# The Composite Pattern

## References

In this material, we will explore, via an example, the Composite pattern. Let us say that we are aiming to work with arithmetic expressions built out of +,-, *, / and constants. Thus, the expressions that we are interested in are described by the BNF grammar:

 e ::= c (constant) | e1+e2 (addition) | e1-e2 (minus) | e1*e2 (multiplication) | e1/e2 (quotient)

The operations that we desire are as follows:

• Evaluation via an inorder traversal
• Preorder traversal for display
• Postorder traversal for display

We begin by defining an interface that captures the desired functionality:

`public interface Expr {    void preorder();    void postorder();    int evaluate();}`

Now, for each constructor of arithmetic expressions that we are interested in, we have a subclass that implements this interface. The basic idea is to use dynamic binding as a replacement for conditionals that perform switching based on the kind of expression. Thus, we are looking at a picture of the kind: Let us begin by only looking at the constructors of the subclasses. The constructors reflect the arity of the arithmetic operators. Thus, we have:

 Const(int i) Plus(Expr lt, Expr rt) Minus(Expr lt, Expr rt) Mult(Expr lt, Expr rt) Quot(Expr lt, Expr rt)

Note how the constructors as well as the class diagram show that certain kinds of expressions (all except `Const`) are composed from subexpressions. Hence the name Composite Pattern.

Even before we write the code that implements these subclasses, the form of the constructors allows us to explore thre representation of expressions that we have achieved.

 1 new Const(1) 1+2 new Plus(new Const(1), new Const(2)) 3*4 new Mult(new Const(3), new Const(4)) (1+2)-(3*4) new Minus(new Plus(new Const(1), new Const(2)), new Mult(new Const(3), new Const(4))) ((1+2)-(3*4))/5 new Quot(new Minus(new Plus(new Const(1), new Const(2)), new Mult(new Const(3), new Const(4))), new Const(5)

What we are doing is write down the prefix form of the expression tree. The key is that all the above expressions are of type Expr, enabling the following code to typecheck.

`Expr one = new Const(1);Expr onePtwo = new Plus(new Const(1), new Const(2));Expr threeMfour = new Mult(new Const(3), new Const(4));Expr m = new Minus(onePtwo, threeMfour);Expr n = new Quot(m,new Const(5));`

Before we proceed further, let us consider each of the subclasses in turn.

`class Const implements Expr {    private int v;    public Const(int w) { v = w;}    public int evaluate() { return v; }    public void preorder() { System.out.println("CONST" + v); }    public void postorder() { System.out.println("CONST" + v); }}`
```class Plus implements Expr {    Expr lt, rt;    public Plus(Expr l, Expr r) { lt = l; rt =r ; }    public int evaluate() { return lt.evaluate() + rt.evaluate();}    public void preorder() {        System.out.println("PLUS");        lt.preorder();        rt.preorder();    }    public void postorder() {        lt.postorder();        rt.postorder();
System.out.println("PLUS");    }}```
`class Minus implements Expr {    Expr lt, rt;    public Minus(Expr l, Expr r) { lt = l; rt =r ; }    public int evaluate() { return lt.evaluate() - rt.evaluate(); }    public void preorder() {        System.out.println("MINUS");        lt.preorder();        rt.preorder();    }    public void postorder() {        lt.postorder();        rt.postorder();        System.out.println("MINUS");    }}`
`class Mult implements Expr {    Expr lt, rt;    public Mult(Expr l, Expr r) { lt = l; rt =r ; }    public int evaluate() { return lt.evaluate() *rt.evaluate();}    public void preorder() {        System.out.println("MULT");        lt.preorder();        rt.preorder();    }    public void postorder() {        lt.postorder();        rt.postorder();        System.out.println("MULT");    }}`
`class Quot implements Expr {    Expr lt, rt;    public Quot(Expr l, Expr r) { lt = l; rt =r ; }    public int evaluate() { return lt.evaluate() / rt.evaluate();}    public void preorder() {        System.out.println("QUOT");        lt.preorder();        rt.preorder();    }    public void postorder() {        lt.postorder();        rt.postorder();        System.out.println("QUOT");    }}`

The striking feature of the above code is the total absence of conditionals. Indeed, a main theme of the lectures will be "Inheritance is a succinct way to express conditionals". To make this discussion concrete, considera typical recursive evaluation function that would be written in a standard imperative language, where we have a record type called Node, with a Type function that yields the type of node. The evaluation makes a case-switch based on the type.

`public int evaluate(Node n) {    int r;    if (Type(Node) == CONSTANT) {        r = n.contents();    } else if (Type(Node) == PLUS) {        r = evaluate(n.left) + evaluate(n.right);    }    ...}`

In our presentation, the effect of the conditional is achieved by dynamic dispatch. For example, consider the following code.

`Expr one = new Const(1);Expr onePtwo = new Plus(new Const(1), new Const(2));Expr threeMfour = new Mult(new Const(3), new Const(4));Expr m = new Minus(onePtwo, threeMfour);Expr n = new Quot(m,new Const(5));System.out.println(n.evaluate());`

We will trace the execution a little to see the way in which dynamic dispatch works.

• n.evaluate() --- n is an object of Class Quot, so the evaluate method invoked is that of Class Quot. This call invokes evaluate on m and new Const(5). Note that the code of class Quot (indeed of all subclasses) is not aware of the type of nodes of its two components.
• new Const(5).evaluate() --- new Const(5) is an object of class Const. Thus, the evaluate method invoked is of the subclass Const and this call returns the value 5.
• m.evaluate() --- m is an object of class Minus. Thus, the evaluate method invokes in of the subclass Minus.

## Issues to explore and think about

• The presentation above is very fragile with respect to the addition of new operations at the parent class. Thus, any such change has to be carried out systematically across all subclasses. In future lectures, we will examine the use of another design pattern, the Visitor, to localize such changes.
• Those familiar with recursive data structures and pattern matching constructs in languages such as ML, will recognize the Composite pattern as a way to describe recursive data structures. The Visitor pattern will enable us to build pattern matching in the context of the OO paradigm.
• In the presentation above, there was no need for parent pointers. In general applications of the Composite pattern, maintaining references from children to parents can facilitate the traversal of the data strucuture --- for a more detailed treatment, see page 166 of Gamma et al.
• The primary aim of the Composite pattern is to insulate clients from the different kinds of components. in our example, the clients are only aware of Expr, and are insulated from all the different ways of building arithmetic expressions. There is some tension between this aim and the possible variety of node types. For example, in our example, the Constant nodes differ from others in that they are base case; in all other cases, they form inductive steps of expression building, and build bigger expressions out of smaller ones. Thus, in all these latter cases, they need data strcuctures to store and manipulate references to children. In our example, these manipulations are localized to such nodes and are not expressed in the Expr interface. In general, this may not be possible. For a detailed discussion, refer to pages 167--170 of Gamma et al.
• The use of the Composite pattern to define a language is called the Interpreter pattern.