remove EvaluationResult::EvaluatedTo{Unknown,Recur}

[aliemjay]

I was confused by them while reviewing #114023.

EvaluatedToRecur was introduced in #42840, seemingly with intent to support inductive cycles in the current solver, but that's not something we plan to have in the old solver. It is broken anyway because it's never converted into EvaluatedToErr in the root predicate, see the FIXME in the docs.

EvaluatedToUnknown is not handled any differently than EvaluatedtoAmbig.

r? @lcnr@compiler-errors

[compiler-errors]

How is it sound to remove this? Aren't we caching things that should not be cached?

[aliemjay]

as @compiler-errors noted, this example breaks under this PR because fn match_fresh_trait_refs is incomplete. See also rust-lang/trait-system-refactor-initiative#56.
Box<u16>: Trait<_> is cached as ambiguous although it clearly doesn't hold.

traitTrait<J>{}// This impl doesn't holdimpl<J>Trait<J>forBox<u32>whereBox<u16>:Trait<J>,Option<u8>:Trait<J>,{}// This impl holdsimplTrait<()>forBox<u8>{}// --- Trait2traitTrait2<J>{}// This impl doesn't holdimpl<J>Trait2<J>forBox<u32>whereBox<u16>:Trait<J>{}// This impl holdsimplTrait2<()>forBox<u8>{}fnimpls_trait1<T:Trait<J>,J>(){}fnimpls_trait2<T:Trait2<J>,J>(){}fnmain(){impls_trait1::<Box<_>,_>;impls_trait2::<Box<_>,_>;}

@rustbot author

[bors]

☔ The latest upstream changes (presumably #114023) made this pull request unmergeable. Please resolve the merge conflicts.

[Dylan-DPC]

@aliemjay any updates on this?

[Dylan-DPC]

Closing this as inactive. Feel free to reöpen this pr or create a new pr if you get the time to work on this. Thanks