Beans set injection

There is one great (and new) feature of Petite container: it is possible to inject set of beans that are of the same type. All beans registered in the Petite container that implements some interface can be injected as a Set collection into a target bean.


Since this may sound a bit abstract at first sight, here is a simple example. Lets say we have an interface SuperHero:

    public interface SuperHero {
        String getHeroName();

Here are two SuperHero implementations marked as Petite beans:

    public class Batman implements SuperHero {
        public String getHeroName() {
            return "Batman";

    public class Batgirl implements SuperHero {
        public String getHeroName() {
            return "Batgirl";

We can assume that there will be more SuperHero implementations, but at this moment we are not yet sure how many.

Now, we need to call all our superheros at one place, no matter how many implementations there are. All you have to do is specify the injection target as a Set field that has generics type of beans that has to be injected:

    public class GothamCity {

        private Set<SuperHero> superHeros;

        public void callForHelp() {
            for (SuperHero superHero : superHeros) {

After retrieving GothamCity from container, the field superHeroes will contain an Set instance. Set will contain all Petite beans that are of type SuperHero. Easy as that!

Some features

  • It works only for fields. Due to Java limitation (generic type erasure) it is possible to read generic type only of fields. So set injection will not work for methods and parameters.
  • field names do not play any role in injection, just generic type.
  • Set instance will be always created, even if there are no matching beans. This prevents null-checking.
  • Only Collection, Set and HashSet can be used as field type. Although internal code practically allows all Collection types, we wanted to prevent possible vagueness of having other types (e.g. having a List with no defined beans order).

For What It's Worth

We can think of interface methods as messages and target beans as message destinations (hey, who said Event Bus ? :) Let's see this idea in action.

Let's imagine that there is a requirement to send an notification e-mail every time when some user interaction occurs - e.g. when user confirms some payment. We don't have to be theoretical physicists to come with an interface like this one:

    public interface PaymentEvent extends AppEvent {
        void paymentFinished(User user, Payment payment);

Now we can code an implementation, EmailPaymentEvent. This one knows how to send an e-mail that contains all payment data to the user.

Let's now focus on usage in PaymentService - business class that actually performs the payment and that should call the event after successful payment.

Common approach

Usually, we could simply wire these beans, i.e. inject EmailPaymentEvent into the PaymentService:

    public class PaymentService {

        EmailPaymentEvent emailPaymentEvent;

        public void processOrder(User user, Payment payment) {
            ...    // business code
            emailPaymentEvent.paymentFinished(user, payment);

And this would work. But…

Some problems

Sending emails can be a long process - especially if some reports has to be generated in PDF, etc.. Why stopping the user request thread for this? We could change the PaymentEvent code to support executors thread pool - but, to be hones, that is not the right place for that. Event doesn't have to be aware of the way how it is invoked.

Moreover, lets say that we need to send a SMS to the user, too, indicating successful payment. So we can code another class that sends an e-mail and wire it to the services. That would be too much error-prone work.

The new way

The solution is obvious - we can build a façade layer that will dispatch event call to all our implementations. However, we still have to manually wire all our implementations in the façade. So, the event delegation and dispatching logic would be hard-wired to the event implementations.

Instead, we can have the following code:

    public class PaymentEventDispatcher implements PaymentEvent {

        Set<PaymentEvent> paymentEvents;

        void paymentFinished(User user, Payment payment){
            // process paymentEvents any way you need:
            // using thread pools, iterative etc.

So adding a new payment event is just about writing the implementation - and that's it! Everything will continue working as before.

Someone may noticed that PaymentEventDispatcher is also a PaymentEvent - since it is a dispatcher and façade, it make sense. But wait, since PaymentEventDispatcher is also part of the Petite context, wouldn't it be also inside the injected set? Good thinking, but no - Petite is smart, so it will not put target injection instance in the set.

This gives us a new freedom - we can have independent dispatcher logics that would work for different events. For example, we can have a parallel executor, or iterative one, already implemented as an abstract dispatcher class that can be reused for your needs.

New concept - your ideas?

This is a new concept in Petite. Although the explained functionality will stay, we feel there is more behind this idea that can enrich this concept.

Please feel free to contact us with your ideas about this interesting subject;)