  = note: this error originates in a macro outside of the current crate