Atomic* should not be copyable

[brson]

Most of the types in unstable::atomics are implicitly copyable. This results in some confusing situations and is also incorrect since it's doing non-atomic copies.

[Aatch]

I don't think an "atomic copy" is a thing? Since all the atomic operations essentially just work on the self pointer, copying an atomic type results in an entire new value, in a new memory location. I'm pretty sure that an "atomic copy" is impossible since it requires a read and then store to 2 separate memory locations. The only thing I can think of is the lack of an atomic load.

I'm also not sure that being implicitly copyable is an issue, other the above point, and would prefer to keep this way to be honest. There is no semantic difference between using the Atomic* types and the methods on them and using the atomic intrinsics directly pointers to the values you want to manipulate. The only real advantage of the atomic types is a nicer API. The way it currently is means that the atomic types are identical (except for AtomicBool) to their non-atomic counterparts.

[brson]

It's doing a copy of the value at a memory location that is only accessed through atomics without doing an atomic load.

It's not true that they are identical to non-atomic types. They encapsulate a memory location to which they only allow access via atomics - except for this non-atomic load.

[bblum]

clone() would make sense, I think.

[catamorphism]

Nominating