How to define a subtrait as a version of its supertrait with extra constraints?

help
1

Hello. Just wondering, is it possible to define a subtrait as a version of its supertrait with extra constraints on the supertrait's generic associated type? I'm thinking of something like this, but this seems to be invalid syntax.

2

You can make the super-trait be more specific.

pub trait ClonableFn<'a, B>: Kind1L2T + Sized
where
    <Self as Kind1L2T>::Output<'a, Self, B>:,
{
}

Not sure if it will match what you may be trying to get to.

"extra constraints" in general is a term I would not use describing super-traits; as the purpose of traits is one of extending functionality.

3

You can have constrained subtraits already. The barriers to your particular example have more to do with the limitations of higher-ranked trait bounds (where for<...>).

Higher-ranked trait bounds over types requires non-lifetime binders (for<T>), which are unstable. You also want conditional binders (for<'a where 'b: 'a>), which don't exist so far (and it's unsure what form they will take). So it will be some time before a constraint close to that will be directly possible (if it does become possible).

There's another problem with your particular example: Kind1L2T requires that Output<'a, A, B> is defined for all 'a, A, and B -- not just when A: 'a and B: 'a. So no implementor can actually meet your bounds, given the GAT in that form.

The poor interaction between GATs and HRTBs is a well known problem. Here's an article about working around those problems for lifetime GATs. There's a section about HRTB implicit bounds, which is a way to emulate conditional binders. However, to be applicable to your use case (with a type GAT), you would also need non-lifetime binders.

It might look something like this:

(I didn't try making actual implementations and testing the bound when calling a function. And as the warning says, that feature is incomplete.)

Edit: The incomplete non-lifetime binder feature doesn't impose an implicit Sized bound on the introduced type generics, which may make it impractical unless something else changes (conditional binders, a way to emulate them for trait bounds, support for implied Sized specifically, ...). But here's the example in action.

1 Like
4

This topic was automatically closed 90 days after the last reply. We invite you to open a new topic if you have further questions or comments.