Replace Conditionals With Composition And Polymorphism
Conditionals are an integral part of any programming language. We use them every day, so why in OOP they are considered as a code smell?
Conditioт becomes a code smell when we have to check an object’s type in order to make some logic or behavior decision. It doesn’t matter whether it is a stack of if/else
block or a switch
statement. Consider this StatisticsReport
class, which is used to generate reports in different formats:
Here we have getData
method, that returns statistics in a specified format. Our report can be returned as a cvs string, as an array or even as HTML string. Then lately the business can tell us that they need reports in pdf to send them by email. What shall we do then? Of cource we can add two more case
blocks to switch
statement. Our getData
will grow and grow, even if we refactor it and put formatting logic in different methods, this doesn’t change the whole picture:
This class violates Open-Closed Principle, which says:
software entities (classes, modules, functions, etc.) should be open for extension, but closed for modification
But in our case, our class StatisticsReport
is closed for extension, because every time when we need to add/extend some functionality, we should go and change its source code. Now it is clear, that condition here is a code smell. So how to fix it? How can we make this class opened for extension, and add new functionality to it without changing its source code?
There are several refactoring recipes for removing conditionals from your code. On is Replace Conditional With State/Strategy and Replace Conditional With Polymorphism. The first one places conditional branches into new objects, one of which is selected and used at a runtime. This recipe uses composition. The second one removes conditionals be creating a class hierarchy with a base class for the default condition branch and subclasses for each specialization. And again the required object is chosen at a runtime. This recipe uses inheritance.
There is no right or wrong recipe. It always depends on your current application context. Both recipes lead to new objects in your system, that hold conditional logic from each conditional branch.
Lets try both recipes with our example of the StatisticsReport
class.
Replace Conditionals With Polymorphism
In OOP world polymorphism is a very simplistic translation means same name, different logic. This word consists of two greek words: polys which means “many” and morph which means form or shape. In most cases, when we replace conditional with polymorphism, we deal with a subtype polymorphism. This type of polymorphism in OOP means the ability to change the behavior of the method by providing a method with the same name in a child class. Let’s consider the most known example from OOP tutorials:
The code above very clearly demonstrates the subtype polymorphism. We have a base type Shape
and two subtypes (specifications): Square
and Triangle
. Method getArea()
here represents the definition of polymorphism: “same name, different logic”.
Now let’s move back to our StatisticsReport
class again:
It looks like we have a base array
behavior, and several specializations: csv
, html
, pdf
and json
. Now we map these concepts to classes. We leave StatisticsReport
class with an array
implementation and extract several classes each for every format: CvsStatisticsReport
, HtmlStatisticsReport
and PdfStatisticsReport
. And then in these specialization classes, we override base array
behavior in getData()
method:
With this hierarchy, our StatisticsReport
class now is open for extension. Every time our business needs a report in a new format, we can extend this class and create new subtypes for every new report format. The only thing left is how we create all these classes? Where to put this creation logic?
You can notice that we actually didn’t replace the condition, instead, we have moved it to another place. But the key thing here was to make StatisticsReport
class opened for extension. Before refactoring it was impossible to extend this class without touching its source code. Before refactoring this class made a decision how to behave according to some condition. Now we have small classes each with its own behavior. The conditional logic now is used for creation, not for choosing the right behavior.
So, now we are faced with another problem. Where to put this creation logic? How we can get the required object? For this purpose, we can use Factory design patterns. The main responsibility for factories is to create objects. In our case, we need different objects according to different formats:
Next time, when our manager comes to us and asks to build API for reports, so these reports now should be available in json format, it can be done easily. And without touching existing classes, of course except for the factory. We simply go and create a new JsonStatisticsReport
and add a new case
statement for it:
And then somewhere in api controller in our application:
Replace Conditionals With Composition
First of all, let’s remember, what does the composition mean in object-oriented programming. Composition combines different simple, transparent and independent objects into one complex whole thing.
With composition, we have two options here. We can replace conditional with Strategy or with State. These two patterns are closely related. In both patterns, we inject some encapsulated behavior in the original object. In State in choose behavior according to an object’s internal state (one or many property values). And in Strategy, we make a decision what kind of behavior we need according to how we want things to be processed.
StatisticsReport
class doesn’t have any internal state based on its property values. Instead, we want different strategies to process data according to some format value. So, in our case, we can use Strategy pattern to remove conditionals. We will encapsulate each algorithm for every report format in its own object and then unify all of them with one common interface.
Strategy
In Wikipedia:
the strategy pattern (also known as the policy pattern) is a behavioral software design pattern that enables an algorithm’s behavior to be selected at runtime.
The strategy pattern:
- defines a family of algorithms,
- encapsulates each algorithm, and
- makes the algorithms interchangeable within that family.
In our case, that means:
- common interface for different formats logic
- classes for every format
- a factory and a setter for injecting a behavior object
The common interface will be very simple and consist of the only one method:
Notice: I think that it is not a good idea to have design pattern words in your classes and interfaces. But when we have these out of other code examples, I think it is better to have very explicit class names. Also some classes, for example, StatisticsPdfReportStrategy
and StatisticsHtmlReportStrategy
may have their own dependencies for creating pdf and html documents, but they were omitted in this example, because I want to pay your attention to the classes design, but not to the different formats algorithms.
The last step is to provide our StatisticsReport
with a required behavior. Here comes a factory, the only place with conditionals in our design. It will look very similar to StatisticsReportFactory
from the Inheritance chapter. The only difference is that now our factory returns a behavior (strategy) for every format:
Then we need a setter for a strategy to be injected in our StatisticsReport
class. And also we need to update its getData
method. Now this method will use the provided strategy for formatting:
The key difference with the inheritance approach is that our strategies are reusable within our application, and are not coupled to statistics and reports logic. Strategies know nothing about the objects that use them.