angular-cn/public/docs/ts/latest/guide/dependency-injection.jade

include ../../../../_includes/_util-fns
:markdown
  Dependency Injection is an important application design pattern.
  Angular has its own Dependency Injection framework and 
  we really can't build an Angular application without it.

  In this chapter we'll learn what Dependency Injection is, why we want it, and how to use it.
  <a name="why-di"></a>
.l-main-section
:markdown
  ## Why Dependency Injection?
  
  Let's start with the following code. 
  
  ```
  class Engine {}
  
  class Tires {}
  
  class Car {
    private engine: Engine;
    private tires: Tires;
  
    constructor() {
      this.engine = new Engine();
      this.tires = new Tires();
    }
    // Method using the engine and tires
    drive() {}
  }
  ```
  
  Our `Car` creates everything it needs inside its constructor. 
  What's the problem? 
  
  The problem is that our `Car` class is brittle, inflexible, and hard to test.
  
  Our `Car` needs an engine and tires. Instead of asking for them,
  the `Car` constructor creates its own copies by "new-ing" them from
  the very specific classes, `Engine` and `Tires`.
  
  What if the `Engine` class evolves and its constructor requires a parameter?
  Our `Car` is broken and stays broken until we rewrite it along the lines of 
  `this.engine = new Engine(theNewParameter)`. 
  We didn't care about `Engine` constructor parameters when we first wrote `Car`.
  We don't really care about them now. 
  But we'll *have* to start caring because 
  when the definion of `Engine` changes, our `Car` class must change.
  That makes `Car` brittle.
  
  What if we want to put a different brand of tires on our `Car`. Too bad. 
  We're locked into whatever brand the `Tires` class creates. That makes our `Car` inflexible.
  
  Right now each new car gets its own engine. It can't share an engine with other cars. 
  While that makes sense for an automobile engine, 
  we can think of other dependencies that should be shared ... like the onboard
  wireless connection to the manufacturer's service center. Our `Car` lacks the flexibility
  to share services that have been created previously for other consumers.
  
  When we write tests for our `Car` we're at the mercy of its hidden dependencies.
  Is it even possible to create a new `Engine` in a test environment? 
  What does `Engine`itself depend upon? What does that dependency depend on?
  Will a new instance of `Engine` make an asynchronous call to the server? 
  We certainly don't want that going on during our tests.
  
  What if our `Car` should flash a warning signal when tire pressure is low.
  How do we confirm that if actually does flash a warning
  if we can't swap in low-pressure tires during the test?
  
  We have no control over the car's hidden dependencies. 
  When we can't control the dependencies, a class become difficult to test.
  
  How can we make `Car` more robust, more flexible, and more testable? 
  
  That's super easy. We probably already know what to do. We change our `Car` constructor to this:
  <a name="ctor-injection"></a>
  ```
  constructor(engine: Engine, tires: Tires) {
    this.engine = engine;
    this.tires = tires;
  }
  ```
  See what happened? We moved the definition of the dependencies to the constructor. 
  Our `Car` class no longer creates an engine or tires.
  It just consumes them.
  
  Now we create a car by passing the engine and tires to the constructor. 
  ```
  var car = new Car(new Engine(), new Tires());
  ```
  How cool is that? 
  The definition of the engine and tire dependencies are decoupled from the `Car` class itself.
  We can pass in any kind of engine or tires we like, as long as they
  conform to the general API requirements of an engine or tires.
  
  If someone extends the `Engine` class, that is not `Car`'s problem.
.l-sub-section
  :markdown
    The consumer of `Car` has the problem. The consumer must update the car creation code to
    something like:
    ```
    var car = new Car(new Engine(theNewParameter), new Tires());
    ```
    The critical point is this: `Car` itself did not have to change. 
    We'll take care of the consumer's problem soon enough.
  
:markdown
  The `Car` class is much easier to test because we are in complete control 
  of its dependencies.
  We can pass mocks to the constructor that do exactly what we want them to do
  during each test:
  ```
  var car = new Car(new MockEngine(), new MockLowPressureTires());
  ```
  
  **We just learned what Dependency Injection is**. 
  
  It's a coding pattern in which a class receives its dependencies from external
  sources rather than creating them itself.
  
  Cool! But what about that poor consumer?
  Anyone who wants a `Car` must now
  create all three parts: the `Car`, `Engine`, and `Tires`.
  The `Car` class shed its problems at the consumer's expense.
  We need something that takes care of assembling these parts for us.
  
  We could write a giant class to do that:
  ```
  class SuperFactory {
    createEngine = () => new Engine();
    createTires =  () => new Tires();
    createCar = () => new Car(this.createEngine(), this.createTires());
  }
  ```
  It's not so bad now with only three creation methods.
  But maintaining it will be hairy as the application grows.
  This `SuperFactory` is going to become a huge spider web of
  interdependent factory methods!
  
  Wouldn't it be nice if we could simply list the things we want to build without 
  having to define which dependency gets injected into what?
  
  This is where the Dependency Injection Framework comes into play.
  Imagine the framework had something called an `Injector`. 
  We register some classes with this `Injector` and it figures out how to create them.
  
  When we need a `Car`, we simply ask the `Injector` to get it for us and we're good to go.
  ```
  function main() {
    var injector = new Injector([Car, Engine, Tires, Logger]);
    var car = injector.get(Car);
    car.drive();
  }
  ```
  Everyone wins. The `Car` knows nothing about creating an `Engine` or `Tires`.
  The consumer knows nothing about creating a `Car`. 
  We don't have a gigantic factory class to maintain.
  Both `Car` and consumer simply ask for what they need and the `Injector` delivers.

  This is what a **Dependency InjectionFramework** is all about. 

  Now that we know what Dependency Injection is and appreciate its benefits, 
  let's see how it is implemented in Angular.

.l-main-section
:markdown
  ## Angular Dependency Injection
  
  Angular ships with its own Dependency Injection framework. This framework can also be used
  as a standalone module by other applications and frameworks. 
  
  That sounds nice. What does it do for us when building components in Angular?
  Let's see, one step at a time. 
  
  We'll begin with a simplified version of the `HeroesComponent` 
  that we built in the [The Tour of Heroes](../tutorial/).
  ```
  import {Component} from 'angular2/angular2';
  import {Hero} from './hero';
  import {HEROES} from './mock-heroes';
  
  @Component({
    selector: 'my-heroes'
    templateUrl: 'app/heroes.component.html'
  })
  export class HeroesComponent {
  
    heroes: Hero[] = HEROES;
  
  }
  ```
  It assigns a list of mocked heroes to its `heroes` property for binding within the template.
  Pretty straight forward. 
  
  Those heroes are currently a fixed, in-memory collection, defined in another file and imported by the component. 
  That works in the early stages of development but it's far from ideal.
  As soon as we try to test this component or want to get our heroes data from a remote server, 
  we'll have to change this component's implementation of `heroes` and 
  fix every other use of the `HEROES` mock data.
   
  Let's make a service that hides how we get Hero data.
.l-sub-section
  :markdown
    Write this service in its own file. See [this note](#forward-ref) to understand why.
:markdown
  ```
  import {Hero} from './hero';
  import {HEROES} from './mock-heroes';
  
  class HeroService {
  
    heroes: Hero[];
  
    constructor() {
      this.heroes = HEROES;
    }
  
    getHeroes() {
      return this.heroes;
    }
  }
  ```
  Our `HeroService` exposes a `getHeroes()` method that returns 
  the same mock data as before but none of its consumers need to know that. 
    
  A service is nothing more than a class in Angular 2. 
  It remains nothing more than a class until we register it with
  the Angular injector.
  
  ### Configuring the Injector
  
  We don't have to create the injector.
  <a name="bootstrap"></a>
  Angular creates an application-wide injector for us during the bootstrap process.
  ```
  bootstrap(HeroesComponent);
  ``` 
  
  Let’s configure the injector at the same time that we bootstrap by adding 
  our `HeroService` to an array in the second argument. 
  We'll explain that array when we talk about [providers](#providers) later in this chapter.
  ```
  bootstrap(AppComponent, [HeroService]);
  ```  
  That’s it! The injector now knows about the `HeroService` which is available for injection across our entire application.
    
  ### Preparing the `HeroesComponent` for injection
    
  The `HeroesComponent` should get its heroes from this service.
  Per the dependency injection pattern, the component must "ask for" the service in its constructor [as we explained
  earlier](#ctor-injection)".
  
  ```
  constructor(heroService: HeroService) {
    this.heroes = heroService.getHeroes();
  }
  ```
  <a name="di-metadata"></a>
.l-sub-section
  :markdown
    Adding a parameter to the constructor isn't all that's happening here.
    
    We are writing the app in TypeScript and have followed the parameter name with a type notation, `:HeroService`.
    The class is also decorated with the `@Component` decorator (scroll up to confirm that fact).
    
    When the TypeScript compiler evaluates this class, it sees the decorator and adds class metadata
    into the generated JavaScript code. Within that metadata lurks the information that
    associates the `heroService` parameter with the `HeroService` class. 
    
    That's how the Angular injector will know to inject an instance of the `HeroService` when it
    creates a new `HeroesComponent`.
:markdown
  ### Creating the `HeroesComponent` with the injector (implicitly)
  When we introduced the idea of an injector above, we showed how to create
  a new `Car` with that injector.
  ```
  var car = injector.get(Car);
  ```
  Search the entire Tour of Heroes source. We won't find a single line like
  ```
  var hc = injector.get(HeroesComponent);
  ```
  We *could* write code like that if we wanted to. We just don't have to.
  Angular does that for us when it renders a `HeroesComponent` 
  whether we ask for it in an HTML template ...
  ```
  <my-heroes></heroes>
  ```
  ... or navigate to a `HeroesComponent` view with the [router](./router.html).
  
  ### Singleton services  
  We might wonder what happens when we inject the `HeroService` into other components. 
  Do we get the same instance every time?
   
  Yes we do. Dependencies are singletons. 
  We’ll discuss that later in our chapter about 
  [Hierarchical Injectors](./hierarchical-dependency-injection.html).

  ### Testing the component
  We emphasized earlier that designing a class for dependency injection makes it easier to test.
  
  Mission accomplished!  We don't even need the Angular Dependency Injection system to test the `HeroesComponent`.
  We simply create a bew `HeroesComponent` with a mock service and poke at it:
  ```
  it("should have heroes when created", () => {
    let hc = new HeroesComponent(mockService);
    expect(hc.heroes.length).toEqual(mockService.getHeroes().length);
  })
  ```
  ### When the service needs a service 
  Our `HeroService` is very simple. It doesn't have any dependencies of its own. 
  
  
  What if it had a dependency? What if it reported its activities through a logging service?
  We'd apply the same "constructor injection" pattern. 
  
  Here's a rewrite of `HeroService` with a new constructor that takes a `logger` parameter.
  ```
  import {Hero} from './hero';
  import {HEROES} from './mock-heroes';
  import {Logger} from './logger';
  
  @Injectable()
  class HeroService {
  
    heroes: Hero[];
  
    constructor(private logger: Logger) {
      this.heroes = HEROES;
    }
  
    getHeroes() {
      this.logger.log('Getting heroes ...')
      return this.heroes;
    }
  }
  ```
  The constructor now asks for an injected instance of a `Logger` and stores it in a private property called `logger`.
  We call that property within our `getHeroes()` method when anyone asks for heroes.
  
  **The `@Injectable()` decoration catches our eye!**
  
.alert.is-critical
  :markdown
    **Always include the parentheses!** Always call `@Injectable()`. It's easy to forget the parentheses. 
    Our application will fail mysteriously if we do. It bears repeating: **always include the parentheses.**
:markdown
  We haven't seen `@Injectable()` before. 
  As it happens, we could have added it to `HeroService`. We didn't bother because we didn't need it then.
  
  We need it now ... now that our service has an injected dependency. 
  We need it because Angular requires constructor parameter metadata in order to inject a `Logger`. 
  As [we mentioned earlier](#di-metadata), TypeScript *only generates metadata for classes that have a decorator*. .
  
  The `HeroesComponent` has an injected dependency too. Why don't we add `@Injectable()` to the `HeroesComponent`? 
  We *can* add it if we really want to. It isn't necessary because the `HeroesComponent` is already decorated with `@Component`.
  TypeScript generates metadata for *any* class with a decorator and *any* decorator will do.
 
.l-main-section
:markdown
  <a name="providers"></a>
  ## Injector Providers
  
  Remember when we added the `HeroService` to an array in the [bootstrap](#bootstrap) process? 
  ```
  bootstrap(AppComponent, [HeroService]);
  ```
  That list of classes is actually a list of **providers**.  
  
  "Providers" create the instances of the things that we ask the injector to inject.
  There are many ways ways to "provide" a thing that has the necessary shape and behavior to serve as a `HeroService`.
  A class is a natural provider - it's meant to be created. But it's not the only way
  to produce something injectable. We could hand the injector an object to return. We could give it a factory function to call. 
  Any of these approaches might be a good choice under the right circumstances. 
  
  What matters is that the injector knows what to do when something asks for a `HeroService`.

  ### The Provider Class
  
  When we wrote ...
  ```
  [HeroService];
  ```
  we used a short-hand expression for provider registration. 
  Angular expanded that short-hand into a call to the Angular `provide` method
  ```
  [provide(HeroService, {useClass:HeroService})]; 
  ```
  and the `provide` method in turn creates a new instance of the Angular 
  [Provider class](http://localhost:3000/docs/ts/latest/api/core/Provider-class.html):
  ```
  [new Provider(HeroService, {useClass:HeroService})]
  ```
  This provider instance associates a `HeroService` *token* 
  with code that can create an *instance* of a `HeroService`.

  The first parameter is the [token](#token) that serves as the key for both locating a dependency value
  and registering the provider. 
  
  The second parameter is a provider definition object
  which we think of as a "recipe" for creating the dependency value. 
  There are many ways to create dependency values ... and many ways to write a recipe.
  
  ### Alternative Class Providers

  Occasionally we'll ask a different class to  provide the service.
  
  We do that regularly when testing a component that we're creating with dependency injection.
  In this example, we tell the injector
  to return a `MockHeroService` when something asks for the `HeroService`.
  ```
  beforeEachProviders(() => [
    provide(HeroService, {useClass: MockHeroService});
  ]);
  ```
  ### Value Providers
  
  Sometimes it's easier to provide a ready-made object rather than ask the injector to create it from a class.
  
  We do that a lot when we write tests. We might write the following test setup
  for  tests that explore how the `HeroComponent` behaves when the `HeroService` 
  returns an empty hero list.
  ```
  beforeEachProviders(() => {
    
    let emptyHeroService = { getHeroes: () => [] };
    
    return [ provide(HeroService, {useValue: emptyHeroService}) ];
  });
  ```
  Notice we defined the recipe with `useValue` instead of `useClass`.
 
  ### Factory Providers
  
  Sometimes the best choice for a provider is neither a class nor a value.
  
  Suppose our HeroService has some cool new feature that we're only offering to "special" users.
  The HeroService shouldn't know about users and 
  we won't know if the current user is special until runtime anyway.
  We decide to extend our `HeroService` constructor to accept a `useCoolFeature` flag 
  that toggles the feature on or off.
  We rewrite the `HeroService` again as follows.
  ```
  @Injectable()
  class HeroService {
  
    heroes: Hero[];
  
    constructor(private logger: Logger, private useCoolFeature: boolean) {
      this.heroes = HEROES;
    }
  
    getHeroes() {
      let msg = this.useCoolFeature ? 'the cool new way' : 'the old way';
      this.logger.log('Getting heroes ...' + msg)
      return this.heroes;
    }
  }
  ```
  The feature flag is a simple boolean value. We'd like to inject the flag but it seems silly to write an entire class for a 
  simple flag.
  
  We can replace the `HeroService` provider with a factory function that creates a properly configured `HeroService` for the current user.
  We'll' build up to that result, beginning with our definition of the factory function:
  ```
  let heroServiceFactory = (logger: Logger, userService: UserService) => {
    return new HeroService(logger, userService.user.isSpecial);
  }
  ```
.l-sub-section
  :markdown
    The factory takes two parameters: the logger service and a user service. 
    The logger we pass straight to the constructor as we did before.
    
    We'll know to use the cool new feature if the `userService.user.isSpecial` flag is true, 
    a fact we can't know until runtime.
:markdown
  We use dependency injection everywhere so of course the factory function depends on 
  two injected services: `Logger` and `UserService`. 
  We declare those requirements in our provider definition object:
  ```
  let heroServiceDefinition = {
     useFactory: heroServiceFactory, 
     deps: [Logger, UserService]
  };
  ```
.l-sub-section
  :markdown
    The `useFactory` field tells Angular that the provider is a factory function and that its implementation is the `heroServiceFactory`.
    
    The `deps` property is an array of [provider tokens](#token). 
    The `Logger` and `UserService` classes serve as tokens for their own class providers.
:markdown
  Finally, we create the provider and adjust the bootstrapping to include that provider 
  among its provider registrations.
  ```
  let heroServiceProvider = provide(HeroService, heroServiceDefinition);
  
  bootstrap(AppComponent, [heroServiceProvider, Logger, UserService]);
  ```
 
  ### String tokens
  
  Sometimes we have an object dependency rather than a class dependency.
  
  Applications often define configuration objects with lots of small facts like the title of the application or the address of a web api endpoint.
  These configuration objects aren't always instances of a class. They're just objects ... like this one:
  ```
  let config = {
    apiEndpoint: 'api.heroes.com',
    title: 'The Hero Employment Agency'
  }; 
  ```
  We'd like to make this `config` object available for injection. 
  We know we can register an object with a "Value Provider". But what do we use for the token?

  <a id="token"></a>
  Until now, we've always asked the class to play the token role 
  whether we wrote a provider with a class, value, or factory recipe.
  This time we don't have a class to serve as a token. There is no `Config` class.
  
  Fortunately, the token can be a string, a class type, or an 
  [OpaqueToken](http://localhost:3000/docs/ts/latest/api/core/OpaqueToken-class.html). 
  Internally, the `Provider` turns the string and class parameter into an `OpaqueToken`;
  the injector locates dependency values and providers by this token.

  We'll register our configuration object with a string-based token!
  ```
  bootstrap(AppComponent, [
    // other providers //
    provide('App.config', {useValue:config})
  ]);
  ```

  Let's apply what we've learned and update the `HeroesComponent` constructor so it can display the configured title.  
  Right now the constructor signature is
  ```
  constructor(heroService: HeroService)
  ```
  We might think we can add the `config` dependency by writing:
  ```
  // FAIL!
  constructor(heroService: HeroService, config: config)
  ```
  That's not going to work. There is no type called `config` and we didn't register the `config` object under that name anyway.
  We'll need a little help from another Angular decorator called `@Inject`.
  ```
  import {Inject} from 'angular2/angulare2'
  
  constructor(heroService: HeroService, @Inject('app.config') config)
  
  ```

.l-main-section
:markdown
  # Next Steps
  We learned the basics of Angular Dependency Injection in this chapter.
  
  The Angular Dependency Injection is more capable than we've described. 
  We can learn more about its advanced features, beginning with its support for
  a hierarchy of nested injectors in the next 
  [Dependency Injection chapter](./hierarchical-dependency-injection.html)
  
.l-main-section
<a name="forward-ref"></a>
:markdown
  ### Appendix: Why we recommend one class per file
  Developers expect one class per file. Multiple classes per file is confusing and is best avoided. 
  If we define every class in its own file, there is nothing in this note to worry about.
  Move along!
  
  If we scorn this advice
  and we add our `HeroService` class to the `HeroesComponent` file anyway, 
  **define the `HeroesComponent` last!** 
  If we put it define component before the service,
  we'll get a runtime null reference error.

  To understand why, paste the following incorrect, ultra-simplified rendition of these two
  classes into the [TypeScript playground](http://www.typescriptlang.org/Playground).
 
  ```
  class HeroesComponent {
    static $providers=[HeroService]
  }

  class HeroService { }

  alert(HeroesComponent.$providers)
  ```
.l-sub-section
  :markdown
    The `HeroService` is incorrectly defined below the `HeroComponent`.
    
    The `$providers` static property represents the metadata about the injected `HeroService`
    that TypeScript compiler would add to the component class.
    
    The `alert` simulates the action of the Dependency Injector at runtime 
    when it attempts to create a `HeroesComponent`.
:markdown
  Run it. The alert appears but displays nothing. 
  This is the equivalent of the null reference error thrown at runtime.
  
  We understand why when we review the generated JavaScript which looks like this:
  ```
  var HeroesComponent = (function () {
      function HeroesComponent() {
      }
      HeroesComponent.$providers = [HeroService];
      return HeroesComponent;
  })();
  
  var HeroService = (function () {
      function HeroService() {
      }
      return HeroService;
  })();
  
  alert(HeroesComponent.$providers);
  ```

  Notice that the TypeScript compiler turns classes into function expressions 
  assigned to variables. The value of the captured `HeroService` variable is undefined
  when the `$providers` array is assigned. The `HeroService` variable gets its value too late
  to be captured.
  
  Reverse the order of class definition so that the  `HeroService`
  appears before the `HeroesComponent` that requires it.
  Run again. This time the alert displays the `HeroService` function definition.

  If we insist on defining the `HeroService` in the same file and insist on
  defining the component first, Angular offers a way to make that work.
  The `forwardRef()` method let's us reference a class 
  before it has been defined. 
  Learn more about this problem and the `forwardRef()`
  in this [blog post](http://blog.thoughtram.io/angular/2015/09/03/forward-references-in-angular-2.html).