Unboxed closures with all Clone members should implement Clone

[seanmonstar]

Similar to #19128

#![feature(core, unboxed_closures)]
use std::ops::Fn;

#[derive(Clone)]
struct NotCopy;

#[derive(Clone)]
struct Derp {
    foo: NotCopy
}

impl Fn<()> for Derp {
    type Output = ();
    extern "rust-call" fn call(&self, _: ()) {
    }
}

fn with_clone<F>(_f: F) where F: Fn() + Clone {

}

fn main() {
    let foo = NotCopy;
    with_clone(Derp { foo: NotCopy }); // works
    with_clone(move || { // errors
        let _ = foo; // <- should be like Derp, accessing self.foo
    });
}

[bluss]

This is important for iterators -- with this we can clone most iterator pipelines. In fact, most don't even need captures at all and would be the cheapest to clone.

[withoutboats]

This would be good for situations in which one wants to box and store a closure in a small minority of cases. Would be good to be able to have something like Cow<'a, Fn(...)>.

[arielb1]

Sounds like an nice idea. Unfortunately it would require Clone to be a lang-item, but it is already half a lang-item as a supertrait to Copy.

[seanmonstar]

The expanded closure struct couldn't have the equivalent of 'derive(Clone)'
added to it?

On Sat, Nov 14, 2015, 9:32 AM arielb1 notifications@github.com wrote:

Sounds like an nice idea. Unfortunately it would require Clone to be a
lang-item, but it is already half a lang-item as a supertrait to Copy.

—
Reply to this email directly or view it on GitHub
#23501 (comment).

[arielb1]

@seanmonstar

derive(Clone) runs before upvars are known.

[oli-obk]

I think this can be closed now that we have #44490