Best Practices for Effective Usage of Contexts Dependency Injection (CDI) in Java Applications

Looking at the web, we don’t see many articles talking about Contexts Dependency Injection’s best practices. Hence, I have made the decision to discuss the utilization of Contexts Dependency Injection (CDI) using best practices, providing a comprehensive guide on its implementation.

The CDI is a Jakarta specification in the Java ecosystem to allow developers to use dependency injection, managing contexts, and component injection in an easier way. The article https://www.baeldung.com/java-ee-cdi defines the CDI as follows:

CDI turns DI into a no-brainer process, boiled down to just decorating the service classes with a few simple annotations, and defining the corresponding injection points in the client classes.

If you want to learn the CDI concepts you can read Baeldung’s post and Otavio Santana’s post. Here, in this post, we will focus on the best practices topic.

In fact, CDI is a powerful framework and allows developers to use Dependency Injection (DI) and Inversion of Control (IoC). However, we have one question here. How tightly do we want our application to be coupled with the framework? Note that I’m not talking you cannot couple your application to a framework, but you should think about it, think about the coupling level, and think about the tradeoffs. For me, coupling an application to a framework is not wrong, but doing it without thinking about the coupling level and the cost and tradeoffs is wrong.

It is impossible to add a framework to your application without minimally coupling your application. Even though your application does not have a couple expressed in the code, probably you have a behavioral coupling, that is, a behavior in your application depends on a framework’s behavior, and in some cases, you can not guarantee that other framework will provide a similar behavior, in case of changes.

Best Practices for Injecting Dependencies

When writing code in Java, we often create classes that rely on external dependencies to perform their tasks. To achieve this using CDI, we employ the @Inject annotation, which allows us to inject these dependencies. However, it’s essential to be mindful of whether we are making the class overly dependent on CDI for its functionality, as it may limit its usability without CDI. Hence, it’s crucial to carefully consider the tightness of this dependency. As an illustration, let’s examine the code snippet below. Here, we encounter a class that is tightly coupled to CDI in order to carry out its functionality.

public class ImageRepository {
    @Inject
    private StorageProvider storageProvider;

    public void saveImage(File image){
        //Validate the file to check if it is an image.
        //Apply some logic if needed
        storageProvider.save(image);
    }
}

As you can see the class ImageRepository has a dependency on StorageProvider, that is injected via CDI annotation. However, the storageProvider variable is private and we don’t have setter method or a constructor that allows us to pass this dependency by the constructor. It means this class cannot work without a CDI context, that is, the ImageRepository is tightly coupled to CDI.

This coupling doesn’t provide any benefits for the application, instead, it only causes harm both to the application itself and potentially to the testing of this class.

Look at the code refactored to reduce the couple to CDI.

public class ImageRepository implements Serializable {

    private StorageProvider storageProvider;

    @Inject
    public ImageRepository(StorageProvider storageProvider){
        this.storageProvider = storageProvider;
    }

    public void saveImage(File image){
        //Validate the file to check if it is an image.
        //Apply some logic if needed
        storageProvider.save(image);
    }
}

As you can see, the ImageRepository class has a constructor that receives the StorageProvider as a constructor argument. This approach follows what is said in the Clean Code book.

“True Dependency Injection goes one step further. The class takes no direct steps to resolve its dependencies; it is completely passive. Instead, it provides setter methods or constructor arguments (or both) that are used to inject the dependencies.”

(from “Clean Code: A Handbook of Agile Software Craftsmanship” by Martin Robert C.)

Without a constructor or a setter method, the injection depends on the CDI. However, we still have one question about this class. The class has a CDI annotation and depends on the CDI to be compiled. I’m not saying it is always a problem, but it can be a problem, especially if you are writing a framework. Coupling a framework with another framework can be a problem in cases you want to use your framework with another mutually exclusive one. In general, it should be avoided by frameworks. Thus, how can we fully decouple the ImageRepository class from CDI?

CDI Producer Method

The CDI producer is a source of an object that can be used to be injected by CDI. It is like a factor of a type of object. Look at the code below:

public class ImageRepositoryProducer {

    @Produces
    public ImageRepository createImageRepository(){
        StorageProvider storageProvider = CDI.current().select(StorageProvider.class).get();
        return new ImageRepository(storageProvider);
    }
}

Please note that we are constructing just one object, but the StorageProvider‘s object is read by CDI. You should avoid constructing more than one object within a producer method, as this interlinks the construction of these objects and may lead to complications if you intend to designate distinct scopes for them. You can create a separated producer method to produce the StorageProvider.

This is the ImageRepository class refactored.

public class ImageRepository implements Serializable {

    private StorageProvider storageProvider;

    public ImageRepository(StorageProvider storageProvider){
        this.storageProvider = storageProvider;
    }

    public void saveImage(File image){
        //Validate the file to check if it is an image.
        //Apply some logic if needed
        storageProvider.save(image);
    }
}

Please note that the ImageRepository class does not know anything about the CDI, and is fully decoupled from CDI. The codes about the CDI are inside the ImageRepositoryProducer, which can be extracted to another module if needed.

CDI Interceptor

The CDI Interceptor is a very cool feature of CDI that provides a nice CDI-based way to work with cross-cutting tasks (such as auditing). This is a little definition said in my book:

“A CDI interceptor is a class that wraps the call to a method — this method is called target method — that runs its logic and proceeds the call either to the next CDI interceptor if it exists, or the target method.”

(from “Jakarta EE for Java Developers” by Rhuan Rocha.)

The purpose of this article is not to discuss what a CDI interceptor is, but to discuss CDI best practices. So if you want to read more about CDI interceptor, check out the book Jakarta EE for Java Developers.

As said, the CDI interceptor is very interesting. I am quite fond of this feature and have incorporated it into numerous projects. However, using this feature comes with certain trade-offs for the application.

When you use the CDI interceptor you couple the class to the CDI, because you should be annotating the class with a custom annotation that is a interceptor binding. Look at the example below shown on the Jakarta EE for Java Developers book:

@ApplicationScopedpublic class SecuredBean{
   @Authentication
   public String generateText(String username) throws AutenticationException{
       return "Welcome "+username;
   }
}

As you can see we should define a scope, as it should be a bean managed by CDI, and you should be annotating the class with the interceptor binding. Hence, if you eliminate CDI from your application, the interceptor’s logic won’t execute, and the class won’t be compiled. With this, your application has a behavioral coupling, and a dependency on the CDI lib jar to compile.

As said, it is not necessarily bad, however, you should think if it is a problem in your context.

CDI Event

The CDI Event is a great feature within the CDI framework that I have employed extensively in various applications. This functionality provides the implementation of the Observer Pattern, enabling us to emit events that are then observed by observers who execute tasks asynchronously. However, if we add the CDI codes inside our class to emit events we will couple the class to the CDI. Again, this is not an error, but you should be sure it is not a problem with your solution. Look at the example below.

import jakarta.enterprise.event.Event;

public class User{

 private Event<Email> emailEvent;

 public User(Event<Email> emailEvent){
   this.emailEvent = emailEvent;
 }

 public void register(){
   //logic
   emailEvent.fireAsync(Email.of(from, to, subject, content));
 }
}

Note we are receiving the Event class, which is from CDI, to emit the event. It means this class is coupled to CDI and depends on it to work. One way to avoid it is creating your own class to emit the event, and abstract the details about what is the mechanism (CDI or other) that is emitting the event. Look at the example below.

import net.rhuan.example.EventEmitter;

public class User{

 private EventEmiter<Email> emailEventEmiter;

 public User(EventEmiter<Email> emailEventEmiter){
   this.emailEventEmiter = emailEventEmiter;
 }

 public void register(){
   //logic
   emailEventEmiter.emit(Email.of(from, to, subject, content));
 }
}

Now, your class is agnostic to the emitter of the event. You can use CDI or others, according to the EventEmiter implementation.

Conclusion

The CDI is an amazing specification from Jakarta EE widely used in many Java frameworks and Java applications. Carefully determining the degree of integration between our application and the framework holds immense significance. This intentional decision becomes an important factor in proactively mitigating challenges during the solution’s evolution, especially when working on the development of a framework.

If you have a question or want to share your thoughts, feel free to add comments or send me messages about it. 🙂