Type Wildcard in Java Generics - A Tutorial

Overview

Type wildcard is the most tricky part of Java Generics. The readers are assumed having the basic knowledge of Java Generics. For a quick review of the basics of Java Generics, see my other post, The Basics of Java Generics. Type wildcard is only used in parameterized types, as the type arguments. A type wildcard is never used in a definition of generic type or generic method . In practice, such parameterized types are commonly used in method definitions, serving as the type of the formal parameters of the methods.

A parameterized type may take one of four forms:

1. List<String>, here String is a concrete type, serving as the type argument for the parameterized type List<String>
2. List<?>, where ? is a type wildcard without a bound, serving as the type argument for the parameterized type List<?> 
3. List<? extends Number>, where ? is a type wildcard with a upper bound, serving as the type argument for the parameterized type List<? extends Number>, and Number is the upper bound of the wildcard.
4. List<? supper Number>, where ? is a type wildcard with a lower bound, serving as the type argument for the parameterized type List<? super Number>, and Number is the lower bound of the wildcard.

For a parameterized type with a type wildcard as a type argument, the wildcard may have either a single upper or lower bound, but not both.

In a World without Type Wildcard

To understand why Java introduces the type wildcard for parameterized type, let’s look at an example. This example is about a circus. In the circus, there are two kinds of animals, birds and dogs. Brids can fly. Dogs can bark. A special kind of birds, nightingares, can also sing. Class hierarchy of animals used in this example are shown in Figure 1. Code for all kinds of animals are shown in Listing 1-1 to 1-4.

Figure 1 - Animal Hierarchy

Listing 1-1 - Animal

package animals;

public class Animal {
    private String name;
    
    public Animal(String name) {
        this.name = name;
    }
    
    public void jump() {
        System.out.println(getName() + " is jumping.");
    }
    
    public String getName() {
        return name;
    }
}

Listing 1-2 -Brid

package animals;

public class Bird extends Animal {    
    public Bird(String name) {
        super(name);
    }
    
    public void fly() {
        System.out.println(getName() + " is flying.");
    }
}

Listing 1-3 -  Nightingale

package animals;

public class Nightingale extends Bird {
    public Nightingale(String name) {
        super(name);
    }
    
    public void sing() {
        System.out.println(getName() + " is singing.");
    }
}

Listing 1-4 - Dog

package animals;

public class Dog extends Animal {   
    public Dog(String name) {
        super(name);
    }
    
    public void bark() {
        System.out.println(getName() + " is barking.");
    }
}

There is a kind of actors in the circus called AnimalTrainer. When given a collection of animals, the AnimalTrainer commands them to jump. The first version of AnimalTrainer is shown in Listing 2-1.

Listing 2-1 - AnimalTrainer Version 1 - Without Type Wildcard

package nowildcard;
import java.util.List;
import animals.Animal;

public class AnimalTrainer {
    public void act(List<Animal> animalList) {
        for (Animal animal : animalList) {
            animal.jump();
        }
    }
}

The first version of the main class, Circus, is shown in Listing 2-2. When the Circus program is launched, it will at first create an AnimalTrainer. Then a list of Animals is created, and two animals are added into the list. It finally invokes the act method, passing the animal list, on the AnimalTrainer. The AnimalTrainer acts by command every animal to jump. So far so good.

Listing 2-2 - Circus Version 1

package nowildcard;
import java.util.ArrayList;
import java.util.List;
import animals.Animal;
import animals.Bird;
import animals.Dog;

public class Circus {
    public static void main(String[] args) {
        AnimalTrainer animalTrainer = new AnimalTrainer();
        
        List<Animal> animalList = new ArrayList<Animal>();
        animalList.add(new Dog("Bob"));
        animalList.add(new Dog("Amy"));
                
        animalTrainer.act(animalList);
    }
}

The second version of the main class, Circus, is shown in Listing 2-3. The AnimalTrainer commands a list of birds to jump. We reason that it will be fine since a bird is actually an animal and can jump. This version of the Circus class, however, does not compile. The line with trouble is  animalTrainer.act(birdList); (Line 15) The error message is:

The method act(List<Animal>) in the type AnimalTrainer is not applicable for the arguments (List<Bird>)

Listing 2-3 - Circus Version 2

package nowildcard;
import java.util.ArrayList;
import java.util.List;

import animals.Bird;

public class Circus {
    public static void main(String[] args) {
        AnimalTrainer animalTrainer = new AnimalTrainer();
        
        List<Bird> birdList = new ArrayList<Bird>();
        birdList.add(new Bird("Tim"));
        birdList.add(new Bird("Nancy"));

        animalTrainer.act(birdList);
    }
}

In fact, Java does not regard List<Bird> as a subtype of List<Animal>. So the act method only accepts List<Animal> as argument, rejecting List<Bird>. You might wonder why Java has such an “unreasonable” rule. Actually, this rule exists for a very good reason. Let’s see why.


What if List<Bird> Were Regarded as a Subtype of List<Animal>?

If Java regarded List<Bird> as a subtype of List<Animal>, we were going to have trouble. To show the trouble, let’s update the Circus class as in Listing 3-3. At first lets look at the two new classes newly added, Magician as in Listing 3-1, and BirdTrainer as in Listing 3-2. The act method of Magician takes a list of Animals, and replaces all Animals in the list by two Dogs. The Magician class compiles perfectly. The act method of BirdTrainer takes a list of Birds and commands each of them to fly. The BirdTrainer class as well compiles perfectly. The Circus class (Version 3, Listing 3-3) does not compile for the same reason as in the Version 2 (Listing 2-3). Line 16 fails. The error message is:
The method act(List<Animal>) in the type Magician is not applicable for the arguments (List<Bird>)  

The Java compiler rejects Line 16 for a good reason. If Java regarded List<Bird> as a subtype of List<Animal> and allowed Line 16 compile, we were going to have run time exception because at run time, when the BirdTrainer got a list, the list actually would contain two Dogs. The Dogs would be commanded to fly but they cannot (a Dog object does not have a fly method). In other words, the Java type integrity would be broken.

To prevent trouble of this nature, Java does not regard List<Bird> as a subtype of List<Animal>. In a generic term of generics, it is said that the type parameter T in List<T> is non-variant.

Listing 3-1 - Magician Version 1

package whatif;
import java.util.List;

import animals.Animal;
import animals.Dog;

public class Magician {
    public void act(List<Animal> animalList) {
        for (int i = 0; i < animalList.size(); i++) {
            animalList.remove(i);
        }
        
        animalList.add(new Dog("Green"));
        animalList.add(new Dog("Red"));        
    }
}

Listing 3-2 - BirdTrainer Version 1

package whatif;
import java.util.List;

import animals.Bird;

public class BirdTrainer {
    public void act(List<Bird> birdList) {
        for (Bird bird : birdList) {
            bird.fly();
        }
    }
}

Listing 3-3 - Circus Version 3

package whatif;
import java.util.ArrayList;
import java.util.List;

import animals.Bird;

public class Circus {
    public static void main(String[] args) {
        BirdTrainer birdTrainer = new BirdTrainer();
        Magician magician = new Magician();
        
        List<Bird> birdList = new ArrayList<Bird>();
        birdList.add(new Bird("Tim"));
        birdList.add(new Bird("Nancy"));
        
        magician.act(birdList);    // this line won't compile.
        
        birdTrainer.act(birdList);
    }
}

Type Wildcard with a Upper Bound

Now since we understand why List<Bird> is not regarded as a subtype of List<Animal>, we can come back to seek a solution for our situation: How to enable an AnimalTrainer to accept a List<Bird> as well as a List<Animal>. Java’s solution is type wildcard. Specifically, we can change the AnimalTrainer class to be like in Listing 4-1 (Line 7: act(List<? extends Animal> animalList)). The ? in Line 7 is called a type wildcard, and ? extends Animal says Animal is the upper bound of the type wildcard.
With this version of AnimalTrainer, the main class, Circus, as in Listing 4-2 (it is essentially the same as in Listing 2-3) compiles and runs successfully.

Listing 4-1 - AnimalTrainer Version 2, With Type Wildcard With Upper Bound

package upperbound;
import java.util.List;

import animals.Animal;

public class AnimalTrainer {
    public void act(List<? extends Animal> animalList) {
        for (Animal animal : animalList) {
            animal.jump();
        }
    }
}

Listing 4-2 - Circus Version 4

package upperbound;
import java.util.ArrayList;
import java.util.List;

import animals.Bird;

public class Circus {
    public static void main(String[] args) {
        AnimalTrainer animalTrainer = new AnimalTrainer();
        
        List<Bird> birdList = new ArrayList<Bird>();
        birdList.add(new Bird("Tim"));
        birdList.add(new Bird("Nancy"));
        
        animalTrainer.act(birdList);
    }
}

In general, there are three rules about type wildcard with a upper bound, given a generic type G (e.g. List<E>), and two concrete types X and Y where Y is a subtype of X (e.g. Bird is a subtype of Animal):

1. G<? extends Y> is subtype of G<? extends X> (e.g. List<? extends Bird> is a subtype of List<? extends Animal>
2. G<X> is subtype of G<? extends X> (e.g. List<Bird> is a subtype of List<? extends Bird>)
3. G<?> is just a shorthand for G<? extends Object> (e.g. List<?> is a shorthand for List<? extends Object>)

(For the formal specification, see 4.5.1.1 Type Argument Containment and Equivalence, The Java Language Specification, Third Edition)



(Those rules have something to do with the idea of covariance in the theory of generics. They however do not align very well with the theory.)


In the context of our example, Bird is a subtype of Animal, so List<? extends Bird> is a subtype of List<? extends Animal>. Meanwhile, List<Bird> is a subtype of List<? extends Bird>. Therefore List<Bird> is a subtype of List<? extends Animal>. That 
is the reason why the act method has a formal parameter of the type List<? extends Animal> and can accept List<Bird> and List<Animal> as argument.

You might wonder whether the Magician class can also be updated with type wildcard and bring back the trouble of messing up birds by dogs. It cannot. The catch is that when the act method of the Magician is updated to take a parameter of the type List<? extends Animal>, as in Listing 4-3 (Line 8), it is no longer allowed to call the add method of the List class. The compile error message is "The method add(capture#3-of ? extends Animal) in the type List<capture#3-of ? extends Animal> is not applicable for the arguments (Dog)". Therefore, it cannot add dogs into the list.  It fails because the add method here expects an argument of the type “? extends Animal” but Dog is not a subtype of it (Note: even though List<Dog> is a subtype of List<? extends Animal>, Dog is not a subtype of "? Extends Animal". More generally, no class is regard as a subtype of “? Extends Animal”.  It is said that “? Extends Animal” is an undefined type (Note: meanwhile, List<? Extends Animal> is a well defined parameterized type.)

Listing 4-3 Magician Version 2

package upperbound;
import java.util.List;

import animals.Animal;
import animals.Dog;

public class Magician {
    public void act(List<? extends Animal> animalList) {
        for (int i = 0; i < animalList.size(); i++) {
            animalList.remove(i);
        }
        
        animalList.add(new Dog("Green")); // now these two lines do not compile
        animalList.add(new Dog("Red"));    
    }
}

Based on my experience, the general rule is like this: When a method with a parameter of a parameterized type that is one with type wildcard with a upper bound (e.g. act(List<? extends Animal>)), inside the body of the method, it is not allowed to call any method on the parameter object unless the method is parameterless (e.g. add(E element) on List<E>). I tried to find something about it in the Java Language Specification but found nothing. However, it seems that all Java compilers that I saw are implemented in this way. I could not figure out how to derive this rule from other more basic rules in the Java Specification.

Type Wildcard with a Lower Bound

Similar to type wildcard with a upper bound, a type wildcard may have a lower bound. (However, a type wildcard cannot have both upper and lower bound, nor have more than one upper or lower bounds). The syntax is like this: List<? super Bird> where Bird is the lower bound of the type wildcard.

In general, there are also three rules about type wildcard with a lower bound, given a generic type G (e.g. List<E>), and two concrete types X and Y where X is a subtype of Y (e.g. Bird is a subtype of Animal):

1. G<? super Y> is a subtype of G<? super X> (e.g. List<? super Animal> is a subtype of List<? super Bird>
2. G<X> is subtype of G<? super X> (e.g. List<Bird> is a subtype of List<? super Bird>)

(Again, for the formal specification, see 4.5.1.1 Type Argument Containment and Equivalence, The Java Language Specification, Third Edition)

(Those rules have something to do with the idea of contra-variance in the theory of generics. They however do not align very well with the theory.)

Let's illustrate the uses of type wildcard with a lower bound with another version of our Circus program as in Listing 5-1. In this version of the Circus class, an animalList (of type List<Animal>) and a birdList (of type List<Bird>) are created in the main method. Instead of adding animals or birds to the lists in the main method, the animalList is passed to the addBirds method of a BirdKeeper object, and the addDogs method of a DogKeeper object, to add birds and dogs into it. Then an AnimalTrainer acts on the animalList by commanding the animals (birds and dogs) in the list to jump. Also, the birdList is passed to the addBirds method of the BirdKeeper object to add birds to it.  And a BirdTrainer acts on the list by commanding the birds in the birdList to fly. The pertinent version of AnimalTrainer and BirdTrainer are shown in Listing 5-2 and 5-3.

Notice the following points in the above example. For the animalList to contain both birds and dogs, it must be defined as of type List<Animal>, instead of List<Bird> or List<Dog>. For the BirdTrainer to command birds in the birdList to fly, the birdList must be defined as of type List<Bird>, instead of List<Animal>.

For the program to work, the addBirds method of the BirdKeeper class must be able to:

1. accept a List<Animal> as argument (Line 21, the Circus class)
2. accept a List<Bird> as argument (Line 26, the Circus class)
3. add birds into the List<Animal> and the List<Bird> (Line 7-8, the BirdKeeper class)

The solution is shown in Listing 5-4. The type of the formal parameter to the addBirds method is List<? super Bird>. The parameterized type that servers as the type of the formal parameter is a type wildcard with a lower bound. Other choices won't work. If we had the method header as public void addBirds(List<Animal> animalList), the method would not accept List<Bird> as argument; if we If we had the method header as public void addBirds(List<Bird> animalList), the method would not accept List<Animal> as argument; If we had the method header as public void addBirds(List<? extends Animal> animalList), the method would not be able to add anything to the list (i.e. animalList). A type wildcard with a lower bound is the necessary solution to this case.  

Notice that it is allowed to call a method with parameters (e.g the add method of List<E>) on an object referenced by a formal parameter of type wildcard with a lower bound (e.g. the parameter named animalList in the addBirds method of the BirdKeeper class), while it is not allowed if the wildcard has a upper bound.

Similarly, the DogKeeper class is shown in Listing 5-5.

Listing 5-1 - Circus Version 5

package lowerbound;
import java.util.ArrayList;
import java.util.List;

import animals.Animal;
import animals.Bird;

public class Circus {
    public static void main(String[] args) {
        AnimalTrainer animalTrainer = new AnimalTrainer();
        BirdTrainer birdTrainer = new BirdTrainer();
        
        List<Animal> animalList = new ArrayList<Animal>();
        
        BirdKeeper birdKeeper = new BirdKeeper();
        birdKeeper.addBirds(animalList);
               
        DogKeeper dogKeeper = new DogKeeper();        
        dogKeeper.addDogs(animalList);
                
        animalTrainer.act(animalList);
        
        List<Bird> birdList = new ArrayList<Bird>();
        birdKeeper.addBirds(birdList);
        
        birdTrainer.act(birdList);
    }
}

Listing 5-2 - AnimalTrainer Version 3

package lowerbound;
import java.util.List;
import animals.Animal;

public class AnimalTrainer {
    public void act(List<Animal> animalList) {
        for (Animal animal : animalList) {
            animal.jump();
        }
    }
}

Listing 5-3 - BirdTrainer Version 2

package lowerbound;
import java.util.List;
import animals.Bird;

public class BirdTrainer {
    public void act(List<Bird> birdList) {
        for (Bird bird : birdList) {
            bird.fly();
        }
    }
}

Listing 5-4 - BirdKeeper

package lowerbound;
import java.util.List;
import animals.Bird;

public class BirdKeeper {
    public void addBirds(List<? super Bird> animalList) {
        animalList.add(new Bird("Tim"));
        animalList.add(new Bird("Nancy"));
    }
}

Listing 5-5 - DogKeeper

package lowerbound;
import java.util.List;
import animals.Dog;

public class DogKeeper {
    public void addDogs(List<? super Dog> animalList) {
        animalList.add(new Dog("Bob"));
        animalList.add(new Dog("Amy"));
    }
}

Finally, notice that the addBirds method of the BirdKeeper class won't accept an argument of the type List<Nightingale> because List<Nightingale> is not regarded as a subtype of List<? super Bird>. The version of the Circus class as shown in Listing 5-5 won't compile. Line 30, birdKeeper.addBirds(nightingaleList);, will fail. The compile error is:
The method addBirds(List<? super Bird>) in the type BirdKeeper is not applicable for the arguments (List<nightingale>)

Listing 5-5 - Circus Version 6

package lowerbound;
import java.util.ArrayList;
import java.util.List;
import animals.Animal;
import animals.Bird;
import animals.Nightingale;

public class Circus {
    public static void main(String[] args) {
        AnimalTrainer animalTrainer = new AnimalTrainer();
        BirdTrainer birdTrainer = new BirdTrainer();
        
        List<Animal> animalList = new ArrayList<Animal>();
        
        BirdKeeper birdKeeper = new BirdKeeper();
        birdKeeper.addBirds(animalList);
        
        
        DogKeeper dogKeeper = new DogKeeper();        
        dogKeeper.addDogs(animalList);
                
        animalTrainer.act(animalList);
        
        List<Bird> birdList = new ArrayList<Bird>();
        birdKeeper.addBirds(birdList);
        
        birdTrainer.act(birdList);
        
        List<Nightingale> nightingaleList = new ArrayList<Nightingale>();
        birdKeeper.addBirds(nightingaleList); // does not compile
    }
}

Conclusion

1. Type wildcard is used to increase flexibility of methods that take parameters of parameterized types
2. It is not allowed to call a method on an object referenced by a parameter of a parameterized type with type wildcard with a upper bound unless the method is parameter less.
3. It is  allowed to call any method on an object referenced by a parameter of a parameterized type with type wildcard with a lower bound.

References 

• The Java Language Specification, Third Edition, by James Gosling, Bill Joy, Guy Steele, and Gilad Bracha, Addison Wesley Professional, 2005
• The Java Programming Language, 4th Edition, by Ken Arnold, James Gosling, and David Holmes, Prentice Hall PTR, 2005

Sonar Review

Overview

In short, Sonar provides great convenience in source code analysis and unit testing. Sonar works as a facade to various open source tools aiming at code qualities, such as Findbugs, PMD, Checkstyle, Cobertura, Clover, Surefire, etc. The Sonar-Maven integration is excellent. Executing a single Maven goal sonar will run all code analysis and tests, and will store the results in a database.

A dashboard displays summary of various analysis. The summary can be either per project or per Java package. From there one can drill down into various issues with details like where in the source code the issues occur. Often the code with issues is visually marked out in its context displayed on the GUI. For each rule violation, Sonar also provides a brief explanation on why it is an issue. For programmers who do not have any idea about the potential harm of the code with the issue, the explanation is a good starting point to learn.


Features

Rule Violations

This is one of the most useful features of Sonar. It lists the total number of rule violations, and the subtotal numbers in the groups: blocker, critical, major, minor, and info (they are in the descendant order of severity). It also gives the percentage of rule compliance. Below are two bugs caught by Sonar in real world enterprise projects that I saw in person.

Bug 1 – An Infinite Loop

public Table() {
    Table myTable = new Table();
}

Sonar's remark: Correctness - An apparent infinite recursive loop.

The constructor calls itself recursively and forms an infinite loop.




Bug 2 – Mistaking & for &&

if ( books != null & !books.isEmpty() ) {
     return books.toArray(new Book[]{});
} 

Sonar's remark: Correctness - Possible null pointer dereference

The code author intended to prevent NullpointerException but the actually code does not do what he meant. & is actually bit shift rather than logic “and”. The author should have used &&, which is logic “and” instead of &.


Duplications

Sonar is quite good at identifying code duplications. Not just identical code blocks are identified as duplications, but also code blocks similar enough.

The duplication check is very helpful for programmers to minimize code duplications. For an excellent explanation of the harm of code duplications, please see The Evils of Duplication in the must-read The Pragmatic Programmer: From Journeyman to Master by Andrew Hunt and David Thomas.

Complexity

The complexity is measured by the cyclomatic complexity number for the classes and methods. Cyclomatic complexity is a mature and well-accept software quality metric, backed by empirical data. This feature is very helpful to identify hotspots where the code is too complex.

Package tangle index

It is said that there is a package dependency from Package A to Package B if some Java classes in Package A depend on Java classes in Package B. The package tangle index indicates the severity of cyclical package dependency, for example Package A depends on Package B, and meanwhile Package B also depends on Package A.

The layer architecture pattern is the most fundamental architecture pattern. (For an excellent discussion of the layer architecture pattern, see Layer in Pattern-Oriented Software Architecture Volume 1 by Frank Bushchmann and others).  Classes responsible for different layers should be placed in different packages. A cyclical package dependency may indicate that a lower layer  depends on a upper layer, or even worse, the software does not have well defined layers. We may call it anti-layer pattern.

Some cyclical package dependencies can be eliminated by simply moving classes from one package to another.

Comments

In the comments section, sonar shows the percentage of API with javadoc and how many public classes or public methods do not have any companion javadoc. The GUI allows user to drill down into the source code to see which classes/methods are not documented.

Sonar, however, cannot evaluate the quality of the javadoc. It means that one may have 100% API documented but the document may be totally trash.

On the other hand, Sonar cannot tell that certain methods like getters and setters do not really need javadoc.
The analysis score of comments must be taken with a grain of salt.

Code coverage

Sonar provides a summary on the percentage of code and branches covered by unit tests. The GUI also makes it easy to locate code that is not covered by unit tests. However, one must be aware that a 100% code coverage is far from a 100% correctness. 100% coverage only means that 100% code is executed during unit tests. Bugs won't be caught by just executing the code where they reside.

Lack of cohesion of methods (LCOM4) and Response for Class (RFC)

These are two of the six metrics in the Chidamber & Kemerer metrics suite. (LCOM4 is an improved versio. The the original LCOM of the Chidamber & Kemerer metrics suite is usually referred as LCOM1).

Personally, I have not yet seen much value of these two metrics.



Issues

Performance

Sonar is quite slow. Some time it is very slow.

Data Corruption

A bug frequently corrupts data in database and causes sonar execution (analysis and testing) to fail.

Example - Error Message from Failing Sonar Execution (via Maven)

[INFO] ------------------------------------------------------------------------
[ERROR] BUILD ERROR
[INFO] ------------------------------------------------------------------------
[INFO] Can not execute Sonar

Embedded error: PicoLifecycleException: method 'public void org.sonar.batch.ProjectTree.start() throws java.io.IOException', instance 'org.sonar.batch.ProjectTr
ee@a9be37, java.lang.RuntimeException: wrapper
result returns more than one elements
[INFO] ------------------------------------------------------------------------
[INFO] For more information, run Maven with the -e switch
[INFO] ------------------------------------------------------------------------
[INFO] Total time: 1 minute 17 seconds
[INFO] Finished at: Fri Sep 16 10:43:21 EDT 2011
[INFO] Final Memory: 103M/119M
[INFO] ------------------------------------------------------------------------

Sonar expects any two records in the snapshots table with 1 as the value of the islast field to have different project_id. The bug causes more than one record to have the same project_id while the value of the islast field is 1. When it happens, Sonar execution (analysis and testing) will fail.

To resolve the problem, one has to run a certain SQL script in the database to delete the problematic snapshots records. It will only temporarily solve the problem, which will re-occur again.

Documentation

Sonar documentation is pretty good.

1. Sonar documentation at the home site

The Basics of Java Generics

Overview

For the sake of generics, Java types (classes and interfaces) can be grouped into three categories:

• Ordinary type, e.g. String, Integer
• Generic type, e.g. java.lang.Comparable<T>, java.util.List<E> , and java.util.ArrayList<E>
• Parameterized type, e.g. java.lang.Comparable<Integer>, java.util.List<String>, and java.util.ArrayList<String>

In Java, there are four kinds of generic constructs:

• generic interface
• generic class
• generic method
• generic constructor

Constructors are very much like methods, except that there is not any return for constructors. For this reason, we are going to omit any discussion about generic constructors since all discussions about generic methods, except what about method returns, also apply to generic constructors.

Coding with generics usually involves one or more of the following:

• Defining a generic interface, class, or method
• Invoking a generic interface, or class
• Invoking a generic method
• Defining a non-generic method with at least a parameter of a generic type, or with the return of a generic type. Such a method must be a member of a generic type. 
• Defining a method with at least a parameter of a parameterized type, or with the return of a parameterized type
• Invoking a method with at least a parameter of a parameterized type
• Invoking a method with the return of a parameterized type

Defining Generic Types

Example 1 – Defining a generic interface

public interface Iterable<T>

The T in Iterable<T> is called a type parameter. In the language of Java generics, we say that the generic type Iterable takes a type parameter, T. Conventionally, a single upper case T is used as identifier for a type parameter (T stands for type).

Example 2 – Defining a generic interface that extends another generic interface

public interface List<E> extends Collection<E>

Here the E in List<E> is the type parameter. Conventionally, E is used as identifier for type parameter of collections. (E stands for element)

Example 3 – Defining a generic class that implements a generic interface

public class ArrayList<E> implements List<E>


Defining Generic Method

Example  4 – Defining a generic method

<T> T[] toArray(T[] a);

Above is the definition of a toArray method in the body of java.util.List. The first <T> tells that this is a generic method and the method takes a type parameter T. This means that this method has a hole that will be filled later with a concrete type. Then it also tells that the type of the method return is T[],  and the type of the method parameter is a T[] (array of T). Essentially, the type parameter of this method establishes a constrain, in term of type, between the method return and parameters. If we want to turn a list into a String [], we must pass to the toArray method a String[].

Please note that a method whose parameters or return is of a type parameter is not necessarily a generic method unless in its definition  <T>  is placed before its return type (or void).  For example, the methods shown in Example 5 below are not generic methods.

On the other hand, even a non-generic type may have a generic method as its member.


Type Parameter

A type parameter is a placeholder for a concrete type.  It is important to understand that a type parameter is either taken by a generic type,  a generic method, or a generic constructor. On the other hand a generic type or method takes at least one type parameter.

Inside the body of a generic interface or class, a type parameter taken by the interface or class, can server as the type of parameters, the type of return, or the type of local variables, of an instance method. It can also server as the type of instance fields.

A type parameter taken by a generic method can server as type of its parameters, type of its return, or type of its local variables. (Note: It is legal for a generic method to be a static member of a class or interface)

A type parameter taken by a class or interface cannot be:

• Type of its static fields (because there is only one class vs. many different T)
• Anywhere in its static member methods (same reason)
• In a static initial block (same reason)

In addition, none type parameter can be

• used in new T() statement  to create a new object (because erasure)
• used in new T[size]() to create a new array of objects (because erasure)

Example 5 - Defining methods with parameters or return of type parameter

boolean add(E e);
E get(int index);

The above two methods are defined in the body of generic List<E>, the type parameter E servers as the type of parameter named e for the method named add, and the return type of the method named get. These two methods are not generic methods. The type parameter E is not taken by the methods but by their owner type (i.e. List<E>).


Generic Type v.s.  Parameterized Type

It is critical to understand the difference and relationship between generic type and parameterized type. For example, ArrayList<E> is a generic type and  ArrayList<String> is a parameterized type.  They differ in the following aspects:

• E in  ArrayList<E>  is a type parameter, and String  in  ArrayList<String> is an concrete type (particularly, a ordinary class). In regard to  ArrayList<String>, the concrete type String servers as the type argument, to fill the place held by type parameter E, which is taken by  ArrayList<E>.
• A parameterized type, like an ordinary type, is a concrete type, while a generic type is an abstract type
• It is legal to create an object of ArrayList<String> via statement new ArrayList<String>();, statement new ArrayList<E>(); is, however, illegal.
• More generally, the usage of a parameterized type is exactly the same as an ordinary type. A parameterized type can be used at any place where an ordinary type is to be used, i.e. to be used as the type of a variable or a method return. The variable may be a method parameter, a local variable, or a field.

A parameterized type always has a special relationship with a generic type: a parameterized type is always instantiated out of a generic type. For example, ArrayList<String> is instantiated out of ArrayList<E>, by replacing a type argument, String, for the type parameter, E. In order for ArrayList<String> to exist, ArrayList<E> must exist first.

A type parameter is like a hole. When the hole is filled with a concrete type, a parameterized type comes out of the generic type. Replacing a type parameter by a concrete type is called invocation of a generic type. While one can invoke a method passing arguments, one can invoke a generic type passing type arguments.

In a parameterized type a type argument takes all places used to be held by its corresponding type parameter.  For example, in List<String>, there are effectively

boolean add(String e);

String get(int index);

(For the formal specification, see 4.5.2 Members and Constructors of Parameterized Types, The Java Language Specification, Third Edition, Addision Wesley, 2004)


Bounded Type Parameter

In a generic type definition, a type parameter may be given an upper bound.

Example 6 – Defining a generic type named SortedSet which is a set with elements sorted
 
public interface SortedSet<E extends Comparable<E>> extends Set<E>
 
Here <E extends Comparable<E>> indicates that E is a bounded type parameter and Comparable<E> is the upper bound. Any parameterized type out of this generic type must have the type argument as a sub-type of Comparable<E>.  For example, we may have a parameterized type SortedSet<Integer>. It is OK since Integer implements Comparable<Integer>. However, we cannot have a parameterized type SortedSet<java.io.File> because File does not implement Comparable<File>. In short, the bound of a type parameter is used to restrict the type arguments to the generic type. Without a bound, any type will be accepted as legal type argument at compile time. Some of them may lead to runtime exception.
 
A few more words about the example, the upper bound, java.lang.Comparable<E>,  is also a generic type, and its type parameter is E, the same as of SortedSet.

If there are multiple such bounds, separate them by & in the generic type definition.



Calling a Generic Method


Example 7 - Calling a generic method

         List<String> list = new ArrayList<String>();
         list.add("One");
         list.add("Two");
       
         String[] stringArray = list.toArray(new String[]{});
       
         System.out.println(stringArray[0]);
         System.out.println(stringArray[1]);

Usually, it is not required to specifying the type argument (i.e. the concrete type to take the place of the type parameter) when call a generic method, as the example above shows, because the compiler can infer the type argument from the type of the argument to the method (i.e. String[] in the example). That is however, not always the case. In some cases, the compiler cannot determine the concrete type by inference. Then the type argument has to be explicitly specified. The right syntax to specify the type argument to a generic method is show in the example below:

String[] stringArray = list.<String>toArray(new String[]{});
 
The type argument (e.g. String in the example above) is place between < and >, and immediately before the name of the generic method being called.

By the way, be aware that the toArray method does not bring complete type-safety. The following code fragment compiles but causes run time exception.

        List<String> list = new ArrayList<String>();
        list.add("One");
        list.add("Two");
        
        Integer[] intArray = list.toArray(new Integer[]{});

Type Wildcard

Type wildcard in Java Generics is a complex topic. It is discussed in my other post Type Wildcard in Java Generics.

StringTemplate 4 Note for Java Programmers

StringTemplate is a very simple and powerful template engine. The existing documents are fairly complete. As long as one knows where to find those documents, learning StringTemplate is quite easy. This note is intended to help users to quickly find needed documents.

To learn StringTemplate means to learn the following three aspects of it:

1. The core concepts and the relationship among them
2. Syntax of templates and template groups
3. API

StringTemplate 4 Document on the StringTemplate Home Site

In this document, users can find:

• A brief introduction to StringTemplate 4
• Instructions to setup Java programs to use StringTemplate 4
• Syntax of templates and groups  
• StringTemplate 4 Java API

The syntax documentation is quite formal and is in (a variation of?) the Backus-Naur Form notation. Programmers who are not used to such formal notation may feel that the syntax documentation is hard to understand. For those programmers, my recommendation is to spend one hour to learn the Backus-Naur Form notation. Of course, examples also help.

Enforcing Strict Model-View Separation in Template Engines

This article from the creator of StringTemplate provides:

• An overview of StringTemplate
• The philosophy of StringTemplate
• The theoretical foundation of StringTemplate
• A few examples showing the main features of StringTemplate, namely:
• attribute (and attribute property) reference
• map operation (i.e. applying a template to an attribute that is a list of objects or applying a list of templates alternatively to an attribute that is a list of objects), 
• conditional include
• recursive template.

Many template users say that certain other template engines are more powerful than StringTemplate. This article helps users to understand why those features existing in other template engines are purposely excluded from StringTemplate for very good reasons.

A Functional Language for Generating Structured Text 

In this article by the creator of StringTemplate, the author provides a comprehensive introduction, explaining the major concepts in StringTemplate with examples:

• Template
• Template group
• Expression
• Attribute
• Multi-valued attribute
• Implicitly set attribute
• Template include
• Conditional include
• Template application, to a single or multiple attributes
• Anonymous inline template
• Recursive template
• Group inheritance and overriding
• Template region
• Group interface
• Map (dictionary) and list
• Renderer

This is the deepest document about core concepts in StringTemplate. Consider this article as the must-read in order to really mater StringTemplate. The syntax of template is up to date. However,  the Java API has been changed since the publication of this article. Therefore the Java code examples in this article are out of date.

Classpath and Resource Files in Java Programs

(Last updated on February 2, 2013)

Frequently a Java program needs to read some resource files in the file system. Such a resource file may be a .properties file or a .xml file for program configuration. Often it is not practical to hardcode the full path to such a file in the program because if we do so, we will not be able to execute the program correctly except from a specific location in the file system. That is highly undesirable. 

A popular practice is to place such a resource file on a classpath and code the program to search the classpath for the resource file. A programmer who adopts this practice must understand well what a classpath is and how to discover it programmatically.

To understand classpath, one must at first understand the concept of class loader. According to the Java API document, “A class loader is an object that is responsible for loading classes”. In general, a Java program uses multiple class loaders, instead of a single one, to load classes. A classpath is the search path of a class loader. In other words, a Java program usually has multiple classpaths. (It is inappropriate to talk about the classpath of a Java program because a Java program has multiple classpaths. On the other hand, it is all right to talk about the classpath of a class loader.) The fact is that some of those classpaths can be discovered programmatically, some simply cannot.

For a typically Java program, the class loaders form a hierarchy.  When a class loader is requested to find a class or a resource file, it will at first recursively delegate the request to its parent class loader. Only when its parent class loader cannot find the class or resource file, it will search it on its own search path.

Class Loader Hierarchy

On the top of the class loader hierarchy is the JVM’s built-in bootstrap class loader, which loads standard JDK classes. The search path of the bootstrap class loader can be found programmatically via a call to System.getProperty("sun.boot.class.path"). The search path is platform specific. On a Windows machine, it is <JAVA-HOME>/jre/lib.  Typical jar files on this search path are rt.jar, jsse.jar, jce.jar etc.

As the child of the bootstrap class loader is the extension class loader, which loads JDK extension classes. The search path of the extension class loader can be found programmatically via a call to System.getProperty("java.ext.dirs").  The path is platform specific. On a Windows machine, it is <JAVA-HOME>/jre/lib/ext.  An example of such JDK extension is sunjce_provider.jar. 

As the child of the extension class loader is the system class loader, which loads classes on the path specified by OS environment variable CLASSPATH or the –classpath option to the JVM. The search path of the system class loader can be found programmatically via a call to System.getProperty("java.class.path").

If the Java program does not create its own user class loader, all non-JDK-standard/extension classes will be loaded by the system class loader. As we just said, the search path of this class loader can be easily found programmatically.

If the Java program creates its own user class loaders, unless the class loaders are of a custom class loader class with a method to retrieve the search path, there is no way to find the search path programmatically. If the user classes are executed in a JEE container, the JEE container is the bootstrap program and it always creates user class loaders to load user classes (usually one for each war or ear). Similarly, Maven always load plugin classes with user class loaders. Therefore,  if the user classes are executed as a Maven plugin, don’t expect to find the search path for those classes by calling System.getProperty("java.class.path"). It is worth to notice that many applications and application servers actually use instances of the java.net.URLClassLoader as their user class loaders. In such cases, the classpath of the class loaders can be found by call the getURLs() method on the classloader instances (after casting them from java.lang.ClassLoader to java.net.URLClassLoader). The getURLs() method returns an array or URLs. Each URL returned is a directory on the classpath. In other words, the class loader will search those directories to find the classes wanted.

No matter the search path can be found programmatically or not, a program can always asks a class loader to find a resource file by calling the getResource(String name) method on the classloader object (or to get an InputStream connected to the resource file by calling the getResourceAsStream(String name) method on the classloader object). It will found the resource file if it is on the search path or in a jar file on the search path. By the way, the getSystemResource(String name) and getSystemResourceAsStream(String name) methods are to find the resource on the search path of the system class loader.

By the way, given any object, one can call the getClass() method on it to find the Class object representing its class. Then one can find the class loader by calling the getClassLoader() method on the Class object. In short, obj.getClass().getClassLoader() will return the class loader that loaded the class of obj, an object.

It is usually preferable to use the getResource(String name) method rather than the getResourceAsStream(String name) method for the reason of logging. The returned URL can be logged. In case there are inadvertently multiple resource files with the same name on different locations on the search path, the logged URL can help to debug. Even if there is only one resource file with the name, if there is some difficult to read from or write to it, the logged URL can point the programmer quickly to the file that needs a fix.

Type-Safety in Common Degister 3

Commons Digester is a Java library to parse XML data into Java objects. Calling the parse method on a Digester object with an file containing the XML data as argument returns the object on the top of the object tree corresponding to the XML data. The XML data may be arbitrary, so does the the type of the returned object. For this obvious reason, the parse method return type, in Digester 1 and 2, is Object (java.lang.Object). Clients of the library have to cast the returned objects into whatever types they actually are. This approach lacks type-safety. At compile time, no incorrect casting may be caught. A incorrect casting will throw a ClassCastException at runtime.

After generic was introduced into Java in Java 5, many libraries were improved with generic to increase type-safety. There were such efforts in Commons Digester 3, too. In Digester 3, a parse method of the Digester class is a generic method. Each parse method has a type parameter (<T>) and the return type is the abstract type T instead of Object. When the return is assigned to a variable, Java compiler can infer the type for the return. Class casting is no longer needed in the client code. This may give an impression that the risk of runtime ClassCastException is prevented. That is really a false insurance. In this case, what the generic method provides is just what is called auto-casting, that is, instead of the client code, the service code does the casting, with help of generic. All the benefit here is the convenience that the client does not need casting anymore. No risk of casting is prevented or reduced.

It is worth to point out that the Digester class is a non-generic class while the parse methods are generic methods. If an abstract type is at the same time of the type of more than one parameters and return of a generic method, the compiler can check the arguments and the variable to assign return to infer the concrete type and ensure the arguments and the return are assignable to the concrete type inferred. That increases type-safety comparing with using Object as the type for method parameters and return. It is not the case for the parse methods of the Digester class. In the case of the parse methods of the Digester class. The abstract type is only the type of the return. There is no way for the compiler to infer the right concrete type.

Object Attributes and States, From Modeling to Programming

Regarding things as objects with attributes is a powerful thinking technique,  learned by some of us as early as in elementary school. As an example, in the popular logic game for children, Zoombinis Logical Journey, every zoombini has four attributes. (In the game, the attributes are visually identified without a text name. For our convenience of discussion, we just name them hair, eyes, nose, and feet.)

Zoombini Attributes

An attribute has a value. Legitimate values for a specific attribute may be elements of a finite or infinite set (the domain of the attribute). In the Zoombinis example, each attribute may have one out of five possible values. An object's state is the combination of its attribute values. In the zoombinis example, the particular zoombini in the picture above is in a state that can be described as

A Zoombini State

All possible states of an attribute together form the state space of the attribute. All possible states of all attributes of a object form the state space of the object. All possible states of all objects of a software system form state space of the software system.

An object may be mutable. In that case, its state can be changed. An object may be immutable. In that case, its state cannot be changed.

UML, as the claimed universal modeling language, directly supports the object/attributes thinking techniques. In UML, an object may have many attributes. In UML, we may model zoombinis as objects of a Zoombini class:

Zoombini Class in UML

 Unfortunately, UML does not offer a notation to distinct mutable and immutable attributes.

When we start coding zoombinis in a programming language, we often encounter a problem - most popular programming languages do not have the concept of attributes. For example, in Java, a object may have member fields and member methods, but not attributes. Because attribute is such a powerful concept in our rational thinking, we are unwilling to give it up so easily. What we can do is what Steve McConnell called "programming into a language". Actually, the JavaBean property concept is the same as object attribute. So we can code an object attribute as an object property in Java. For an immutable attribute, just omit the mutator method of the property. However, we can also code an object's mutable attribute as a public field of the object.