Dependency injection and programmatic lookup

One of the most significant features of CDI—certainly the most recognized—is dependency injection; excuse me, typesafe dependency injection.

Injection points

The @Inject annotation lets us define an injection point that is injected during bean instantiation. Injection can occur via three different mechanisms.

Bean constructor parameter injection:

public class Checkout {

   private final ShoppingCart cart;

   @Inject
   public Checkout(ShoppingCart cart) {
      this.cart = cart;
   }

}

A bean can only have one injectable constructor.

Initializer method parameter injection:

public class Checkout {

   private ShoppingCart cart;

   @Inject
   void setShoppingCart(ShoppingCart cart) {
      this.cart = cart;
   }

}
Note
A bean can have multiple initializer methods. If the bean is a session bean, the initializer method is not required to be a business method of the session bean.

And direct field injection:

public class Checkout {

   private @Inject ShoppingCart cart;

}
Note
Getter and setter methods are not required for field injection to work (unlike with JSF managed beans).

Dependency injection always occurs when the bean instance is first instantiated by the container. Simplifying just a little, things happen in this order:

  • First, the container calls the bean constructor (the default constructor or the one annotated @Inject), to obtain an instance of the bean.

  • Next, the container initializes the values of all injected fields of the bean.

  • Next, the container calls all initializer methods of bean (the call order is not portable, don’t rely on it).

  • Finally, the @PostConstruct method, if any, is called.

(The only complication is that the container might call initializer methods declared by a superclass before initializing injected fields declared by a subclass.)

Note
One major advantage of constructor injection is that it allows the bean to be immutable.

CDI also supports parameter injection for some other methods that are invoked by the container. For instance, parameter injection is supported for producer methods:

@Produces Checkout createCheckout(ShoppingCart cart) {
    return new Checkout(cart);
}

This is a case where the @Inject annotation is not required at the injection point. The same is true for observer methods (which we’ll meet in events ) and disposer methods.

What gets injected

The CDI specification defines a procedure, called typesafe resolution, that the container follows when identifying the bean to inject to an injection point. This algorithm looks complex at first, but once you understand it, it’s really quite intuitive. Typesafe resolution is performed at system initialization time, which means that the container will inform the developer immediately if a bean’s dependencies cannot be satisfied.

The purpose of this algorithm is to allow multiple beans to implement the same bean type and either:

  • allow the client to select which implementation it requires using a qualifier or

  • allow the application deployer to select which implementation is appropriate for a particular deployment, without changes to the client, by enabling or disabling an alternative, or

  • allow the beans to be isolated into separate modules.

Obviously, if you have exactly one bean of a given type, and an injection point with that same type, then bean A is going to go into slot A. That’s the simplest possible scenario. When you first start your application, you’ll likely have lots of those.

But then, things start to get complicated. Let’s explore how the container determines which bean to inject in more advanced cases. We’ll start by taking a closer look at qualifiers.

Qualifier annotations

If we have more than one bean that implements a particular bean type, the injection point can specify exactly which bean should be injected using a qualifier annotation. For example, there might be two implementations of PaymentProcessor:

@Synchronous
public class SynchronousPaymentProcessor implements PaymentProcessor {
   public void process(Payment payment) { ... }
}
@Asynchronous
public class AsynchronousPaymentProcessor implements PaymentProcessor {
   public void process(Payment payment) { ... }
}

Where @Synchronous and @Asynchronous are qualifier annotations:

@Qualifier
@Retention(RUNTIME)
@Target({TYPE, METHOD, FIELD, PARAMETER})
public @interface Synchronous {}
@Qualifier
@Retention(RUNTIME)
@Target({TYPE, METHOD, FIELD, PARAMETER})
public @interface Asynchronous {}

A client bean developer uses the qualifier annotation to specify exactly which bean should be injected.

Using field injection:

@Inject @Synchronous PaymentProcessor syncPaymentProcessor;
@Inject @Asynchronous PaymentProcessor asyncPaymentProcessor;

Using initializer method injection:

@Inject
public void setPaymentProcessors(@Synchronous PaymentProcessor syncPaymentProcessor,
                                 @Asynchronous PaymentProcessor asyncPaymentProcessor) {
   this.syncPaymentProcessor = syncPaymentProcessor;
   this.asyncPaymentProcessor = asyncPaymentProcessor;
}

Using constructor injection:

@Inject
public Checkout(@Synchronous PaymentProcessor syncPaymentProcessor,
                @Asynchronous PaymentProcessor asyncPaymentProcessor) {
   this.syncPaymentProcessor = syncPaymentProcessor;
   this.asyncPaymentProcessor = asyncPaymentProcessor;
}

Qualifier annotations can also qualify method arguments of producer, disposer and observer methods. Combining qualified arguments with producer methods is a good way to have an implementation of a bean type selected at runtime based on the state of the system:

@Produces
PaymentProcessor getPaymentProcessor(@Synchronous PaymentProcessor syncPaymentProcessor,
                                     @Asynchronous PaymentProcessor asyncPaymentProcessor) {
   return isSynchronous() ? syncPaymentProcessor : asyncPaymentProcessor;
}

If an injected field or a parameter of a bean constructor or initializer method is not explicitly annotated with a qualifier, the default qualifier,@Default, is assumed.

Now, you may be thinking, "What’s the different between using a qualifier and just specifying the exact implementation class you want?" It’s important to understand that a qualifier is like an extension of the interface. It does not create a direct dependency to any particular implementation. There may be multiple alternative implementations of @Asynchronous PaymentProcessor!

The built-in qualifiers @Default and @Any

Whenever a bean or injection point does not explicitly declare a qualifier, the container assumes the qualifier @Default. From time to time, you’ll need to declare an injection point without specifying a qualifier. There’s a qualifier for that too. All beans have the qualifier @Any. Therefore, by explicitly specifying @Any at an injection point, you suppress the default qualifier, without otherwise restricting the beans that are eligible for injection.

This is especially useful if you want to iterate over all beans with a certain bean type. For example:

import jakarta.enterprise.inject.Instance;

...

@Inject
void initServices(@Any Instance<Service> services) {
   for (Service service: services) {
      service.init();
   }
}

Qualifiers with members

Java annotations can have members. We can use annotation members to further discriminate a qualifier. This prevents a potential explosion of new annotations. For example, instead of creating several qualifiers representing different payment methods, we could aggregate them into a single annotation with a member:

@Qualifier
@Retention(RUNTIME)
@Target({METHOD, FIELD, PARAMETER, TYPE})
public @interface PayBy {
   PaymentMethod value();
}

Then we select one of the possible member values when applying the qualifier:

private @Inject @PayBy(CHECK) PaymentProcessor checkPayment;

We can force the container to ignore a member of a qualifier type by annotating the member @Nonbinding.

@Qualifier
@Retention(RUNTIME)
@Target({METHOD, FIELD, PARAMETER, TYPE})
public @interface PayBy {
   PaymentMethod value();
   @Nonbinding String comment() default "";
}

Multiple qualifiers

An injection point may specify multiple qualifiers:

@Inject @Synchronous @Reliable PaymentProcessor syncPaymentProcessor;

Then only a bean which has both qualifier annotations would be eligible for injection.

@Synchronous @Reliable
public class SynchronousReliablePaymentProcessor implements PaymentProcessor {
   public void process(Payment payment) { ... }
}

Alternatives

Alternatives are beans whose implementation is specific to a particular client module or deployment scenario. This alternative defines a mock implementation of both @Synchronous PaymentProcessor and @Asynchronous PaymentProcessor, all in one:

@Alternative @Synchronous @Asynchronous
public class MockPaymentProcessor implements PaymentProcessor {
   public void process(Payment payment) { ... }
}

By default, @Alternative beans are disabled. We need to enable an alternative in the beans.xml descriptor of a bean archive to make it available for instantiation and injection. However, this activation only applies to the beans in that archive.

<beans
   xmlns="http://xmlns.jcp.org/xml/ns/javaee"
   xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
   xsi:schemaLocation="
      http://xmlns.jcp.org/xml/ns/javaee
      http://xmlns.jcp.org/xml/ns/javaee/beans_1_1.xsd">
   <alternatives>
         <class>org.mycompany.mock.MockPaymentProcessor</class>
   </alternatives>
</beans>

From CDI 1.1 onwards the alternative can be enabled for the whole application using @Priority annotation.

@Priority(100) @Alternative @Synchronous @Asynchronous
public class MockPaymentProcessor implements PaymentProcessor {
   public void process(Payment payment) { ... }
}

When an ambiguous dependency exists at an injection point, the container attempts to resolve the ambiguity by looking for an enabled alternative among the beans that could be injected. If there is exactly one enabled alternative, that’s the bean that will be injected. If there are more beans with priority, the one with the highest priority value is selected.

Fixing unsatisfied and ambiguous dependencies

The typesafe resolution algorithm fails when, after considering the qualifier annotations on all beans that implement the bean type of an injection point and filtering out disabled beans (@Alternative beans which are not explicitly enabled), the container is unable to identify exactly one bean to inject. The container will abort deployment, informing us of the unsatisfied or ambiguous dependency.

During the course of your development, you’re going to encounter this situation. Let’s learn how to resolve it.

To fix an unsatisfied dependency, either:

  • create a bean which implements the bean type and has all the qualifier types of the injection point,

  • make sure that the bean you already have is in the classpath of the module with the injection point, or

  • explicitly enable an @Alternative bean that implements the bean type and has the appropriate qualifier types, using beans.xml.

  • enable an @Alternative bean that implements the bean type and has the appropriate qualifier types, using @Priority annotation.

To fix an ambiguous dependency, either:

  • introduce a qualifier to distinguish between the two implementations of the bean type,

  • exclude one of the beans from discovery (either by means of @Vetoed or beans.xml),

  • disable one of the beans by annotating it @Alternative,

  • move one of the implementations to a module that is not in the classpath of the module with the injection point, or

  • disable one of two @Alternative beans that are trying to occupy the same space, using beans.xml,

  • change priority value of one of two @Alternative beans with the @Priority if they have the same highest priority value.

Just remember: "There can be only one."

On the other hand, if you really do have an optional or multivalued injection point, you should change the type of your injection point to Instance, as we’ll see in Obtaining a contextual instance by programmatic lookup.

Now there’s one more issue you need to be aware of when using the dependency injection service.

Client proxies

Clients of an injected bean do not usually hold a direct reference to a bean instance, unless the bean is a dependent object (scope @Dependent).

Imagine that a bean bound to the application scope held a direct reference to a bean bound to the request scope. The application-scoped bean is shared between many different requests. However, each request should see a different instance of the request scoped bean—the current one!

Now imagine that a bean bound to the session scope holds a direct reference to a bean bound to the application scope. From time to time, the session context is serialized to disk in order to use memory more efficiently. However, the application scoped bean instance should not be serialized along with the session scoped bean! It can get that reference any time. No need to hoard it!

Therefore, unless a bean has the default scope @Dependent, the container must indirect all injected references to the bean through a proxy object. This client proxy is responsible for ensuring that the bean instance that receives a method invocation is the instance that is associated with the current context. The client proxy also allows beans bound to contexts such as the session context to be serialized to disk without recursively serializing other injected beans.

Unfortunately, due to limitations of the Java language, some Java types cannot be proxied by the container. If an injection point declared with one of these types resolves to a bean with any scope other than @Dependent, the container will abort deployment, informing us of the problem.

The following Java types cannot be proxied by the container:

  • classes which don’t have a non-private constructor with no parameters, and

  • classes which are declared final or have a final method,

  • arrays and primitive types.

It’s usually very easy to fix an unproxyable dependency problem. If an injection point of type X results in an unproxyable dependency, simply:

  • add a constructor with no parameters to X,

  • change the type of the injection point to`Instance<X>`,

  • introduce an interface Y, implemented by the injected bean, and change the type of the injection point to Y, or

  • if all else fails, change the scope of the injected bean to @Dependent.

Note

Weld also supports a non-standard workaround for this limitation. See the Configuration chapter for more information.

Obtaining a contextual instance by programmatic lookup

In certain situations, injection is not the most convenient way to obtain a contextual reference. For example, it may not be used when:

  • the bean type or qualifiers vary dynamically at runtime, or

  • depending upon the deployment, there may be no bean which satisfies the type and qualifiers, or

  • we would like to iterate over all beans of a certain type.

In these situations, the application may obtain an instance of the interface Instance, parameterized for the bean type, by injection:

@Inject Instance<PaymentProcessor> paymentProcessorSource;

The get() method of Instance produces a contextual instance of the bean.

PaymentProcessor p = paymentProcessorSource.get();

Qualifiers can be specified in one of two ways:

  • by annotating the Instance injection point, or

  • by passing qualifiers to the select() of Event.

Specifying the qualifiers at the injection point is much, much easier:

@Inject @Asynchronous Instance<PaymentProcessor> paymentProcessorSource;

Now, the PaymentProcessor returned by get() will have the qualifier @Asynchronous.

Alternatively, we can specify the qualifier dynamically. First, we add the @Any qualifier to the injection point, to suppress the default qualifier. (All beans have the qualifier @Any .)

import jakarta.enterprise.inject.Instance;

...

@Inject @Any Instance<PaymentProcessor> paymentProcessorSource;

Next, we need to obtain an instance of our qualifier type. Since annotations are interfaces, we can’t just write new Asynchronous(). It’s also quite tedious to create a concrete implementation of an annotation type from scratch. Instead, CDI lets us obtain a qualifier instance by subclassing the helper class AnnotationLiteral.

class AsynchronousQualifier
extends AnnotationLiteral<Asynchronous> implements Asynchronous {}

In some cases, we can use an anonymous class:

PaymentProcessor p = paymentProcessorSource
   .select(new AnnotationLiteral<Asynchronous>() {});

However, we can’t use an anonymous class to implement a qualifier type with members.

Now, finally, we can pass the qualifier to the select() method of Instance.

Annotation qualifier = synchronously ?
      new SynchronousQualifier() : new AsynchronousQualifier();
PaymentProcessor p = anyPaymentProcessor.select(qualifier).get().process(payment);
Note

Since CDI 2.0, most annotations from jakarta.enterprise package have their AnnotationLiteral implementations. Therefore, in order to programmatically obtain (for instance) @Any annotation, you can simply do Any.Literal.INSTANCE.

Enhanced version of jakarta.enterprise.inject.Instance

Weld also provides org.jboss.weld.inject.WeldInstance - an enhanced version of jakarta.enterprise.inject.Instance. There are three additional methods. The first one - getHandler() - allows to obtain a contextual reference handler which not only holds the contextual reference but also allows to inspect the metadata of the relevant bean and to destroy the underlying contextual instance. Moreover, the handler implements AutoCloseable:

import org.jboss.weld.inject.WeldInstance;

class Foo {

  @Inject
  WeldInstance<Bar> instance;

  void doWork() {
    try (Handler<Bar> barHandler = instance.getHandler()) {
        barHandler.get().doBusiness();
        // Note that Bar will be automatically destroyed at the end of the try-with-resources statement
    }
    Handler<Bar> barHandler = instance.getHandler()
    barHandler.get().doBusiness();
    // Calls Instance.destroy()
    barHandler.destroy();
  }

}

The next method - handlers() - returns an Iterable which allows to iterate over handlers for all the beans that have the required type and required qualifiers and are eligible for injection. This might be useful if you need more control inside the loop:

@ApplicationScoped
class OrderService {

  @Inject
  @Any
  WeldInstance<OrderProcessor> instance;

  void create(Order order) {
    for (Handler<OrderProcessor> handler : instance.handlers()) {
            handler.get().process(order);
            if (Dependent.class.equals(handler.getBean().getScope()) {
                // Destroy only dependent processors
                handler.destroy();
            }
        }
    }
  }

Third method is a twist on the select() method, but it accepts java.lang.reflect.Type as parameter and optionally qualifier(s). This allows for generic selection of instances which can be handy while dealing with third party beans through extensions. However, in order to stay type-safe, this method has a limitation - it can only be invoked on WeldInstance<Object>. Invocation on any other type than Object will result in an IllegalStateException. Please note that the return value if such select will always be WeldInstance<Object> unless you specify it further using <SomeType> before invoking this select(). Let’s look at actual code:

class MyCustomExtension implements Extension {

  @Inject
  @Any
  WeldInstance<Object> instance;

  private Set<Type> allTypes = new HashSet<>();

  public void observe(@Observes ProcessBean<?> bean) {
    // gather all bean types, even those that we do not own
    allTypes.add(bean.getAnnotated().getBaseType());
  }

  public void doWorkWithBeans(@Observes AfterDeploymentValidation adv) {
    for (Type t : allTypes) {
      // now we can select based on Type once we are sure all beans are initialized
      instance.select(t).isResolvable() ? logValidBeanFound(t) : logInvalidBeanFound(t);
    }
  }

WeldInstance is automatically available in Weld SE and Weld Servlet where the Weld API is always on the class path. It is also available in Weld-powered EE containers. In this case, users would have to compile their application against the Weld API and exclude the Weld API artifact from the deployment (e.g. use provided scope in Maven).

The InjectionPoint object

There are certain kinds of dependent objects (beans with scope @Dependent) that need to know something about the object or injection point into which they are injected in order to be able to do what they do. For example:

  • The log category for a Logger depends upon the class of the object that owns it.

  • Injection of a HTTP parameter or header value depends upon what parameter or header name was specified at the injection point.

  • Injection of the result of an EL expression evaluation depends upon the expression that was specified at the injection point.

A bean with scope @Dependent may inject an instance of InjectionPoint and access metadata relating to the injection point to which it belongs.

Let’s look at an example. The following code is verbose, and vulnerable to refactoring problems:

Logger log = Logger.getLogger(MyClass.class.getName());

This clever little producer method lets you inject a JDK Logger without explicitly specifying the log category:

import jakarta.enterprise.inject.spi.InjectionPoint;
import jakarta.enterprise.inject.Produces;

class LogFactory {

   @Produces Logger createLogger(InjectionPoint injectionPoint) {
      return Logger.getLogger(injectionPoint.getMember().getDeclaringClass().getName());
   }

}

We can now write:

@Inject Logger log;

Not convinced? Then here’s a second example. To inject HTTP parameters, we need to define a qualifier type:

@Qualifier
@Retention(RUNTIME)
@Target({TYPE, METHOD, FIELD, PARAMETER})
public @interface HttpParam {
   @Nonbinding public String value();
}

We would use this qualifier type at injection points as follows:

@HttpParam("username") @Inject String username;
@HttpParam("password") @Inject String password;

The following producer method does the work:

import jakarta.enterprise.inject.Produces;
import jakarta.enterprise.inject.spi.InjectionPoint;

class HttpParams

   @Produces @HttpParam("")
   String getParamValue(InjectionPoint ip) {
      ServletRequest request = (ServletRequest) FacesContext.getCurrentInstance().getExternalContext().getRequest();
      return request.getParameter(ip.getAnnotated().getAnnotation(HttpParam.class).value());
   }

}

Note that acquiring of the request in this example is JSF-centric. For a more generic solution you could write your own producer for the request and have it injected as a method parameter.

Note also that the value() member of the HttpParam annotation is ignored by the container since it is annotated @Nonbinding.

The container provides a built-in bean that implements the InjectionPoint interface:

public interface InjectionPoint {
   public Type getType();
   public Set<Annotation> getQualifiers();
   public Bean<?> getBean();
   public Member getMember();
   public Annotated getAnnotated();
   public boolean isDelegate();
   public boolean isTransient();
}