Exception for cyclic-safe traits

[WildCryptoFox]

Thy cyclic supertrait lint and super-trait tracing code should have an exception to allow for cyclic-safe traits.

https://play.rust-lang.org/?gist=cc794942ed0663d530d8&version=nightly

A trait definition like this is currently invalid.

trait Mode<M>: Mode<A> { type Return; }

Instead of being invalid, this should cut the recursion at the point that M == RequiredMode.

Any type that implements Mode<B> must also implement Mode<A> and when selecting specific traits with <T as Trait>, the exact trait should be respected.

<T as Mode<B>>::Return // B
<T as Mode<A>>::Return // A
<<T as Mode<B>> as Mode<A>>::Return // A

As demonstrated in #12511, I would also say that trait T: T {} should be the identity of T, and consequently be the identity of the exception to permit cyclic-safe types.

Cyclic safe traits

trait Trait: Trait { ... } // Trait == Trait, stop recurring // lint redundant trait?

trait Mode<M>: Mode<A> { ... } // When M == A, stop recurring

Cyclic unsafe traits

trait A: B { ... }
trait B: A { ... } 

[eefriedman]

In theory, there isn't any reason we can't support trait A: B { ... } trait B: A { ... }. We can reason about the cycle the same way we can for trait A: A.

[WildCryptoFox]

@eefriedman True but that would require understanding the known state, instead of just with the exact trait definition. I think that would require a little more work, but indeed, doable.

[arielb1]

@james-darkfox

It was an intentional design decision not to allow for cyclic traits. If you desire to change it, you should post an issue at rust-lang/rfcs.