Section 10.5
Writing Generic Classes and Methods
So far in this chapter, you have learned about using the generic classes and methods that are part of the Java Collection Framework. Now, it's time to learn how to write new generic classes and methods from scratch. Generic programming produces highly general and reusable code -- it's very useful for people who write reusable software libraries to know how to do generic programming, since it enables them to write code that can be used in many different situations. Not every programmer needs to write reusable software libraries, but every programmer should know at least a little about how to do it. In fact, just to read the JavaDoc documentation for Java's standard generic classes, you need to know some of then syntax that is introduced in this section.
I will not cover every detail of generic programming in Java in this section, but the material presented here should be sufficient to cover the most common cases.
10.5.1 Simple Generic Classes
Let's start with an example that illustrates the motivation for generic programming. In Subsection 10.2.1, I remarked that it would be easy to use a LinkedList it implement a queue. (Queues were introduced in Subsection 9.3.2.) To ensure that the only operations that are performed on the list are the queue operations enqueue, dequeue, and isEmpty, we can create a new class that contains the linked list as a private instance variable. To implement queues of strings, for example, we can define the class:
class QueueOfStrings {
private LinkedList<String> items = new LinkedList<String>();
public void enqueue(String item) {
items.addLast(item);
}
public String dequeue() {
return items.removeFirst();
}
public boolean isEmpty() {
return (items.size() == 0);
}
}
This is a fine and useful class. But, if this is how we write queue classes, and if we want queues of Integers or Doubles or JButtons or any other type, then we will have to write a different class for each type. The code for all of these classes will be almost identical, which seems like a lot of redundant programming. To avoid the redundancy, we can write a generic Queue class that can be used to define queues of any type of object.
The syntax for writing the generic class is straightforward: We replace the specific type String with a type parameter such as T, and we add the type parameter to the name of the class:
class Queue<T> {
private LinkedList<T> items = new LinkedList<T>();
public void enqueue(T item) {
items.addLast(item);
}
public T dequeue() {
return items.removeFirst();
}
public boolean isEmpty() {
return (items.size() == 0);
}
}
Note that within the class, the type parameter T is used just like any regular type name. It's used to declare the return type for dequeue, as the type of the formal parameter item in enqueue, and even as the actual type parameter in LinkedList<T>. Given this class definition, we can use parameterized types such as Queue<String> and Queue<Integer> and Queue<JButton>. That is, the Queue class is used in exactly the same way as built-in generic classes like LinkedList and HashSet.
Note that you don't have to use "T" as the name of the type parameter in the definition of the generic class. Type parameters are like formal parameters in subroutines. You can make up any name you like in the definition of the class. The name in the definition will be replaced by an actual type name when the class is used to declare variables or create objects. If you prefer to use a more meaningful name for the type parameter, you might define the Queue class as:
class Queue<ItemType> { private LinkedList<ItemType> items = new LinkedList<ItemType>(); public void enqueue(ItemType item) { items.addLast(item); } public ItemType dequeue() { return items.removeFirst(); } public boolean isEmpty() { return (items.size() == 0); } }
Changing the name from "T" to "ItemType" has absolutely no effect on the meaning of the class definition or on the way that Queue is used.
Generic interfaces can be defined in a similar way. It's also easy to define generic classes and interfaces that have two or more type parameters, as is done with the standard interface Map<T,S>. A typical example is the definition of a "Pair" that contains two objects, possibly of different types. A simple version of such a class can be defined as:
class Pair<T,S> {
public T first;
public S second;
public Pair( T a, S b ) { // Constructor.
first = a;
second = b;
}
}
This class can be used to declare variables and create objects such as:
Pair<String,Color> colorName = new Pair<String,Color>("Red", Color.RED);
Pair<Double,Double> coordinates = new Pair<Double,Double>(17.3,42.8);
Note that in the definition of the constructor in this class, the name "Pair" does not have type parameters. You might have expected "Pair<T,S>. However, the name of the class is "Pair", not "Pair<T,S>, and within the definition of the class, "T" and "S" are used as if they are the names of specific, actual types. Note in any case that type parameters are never added to the names of methods or constructors, only to the names of classes and interfaces.
10.5.2 Simple Generic Methods
In addition to generic classes, Java also has generic methods. An example is the method Collections.sort(), which can sort collections of objects of any type. To see how to write generic methods, let's start with a non-generic method for counting the number of times that a given string occurs in an array of strings:
/**
* Returns the number of times that itemToCount occurs in list. Items in the
* list are tested for equality using itemToCount.equals(), except in the
* special case where itemToCount is null.
*/
public static int countOccurrences(String[] list, String itemToCount) {
int count = 0;
if (itemToCount == null) {
for ( String listItem : list )
if (listItem == null)
count++;
}
else {
for ( String listItem : list )
if (itemToCount.equals(listItem))
count++;
}
return count;
}
Once again, we have some code that works for type String, and we can imagine writing almost identical code to work with other types of objects. By writing a generic method, we get to write a single method definition that will work for objects of any type. We need to replace the specific type String in the definition of the method with a the name of a type parameter, such as T. However, if that's the only change we make, the compiler will think that "T" is the name of an actual type, and it will mark it as an undeclared identifier. We need some way of telling the compiler that "T" is a type parameter. That's what the "<T>" does in the definition of the generic class "class Queue<T> { ...". For a generic method, the "<T>" goes just before the name of the return type of the method:
public static <T> int countOccurrences(T[] list, T itemToCount) {
int count = 0;
if (itemToCount == null) {
for ( T listItem : list )
if (listItem == null)
count++;
}
else {
for ( T listItem : list )
if (itemToCount.equals(listItem))
count++;
}
return count;
}
The "<T>" marks the method as being generic and specifies the name of the type parameter that will be used in the definition. Of course, the name of the type parameter doesn't have to be "T"; it can be anything. (The "<T>" looks a little strange in that position, I know, but it had to go somewhere and that's just where the designers of Java decided to put it.)
Given the generic method definition, we can apply it to objects of any type. If wordList is a variable of type String[] and word is a variable of type String, then
int ct = countOccurrences( wordList, word );
will count the number of times that word occurs in wordList. If palette is a variable of type Color[] and color is a variable of type Color, then
int ct = countOccurrences( palette, color );
will count the number of times that color occurs in palette. If numbers is a variable of type Integer[], then
int ct = countOccurrences( numbers, 17 );
will count the number of times that 17 occurs in numbers. This last example uses autoboxing; the 17 is automatically converted to a value of type Integer, as if we had said "countOccurrences( numbers, new Integer(17) )". Note that, since generic programming in Java applies only to objects, we cannot use countOccurrences to count the number of occurrences of 17 in an array of type int[].
A generic method can have one or more type parameters, such as the "T" in countOccurrences. Note that when a generic method is used, as in the function call "countOccurrences(wordlist, word)", there is no explicit mention of the type that is substituted for the type parameter. The compiler deduces the type from the types of the actual parameters in the method call. Since wordlist is of type String[], the compiler can tell that in "countOccurrences(wordlist, word)", the type that replaces T is String. This contrasts with the use of generic classes, as in "new Queue<String>()", where the type parameter is specified explicitly.
The countOccurrences method operates on an array. We could also write a similar method to count occurrences of an object in any collection:
public static <T> int countOccurrences(Collection<T> collection, T itemToCount) {
int count = 0;
if (itemToCount == null) {
for ( T item : collection )
if (item == null)
count++;
}
else {
for ( T item : collection )
if (itemToCount.equals(item))
count++;
}
return count;
}
Since Collection<T> is itself a generic type, this method is very general. It can operate on an ArrayList of Integers, a TreeSet of Strings, a LinkedList of JButtons, ....
10.5.3 Type Wildcards
There is a limitation on the sort of generic classes and methods that we have looked at so far: The type parameter in our examples, usually named T, can be any type at all. This is OK in many cases, but it means that the only things that you can do with T are things that can be done with every type, and the only things that you can do with objects of type T are things that you can do with every object. With the techniques that we have covered so far, you can't, for example, write a generic method that compares objects with the compareTo() method, since that method is not defined for all objects. The compareTo() method is defined in the Comparable interface. What we need is a way of specifying that a generic class or method only applies to objects of type Comparable and not to arbitrary objects. With that restriction, we should be free to use compareTo() in the definition of the generic class or method.
There are two different but related syntaxes for putting restrictions on the types that are used in generic programming. One of these is bounded type parameters, which are used as formal type parameters in generic class and method definitions; a bounded type parameter would be used in place of the simple type parameter T in "class GenericClass<T> ..." or in "public static <T> void genericMethod(...". The second syntax is wildcard types, which are used as type parameters in the declarations of variables and of formal method parameters; a wildcard type could be used in place of the type parameter String in the declaration statement "List<String> list;" or in the formal parameter list "void max(Collection<String> c)". We will look at wildcard types first, and we will return to the topic of bounded types later in this section.
Let's start with a simple example in which a wildcard type is useful. Suppose that Shape is a class that defines a method public void draw(), and suppose that Shape has subclasses such as Rect and Oval. Suppose that we want a method that can draw all the shapes in a collection of Shapes. We might try:
public static void drawAll(Collection<Shape> shapes) {
for ( Shape s : shapes )
s.draw();
}
This method works fine if we apply it to a variable of type Collection<Shape>, or ArrayList<Shape>, or any other collection class with type parameter Shape. Suppose, however, that you have list of Rects stored in a variable named rectangles of type Collection<Rect>. Since Rects are Shapes, you might expect to be able to call drawAll(rectangles). Unfortunately, this will not work; a collection of Rects is not considered to be a collection of Shapes! The variable rectangles cannot be assigned to the formal parameter shapes. The solution is to replace the type parameter "Shape" in the declaration of shapes with the wildcard type "? extends Shape":
public static void drawAll(Collection<? extends Shape> shapes) {
for ( Shape s : shapes )
s.draw();
}
The wildcard type "? extends Shape" means roughly "any type that is either equal to Shape or that is a subclass of Shape". When the parameter shapes is declared to be of type Collection<? extends Shape>, it becomes possible to call the drawAll method with an actual parameter of type Collection<Rect> since Rect is a subclass of Shape and therefore matches the wildcard "? extends Shape". We could also pass actual parameters to drawAll of type ArrayList<Rect> or Set<Oval> or List<Oval>. And we can still pass variables of type Collection<Shape> or ArrayList<Shape>, since the class Shape itself matches "? extends Shape". We have greatly increased the usefulness of the method by using the wildcard type.
(Although it is not essential, you might be interested in knowing why Java does not allow a collection of Rects to be used as a collection of Shapes, even though every Rect is considered to be a Shape. Consider the rather silly but legal method that adds an oval to a list of shapes:
static void addOval(List<Shape> shapes, Oval oval) {
shapes.add(oval);
}
Suppose that rectangles is of type List<Rect>. It's illegal to call addOval(rectangles,oval), because of the rule that a list of Rects is not a list of Shapes. If we dropped that rule, then addOval(rectangles,oval) would be legal, and it would add an Oval to a list of Rects. This would be bad: Since Oval is not a subclass of Rect, an Oval is not a Rect, and a list of Rects should never be able to contain an Oval. The method call addOval(rectangles,oval) does not make sense and should be illegal, so the rule that a collection of Rects is not a collection of Shapes is a good rule.)
As another example, consider the method addAll() from the interface Collection<T>. In my description of this method in Subsection 10.1.4, I say that for a collection, coll, of type Collection<T>, coll.addAll(coll2) "adds all the objects in coll2 to coll. The parameter, coll2, can be any collection of type Collection<T>. However, it can also be more general. For example, if T is a class and S is a sub-class of T, then coll2 can be of type Collection<S>. This makes sense because any object of type S is automatically of type T and so can legally be added to coll." If you think for a moment, you'll see that what I'm describing here, a little awkwardly, is a use of wildcard types: We don't want to require coll2 to be a collection of of objects of type T; we what to allow collections of any subclass of T. To be more specific, let's look at how a similar addAll() method could be added to the generic Queue class that was defined earlier in this section:
class Queue<T> {
private LinkedList<T> items = new LinkedList<T>();
public void enqueue(T item) {
items.addLast(item);
}
public T dequeue() {
return items.removeFirst();
}
public boolean isEmpty() {
return (items.size() == 0);
}
public void addAll(Collection<? extends T> collection) {
// Add all the items from the collection to the end of the queue
for ( T item : collection )
enqueue(item);
}
}
Here, T is a type parameter in the generic class definition. We are combining wildcard types with generic classes. Inside the generic class definition, "T" is used as if it is a specific, though unknown, type. The wildcard type "? extends T" means some type that extends that specific type. When we create a queue of type Queue<Shape>, "T" refers to "Shape", and the wildcard type "? extends T" in the class definition means "? extends Shape", meaning that the addAll method of the queue can be applied to collections of Rects and Ovals as well as to collections of Shapes.
The for-each loop in the definition of addAll iterates through the collection using a variable, item, of type T. Now, collection can be of type Collection<S>, where S is a subclass of T. Since item is of type T, not S, do we have a problem here? No, no problem. As long as S is a subclass of T, a value of type S can be assigned to a variable of type T. The restriction on the wildcard type makes everything work nicely.
The addAll method adds all the items from a collection to the queue. Suppose that we wanted to do the opposite: Add all the items that are currently on the queue to a given collection. An instance method defined as
public void addAllTo(Collection<T> collection)
would only work for collections whose base type is exactly the same as T. This is too restrictive. We need some sort of wildcard. However, "? extends T" won't work. Suppose we try it:
public void addAllTo(Collection<? extends T> collection) {
// Remove all items currently on the queue and add them to collection
while ( ! isEmpty() ) {
T item = dequeue(); // Remove an item from the queue.
collection.add( item ); // Add it to the collection. ILLEGAL!!
}
}
The problem is that we can't add an item of type T to a collection that might only be able to hold items belonging to some subclass, S, of T. The containment is going in the wrong direction: An item of type T is not necessarily of type S. For example, if we have a queue of type Queue<Shape>, it doesn't make sense to add items from the queue to a collection of type Collection<Rect>, since not every Shape is a Rect. On the other hand, if we have a Queue<Rect>, it would make sense to add items from that queue to a Collection<Shape> or indeed to any collection Collection<S> where S is a superclass of Rect.
To express this type of relationship, we need a new kind of type wildcard: "? super T". This wildcard means, roughly, "either T itself or any class that is a superclass of T." For example, Collection<? super Rect> would match the types Collection<Shape>, ArrayList<Object>, and Set<Rect>. This is what we need for our addAllTo method. With this change, our complete generic queue class becomes:
class Queue<T> {
private LinkedList<T> items = new LinkedList<T>();
public void enqueue(T item) {
items.addLast(item);
}
public T dequeue() {
return items.removeFirst();
}
public boolean isEmpty() {
return (items.size() == 0);
}
public void addAll(Collection<? extends T> collection) {
// Add all the items from the collection to the end of the queue
for ( T item : collection )
enqueue(item);
}
public void addAllTo(Collection<? super T> collection) {
// Remove all items currently on the queue and add them to collection
while ( ! isEmpty() ) {
T item = dequeue(); // Remove an item from the queue.
collection.add( item ); // Add it to the collection.
}
}
}
In a wildcard type such as "? extends T", T can be an interface instead of a class. Note that the term "extends" (not "implements") is used in the wildcard type, even if T is an interface. For example, recall that Runnable is an interface that defines the method public void run(). Runnable objects are usually associated with threads (see Section 8.5). Here is a method that runs all the objects in a collection of Runnables in parallel, by creating a separate thread for each object:
public static runAllInParallel( Collection<? extends Runnable> runnables ) {
for ( Runnable runnable : runnables ) {
Thread runner; // A thread to run the method runnable.run()
runner = new Thread( runnable ); // Create the thread.
runner.start(); // Start the thread running.
}
}
Wildcard types are used only as type parameters in parameterized types, such as Collection<? extends Runnable>. The place where a wildcard type is most likely to occur, by far, is in a formal parameter list, where the wildcard type is used in the declaration of the type of a formal parameter. However, they can also be used in a few other places. For example, they can be used in the type specification in a variable declaration statement.
One final remark: The wildcard type "<?>" is equivalent to "<? extends Object>". That is, it matches any possible type. For example, the removeAll() method in the generic interface Collections<T> is declared as
public boolean removeAll( Collection<?> c ) { ...
This just means that the removeAll method can be applied to any collection of any type of object.
10.5.4 Bounded Types
Wildcard types don't solve all of our problems. They allow us to generalize method definitions so that they can work with collections of objects of various types, rather than just a single type. However, they do not allow us to restrict the types that are allowed as type parameters in a generic class or method definition. Bounded types exist for this purpose.
We start with a small, not very realistic example. Suppose that you would like to create groups of GUI components using a generic class named ComponentGroup. For example, the parameterized type ComponentGroup<JButton> would represent a group of JButtons, while ComponentGroup<JPanel> would represent a group of JPanels. The class will include methods that can be called to apply certain operations to all components in the group at once. For example, there will be a instance method of the form
public void repaintAll() {
.
. // Call the repaint() method of every component in the group.
.
}
The problem is that the repaint() method is defined in any JComponent object, but not for objects of arbitrary type. It wouldn't make sense to allow types such as ComponentGroup<String> or ComponentGroup<Integer>, since Strings and Integers don't have repaint() methods. We need some way to restrict the type parameter T in ComponentGroup<T> so that only JComponent and subclasses of JComponent are allowed as actual type parameters. We can do this by using the bounded type "T extends JComponent" instead of a plain "T" in the definition of the class:
public class ComponentGroup<T extends JComponent> {
private ArrayList<T> components; // For storing the components in this group.
public void repaintAll() {
for ( JComponent c : components )
if (c != null)
c.repaint();
}
public void setAllEnabled( boolean enable ) {
for ( JComponent c : components )
if (c != null)
c.setEnabled(enable);
}
}
public void add( T c ) { // Add a value c, of type T, to the group.
components.add(c);
}
.
. // Additional methods and constructors.
.
}
The restriction "extends JComponent" on T makes it illegal to create the parameterized types ComponentGroup<String> and ComponentGroup<Integer>, since the actual type parameter that replaces "T" is required to be either JComponent itself or a subclass of JComponent. With this restriction, we know -- and, more important, the compiler knows -- that the objects in the group are of type JComponent and the operations c.repaint() and c.setEnabled() are defined for any c in the group.
In general, a bounded type parameter "T extends SomeType" means roughly "a type, T, that is either equal to SomeType or is a subclass of SomeType", and the upshot is that any object of type T is also of type SomeType, and any operation that is defined for objects of type SomeType is defined for objects of type T. The type SomeType doesn't have to be the name of a class. It can be anything that represents an actual type. For example, it can be an interface or even a parameterized type.
Bounded types and wildcard types are clearly related. They are, however, used in very different ways. A bounded type can be used only as a formal type parameter in the definition of a generic method, class, or interface. A wildcard type is used most often to declare the type of a formal parameter in a method and cannot be used as a formal type parameter. One other difference, by the way, is that, in contrast to wildcard types, bounded type parameters can only use "extends", never "super".
Bounded type parameters can be used when declaring generic methods. For example, as an alternative to the generic ComponentGroup class, one could write a free-standing generic static method that can repaint any collection of JComponents as follows:
public static <T extends JComponent> void repaintAll(Collection<T> comps) {
for ( JComponent c : comps )
if (c != null)
c.repaint();
}
Using "<T extends JComponent>" as the formal type parameter means that the method can only be called for collections whose base type is JComponent or some subclass of JComponent. Thus, it is legal to call repaintAll(coll) where coll is of type List<JPanel> but not where coll is of type Set<String>.
Note that we don't really need a generic type parameter in this case. We can write an equivalent method using a wildcard type:
public static void repaintAll(Collection<? extends JComponent> comps) {
for ( JComponent c : comps )
if (c != null)
c.repaint();
}
In this situation, the version that uses the wildcard type is to be preferred, since the implementation is simpler. However, there are some situations where a generic method with a bounded type parameter cannot be rewritten using a wildcard type. Note that a generic type parameter gives a name, such as T, to the unknown type, while a wildcard type does not give a name to the unknown type. The name makes it possible to refer to the unknown type in the body of the method that is being defined. If a generic method definition uses the generic type name more than once or uses it outside the formal parameter list of the method, then the generic type cannot be replaced with a wildcard type.
Let's look at a generic method in which a bounded type parameter is essential. In Subsection 10.2.1, I presented a code segment for inserting a string into a sorted list of strings, in such a way that the modified list is still in sorted order. Here is the same code, but this time in the form of a method definition (and without the comments):
static void sortedInsert(List<String> sortedList, String newItem) {
ListIterator<String> iter = sortedList.listIterator();
while (iter.hasNext()) {
String item = iter.next();
if (newItem.compareTo(item) <= 0) {
iter.previous();
break;
}
}
iter.add(newItem);
}
This method works fine for lists of strings, but it would be nice to have a generic method that can be applied to lists of other types of objects. The problem, of course, is that the code assumes that the compareTo() method is defined for objects in the list, so the method can only work for lists of objects that implement the Comparable interface. We can't simply use a wildcard type. Suppose we try to do it, by replacing List<String> with List<? extends Comparable>:
static void sortedInsert(List<? extends Comparable> sortedList, ???? newItem) { ListIterator<????> iter = stringList.listIterator(); ...
We immediately run into a problem, because we have no name for the unknown type represented by the wildcard. We need a name for that type because the type of newItem and of iter should be the same as the type of the items in the list. The problem is solved if we write a generic method with a bounded type parameter, since then we have a name for the unknown type, and we can write a valid generic method:
static <T extends Comparable> void sortedInsert(List<T> sortedList, T newItem) {
ListIterator<T> iter = sortedList.listIterator();
while (iter.hasNext()) {
T item = iter.next();
if (newItem.compareTo(item) <= 0) {
iter.previous();
break;
}
}
iter.add(newItem);
}
There is still one technicality to cover in this example. Comparable is itself a parameterized type, but I have used it here without a type parameter. This is legal but the compiler might give you a warning about using a "raw type." In fact, the objects in the list should implement the parameterized interface Comparable<T>, since they are being compared to items of type T. This just means that instead of using Comparable as the type bound, we should use Comparable<T>:
static <T extends Comparable<T>> void sortedInsert(List<T> sortedList, ...
With this example, I will leave the topic of generic types and generic programming. In this chapter, I have occasionally used terms such as "strange" and "weird" to talk about generic programming in Java. I will confess that I have some affection for the more simple-minded generic programming style of Smalltalk. Nevertheless, I recognize the power and increased robustness of generics in Java. I hope that I have convinced you that using the Java Collection Framework is reasonably natural and straightforward, and that using it can save you a lot of time and effort compared to repeatedly recoding the same data structures and algorithms from scratch. Things become a little more technical when you start writing new generic classes and methods of your own, and the syntax is (as I've said) a little strange. But with some practice, you'll get used to the syntax and will find that it's not that difficult after all.