Cyclic dependency bug processing RPIT `impl Trait`

[alexreg]

trait Foo: Fn(i32) + Send {}

impl<T: ?Sized + Fn(i32) + Send> Foo for T {}

fn gives_foo() -> impl Foo {
    let f = |x| println!("{}", x);
    f
}

fn main() {
    gives_foo()(42);
}

yields

error[E0391]: cycle detected when processing `gives_foo::{{impl-Trait}}`
 --> test.rs:5:19
  |
5 | fn gives_foo() -> impl Foo {
  |                   ^^^^^^^^
  |
note: ...which requires processing `gives_foo::{{impl-Trait}}`...
 --> test.rs:5:19
  |
5 | fn gives_foo() -> impl Foo {
  |                   ^^^^^^^^
note: ...which requires processing `gives_foo::{{impl-Trait}}`...
 --> test.rs:5:19
  |
5 | fn gives_foo() -> impl Foo {
  |                   ^^^^^^^^
  = note: ...which again requires processing `gives_foo::{{impl-Trait}}`, completing the cycle
note: cycle used when processing `gives_foo`
 --> test.rs:5:1
  |
5 | fn gives_foo() -> impl Foo {
  | ^^^^^^^^^^^^^^^^^^^^^^^^^^

I originally thought this was my PR #55101 that was at fault, but I've now confirmed this bug occurs on master, so it looks like it isn't... not in the latest nightly though, as of now, so it must have been a recent PR.

CC @nikomatsakis @oli-obk

[alexreg]

Okay, weirdly this is only happening when I have RUST_LOG="rustc_typeck::astconv=debug".

[oli-obk]

Yea, the logging output might call type_of on the item to be printed. If that happens while processing type_of, you get a cycle. Not sure how to resolve that

[alexreg]

Yep I think that’s it. Hmm.

[oli-obk]

Hmm. why does printing cause type_of to be called on the TyAnon? Can you investigate and remove the call? it should not be necessary.

[steveklabnik]

Triage: given

Okay, weirdly this is only happening when I have RUST_LOG="rustc_typeck::astconv=debug".

and how much the compiler has changed since nov of 2018 (even though that module still exists), and with no other comments, I am gonna guess that this may not reproduce, and even if it did, wouldn't affect users. So I'm gonna give it a close. @alexreg if this still reproduces for you today, please feel free to re-open.

[alexreg]

@steveklabnik I believe I figured out the cause of this while working on another PR, and solved it. (Completely forgot about this issue!) In any case, seems reasonable to close.