Fix nonsense non-tupled Fn trait error

[compiler-errors]

Given this code:

#![feature(unboxed_closures)]

fn a<F: Fn<usize>>(f: F) {}

fn main() {
    a(|_: usize| {});
}

We currently emit this error:

error[E0631]: type mismatch in closure arguments
 --> src/main.rs:6:5
  |
6 |     a(|_: usize| {});
  |     ^ ---------- found signature of `fn(usize) -> _`
  |     |
  |     expected signature of `fn(usize) -> _`
  |
note: required by a bound in `a`
 --> src/main.rs:3:9
  |
3 | fn a<F: Fn<usize>>(f: F) {}
  |         ^^^^^^^^^ required by this bound in `a`

For more information about this error, try `rustc --explain E0631`.
error: could not compile `playground` due to previous error

Notably, it says the same thing for "expected" and "found"!

Fix the output so that we instead emit:

error[E0308]: mismatched types
 --> /home/gh-compiler-errors/test.rs:6:5
  |
6 |     a(|_: usize| {});
  |     ^ types differ
  |
  = note: expected trait `Fn<usize>`
             found trait `Fn<(usize,)>`
note: required by a bound in `a`
 --> /home/gh-compiler-errors/test.rs:3:9
  |
3 | fn a<F: Fn<usize>>(f: F) {}
  |         ^^^^^^^^^ required by this bound in `a`

error: aborting due to previous error

The error could still use some work, namely the "mismatched types" part, but I'm leaving it a bit rough since the only way you'd ever get this error is when you're messing with #![feature(unboxed_closures)].

Simply making sure we actually print out the difference in trait-refs is good enough for me. I could probably factor in some additional improvements if those are desired.

[rust-highfive]

r? @cjgillot

(rust-highfive has picked a reviewer for you, use r? to override)

[cjgillot]

Sorry for the delay @compiler-errors.
When we have Fn<A> with a concrete type A, we should probably report that a tuple is expected. The diagnostic improvement for generic A is great though.

[cjgillot]

Shouldn't we just report that a tuple is expected and bail out?

[compiler-errors]

Well making this a hard error in #99943, so I don't want to just introduce a special case error logic here.

[compiler-errors]

Also, reporting that a tuple is expected and bailing out is not gonna work when you have a generic Fn<T> where T is a type param.

[cjgillot]

Why do we keep a single ArgKind::empty(), like 1-tuples, instead of an empty vec?

[compiler-errors]

No particular reason, I can fix that I guess.

[compiler-errors]

Oh, I know why. This code also handles Generator<T> which takes a generic argument that isn't a tupled set of arguments.

[compiler-errors]

@rustbot author

[compiler-errors]

I actually don't know if I have anything to change here. See replied comments.

@rustbot ready

[cjgillot]

@bors r+

[bors]

📌 Commit 9e52d22 has been approved by cjgillot

It is now in the queue for this repository.

[matthiaskrgr]

@bors r-

     Checking rustc_session v0.0.0 (/checkout/compiler/rustc_session)
    Checking rustc_attr v0.0.0 (/checkout/compiler/rustc_attr)
    Checking rustc_query_system v0.0.0 (/checkout/compiler/rustc_query_system)
    Checking rustc_parse v0.0.0 (/checkout/compiler/rustc_parse)
    Checking rustc_middle v0.0.0 (/checkout/compiler/rustc_middle)
    Checking rustc_ast_passes v0.0.0 (/checkout/compiler/rustc_ast_passes)
    Checking rustc_expand v0.0.0 (/checkout/compiler/rustc_expand)
    Checking rustc_builtin_macros v0.0.0 (/checkout/compiler/rustc_builtin_macros)
error: passing `TypeError<'_>` by reference
  --> compiler/rustc_middle/src/ty/error.rs:78:29
   |
78 |     pub fn involves_regions(&self) -> bool {
   |                             ^^^^^ help: try passing by value: `TypeError<'_>`
   |
   = note: `-D rustc::pass-by-value` implied by `-D warnings`

error: could not compile `rustc_middle` due to previous error
warning: build failed, waiting for other jobs to finish...
error: could not compile `rustc_middle` due to previous error
Build completed unsuccessfully in 0:02:05

#100593 (comment)

[compiler-errors]

@bors r=cjgillot

[bors]

📌 Commit 1c084f1 has been approved by cjgillot

It is now in the queue for this repository.