1

I'm seeing a difference in isolation checking between accessing a computed property vs invoking a method and I'm not sure if it's correct or not.

Accessing an async computed property of a non-sendable type in an isolated context (e.g. @MainActor) produces an error, whereas invoking an async method on the non-sendable type does not:

class NonSendable {
  var foo: Int { get async { 42 } }
  func foo() async {}
}

@MainActor
func foo() async {
  // 🛑 Non-sendable type 'NonSendable' exiting main
  //    actor-isolated context in call to non-isolated
  //    property 'foo' cannot cross actor boundary
  _ = await NonSendable().foo

  // OK ✅
  await NonSendable().foo()
}

I would think both should cause failures. It also doesn't fail if you pass NonSendable() to an async free function, and I would expect that to fail too.

This happens in Xcode 16 beta 4 as well as a snapshot from 7-16. Haven't tried a newer snapshot yet.

2

I think this actually shouldn't raise an error for the property in that example. Due to the region-based isolation, NonSendable in your example is in disconnected region, so it is safe to pass it to another isolation via async call. I think there was a thread on false diagnostic for computed properties...

UPD. That one: https://forums.swift.org/t/why-does-sending-a-sendable-value-risk-causing-data-races