type aliases cannot be used for traits note is printed wrongly

[nagisa]

src/librustc/session/mod.rs:63:39: 63:67 error: `mir::transform::Passes` is not a trait [E0404]
src/librustc/session/mod.rs:63     pub mir_passes: RefCell<for<'tcx> mir::transform::Passes<'tcx>>,
                                                                     ^~~~~~~~~~~~~~~~~~~~~~~~~~~~
src/librustc/session/mod.rs:63:39: 63:67 help: run `rustc --explain E0404` to see a detailed explanation
src/librustc/session/mod.rs:63:39: 63:67 note: `type` aliases cannot be used for traits
src/librustc/session/mod.rs:63     pub mir_passes: RefCell<for<'tcx> mir::transform::Passes<'tcx>>,
                                                                     ^~~~~~~~~~~~~~~~~~~~~~~~~~~~

The error here is pretty clear, but the note is completely wrong as the relevant snippet appears in the type of the struct field and not type item.

[nagisa]

A bit relatedly, the E0404 is explaining something about implementing non-traits. The error explanation should either be expanded to include the current case or a different error code should be used.

cc @GuillaumeGomez on the thing described in this comment, since you originally authored the explanation for the code.

EDIT: err, I’ll just fill another issue (#31687).

[Manishearth]

The note should use fileline_note instead, and we should have an extra note in case the type alias has a span (which will not be the case if it's from a foreign mod)

[Manishearth]

Mentoring this.

Relevant code
Change the span_note to fileline_note.

You will also need to use resolver.ast_map.span_if_local(def_id) on the defid contained in the TyAlias to find the span for the alias (to use for outputting the extra note)

Let me know if you have any questions!

[nagisa]

@Manishearth while I do not disagree with general suggestion to use fileline_note but I think that won’t resolve the issue, because mir::transform::Passes<'tcx> is not a trait in a first place and the failing code snippet not appear in a typedef.

---

#31687 (comment) has a code sample which reproduces the issue, by the way.

[Manishearth]

As I see it there are two distinct issues. One is that there should be better diagnostics pointing to the typedef. That's what this bug is about. The other is that sometimes it errors for struct types instead (I find this unlikely?). Could you provide an example which reproduces this? The one you link to doesn't. http://is.gd/VKbcPH

[nagisa]

Oh, it doesn’t fail on nightly. I guess, then, original issue I reported is not really an issue anymore.

[vegai]

Ongoing work at #32005