Object-oriented design is like an alloy consisting of a solid grounding in the object-oriented (OO) approach and implementing the best OO practices heavily laced with how to sidestep the OO pitfalls. The design process isn't only a matter of applying basic OO principles. One should go further and achieve a reasonable tradeoff between OO design principles and applicable design patterns. This article will discuss the relationship between three sources of a solid OO design and offer a starting point to understanding what is, in effect, a complex process. It will also serve as a review for experienced designers.

The OO Design Pyramid

Once people get a sense of the basic principles, they start feeling the power of the OO approach. In our experience this is a dramatic moment for every OO designer and developer. The euphoria of possessing a powerful methodology can make them feel completely self-sufficient and they stop and don't try for perfection. It takes time to realize that there are significant issues beyond the basics that can be found in the accumulated wisdom of the OO developer community and are expressed in a set of design heuristics and scientifically proven OO design guidelines. Some OO designers jump to using design patterns everywhere like a cookie recipe. Although design patterns are important and useful tools in building OO systems, they're just templates that define a common design language and significantly improve design quality, when properly used.

To build contemporary real-world enterprise-class systems, OO designers should (a) be proficient in basic OO principles, (b) master the principles of OO design, and (c) understand design patterns. Otherwise the chance of design flaws and total cost of system ownership increases. And, with growing adoption of Service Oriented Architecture (SOA), the robustness of the SOA services can depend directly on the solidity of the OO design of underlying components.

The pyramid in Figure 1 represents three sources of a solid OO design. It also defines a systematic view of design and uncovers dependencies between these sources.

Now let's discuss each level of this pyramid from an architectural and design point of view. Of course there's no way to explain these topics completely in such short article, but we'll try to highlight the necessity of paying attention to the dependencies and their impact on the quality of OO design.

Basic OO principles

The most basic OO principles include encapsulation, inheritance, and polymorphism. Along with abstraction, association, aggregation, and composition, they form the foundation of the OO approach. These basic principles rest on a concept of objects that depicts real-world entities such as, say, books, customers, invoices, or birds. These classes can be considered templates for object instantiation, or object types. In other words, class specifies what an object can do (behavior) and defines patterns for its data (state). Modeling the real world in terms of an object's state and behavior is the goal of the OO approach. In this case, state is represented by a set of object attributes, or data, and behavior is represented by the object's methods.

Now we'll discuss the application of basic OO principles to modeling.

Encapsulation encloses data and behavior in a programming module. Encapsulation is represented by the two close principles of information hiding and implementation hiding. Information hiding restricts access to the object data using a clearly defined "interface" that hides the internal details of the object's structure. For each restricted class variable, this interface appears as a pair of "get" and "set" methods that define read-and-write access to the variable. Implementation hiding defines the access restrictions to the object methods also using a clearly defined interface that hides the internal details of object implementation and exposes only the methods that comprise object behavior. Both information and implementation hiding serve the main goal - assuring the highest level of decoupling between classes.

Inheritance is a relationship that defines one entity in terms of another. It designates the ability to create new classes (types) that contain all the methods and properties of another class plus additional methods and properties. Inheritance combines interface inheritance and implementation inheritance. In this case, interface inheritance describes a new interface in terms of one or more existing interfaces, while implementation inheritance defines a new implementation in terms of one or more existing implementations. Both interface inheritance and implementation inheritance are used to extend the behavior of a base entity.

Polymorphism is the ability of different objects to respond differently to the same message. Polymorphism lets a client make the same request of different objects and assume that the operation will be appropriate to each class of object. There are two kinds of polymorphism - inheritance polymorphism, which works on an inheritance chain, and operational polymorphism, which specifies similar operations for non-related out-of-inheritance classes or interfaces. Because inheritance polymorphism lets a subclass (subtype) override the operation that it inherits from its superclass (supertype), it creates a way to diversify the behavior of inherited objects in an inheritance chain, while keeping their parent-objects intact. Polymorphism is closely related to inheritance as well as to encapsulation.

Naturally, some OO principles are controversial in the sense that they contradict one another. For example, to be able to inherit from a class, one should know the internal structure of that class, while encapsulation's goal is exactly the opposite - it tries to hide as much of the class structure as possible. In real life the tradeoff between these two principles is a fine line that can't be established without stepping up to the next level in the design pyramid.

OO Design Principles and Heuristics

There are about a dozen OO design principles and four dozens OO design heuristics identified over the years by OO evangelists such as Grady Booch, Bertrand Meyer, Robert C. Martin, Barbara Liskov, and others. OO design principles define the most common scientifically derived approaches for building robust and flexible systems. These approaches proved to be the best tools in solving numerous OO design issues that can't be captured by fundamental OO principles.

The class structure and relationships group consists of the following design principles: the Single Responsibility Principle (SRP), the Open/Closed Principle (OCP), the Liskov Substitution Principle (LSP), the Dependency Inversion Principle (DIP), and the Interface Segregation Principle (ISP).

The Single Responsibility Principle specifies that class should have only one reason to change. It's also known as the cohesion principle and dictates that class should have only one responsibility, i.e., it should avoid putting together responsibilities that change for different reasons.

The Open/Closed Principle dictates that software entities should be open to extension but closed to modification. Modules should be written so that they can be extended without being modified. In other words, developers should be able to change what the modules do without changing the modules' source code.

The Liskov Substitution Principle says that subclasses should be able to substitute for their base classes, meaning that clients that use references to base classes must be able to use the objects of derived classes without knowing them. This principle is essentially a generalization of a "design by contract" approach that specifies that a polymorphic method of a subclass can only replace its pre-condition by a weaker one and its post-condition by a stronger one.

Dependency Inversion Principle says high-level modules shouldn't depend on low-level modules. In other words, abstractions shouldn't depend on details. Details should depend on abstractions.

The Interface Segregation Principle says that clients shouldn't depend on the methods they don't use. It means multiple client-specific interfaces are better than one general-purpose interface.