Diagnostics: Nested higher-ranked lifetimes are incorrectly printed

[fmease]

For

type One = fn(HelperOne);
type HelperOne = for<'a> fn(&'a (), &'a ());

type Two = for<'o> fn(HelperTwo<'o>);
type HelperTwo<'x> = for<'a> fn(&'a (), &'x ());

fn accept(x: One) -> Two { x }

the output is

error[E0308]: mismatched types
 --> src/lib.rs:7:28
  |
7 | fn accept(x: One) -> Two { x }
  |                      ---   ^ one type is more general than the other
  |                      |
  |                      expected `for<'o> fn(for<'a> fn(&'a (), &'o ()))` because of return type
  |
  = note: expected fn pointer `for<'o> fn(for<'o, 'a> fn(&'a (), &'o ()))`
             found fn pointer `fn(for<'a> fn(&'a (), &'a ()))`

Here the type for<'o> fn(for<'o, 'a> fn(&'a (), &'o ())) shown after “expected fn pointer” contains
for<'o, 'a> which doesn't make sense. The binder contains the lifetime bound in the outer binder.
The inner binder should just be for<'a> just like in the type shown in the label.
Note that this doesn't happen when -Zverbose is passed ¹.

Further, if we change the definition of Two to be type Two = for<'a> fn(HelperTwo<'a>); ('o => 'a),
then the pretty-printer does not try to avoid shadowing through renaming of one of the lifetimes or
through other means. Output:

error[E0308]: mismatched types
 --> src/lib.rs:7:28
  |
7 | fn accept(x: One) -> Two { x }
  |                      ---   ^ one type is more general than the other
  |                      |
  |                      expected `for<'a> fn(for<'a> fn(&'a (), &'a ()))` because of return type
  |
  = note: expected fn pointer `for<'a> fn(for<'a, 'a> fn(&'a (), &'a ()))`
             found fn pointer `fn(for<'a> fn(&'a (), &'a ()))`

Ideally it should look something like

  error[E0308]: mismatched types
   --> src/lib.rs:7:28
    |
  7 | fn accept(x: One) -> Two { x }
    |                      ---   ^ one type is more general than the other
    |                      |
-   |                      expected `for<'a> fn(for<'a> fn(&'a (), &'a ()))` because of return type
+   |                      expected `for<'r> fn(for<'a> fn(&'a (), &'r ()))` because of return type
    |
-   = note: expected fn pointer `for<'a> fn(for<'a, 'a> fn(&'a (), &'a ()))`
+   = note: expected fn pointer `for<'r> fn(for<'a> fn(&'a (), &'r ()))`
               found fn pointer `fn(for<'a> fn(&'a (), &'a ()))`

Here is a playground for a smaller reproducer of the last issue. In this case, the compiler could either
perform renaming or it could omit the inner binder entirely since the lifetime it binds is not referenced
in the inner type.

I've looked through all relevant A-diagnostics A-lifetimes issues and through some others and couldn't find anything that matches 100% what I was looking for (out of those maybe kinda #29061, #56423, #73457, #92281 but not really).

@rustbot label T-compiler A-diagnostics A-lifetimes D-incorrect

Footnotes

1. It's for<Region(BrNamed(DefId(0:7 ~ repro[8603]::Two::'o), 'o))> fn(for<Region(BrNamed(DefId(0:10 ~ repro[8603]::HelperTwo::'a), 'a))> fn(..., ...)). ↩

[compiler-errors]

Yeah, this has to do with some irresponsible binder-skipping in TypeErrCtxt::cmp I'm pretty sure.