09-07-2012, 11:28 AM
Limitations and Alternatives
When using inheritance extensively in designing a program, one should note certain constraints that it imposes.
For example, consider a class Person that contains a person's name, address, phone number, age, gender, and race. We can define a subclass of Person called Student that contains the person's grade point average and classes taken, and another subclass of Person called Employee that contains the person's job-title, employer, and salary.
In defining this inheritance hierarchy we have already defined certain restrictions, not all of which are desirable:
Design Constraints
• Singleness:
using single inheritance, a subclass can inherit from only one superclass. Continuing the example given above, Person can be either a Student or an Employee, but not both. Using multiple inheritance partially solves this problem, as one can then define a StudentEmployee class that inherits from both Student and Employee. However, it can still inherit from each superclass only once; this scheme does not support cases in which a student has two jobs or attends two institutions.
• Static:
the inheritance hierarchy of an object is fixed at instantiation when the object's type is selected and does not change with time. For example, the inheritance graph does not allow a Student object to become a Employee object while retaining the state of its Person superclass. (Although similar behavior can be achieved with the decorator pattern.) Some have criticized inheritance, contending that it locks developers into their original design standards.[11]
• Visibility:
whenever client code has access to an object, it generally has access to all the object's superclass data. Even if the superclass has not been declared public, the client can still cast the object to its superclass type. For example, there is no way to give a function a pointer to a Student's grade point average and transcript without also giving that function access to all of the personal data stored in the student's Person superclass. Many modern languages, including C++ and Java, provide a "protected" access modifier that allows subclasses to access the data, without allowing any code outside the chain of inheritance to access it. This largely mitigates this issue.
The Composite reuse principle is an alternative to inheritance. This technique supports polymorphism and code reuse by separating behaviors from the primary class hierarchy and including specific behavior classes as required in any business domain class. This approach avoids the static nature of a class hierarchy by allowing behavior modifications at run time and allows a single class to implement behaviors buffet-style, instead of being restricted to the behaviors of its ancestor classes.
Roles and Inheritance
Sometimes inheritance-based design is used instead of roles. A role, say Student role of a Person describes a characteristic associated to the object that is present because the object happens to participate in some relationship with another object (say the person in student role -has enrolled- to the classes). Some object-oriented design methods do not distinguish this use of roles from more stable aspects of objects. Thus there is a tendency to use inheritance to model roles, say you would have a Student role of a Person modelled as a subclass of a Person. However, neither the inheritance hierarchy nor the types of the objects can change with time. Therefore, modelling roles as subclasses can cause the roles to be fixed on creation, say a Person cannot then easily change his role from Student to Employee when the circumstances change. From modelling point of view, such restrictions are often not desirable, because this causes artificial restrictions on future extensibility of the object system, which will make future changes harder to implement, because existing design needs to be updated. Inheritance is often better used with a generalization mindset, such that common aspects of instantiable classes are factored to superclasses; say having a common superclass 'LegalEntity' for both Person and Company classes for all the common aspects of both. The distinction between role based design and inheritance based design can be made based on the stability of the aspect. Role based design should be used when it's conceivable that the same object participates in different roles at different times, and inheritance based design should be used when the common aspects of multiple classes (not objects!) are factored as superclasses, and do not change with time.
One consequence of separation of roles and superclasses is that this cleanly separates compile-time and run-time aspects of the object system. Inheritance is then clearly a compile-time construct. it does influence the structure of many objects at run-time, but the different kinds of structure that can be used are already fixed at compile-time.
To model the example of Person as an employee with this method, the modelling ensures that a Person class can only contain operations or data that are common to every Person instance regardless of where they are used. This would prevent use of a Job member in a Person class, because every person does not have a job, or at least it is not known that the Person class is only used to model Person instances that have a job. Instead, object-oriented design would consider some subset of all person objects to be in an "employee" role. The job information would be associated only to objects that have the employee role. Object-oriented design would also model the "job" as a role, since a job can be restricted in time, and therefore is not a stable basis for modelling a class.
The corresponding stable concept is either "WorkPlace" or just "Work" depending on which concept is meant. Thus, from object-oriented design point of view, there would be a "Person" class and a "WorkPlace" class, which are related by a many-to-many associatation "works-in", such that an instance of a Person is in employee role, when he works-in a job, where a job is a role of his work place in the situation when the employee works in it.
Note that in this approach, all classes that are produced by this design process form part of the same domain, that is, they describe things clearly using just one terminology. This is often not true for other approaches.
The difference between roles and classes is especially difficult to understand, if one assumes referential transparency, because roles are types of references and classes are types of the referred-to objects.