What to do with impl Interface1 as Interface2?

[danakj]

Summary of issue:

Given this code:

interface Eats {}
interface Goat {}

impl Goat as Eats {}

fn F(e: Eats) {}
fn G() { F(Goat); }

The impl Goat as Eats looks a lot like a blanket impl of Eats for Goat types, but it's not. It's an impl on the Goat interface itself.

@zygoloid expressed that maybe this should be a warning. If it's a warning, for many projects it will become an error, essentially banning it from the language. Is this useful to keep? Should it be diagnosed in some way?

Details:

No response

Any other information that you want to share?

No response

[zygoloid]

A potential use case for this kind of impl would be as part of a metaprogramming system to synthesize impls for an interface. Eg:

interface Compare { ... }
impl Compare as Derivable {
  fn Derive(template T:! type) -> Meta.Impl {
    // Generate impl for `Compare` for `T`.
  }
}

class C {
  // Or whatever attribute syntax we use.
  @Derive impl as Compare;
}