Unboxed closures with no or only-Copy captures should be Copy

[reem]

Right now no unboxed closures are Copy, but there's no reason that 0-size closures or those with only copy captures couldn't be Copy.

[japaric]

This is solved, unboxed closures are currently copyable:

fn is_copy<T: Copy>(_: T) {}

fn main() {
    let i = 5i;
    is_copy(move |:| i);
    is_copy(|&:| {});
    is_copy(|&mut :| {});
    is_copy(|:| i);
    is_copy(|:| {});
}

But according to #19817 and #19889, the plan is that they shouldn't be implicitly copyable (which I don't quite agree with).

So apparently the changes to make the trait Copy opt-in for closures have been implemented, which means that the code above doesn't work any more.
But can somebody explain how I can actually opt-in for the Copy trait when declaring a closure?

[steveklabnik]

Triage: closures aren't copyable again, but I am not sure if they're intended to be or not. @rust-lang/lang?

[bluss]

They were deliberately made non-copy in issue #19817

[huonw]

But can somebody explain how I can actually opt-in for the Copy trait when declaring a closure?

There's a few possible ways to have duplicatable closures (all aren't great, unfortunately):

• pass a & reference instead of the closure itself (&F implements Fn when F itself does), which implements Copy
• use Rc<F> (possibly a trait object), similar to the above, but this can be stored/returned (only implements Clone
• write fn foo<F, G>(f: F) where F: Fn() -> G, G: Fn(...) -> ... instead of fn foo<F>(f: F) where F: Fn(...) -> ..., where the F closure creates new instances of the actual closure of interest (G), effectively behaving as the clone function. I suspect this often works as just foo(|| |actual| closure), i.e. just add || at the start of the closure.

These have various downsides e.g. the first two only work for Fn (not FnMut or FnOnce), the first cannot be returned from functions/stored inside return values, the second doesn't integrate with generics quite as well as others, and the third is rather ugly.

[steveklabnik]

Thanks @bluss !