DistributedTargetInvocationEncoder

Forming an invocation

On the sending-side an instance of an invocation is constructed by the runtime, and calls to: recordGenericSubstitution, recordArgument, recordReturnType, recordErrorType, and finally doneRecording are made (in this order).

If the return type of the target is Void the recordReturnType is not invoked.

If the error type thrown by the target is not defined the recordErrorType is not invoked.

An invocation implementation may decide to perform serialization right-away in the record... invocations, or it may choose to delay doing so until the invocation is passed to the remoteCall. This decision largely depends on if serialization is allowed to happen on the caller’s task, and if any smarter encoding can be used once all parameter calls have been recorded (e.g. it may be possible to run-length encode values of certain types etc.)

Once encoded, the system should use some underlying transport mechanism to send the bytes serialized by the invocation to the remote peer.

Protocol requirements

Similar to the DistributedActorSystem and its remoteCall and remoteCallVoid protocol requirements, the DistributedTargetInvocationEncoder contains a few methods which are not possible to express in source due to advanced use of generics combined with associated types. Specifically, the recordArgument and recordReturnType methods are not expressed in source as protocol requirements, but will be treated by the compiler as-if they were.

In addition to the compiler offering compile errors if those witnesses are missing in an adopting type, we present their signatures here for reference:

 /// Record an argument of `Argument` type.
 /// This will be invoked for every argument of the target, in declaration order.
mutating func recordArgument<Value: SerializationRequirement>(
    _ argument: DistributedTargetArgument<Value>
) throws

 /// Ad-hoc requirement
 ///
 /// Record the return type of the distributed method.
 /// This method will not be invoked if the target is returning `Void`.
 mutating func recordReturnType<R: SerializationRequirement>(_ type: R.Type) throws

Decoding an invocation

Since every actor system is going to deal with a concrete invocation type, they may implement decoding them whichever way is most optimal for the given system.

Once decided, the invocation must be passed to executeDistributedTarget which will decode the substitutions, argument values, return and error types (in that order).

Note that the decoding will be provided the specific types that the sending side used to preform the call, so decoding can rely on simply invoking e.g. Codable (if that is the SerializationRequirement) decoding entry points on the provided types.