alloc: Make deriving more friendly with Arc

[alexcrichton]

This adds impls of Eq/Ord/PartialEq/PartialOrd/Show/Default to Arc, and it
also removes the Send + Sync bound on the Clone impl of Arc to make it more
deriving-friendly. The Send + Sync requirement is still enforce on
construction, of course!

[rust-highfive]

Warning

• These commits modify unsafe code. Please review it carefully!

[gereeter]

Is the Send + Sync bound on construction just there so that people use Rc for non-threadsafe data? It feels to me that this should be documented or the bound should be removed.

[alexcrichton]

Hm, it took me a second to think about this.

If you remove Send + Sync from all bounds, then it's safe to create Arc<Cell<int>> and share it among threads. Each instance of Arc<T> will be Send so long as T is Send (which Cell<int> is), but being this means that T needs to adhere to Sync if T is Send.

I think that we may be able to remove the Send bounds, but we may not be able to remove the Sync bounds.

I do agree, however, that the documentation of Arc could use some improvement!

[gereeter]

I would think that Arc<T> would only be Send if T was Send + Sync, which would fix the problem.

[alexcrichton]

I think we may be able to expose that with opt-in-builtin-traits, but until that time I'm not sure it's expressible. (but it'd be nice!)