Refactor the code that generates SIL to call into the distributed actor
transport to eliminate duplication and better cope with concrete actor
transports. Centralize the knowledge of which actor transport is used
with a given distributed actor type.
Immediately after the hop_to_executor in an async, distributed
actor init, we need to notify the transport that the actor is ready.
This patch does not yet account for more complex cases. In particular,
we will need a mechanism to prevent multiple calls to actorReady,
which can happen if a loop appears in the init:
distributed actor Dactor {
var x: Int
init(tport: ActorTransport, count: Int) async {
var i = count
repeat {
self.x = count
// hop is injected here
i -= 1
} while i > 0
}
}
We need to be able to inject a call to a distributed actor's
transport.actorReady, passing the actor instance to it,
during definite initialization. This means that its dependence
on SILGenFunction must be broken, hence this refactoring as
a SILOptimizer utility.
