Debug impl of Any could be improved

[mzabaluev]

The Debug impl of Any is currently very simple, it always returns "Any":

use std::any::Any;

/// I don't care if this downcasts to a string,
/// it's all "Any" to me.
fn print_any(something: &Any) {
    println!("{:?}", something);
}

fn main() {
    print_any(&String::from("Hello"));
}

This shows up in panic messages when e.g. unwrap() is called on a thread::Result of a panicked thread. The implementation could do what the default panic hook already does and try to downcast the reference to &'static str and String (and maybe also Cow<'static, str>, though that might be very unusual) and, if successful, include the contents into the output or simply delegate to the string's Debug.

[bluss]

panic! and assert! produce (usually) &str and String, so those would be good to cover.

[mzabaluev]

I could try to create a pull request. What format does the audience prefer? Delegate to the strings' Debug, or should Any get a mention in the output? Box delegates, and I think Any should, too.

[mzabaluev]

One problem is that String is an allocating container defined in liballoc, so it can't be known to the impls in libcore without some low-level hacking. A Debug implementation that only downcasts to static strings would leave out any dynamically formatted panic messages and would look more inconsistent than it does now.

[mzabaluev]

So perhaps, the String downcasting protocol could be defined in libcore to be filled in by the implementation in alloc or stubs in no-alloc linkages, similar to how it's done with panicking. Basically, it's one callback for checking against a type id, and another for obtaining the string slice from an Any pointer.

[abonander]

More discussion here: rust-lang/rfcs#1389

[mzabaluev]

@abonander Thank you, I failed to search in the RFC issues.
The powers that be can close this Rust issue now, or maybe keep it around as the tracking issue for an eventual RFC.

[abonander]

If we're just talking about changing the internals of an impl, I don't think it needs an RFC. I think most people would be happy with it printing the value if the internal type is String or &str.

[mzabaluev]

The RFC-qualifying issue is that, as libcore does not know anything about owned types, it would need a special-case hook for what liballoc, or conceivably another library in the program's runtime stack, would supply as the typecheck and as_str interface for String. The Debug impl that only prints static panic values would be a glass half empty.

[abonander]

I can't think of a way that wouldn't be incredibly hacky or violate coherence. I think a specialized impl Debug for Box<Any> which can downcast to String would be a good place to start, since both types are in alloc. This would lift the existing code out of the panic framework that's already doing the same.

[Mark-Simulacrum]

I'm going to go ahead and close this as it seems like it's quite hard to do in practice (due to not having the relevant types around). A PR for this would be the best way to land changes here, and discussion can happen on internals.