Fat pointer trait implementations

[Diggsey]

Currently Hash is only implemented for thin pointers, and Eq/PartialEq aren't implemented for any pointers (but == works for all pointers).

All of these traits should be implemented for all pointers.

Without this, it's impossible to implement generic reference-equality semantics that support dynamically sized types, even in unsafe code, thanks to #27570. Even with #27570 it will still be impossible to do safely on stable rust, since raw::TraitObject is unstable, and even if it were stable, the type system isn't powerful enough to express when to use raw::TraitObject vs a thin pointer.

Fixing this properly will likely require improvements to the type system. Any improvement which solves rust-lang/rfcs#1279 should be enough.

[arielb1]

You can "dynamically" mem::transmute by using copy_memory (and size_of).

[Diggsey]

That's still unstable, and it's still nontrivial to implement Hash. You can do it with unsafe code and size_of, but it's not ideal - this sort of thing should be handled by the standard library so that user code doesn't have to depend on implementation details like the layout of fat pointers.

Without sufficiently smart optimizations, it will also generate sub-optimal code.

When it's possible to write a generic ReferenceEquality<T: Deref> wrapper that can be used to key a HashMap without requiring unsafe code, then this can be closed.

[arielb1]

Eq is implemented for all pointers, including fat ones:

fn check_eq<T: Eq>(t: T) { t == t; }
fn main() {
    let x: *mut [u8] = &mut [0,1];
    check_eq(x);
}

Anyway, why is

[Diggsey]

Eq is implemented for all pointers, including fat ones

Ah, it doesn't show up in rustdoc. I guess it's implemented internally in the compiler or something?

Anyway, why is

A truly deep question 💭

[arielb1]

http://doc.rust-lang.org/std/cmp/trait.Eq.html

has

impl<T> Eq for *const T where T: ?Sized

The impl (at https://github.com/rust-lang/rust/blob/master/src/libcore/ptr.rs#L279) just does *self == *other - the real implementation is in trans.

[arielb1]

It would be nice to have better primitives for accessing the fields of fat pointers, and at least implement Hash for them (not in the compiler, obviously, just using memcpy).

[steveklabnik]

Triage: 18 months with no comments. It seems that the original issue is in fact not an issue. Closing!