Documentation

§Dependency Injection

Dependency injection is a widely used design pattern that helps separate your components’ behaviour from dependency resolution. Play supports both runtime dependency injection based on JSR 330 (described in this page) and compile time dependency injection in Scala.

Runtime dependency injection is so called because the dependency graph is created, wired and validated at runtime. If a dependency cannot be found for a particular component, you won’t get an error until you run your application.

Play supports Guice out of the box, but other JSR 330 implementations can be plugged in. The Guice wiki is a great resource for learning more about the features of Guice and DI design patterns in general.

§Motivation

Dependency injection achieves several goals:
1. It allows you to easily bind different implementations for the same component. This is useful especially for testing, where you can manually instantiate components using mock dependencies or inject an alternate implementation.
2. It allows you to avoid global static state. While static factories can achieve the first goal, you have to be careful to make sure your state is set up properly. In particular Play’s (now deprecated) static APIs require a running application, which makes testing less flexible. And having more than one instance available at a time makes it possible to run tests in parallel.

The Guice wiki has some good examples explaining this in more detail.

§How it works

Play provides a number of built-in components and declares them in modules such as its BuiltinModule. These bindings describe everything that’s needed to create an instance of Application, including, by default, a router generated by the routes compiler that has your controllers injected into the constructor. These bindings can then be translated to work in Guice and other runtime DI frameworks.

The Play team maintains the Guice module, which provides a GuiceApplicationLoader. That does the binding conversion for Guice, creates the Guice injector with those bindings, and requests an Application instance from the injector.

There are also third-party loaders that do this for other frameworks, including Scaldi and Spring.

Alternatively, Play provides a BuiltInComponents trait that allows you to create a pure Scala implementation that wires together your app at compile time.

We explain how to customize the default bindings and application loader in more detail below.

§Declaring runtime DI dependencies

If you have a component, such as a controller, and it requires some other components as dependencies, then this can be declared using the @Inject annotation. The @Inject annotation can be used on fields or on constructors. We recommend that you use it on constructors, for example:

import javax.inject._
import play.api.libs.ws._

class MyComponent @Inject() (ws: WSClient) {
  // ...
}

Note that the @Inject annotation must come after the class name but before the constructor parameters, and must have parentheses.

Also, Guice does come with several other types of injections, but constructor injection is generally the most clear, concise, and testable in Scala, so we recommend using it.

Guice is able to automatically instantiate any class with an @Inject on its constructor without having to explicitly bind it. This feature is called just in time bindings is described in more detail in the Guice documentation. If you need to do something more sophisticated you can declare a custom binding as described below.

§Dependency injecting controllers

There are two ways to make Play use dependency injected controllers.

§Injected routes generator

By default (since 2.5.0), Play will generate a router that will declare all the controllers that it routes to as dependencies, allowing your controllers to be dependency injected themselves.

To enable the injected routes generator specifically, add the following to your build settings in build.sbt:

routesGenerator := InjectedRoutesGenerator

When using the injected routes generator, prefixing the action with an @ symbol takes on a special meaning, it means instead of the controller being injected directly, a Provider of the controller will be injected. This allows, for example, prototype controllers, as well as an option for breaking cyclic dependencies.

§Static routes generator

You can configure Play to use the legacy (pre 2.5.0) static routes generator, that assumes that all actions are static methods. To configure the project, add the following to build.sbt:

routesGenerator := StaticRoutesGenerator

We recommend always using the injected routes generator. The static routes generator exists primarily as a tool to aid migration so that existing projects don’t have to make all their controllers non static at once.

If using the static routes generator, you can indicate that an action has an injected controller by prefixing the action with @, like so:

GET        /some/path           @controllers.Application.index

§Component lifecycle

The dependency injection system manages the lifecycle of injected components, creating them as needed and injecting them into other components. Here’s how component lifecycle works:

§Singletons

Sometimes you may have a component that holds some state, such as a cache, or a connection to an external resource, or a component might be expensive to create. In these cases it may be important that there is only be one instance of that component. This can be achieved using the @Singleton annotation:

import javax.inject._

@Singleton
class CurrentSharePrice {
  @volatile private var price = 0

  def set(p: Int) = price = p
  def get = price
}

§Stopping/cleaning up

Some components may need to be cleaned up when Play shuts down, for example, to stop thread pools. Play provides an ApplicationLifecycle component that can be used to register hooks to stop your component when Play shuts down:

import scala.concurrent.Future
import javax.inject._
import play.api.inject.ApplicationLifecycle

@Singleton
class MessageQueueConnection @Inject() (lifecycle: ApplicationLifecycle) {
  val connection = connectToMessageQueue()
  lifecycle.addStopHook { () =>
    Future.successful(connection.stop())
  }

  //...
}

The ApplicationLifecycle will stop all components in reverse order from when they were created. This means any components that you depend on can still safely be used in your component’s stop hook. Because you depend on them, they must have been created before your component was, and therefore won’t be stopped until after your component is stopped.

Note: It’s very important to ensure that all components that register a stop hook are singletons. Any non singleton components that register stop hooks could potentially be a source of memory leaks, since a new stop hook will be registered each time the component is created.

§Providing custom bindings

It is considered good practice to define a trait for a component, and have other classes depend on that trait, rather than the implementation of the component. By doing that, you can inject different implementations, for example you inject a mock implementation when testing your application.

In this case, the DI system needs to know which implementation should be bound to that trait. The way we recommend that you declare this depends on whether you are writing a Play application as an end user of Play, or if you are writing library that other Play applications will consume.

§Play applications

We recommend that Play applications use whatever mechanism is provided by the DI framework that the application is using. Although Play does provide a binding API, this API is somewhat limited, and will not allow you to take full advantage of the power of the framework you’re using.

Since Play provides support for Guice out of the box, the examples below show how to provide bindings for Guice.

§Binding annotations

The simplest way to bind an implementation to an interface is to use the Guice @ImplementedBy annotation. For example:

import com.google.inject.ImplementedBy

@ImplementedBy(classOf[EnglishHello])
trait Hello {
  def sayHello(name: String): String
}

class EnglishHello extends Hello {
  def sayHello(name: String) = "Hello " + name
}

§Programmatic bindings

In some more complex situations, you may want to provide more complex bindings, such as when you have multiple implementations of the one trait, which are qualified by @Named annotations. In these cases, you can implement a custom Guice Module:

import com.google.inject.AbstractModule
import com.google.inject.name.Names

class Module extends AbstractModule {
  def configure() = {

    bind(classOf[Hello])
      .annotatedWith(Names.named("en"))
      .to(classOf[EnglishHello])

    bind(classOf[Hello])
      .annotatedWith(Names.named("de"))
      .to(classOf[GermanHello])
  }
}

If you call this module Module and place it in the root package, it will automatically be registered with Play. Alternatively, if you want to give it a different name or put it in a different package, you can register it with Play by appending its fully qualified class name to the play.modules.enabled list in application.conf:

play.modules.enabled += "modules.HelloModule"

You can also disable the automatic registration of a module named Module in the root package by adding it to the disabled modules:

play.modules.disabled += "Module"

§Configurable bindings

Sometimes you might want to read the Play Configuration or use a ClassLoader when you configure Guice bindings. You can get access to these objects by adding them to your module’s constructor.

In the example below, the Hello binding for each language is read from a configuration file. This allows new Hello bindings to be added by adding new settings in your application.conf file.

import com.google.inject.AbstractModule
import com.google.inject.name.Names
import play.api.{ Configuration, Environment }

class Module(
  environment: Environment,
  configuration: Configuration) extends AbstractModule {
  def configure() = {
    // Expect configuration like:
    // hello.en = "myapp.EnglishHello"
    // hello.de = "myapp.GermanHello"
    val helloConfiguration: Configuration =
      configuration.getConfig("hello").getOrElse(Configuration.empty)
    val languages: Set[String] = helloConfiguration.subKeys
    // Iterate through all the languages and bind the
    // class associated with that language. Use Play's
    // ClassLoader to load the classes.
    for (l <- languages) {
      val bindingClassName: String = helloConfiguration.getString(l).get
      val bindingClass: Class[_ <: Hello] =
        environment.classLoader.loadClass(bindingClassName)
        .asSubclass(classOf[Hello])
      bind(classOf[Hello])
        .annotatedWith(Names.named(l))
        .to(bindingClass)
    }
  }
}

Note: In most cases, if you need to access Configuration when you create a component, you should inject the Configuration object into the component itself or into the component’s Provider. Then you can read the Configuration when you create the component. You usually don’t need to read Configuration when you create the bindings for the component.

§Eager bindings

In the code above, new EnglishHello and GermanHello objects will be created each time they are used. If you only want to create these objects once, perhaps because they’re expensive to create, then you should use the @Singleton annotation as described above. If you want to create them once and also create them eagerly when the application starts up, rather than lazily when they are needed, then you can Guice’s eager singleton binding.

import com.google.inject.AbstractModule
import com.google.inject.name.Names

class Module extends AbstractModule {
  def configure() = {

    bind(classOf[Hello])
      .annotatedWith(Names.named("en"))
      .to(classOf[EnglishHello]).asEagerSingleton

    bind(classOf[Hello])
      .annotatedWith(Names.named("de"))
      .to(classOf[GermanHello]).asEagerSingleton
  }
}

Eager singletons can be used to start up a service when an application starts. They are often combined with a shutdown hook so that the service can clean up its resources when the application stops.

§Play libraries

If you’re implementing a library for Play, then you probably want it to be DI framework agnostic, so that your library will work out of the box regardless of which DI framework is being used in an application. For this reason, Play provides a lightweight binding API for providing bindings in a DI framework agnostic way.

To provide bindings, implement a Module to return a sequence of the bindings that you want to provide. The Module trait also provides a DSL for building bindings:

import play.api.inject._

class HelloModule extends Module {
  def bindings(environment: Environment,
               configuration: Configuration) = Seq(
    bind[Hello].qualifiedWith("en").to[EnglishHello],
    bind[Hello].qualifiedWith("de").to[GermanHello]
  )
}

This module can be registered with Play automatically by appending it to the play.modules.enabled list in reference.conf:

play.modules.enabled += "com.example.HelloModule"

In order to maximise cross framework compatibility, keep in mind the following things:

§Excluding modules

If there is a module that you don’t want to be loaded, you can exclude it by appending it to the play.modules.disabled property in application.conf:

play.modules.disabled += "play.api.db.evolutions.EvolutionsModule"

§Managing circular dependencies

Circular dependencies happen when one of your components depends on another component that depends on the original component (either directly or indirectly). For example:

import javax.inject.Inject

class Foo @Inject() (bar: Bar)
class Bar @Inject() (baz: Baz)
class Baz @Inject() (foo: Foo)

In this case, Foo depends on Bar, which depends on Baz, which depends on Foo. So you won’t be able to instantate any of these classes. You can work around this problem by using a Provider:

import javax.inject.{ Inject, Provider }

class Foo @Inject() (bar: Bar)
class Bar @Inject() (baz: Baz)
class Baz @Inject() (foo: Provider[Foo])

Generally, circular dependencies can be resolved by breaking up your components in a more atomic way, or finding a more specific component to depend on. A common problem is a dependency on Application. When your component depends on Application it’s saying that it needs a complete application to do its job; typically that’s not the case. Your dependencies should be on more specific components (e.g. Environment) that have the specific functionality you need. As a last resort you can work around the problem by injecting a Provider[Application].

§Advanced: Extending the GuiceApplicationLoader

Play’s runtime dependency injection is bootstrapped by the GuiceApplicationLoader class. This class loads all the modules, feeds the modules into Guice, then uses Guice to create the application. If you want to control how Guice initializes the application then you can extend the GuiceApplicationLoader class.

There are several methods you can override, but you’ll usually want to override the builder method. This method reads the ApplicationLoader.Context and creates a GuiceApplicationBuilder. Below you can see the standard implementation for builder, which you can change in any way you like. You can find out how to use the GuiceApplicationBuilder in the section about testing with Guice.

import play.api.ApplicationLoader
import play.api.Configuration
import play.api.inject._
import play.api.inject.guice._

class CustomApplicationLoader extends GuiceApplicationLoader() {
  override def builder(context: ApplicationLoader.Context): GuiceApplicationBuilder = {
    val extra = Configuration("a" -> 1)
    initialBuilder
      .in(context.environment)
      .loadConfig(extra ++ context.initialConfiguration)
      .overrides(overrides(context): _*)
  }
}

When you override the ApplicationLoader you need to tell Play. Add the following setting to your application.conf:

play.application.loader = "modules.CustomApplicationLoader"

You’re not limited to using Guice for dependency injection. By overriding the ApplicationLoader you can take control of how the application is initialized. Find out more in the next section.

Next: Compile time dependency injection


Found an error in this documentation? The source code for this page can be found here. After reading the documentation guidelines, please feel free to contribute a pull request.