Stable type_eq

[addisoncrump]

This is right up there with the evilest code I've ever written.

struct A;
struct B;

fn main() {
    println!("A is A: {}", type_eq::<A, A>()); // true
    println!("A is B: {}", type_eq::<A, B>()); // false
}

This performs a type equality comparison using Copy/Clone specialisation imported from core. In testing, I found that it is optimized out entirely, i.e.:

fn main() {
    if type_eq::<A, A>() {
        abort();
    }
}

Is compiled down to:

> disass copy_specialization::main
Dump of assembler code for function _ZN19copy_specialization4main17h45816c84b6311796E:
   0x000055555555c260 <+0>:     push   rax
=> 0x000055555555c261 <+1>:     call   QWORD PTR [rip+0x4a8f9]        # 0x5555555a6b60 (std::process::abort)
End of assembler dump.

[addisoncrump]

cc @langston-barrett: any immediate soundness concerns come to mind here?

[domenukk]

It's insane, I love it.

[domenukk]

I think all fns should be #[inline], and add a comment/ some documentation why this works

[domenukk]

Any documentation on this?

[domenukk]

Like, anything official/some blog post, forum entry, docstring, ...?

[addisoncrump]

Sanity check here (implied specialization): https://doc.rust-lang.org/stable/std/clone/trait.Clone.html#how-can-i-implement-clone

You can see the specific specialization here: https://doc.rust-lang.org/stable/src/core/slice/mod.rs.html#4656-4686

[domenukk]

Can we add those to the code?

[domenukk]

Feel free to merge :)