Higher ranked lifetimes cannot cross type boundaries

[Diggsey]

This currently does not compile:

struct Message<'a> {
    data: &'a u32,
}

struct Handler<T>(Option<Box<Fn(T)>>);

struct Cfg {
    handler: for<'a> Handler<Message<'a>>, // error: `Handler` is not a trait [E0404]
}

This precludes a lot of useful patterns.

[eefriedman]

What you've written isn't really what you want... what you want is something more like:

struct Message<'a> {
    data: &'a u32,
}

// The second for<> is written out for clarity, but could be elided.
struct Handler<for<'a> T<'a>>(Option<Box<for<'a> Fn(T<'a>)>>);

struct Cfg {
    handler: Handler<Message>,
}

That's completely reasonable... but sort of complicated to implement. Possibly a duplicate of rust-lang/rfcs#324 .

[Diggsey]

Ah, I was wondering if this was the same as HKT or not: the frustrating part about this it's that it's already possible if you get rid of Handler and always write the signature in full, so it feels like a failure of the language that you can't refactor out the common code.

[arielb1]

As this is equivalent to HKT, I am closing (feel free to open an RFC).