Interfaces (part 3)

Objectives

In this lesson you will learn:

• About hierarchical relationships between interfaces, i.e., public interface InterfaceA extends InterfaceB.
• About how a class can implement multiple interfaces, i.e., public class MyClass implements InterfaceA, InterfaceB.
• About default methods in interfaces.
• About “the diamond” problem, i.e., what happens when a class inherits two implementations of the same method.

Recap

Interfaces allow us to define what a type should be able to do, but not how it should do it. We do this through the use of abstract methods that define a method signature, but no method body. Implementing subclasses must including implementations for those methods, and can include additional methods if needed.

In the previous lessons we learned about interfaces—both the general idea, as well as the embodiment of that idea in the form of the interface construct in the Java language.

Continuing with our running example of the Nim game, we came up with the following class design for the game.

The Game class has two Players and a Pile of sticks. The Player is an interface, which means that at run time the Player might be any kind of available Player subtype, and the Game doesn’t know or care which. The Player interface, in other words, sits between the Game and the different player implementations, hiding those implementation details from other parts of the system.

classDiagram
  direction LR
  note for Game "Underlined members are static." 
  Game --> `interface Player` : has two
  Game --> Pile : has a
  `interface Player` <|-- TimidPlayer : is a 
  `interface Player` <|-- GreedyPlayer : is a 
  `interface Player` <|-- RandomPlayer : is a 

  class Game {
    +Player p1
    +Player p2
    +Pile pile
    playGame() void$
  }

  class `interface Player` {
    +getName() String*
    +takeTurn(Pile) int*
  }

  namespace PlayerSubtypes {
    class TimidPlayer { }

    class GreedyPlayer {
      +String jeer
      +jeer() String
    }

    class RandomPlayer { }
  }

  class Pile {
    +int numSticks
    +remove(int) void
    +getSticks() int
  }

For the rest of this lesson, for the sake brevity, we will only include the Player and its subtypes when we consider a diagrammatic view of the class structure.

Interfaces can extend interfaces

In this lesson, we are going to consider how to incorporate yet more features into our Nim game.

In the design above, we have included support for the GreedyPlayer to jeer at their opponent each time they take a turn.

Let’s suppose want to add support for multiple types of “chatty” players in our system. That is, we don’t want only our GreedyPlayer to be able to print a message each time they take a turn—we want to allow any new Player type to optionally have this ability.

Let’s consider some options to accomplish this.

Option 1 — Give all players the ability to print a message

We could give all of our Player subtypes a makeComment() method that prints the Player’s comment to the screen. We can also give them a makeComment(String) overload that prints the specified message to the screen, instead of the Player’s pre-defined message.

An advantage of this approach is that it’s simple. We can stick a makeComment overloaded abstract methods in the Player interface, and all implementing subclasses can print their chosen message.

A disadvantage is that this would reduce the cohesiveness of our Player type. For example, the TimidPlayer has no intention of presenting a message each time they take a turn — what should the TimidPlayer’s makeComment() method return? An empty string ("")? null? Some string that will never be used?

The Player interface is now less cohesive because it includes behaviour (message()) that is not relevant to all Player instances.

Option 2 — Break out the chatty behaviour into another type

Option 2 is to separate the behaviour of playing games and “socializing” — they are separate concerns that don’t need to be implemented together.

We can do this by creating a separate interface for SocialPlayers. The SocialPlayer interface will extend the Player interface.

The SocialPlayer interface looks like the following:

interface SocialPlayer extends Player {
  void makeComment(); // prints some pre-defined comment
  void makeComment(String message); // prints the specified message
}

In the code above, we use the extends keyword to declare that the SocialPlayer interface is a subtype of the Player interface. That is, all SocialPlayers are Players.

The SocialPlayer only declares one (overloaded) behaviour—the makeComment methods. Any class that implements SocialPlayer must now include implementations of Player methods and implementations of SocialPlayer methods.

In our updated Nim game, our GreedyPlayer is a SocialPlayer, which means it can take turns in the game like all Players, but can also send messages.

Here is what the Player type hierarchy looks like now. In the diagram below, we use implements or extends to show the specific keyword used to define the “is a” relationship between a child type and its parent type.

classDiagram
  direction TB
  `interface Player` <|-- `interface SocialPlayer` : extends
  `interface Player` <|-- TimidPlayer : implements 
  `interface Player` <|-- RandomPlayer : implements 
  `interface SocialPlayer` <|-- GreedyPlayer : implements 

  class `interface Player` {
    +getName() String*
    +takeTurn(Pile) int*
  }
  
  class `interface SocialPlayer` {
    +makeComment() void*
    +makeComment(String message) void*
  }

  namespace PlayerSubtypes {
    class TimidPlayer { }

    class GreedyPlayer { }

    class RandomPlayer { }
  }

Our GreedyPlayer implementation would look the same as before (except instead of jeer we now use the more general makeComment methods to chat). But now we can incorporate additional player types that have social behaviours, and treat them all as SocialPlayers in other parts of the system.

public class GreedyPlayer implements SocialPlayer {
  private String name;
  private String jeer; // This player talks smack
  
  public GreedyPlayer(String name, String jeer) {
    this.name = name;
    this.jeer = jeer;
  }

  @Override
  public void makeComment() {
    System.out.println(this.jeer);
  }

  @Override
  public void makeComment(String message) {
    System.out.println(message);
  }

  @Override
  public String getName() {
    return this.name;
  }

  @Override  
  public int takeTurn(Pile pile) {
    int toRemove = 0;
    if (pile.getSticks() >= 3) {
      toRemove = 3;
    } else {
      toRemove = pile.getSticks();
    }
    return toRemove;
  }
}

What’s the benefit of doing this?

The Player interface introduced a uniform set of behaviours that the Game could rely on; i.e., it could expect all Player objects to be able to do things like taking a turn, no matter what kind of player subtype they were.

Similarly, the SocialPlayer interface introduces a uniform set of additional behaviours that some players can perform—they can make comments. This opens up an avenue for a version of the Game that allows some players (those that support social behaviours) to make comments during game play.

Consider this updated play method for the Game class. In the code below, each time a Player takes a turn, we give the opponent the opportunity to print a message.

We use the instanceof operator to check if the opponent is an instance of the SocialPlayer type, and if so, we print the message. The SocialPlayer interface allows us to change our “view” of the opponent object, deciding whether to see it as simply a Player, or as a SocialPlayer, depending on what set of behaviours we mean to invoke.

public static boolean play(Player p, Pile pile, Player opponent) {
  int sticksTaken = p.takeTurn(pile);
  System.out.println("\n" + p.getName() + " takes " + sticksTaken + " sticks.\n" +
    "There are " + pile.getSticks() + " left in the pile.");

  // Each time a player takes a turn, if the opponent is a social player,
  // print a vaguely threatening message.
  if (opponent instanceof SocialPlayer) {
    ((SocialPlayer) opponent).makeComment("Bad move, " + p.getName() + "!");
  }

  if (pile.getSticks() <= 0) {
      return true;
  }

  return false;      
}

Do you think this is a reasonable use of the instanceof operator? Is there any way to print the opponent’s message from the SocialPlayer implementing subclasses directly?

In the example, we are dealing with both “views” of the opponent in the same method.

As a larger example, consider if we wrote a separate, feature-rich “chatroom” module for this Nim game. We can make that module only view players as SocialPlayers, since it is only interested in the behaviours relevant to social interactions, and not game play.

Classes can implement multiple interfaces

What if, as part of our expanded Nim application, we wanted to also support general “socializers”? That is, users who are not players in the Nim game, but are still able to make comments that other users can see?

As things currently stand, to create “socializers”—users that can make comments—we would need to implement the SocialPlayer interface. Unfortunately, this brings with it a fair bit of baggage — if you implement the commenting behaviour from the SocialPlayer, you need to also implement the game play behaviour from the Player interface. This is due to the hierarchical relationship between SocialPlayer and Player: all SocialPlayers are Players.

We can decouple the socializing functionality from the gameplay functionality by removing the hierarchical relationship between those interfaces. That is, instead of wrapping social behaviours into a SocialPlayer interface that is a child type of Player, we can create two separate, unrelated interfaces:

• The Player interface for gameplay behaviours, unchanged from what we have seen so far.
• The Socializer interface for social behaviours, like making comments.

In Java, classes can implement multiple interfaces. These interfaces provide different “views” to the class, or different, well, interfaces through which to interact with the class.

The player subtypes that only support gameplay functionality (like TimidPlayer and RandomPlayer) will implement only the Player interface. The player subtypes that want to support both gameplay functionality and socializing functionality (like GreedyPlayer) will implement both the Player and Socializer interfaces. Finally, the users that want to only support socializing functionality (say, Spectator or Referee objects) will only implement the Socializer interfaces.

Here is what our new class structure would look like. By separating the Player and Socializer interfaces (instead of having them have a hierarchical relationship), we have allowed classes to more flexibly combine or not combine those behaviours.

classDiagram
  direction TB
  `interface Player` <|-- TimidPlayer : implements 
  `interface Player` <|-- RandomPlayer : implements 
  `interface Player` <|-- GreedyPlayer : implements 
  `interface Socializer` <|-- GreedyPlayer : implements 
  `interface Socializer` <|-- Spectator : implements
  `interface Socializer` <|-- Referee : implements

  class `interface Player` {
    +getName() String*
    +takeTurn(Pile) int*
  }

  class `interface Socializer` {
    +makeComment() void*
    +makeComment(String message) void*
  }

  class TimidPlayer { }

  class GreedyPlayer { }

  class RandomPlayer { }

To support the structure above, our GreedyPlayer only needs to change its signature to implement the two interfaces. Everything else in the class would remain the same: it still needs to implement all Player behaviours, and it still needs to implement all Socializer behaviours.

public class GreedyPlayer implements Player, Socializer {
  // Rest of the GreedyPlayer class remains the same
}

default methods

The GreedyPlayer, Spectator, and Referee objects all have to include implementations for the two makeComment methods from the Socializer interface. The overload that has no parameters, i.e., makeComment(), will be different for each implementation.

It makes sense for those three classes to implement their own versions of makeComment().

• The GreedyPlayer prints its jeer.
• The Spectator might print support a message of support for the player that has their allegiance.
• The Referee might print a message after conducting some checks concerning game rules.

However, consider that second overload: makeComment(String). In most cases, the job of that method is to simply print the input it has been given.

Do we really want to duplicate that code in each of those three classes?

Enter default methods. So far, we have only seen abstract methods in interfaces: methods with declarations, but no definitions. Method signatures, but no bodies. These methods must be implemented by a subclass, because otherwise the object cannot actually perform that behaviour.

However, interfaces also allow us to define default methods: these are methods in interfaces that do have implementations. These implementations are inherited by all implementing subclasses, unless the subclass overrides it.¹

Let us consider the Socializer interface as our illustrative example.

public interface Socializer {
  void makeComment();

  default void makeComment(String message) {
    System.out.println(message);
  }
}

Each of the implementing subclasses (GreedyPlayer, Spectator, and Referee) will inherit the existing default implementation of the second makeComment method above. This means that they only need to implement the first makeComment() in order to “fully implement” the Socializer interface. This is great! It saves us from having to duplicate the second makeComment(String) method three times.

Of course, they are free to override the makeComment(String) method if they want to do something different from inherited default method.

For example, suppose the Referee wants to print the word "WHISTLE!" before each comment they print.

In the code below, the Referee class implements one makeComment() method because that has to be implemented — it’s abstract in the Socializer interface. It also implements the makeComment(String) method, this time overriding the default behaviour that was inherited from the Socializer interface.

It prints the word "WHISTLE!" first. Then it uses the super keyword to invoke the parent implementation of the makeComment(String) method.

The super keyword is like the this keyword, except the object refers to itself as its parent type instead of its own type. In the example below, the super.makeComment(message) invokes the makeComment(String) method from the parent type, i.e., the Socializer interface.

public class Referee implements Socializer {
  // Referee's instance variables
  private int game; // The Game that is currently being refereed

  public Referee(Game game) {
    this.game = game;
  }

  @Override
  public void makeComment() {
    System.out.println(game); // The referee just reports the Game state
  }

  @Override
  public void makeComment(String message) {
    System.out.print("WHISTLE! "); // Referee adds its own behaviour here
    super.makeComment(message); // Referee invokes the Socializer's default behaviour
  }
}

In the makeComment(String) method above, what do you think would happen if we called this.makeComment(message) instead of super.makeComment(message)?

Multiple conflicting default methods

So far, we have learned the following facts about interfaces in Java:

• Interfaces can define abstract methods that must be implemented by subclasses.
• Interfaces can also define “concrete” methods using the default keyword — these can be overridden by implementing subclasses, or they can be inherited.
• Classes can implement multiple interfaces.

With the facts above in mind, consider the following example.

Class C implements both interfaces A and B. Interfaces A and B both define the method doStuff as a default method, but both do different things.

classDiagram
  direction TD

  note "Interfaces A and B both define doStuff as a default method."

  `interface A` <|-- C : implements 
  `interface B` <|-- C : implements 

  class `interface A` {
    +doStuff() void
  }

  class `interface B` {
    +doStuff() void
  }

Which of the two doStuff behaviours should class C inherit?

Java doesn’t support multiple inheritance

In a situation like the above, there is simply no way for the compiler to know which doStuff method you want to inherit. So the compiler will show you an error until you give class C its own implementation of doStuff. By implementing its version of the method, there is no ambiguity about which one to inherit—C inherits neither of the parent interface’s doStuff implementations.

What if you wanted one of the specific implementations? Do you just duplicate that code? No.

In that case, you still need to write your own doStuff method, but in the body of the method, you can invoke the specific parent doStuff that you want.

So suppose you want the class C’s doStuff method to do whatever was defined in interface A. You do the following:

public class C implements A, B {
  @Override
  public void doStuff() {
    A.super.doStuff(); // Invoke the interface A's version of the method
  }
}

extends or implements?

So we’ve seen two ways of creating fairly complex class structures. We can either create a “tree like” structure, where one interface extends another interface in order to allow classes to combine those behaviours. Or we can create a “flatter” structure by creating multiple interfaces; classes that want to combine those behaviours simply need to implement both interfaces. And default methods allow us to also introduce some code reuse into this picture.

Interfaces let you create non-hierarchical type frameworks. Not all class organisations lend themselves to tree structures. That is, you may want different combinations of types “mixed together” for specific subclasses. To achieve this flexibly with extends relationships, you would end up with many more “intermediate” layers in your type hierarchy, creating a separate type for each combination of functionality you want to support. With interfaces you have infinite flexibility to enhance class behaviours as needed.

However, it is often easier to reason about tree-like structures, because there is less ambiguity about what classes can perform what behaviours. This “straightline” flow of inherited behaviours can often be a blessing in a large, complex class structure. If you know you’re not likely to add new classes that support some behaviours but not others, it may be worthwhile to commit to a tree-like structure using extends for the time being.

Summary

To sum up, here are some facts about interfaces:

• Interfaces can have abstract methods. These only have method signatures, but not method bodies. There is no special keyword for abstract methods in interfaces.
• A class that implements an interface must include implementations for all abstract methods in a parent interface.
• Interfaces can also have methods that do have implementations. These are created using the default keyword.
• Implementing subclasses can include implementations for the default methods from the parent interface; this will override the interface’s default behaviour for that method.
• Interfaces can have static variables and static methods.
• Interfaces cannot have instance variables.
• Interfaces can extend other interfaces using the extends keyword. If a class declares that it implements an interface, and that interface has “parent” interfaces, the class must implement all abstract methods from the immediate parent as well as all the “grand parent” interfaces.
• Classes can implement more than one interface.