Interpreter<H> isn't safely Send or Sync because of `dyn RegionHandler`

[martin-hughes]

The lines

unsafe impl<H> Send for Interpreter<H> where H: Handler + Send {}
unsafe impl<H> Sync for Interpreter<H> where H: Handler + Send {}

aren't true because dyn RegionHandler might not be Send/Sync.

I can see that facs is safe to declare Send or Sync because only the global_lock field is actually used.

I'm sure there's some generic-types magic to resolve this. Alternatively it might be desirable to constrain the RegionHandlers to be Send + Sync, but that'd technically be a breaking API change.

[IsaacWoods]

This is true - region handlers at the moment are basically at the stage of me sketching them out in the initial rewrite to say that they should exist. The signature will need to change: read and write will need to take an address in the region the IO is happening at, and likely will need to take an &Interpreter to be able to e.g. find a PCI address (unless we want to say region handlers should be storing an Arc<Interpreter> or something behind the scenes (but that pushes users in a direction they may not want to take)).

Overall, RegionHandler should require types to also be Send and I think most likely Sync