doc: Be more specific about how to box closures

[richo]

The old error was still confusing, when you try boxing you still get the same error: eg, http://is.gd/yDwxPF

[rust-highfive]

r? @brson

(rust_highfive has picked a reviewer for you, use r? to override)

[brson]

This revision is less precise than the former, since the former tells you two options, and latter just one. Creating &Trait is not 'boxing' it, since boxing means moving something onto the heap.

I also don't quite see how this description makes the behavior in your example, where boxing the closure doesn't work, less confusing. If I saw the new message I would probably wrote what you wrote that doesn't work.

So I think we can probably do better still. r? @steveklabnik

[richo]

So to be clear, the solution I wound up with here was that:

fn main() {
    let f = match 1 {
        1 => Box::new(|c| c + 1),
        _ => Box::new(|c| c - 1),
    };

    println!("{}", f(1));
}

Doesn't work but:

fn main() {
    let f: Box<Fn(i32) -> i32> = match 1 {
        1 => Box::new(|c| c + 1),
        _ => Box::new(|c| c - 1),
    };

    println!("{}", f(1));
}

Does, eg, it needed to be boxed and coerced to a trait object. It was unclear to me that I could coerce it as a reference, since that would intuitively seem like trying to put DST on the stack.

[steveklabnik]

agree with @brson here, these are two options that are being reduced to one. Boxing is one way of making a trait object, &Trait is another

[richo]

So, ignoring the content of this PR, my gripe is that the help told me to box them, and then after I put them all in boxes it told me the same thing. I'm not really fussed about this specific wording, but I think it should be more explicit.

[steveklabnik]

yeah, I'm also not saying this error message is great, for sure

[bors]

☔ The latest upstream changes (presumably #26575) made this pull request unmergeable. Please resolve the merge conflicts.