There are two fundamental ways to establish relationships between class in object-orientated design: inheritance and composition.
First of all, let’s determine that there is no better or worse principle in object-orientated design. But there is an appropriate design for the concrete task. You should use wisely both inheritance or composition, depending on the situation.
The natural outcome of the inheritance is that it provides tight coupling between a parent class and any child classes. It is a hard-coded dependency via
extends keyword. When we treat abstraction and inheritance in the correct way this will never be a problem.
The abstract parent class should provide the new data type definition. This data type should be present in all of the child instances. But very often inheritance hierarchy is based only on the purposes of the code reuse. One class is inherited from another just to gain access to the desired method. Or we create a parent class only to share this method. Other words it is a code reuse through inheritance.
The main problem here is in the parent class, in its interface. Inheritance, regardless of whether it’s used in order to achieve code reuse or not, introduces fragility into our design. One little change to a parent class automatically effects all the child classes in the entire hierarchy. This change can ripple out and require many changes in many other places of the application.
Actually in a parent class the most fragile thing - is its interface. If the superclass is well-designed and defines a new data type with a clean separation of interface and implementation in the object-oriented style, any changes of the superclass shouldn’t ripple at all. But if the inheritance is used to achieve a code reuse, changes in the parent class can break any code that uses this class or any of its subclasses.
This means that all classes in the hierarchy should be substitutable (Liskov Substitution Principle). All the classes behave as expected. They must implement the same methods as their parent class does, taking the same kinds of arguments and return the same kinds of values. Child classes should not do anything that forces other collaborators to check their type in order to know how to collaborate with them.
Child classes may violate their parent contract by accepting input arguments that have broader restrictions and returning results that have narrower restrictions. In this case, they can be perfectly substitutable for their parent class.
In the code above we have an abstraction - a top level of the hierarchy. Every child class in this hierarchy is a
PromoCode and it is very easy to add a new one specification.
Easy to create child classes. Correctly designed hierarchy is easy to extend. There is an abstraction in it, and every new child class comes with a few concrete differences. In this way adding a new child class requires no changes to the existing code.
Easy to change behavior. Changes made to the methods defined at the top of the hierarchy influence all the child classes down the tree. Big changes in behavior can be achieved via small changes in the existing code.
When your main goal is a code reuse, the composition is a better choice than inheritance. The composition provides the ability to employ switchable behaviors at run-time. It combines different simple, transparent and independent objects into one complex whole thing.
In composition, the larger object is connected to its smaller parts via has-a relationships. The main idea here is not only that a larger object has parts, but it communicates with them via an interface. Any smaller object plays the appropriate role and the larger object can collaborate with any object that plays this role. The composed object depends on the interface of the role.
In terms of inheritance, with the composition the parent class becomes backend and the child class becomes frontend. With inheritance, a child class automatically inherits the implementations of all the non-private methods of its parent. Unlike the inheritance, in the composition the frontend class must explicitly invoke a backend class method with its own implementation (delegation).
With composition approach to code reuse, we have a stronger encapsulation than with inheritance. A change to a parent (backend) class doesn’t break any code, that relies only on the child (frontend) class. This means that ripple effect caused by the changes in the parent (back-end) class stops at the child (frontend) class. We have structural independence, but at the cost of explict method delegation.
Parent class interface change. With composition, it is easier to change the parent (backend) class interface, than with inheritance. A change to backend class interface may require changes in the front-end class but doesn’t require changes in its interface. Code that depends only on the front-end class continues to work. In inheritance changes in the parent class interface ripple down to all the child classes and the code that uses them.
Child class interface change. With composition, it is easier to change a front-end class interface. With inheritance, we can’t make changes in the child classes without checking that their new interface is compatible with the parent one. We can’t in the child class override a method and return a value of different type. On the other hand, composition allows us to make changes in the front-end class without affecting back-end classes.
Changing behavior at a run-time. Composition allows us to delay the creation of the back-end objects until they become required. Also, we can change them dynamically throughout the lifetime of the front-end objects in run-time. With inheritance the parent always class exists as the parent of the child class.
As an example let’s consider an
Order class, that has different statuses. The task is to implement methods, that will respond according to the order status.
Here we have only two methods, but in the real world application, there will be much more. And then the business changes and we should implement more and more. The problem here is that we are checking a state of the object property to determine how to respond. Now, we can refactor it at first with inheritance and create a child class for each state:
This code seems to be an appropriate solution for the problem with status. It is nice, clear and easy to read. When the business tells us to implement a new status, we can simply create a new child class. What’s wrong with this code? We remember that hierarchy means is-a relationship.
Order or it plays
Pending role or it has-a
And here comes a new class
OrderState, that represents a state of the order status. But what if we didn’t notice that and continue with our hierarchy? We have a bunch of child classes - each for every status. But what will happen when our order changes its state? For example, an order was pending and then it becomes dispatched. What objects do we use? We can create a new object from the current, or we can set a wrong status in the current object. Neither solution is nice.
As a result, we ended with the composition. Our order has-a state and delegates status messages to it.
When choosing between composition and inheritance you should always determine the relationship between classes. If it is is-a relationship, then in most cases it should be inheritance: a child class is-a parent class:
Some situations require different objects to play a common role. In addition to the core responsibilities, they might play roles like loggable or printable or any others. There are two ways to recognize the existence of a role:
- when an object plays a role, it is not the object’s main responsibility;
- many other unrelated objects can play this role;
Some roles have only common interfaces, others share common behaviours:
The decision between inheritance and composition lies in is-a versus has-a distinction. The more parts an object has, the more likely it should be designed with composition.