Răzvan Petruescu bio photo

Răzvan Petruescu

Functional programming for the masses

Twitter Google+ LinkedIn Stackoverflow BitBucket Coursera

I will continue the series on Scala’s type system with a discussion about path-dependent types, type projections and structural types.

Path-dependent types and type projections

Types within Scala are referred to via two mechanisms: the dot (.) and hash (#) operators.

The . operator is doing the same for types as it does for members of an object, namely, it refers to a type found on a specific object instance.

This is called a path-dependent type.

Consider this example:

trait MultiCastEventSink[T <: Event] {

  trait Sink[-In] {
    def notify(o: In)

  // notify all internal sinks that an event happened
  def notify(o: T)

  // notify only the internal sinks present in the sequence
  def notifyOnly(o: T, s: Seq[this.Sink[T]])

  // notify the internal sinks present in the sequence, if found
  def notifyAny(o: T, s: Seq[MultiCastEventSink[T]#Sink[T]])

This trait is an abstraction of an event sink, similar to the one in this article. This is a delegating or multi casting event sink that sends all notifications to an internal pool of sinks. In addition, this trait contains the definition of a regular Sink, for reasons that I will describe later. Obviously, most of the implementation details have been omitted.

A couple of MultiCastEventSink implementations follow as example:

val firstSink = new MultiCastEventSink[UserEvent] {

  override def notify(o: UserEvent): Unit = ???

  override def notifyOnly(o: UserEvent, s: Seq[Sink[UserEvent]]): Unit = ???

  override def notifyAny(o: UserEvent, s: Seq[MultiCastEventSink[UserEvent]#Sink[UserEvent]]): Unit = ???

val secondSink = new MultiCastEventSink[UserEvent] {

  override def notify(o: UserEvent): Unit = ???

  override def notifyOnly(o: UserEvent, s: Seq[Sink[UserEvent]]): Unit = ???

  override def notifyAny(o: UserEvent, s: Seq[MultiCastEventSink[UserEvent]#Sink[UserEvent]]): Unit = ???

Then a couple of regular Sinks are created:

val sink1 = new firstSink.Sink[UserEvent] {
  override def notify(o: UserEvent): Unit = ???

val sink2 = new secondSink.Sink[UserEvent] {
  override def notify(o: UserEvent): Unit = ???

Notice the syntax: new firstSink.Sink[UserEvent]. The . operator is used to access a type on an already existing instance.

Consider the method notifyOnly(o: T, s: Seq[this.Sink[T]]). The method will notify only the sinks present in the sequence that an event has happened. For reasons you will understand shortly, there’s a strict guarantee that all the Sink[T] implementations in the sequence will be found in MultiCastEventSink’s internal pool.

The second parameter of this method is a Seq of Sinks, but not of any Sinks. These are Sinks that must have been constructed using a reference of the current object: this.Sink[T]]. This means that you can’t use a type from a different object, of the same class, to satisfy any type constraints made using the dot operator.

You can think of this as if there’s a path of specific object instances connected by the dot operator. For a variable to match your type, it must follow the same object instance path.

Therefore, the statement firstSink.notifyOnly(new UserEvent {}, Seq(sink1)) will compile, as sink1 was constructed using the firstSink instance reference.

The statement secondSink.notifyOnly(new UserEvent {}, Seq(sink1)) will not compile as sink1 was not constructed using the secondSink instance reference.

Also, consider the method: notifyAny(o: T, s: Seq[MultiCastEventSink[T]#Sink[T]]). The second parameter is a Seq of Sinks, but in this case, they could be Sinks that have been constructed using the . operator using any MultiCastEventSink reference.

The # operator is a looser restriction than the . operator. It’s known as a type projection, which is a means of referring to a nested type without requiring a path of object instances. This means that you can reference a nested type as if it wasn’t nested.

Consequently, the statement secondSink.notifyAny(new UserEvent {}, Seq(sink1)) will compile.

Structural types

Imagine the following scenario: you have a trait and you want to make sure that all types which are mixing in that trait must conform to the following requirement: they must have a method with a given signature.

As a more concrete example, I want to embellish an EventSource with the additional capability of self auditing. That could mean that the EventSource will update possible interested entities with details about its internal state, the number of published events, etc.

One way to accomplish this result is to use inheritance. The problem with this approach is that you might not always be able or want to add a trait to the classes you are using.

I’ll use the EventSource from this article as a base for constructing the example.

The definition of the trait follows:

trait AuditableEventSource[T <: Event] extends Source[T] {

  // the event source must be audit-able
  this: { def audit() } =>

If you are wondering what T <: Event means, this article will shed some light.

A type that mixes in the trait has to conform to the structural specification:

class AuditableSystemEventSource extends AuditableEventSource[SystemEvent] {

  // structurally imposed
  def audit() = ???

  // imposed by inheritance
  override def get(): SystemEvent = ???