Rfcs: Local type aliases to concrete types

[ebfull]

struct Foo<T>(T);

fn test<T, U>() {
    type A = Foo<T>;
    type B = Foo<U>;

    let a = A::method();
    let b = B::method();

    // and so on...
}

This is not possible because of an error: can't use type parameters from outer function; try using a local type parameter instead.

There are probably good reasons for this error but I really think there should be a way to accomplish this kind of aliasing. Otherwise it makes some code noisy and hard to read:

let a = Foo::<T>::method();
let b = Foo::<U>::method();

...especially when you use Foo<T> and Foo<U> _a lot_.

[gsingh93]

There's a similar problem for impls. For example, I'd like this code to compile:

use std::marker::PhantomData;

struct Foo<Bar> {
    foo: PhantomData<Bar>,    
}

impl<Bar> Foo<Bar> {
    type Qux = Vec<Bar>;

    fn baz() -> Qux {
        vec![]
    }
}

But I get associated types are not allowed in inherent impls. I'm thinking we'd have to make a new keyword for type aliases to allow this.

[burdges]

Associated constants would be nice in both fn bodies and inherent impls too. I'd think structs could benefit from all three as well.

[withoutboats]

I'm thinking we'd have to make a new keyword for type aliases to allow this.

No, we just have to make associated types allowed in inherent impls. This is actually a part of the RFC which introduced the concept IIRC.

[eddyb]

On the implementation side this has been hard because looking up associated types happens very early on, before a coherent list of impls is even built. We should be able to address this Soon (TM).

[Centril]

This is essentially ScopedTypeVariables afaik.

@eddyb are we there yet? (re. Soon (TM))