Cannot refer to associated type of generics of the same trait (illegal recursive type)

[kennytm]

This code:

#![feature(associated_types)]

trait Factory {
    type Product;
    fn create(&self) -> <Self as Factory>::Product;
}

impl Factory for f64 {
    type Product = f64;
    fn create(&self) -> f64 { *self * *self }
}

impl<A: Factory, B: Factory> Factory for (A, B) {
    type Product = (<A as Factory>::Product, <B as Factory>::Product); // line 14
    fn create(&self) -> (<A as Factory>::Product, <B as Factory>::Product) {
        let (ref a, ref b) = *self;
        (a.create(), b.create())
    }
}

fn main() {
    println!("{}", (4.0f64, 5.0f64).create());
    // ^ should print (16, 25)
}

Failed to compile with error:

1.rs:14:20: 14:70 error: illegal recursive type; insert an enum or struct in the cycle, if this is desired
1.rs:14     type Product = (<A as Factory>::Product, <B as Factory>::Product);
                           ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

I think this should be allowed.

[jroesch]

@nikomatsakis This seems like a bug to me as well. I just encountered it trying to get this code to compile: https://github.com/jroesch/molten/blob/master/src/hlist.rs#L76.

The way I see it the expansion should be as follows.

Given a HList of type: HCons<int, HCons<int, HNil>> a call to take(S { pred: Z }) should result in solving an obligation for an impl of Take<S<Z>> for HCons<int, HCons<int, HNil>> which incurs a sub obligation to produce an instance of Take<Z> for HCons<int, HNil>.

When checking this we should know Take<_>::R always resolve to an HList (in this case to HNil) and then HCons<int, <T as Take<Z>>::R> should be equivalent to HCons<int, H0: HList> for the representability check. So forall T, H: HList. HCons<T, H> can be represented as HCons { head: T, tail: Box<H> } which looks finite to me.