In the work towards 0.9 release we have now done a major change to how Actors are created. In short we have separated identity and value, created an 'ActorRef' that holds the actual Actor instance. This allows us to do many great things such as for example:
* Create serializable, immutable, network-aware Actor references that can be freely shared across the network. They "remember" their origin and will always work as expected.
* Not only kill and restart the same supervised Actor instance when it has crashed (as we do now), but dereference it, throw it away and make it eligible for garbage collection.
* etc. much more
These work very much like the 'PID' (process id) in Erlang.
These changes means that there is no difference in defining Actors. You still use the old Actor trait, all methods are there etc. But you can't just new this Actor up and send messages to it since all its public API methods are gone. They now reside in a new class; 'ActorRef' and use need to use instances of this class to interact with the Actor (sending messages etc.).
Here is a short migration guide with the things that you have to change. It is a big conceptual change but in practice you don't have to change much.