Design Hints for Inheritance

We want to end this chapter with some hints that we have found useful when using inheritance.

1. Place common operations and fields in the superclass.

   This is why we put the name field into the Person class rather than replicating it in the Employee and Student classes.
2. Don’t use protected fields.

   Some programmers think it is a good idea to define most instance fields asprotected, “just in case,” so that subclasses can access these fields if they need to. However, the protected mechanism doesn’t give much protection, for two reasons. First, the set of subclasses is unbounded—anyone can form a subclass of your classes and then write code that directly accessesprotected instance fields, thereby breaking encapsulation. And second, in the Java programming language, all classes in the same package have access to protected fields, whether or not they are subclasses.

   However, protected methods can be useful to indicate methods that are not ready for general use and should be redefined in subclasses. The clonemethod is a good example.
3. Use inheritance to model the “is–a” relationship.

   Inheritance is a handy code-saver, and sometimes people overuse it. For example, suppose we need a Contractor class. Contractors have names and hire dates, but they do not have salaries. Instead, they are paid by the hour, and they do not stay around long enough to get a raise. There is the temptation to form a subclass Contractor from Employee and add anhourlyWage field.

   1 2 3 4

   class Contractor extends Employee {  . . .    private double hourlyWage; }

   This is not a good idea, however, because now each contractor object has both a salary and hourly wage field. It will cause you no end of grief when you implement methods for printing paychecks or tax forms. You will end up writing more code than you would have by not inheriting in the first place.

   The contractor/employee relationship fails the “is–a” test. A contractor is not a special case of an employee.
4. Don’t use inheritance unless all inherited methods make sense.

   Suppose we want to write a Holiday class. Surely every holiday is a day, and days can be expressed as instances of the GregorianCalendar class, so we can use inheritance.

   1	class Holiday extends GregorianCalendar { . . . }

   Unfortunately, the set of holidays is not closed under the inherited operations. One of the public methods of GregorianCalendar is add. And add can turn holidays into nonholidays:

   1 2	Holiday christmas; christmas.add(Calendar.DAY_OF_MONTH, 12);

   Therefore, inheritance is not appropriate in this example.
5. Don’t change the expected behavior when you override a method.

   The substitution principle applies not just to syntax but, more important, to behavior. When you override a method, you should not unreasonably change its behavior. The compiler can’t help you—it cannot check whether your redefinitions make sense. For example, you can “fix” the issue of the addmethod in the Holiday class by redefining add, perhaps to do nothing, or

   1. to throw an exception, or to move on to the next holiday.

      However, such a fix violates the substitution principle. The sequence of statements

      1 2 3 4

      int d1 = x.get(Calendar.DAY_OF_MONTH); x.add(Calendar.DAY_OF_MONTH, 1); int d2 = x.get(Calendar.DAY_OF_MONTH); System.out.println(d2 - d1);

      should have the expected behavior, no matter whether x is of typeGregorianCalendar or Holiday.

      Of course, therein lies the rub. Reasonable and unreasonable people can argue at length what the expected behavior is. For example, some authors argue that the substitution principle requiresManager.equals to ignore the bonus field becauseEmployee.equals ignores it. These discussions are always pointless if they occur in a vacuum. Ultimately, what matters is that you do not circumvent the intent of the original design when you override methods in subclasses.
   2. Use polymorphism, not type information.

      Whenever you find code of the form
      if (x is of type 1)
      action1(x);
      else if (x is of type 2)
      action2(x);

      think polymorphism.

      Do action1 and action2 represent a common concept? If so, make the concept a method of a common superclass or interface of both types. Then, you can simply call

      1	x.action();

      and have the dynamic dispatch mechanism inherent in polymorphism launch the correct action.

      Code using polymorphic methods or interface implementations is much easier to maintain and extend than code that uses multiple type tests.
   3. Don’t overuse reflection.

      The reflection mechanism lets you write programs with amazing generality, by detecting fields and methods at run time. This capability can be extremely useful for systems programming, but it is usually not appropriate in applications. Reflection is fragile—the compiler cannot help you find programming errors. Any errors are found at run time and result in exceptions.