Archive | June 2016

Using Apache Zest to enforce application composition

Layers and modules are well known concepts of application composition. But how to enforce them? Besides the access modifiers for classes and methods there are currently no further means within the language itself. Java 9 will introduce with the “Java Module System” further means to control the usage of packages from other “modules”. Then you can use of course the build system to manage dependencies and therewith configure which module can access which other module. Beyond that tools like “Structure 101” or “degraph” help you with that.

The problem becomes even more obvious when using dependency injection. In this case the injected object can stem from another module or layer and borders between them vanish. Apache Zest therefore provides means of tackling the enforcement of application composition by letting you define to which layers and modules classes belong to. The injection container then has fine grained control from which module or layer injected objects can stem from.

Let’s take a closer look by implementing a small example application. The following dependencies are added to the project:


Now we can construct the application using an anonymous ApplicationAssembler:

Energy4Java e4j = new Energy4Java();
final Application application = e4j.newApplication(new ApplicationAssembler() {
	public ApplicationAssembly assemble(ApplicationAssemblyFactory factory)
		throws AssemblyException {

		ApplicationAssembly assembler = factory.newApplicationAssembly();

		LayerAssembly domainLayer = assembler.layer("domain");
		ModuleAssembly personModule = domainLayer.module("person");

		LayerAssembly persistenceLayer = assembler.layer("persistence");
		ModuleAssembly fileStoreModule = persistenceLayer.module("storage");


		return assembler;

First of all we create the “domain” layer. This layer is supposed to contain the module “person”. A module can contain entities and services that are then added to it. For the entities and services we can specify their visibility. In our simple example the scope layer is enough.

Next to the “domain” layer we also introduce a “persistence” layer with a StorageService that is exposed to the whole application. This is necessary because we want to use it later on in the PersonFactoryService to store and load persons.

The fact that the “domain” layer is using the “persistence” layer can be expressed by invoking the uses() method on the “domain” layer instance.

Finally we activate the application, i.e. the lifecycle of the injection container starts.

The Person entity consists of two interfaces:

public interface Person {
	Property<String> firstName();
	Property<String> lastName();

public interface PersonEntity extends Person, EntityComposite {


The interface Person has two methods that return a Property instance. In contrast to ordinary value object implementations the indirection through Property allows the framework to check if the value is not changed as the annotation Immutable suggests. The actual PersonEntity interface extends both the Person and the EntityComposite interface. The latter marks the interface as an “entity”.

The concrete implementation of the interface PersonEntity is done as a proxy object by the container. This way the container can inject the code to verfiy access to the properties and control the lifecycle of the object. The following method from the PersonFactoryMixin shows how to construct such kind of entity:

public class PersonFactoryMixin implements PersonFactory,
	EntityStore, IdentityGenerator {

	Module module;
	StorageService storageService;

	public Person create(String firstName, String lastName)
		throws IOException {

		UnitOfWork unitOfWork = module.currentUnitOfWork();
		EntityBuilder<Person> personEntityBuilder =

		Person person = personEntityBuilder.instance();


		return personEntityBuilder.newInstance();

The annotation @Structure and @Service let the container inject the appropriate instances of the two interfaces Module and StorageService. The first one is used in the first line of the method to obtain a reference to the current “unit of work”. The term “unit of work” is used in Apache Zest to desribe a “transaction”. This “unit of work” can then be used to retrieve an EntityBuilder.

The EntityBuilder‘s method instance() creates a Person proxy that can be initialized. Once all fields are set, the method can return an immutable instance of the person using the method newInstance() of the EntityBuilder.

Meanwhile the person has also been persisted to the “storage” using the injected StorageService. This works because we have defined at the beginning that the “domain” layer can “use” the “persistence” layer. Would this dependency have been defined the other way round, we would obtain an exception at runtime:

org.qi4j.api.common.ConstructionException: [Module person] Non-optional @Service interface was null in

When we use the create() method of the PersonService and try to change the first name of a person, like in the followig example code, we get an IllegalStateException: java.lang.IllegalStateException: Property [com.wordpress.mdw.zest.entities.Person:firstName] is immutable.:

Module personModule = application.findModule("domain", "person");
try (UnitOfWork unitOfWork = personModule.newUnitOfWork()) {
	PersonFactoryService personFactory = personModule
	Person person = personFactory.create("Homer", "Simpson");
	//try to modify @Immutable property

Conclusion: It was fun to see all the popular approaches like Aspect Oriented Programming, Dependency Injection and Domain Driven Design combined into one framework that even lets you enforce boundaries between layers and modules. As the language itself unfortunately does not (yet) support such kind of features, it is up to frameworks like Apache Zest to retrofit them.