angular-cn/aio/content/guide/testing.md

{@a top}

Testing

This guide offers tips and techniques for unit and integration testing Angular applications.

The guide presents tests of a sample CLI application that is much like the Tour of Heroes tutorial. The sample application and all tests in this guide are available for inspection and experimentation:

• Sample app

• Tests

---

Setup

准备工作

The Angular CLI downloads and install everything you need to test an Angular application with the Jasmine test framework.

The project you create with the CLI is immediately ready to test. Just run this one CLI command:

ng test

The ng test command builds the app in watch mode, and launches the karma test runner.

The console output looks a bit like this:

10% building modules 1/1 modules 0 active ...INFO [karma]: Karma v1.7.1 server started at http://0.0.0.0:9876/ ...INFO [launcher]: Launching browser Chrome ... ...INFO [launcher]: Starting browser Chrome ...INFO [Chrome ...]: Connected on socket ... Chrome ...: Executed 3 of 3 SUCCESS (0.135 secs / 0.205 secs)

The last line of the log is the most important. It shows that Karma ran three tests that all passed.

A chrome browser also opens and displays the test output in the "Jasmine HTML Reporter" like this.

Most people find this browser output easier to read than the console log. You can click on a test row to re-run just that test or click on a description to re-run the tests in the selected test group ("test suite").

Meanwhile, the ng test command is watching for changes.

To see this in action, make a small change to app.component.ts and save. The tests run again, the browser refreshes, and the new test results appear.

Configuration

The CLI takes care of Jasmine and karma configuration for you.

You can fine-tune many options by editing the karma.conf.js file in the project root folder and the test.ts file in the src/ folder.

The karma.conf.js file is a partial karma configuration file. The CLI constructs the full runtime configuration in memory,based on application structure specified in the .angular-cli.json file, supplemented by karma.conf.js.

Search the web for more details about Jasmine and karma configuration.

Other test frameworks

You can also unit test an Angular app with other testing libraries and test runners. Each library and runner has its own distinctive installation procedures, configuration, and syntax.

Search the web to learn more.

Test file name and location

Look inside the src/app folder.

The CLI generated a test file for the AppComponent named app.component.spec.ts.

The test file extension must be .spec.ts so that tooling can identify it as a file with tests (AKA, a spec file).

The app.component.ts and app.component.spec.ts files are siblings in the same folder. The root file names (app.component) are the same for both files.

Adopt these two conventions in your own projects for every kind of test file.

Service Tests

Services are often the easiest files to unit test. Here are some synchronous and asynchronous unit tests of the ValueService written without assistance from Angular testing utilities.

{@a services-with-dependencies}

Services with dependencies

Services often depend on other services that Angular injects into the constructor. In many cases, it easy to create and inject these dependencies by hand while calling the service's constructor.

The MasterService is a simple example:

MasterService delegates its only method, getValue, to the injected ValueService.

Here are several ways to test it.

The first test creates a ValueService with new and passes it to the MasterService constructor.

However, injecting the real service rarely works well as most dependent services are difficult to create and control.

Instead you can mock the dependency, use a dummy value, or create a spy on the pertinent service method.

Prefer spies as they are usually the easiest way to mock services.

These standard testing techniques are great for unit testing services in isolation.

However, you almost always inject service into application classes using Angular dependency injection and you should have tests that reflect that usage pattern. Angular testing utilities make it easy to investigate how injected services behave.

Testing services with the TestBed

Your app relies on Angular dependency injection (DI) to create services. When a service has a dependent service, DI finds or creates that dependent service. And if that dependent service has its own dependencies, DI finds-or-creates them as well.

As service consumer, you don't worry about any of this. You don't worry about the order of constructor arguments or how they're created.

As a service tester, you must at least think about the first level of service dependencies but you can let Angular DI do the service creation and deal with constructor argument order when you use the TestBed testing utility to provide and create services.

{@a testbed}

Angular TestBed

The TestBed is the most important of the Angular testing utilities. The TestBed creates a dynamically-constructed Angular test module that emulates an Angular @NgModule.

The TestBed.configureTestingModule() method takes a metadata object that can have most of the properties of an @NgModule.

To test a service, you set the providers metadata property with an array of the services that you'll test or mock.

Then inject it inside a test by calling TestBed.get() with the service class as the argument.

Or inside the beforeEach() if you prefer to inject the service as part of your setup.

When testing a service with a dependency, provide the mock in the providers array.

In the following example, the mock is a spy object.

The test consumes that spy in the same way it did earlier.

{@a no-before-each}

Testing without beforeEach()

Most test suites in this guide call beforeEach() to set the preconditions for each it() test and rely on the TestBed to create classes and inject services.

There's another school of testing that never calls beforeEach() and and prefers to create classes explicitly rather than use the TestBed.

Here's how you might rewrite one of the MasterService tests in that style.

Begin by putting re-usable, preparatory code in a setup function instead of beforeEach().

The setup() function returns an object literal with the variables, such as masterService, that a test might reference. You don't define semi-global variables (e.g., let masterService: MasterService) in the body of the describe().

Then each test invokes setup() in its first line, before continuing with steps that manipulate the test subject and assert expectations.

Notice how the test uses destructuring assignment to extract the setup variables that it needs.

Many developers feel this approach is cleaner and more explicit than the traditional beforeEach() style.

Although this testing guide follows the tradition style and the default CLI schematics generate test files with beforeEach() and TestBed, feel free to adopt this alternative approach in your own projects.

Testing HTTP services

Data services that make HTTP calls to remote servers typically inject and delegate to the Angular HttpClient service for XHR calls.

You can test a data service with an injected HttpClient spy as you would test any service with a dependency.

The HeroService methods return Observables. Subscribe to the method observable to (a) cause it to execute and (b) assert that the method succeeds or fails.

The subscribe() method takes a success and fail callback. Make sure you provide both callbacks so that you capture errors.

Neglecting to do so produces an asynchronous uncaught observable error that the test runner will likely attribute to a completely different test.

HttpClientTestingModule

Extended interactions between a data service and the HttpClient can be complex and difficult to mock with spies.

The HttpClientTestingModule can make these testing scenarios more manageable.

While the code sample accompanying this guide demonstrates HttpClientTestingModule, this page defers to the Http guide, which covers testing with the HttpClientTestingModule in detail.

This guide's sample code also demonstrates testing of the legacy HttpModule in app/model/http-hero.service.spec.ts.

Component Test Basics

A component, unlike all other parts of an Angular application, combines an HTML template and a TypeScript class. The component truly is the template and the class working together. and to adequately test a component, you should test that they work together as intended.

Such tests require creating the component's host element in the browser DOM, as Angular does, and investigating the component class's interaction with the DOM as described by its template.

The Angular TestBed facilitates this kind of testing as you'll see in the sections below. But in many cases, testing the component class alone, without DOM involvement, can validate much of the component's behavior in an easier, more obvious way.

Component class testing

Test a component class on its own as you would test a service class.

Consider this LightswitchComponent which toggles a light on and off (represented by an on-screen message) when the user clicks the button.

You might decide only to test that the clicked() method toggles the light's on/off state and sets the message appropriately.

This component class has no dependencies. To test a service with no dependencies, you create it with new, poke at its API, and assert expectations on its public state. Do the same with the component class.

Here is the DashboardHeroComponent from the Tour of Heroes tutorial.

It appears within the template of a parent component, which binds a hero to the @Input property and listens for an event raised through the selected @Output property.

You can test that the class code works without creating the the DashboardHeroComponent or its parent component.

When a component has dependencies, you may wish to use the TestBed to both create the component and its dependencies.

The following WelcomeComponent depends on the UserService to know the name of the user to greet.

You might start by creating a mock of the UserService that meets the minimum needs of this component.

Then provide and inject both the component and the service in the TestBed configuration.

Then exercise the component class, remembering to call the lifecycle hook methods as Angular does when running the app.

Component DOM testing

Testing the component class is as easy as testing a service.

But a component is more than just its class. A component interacts with the DOM and with other components. The class-only tests can tell you about class behavior. They cannot tell you if the component is going to render properly, respond to user input and gestures, or integrate with its parent and and child components.

None of the class-only tests above can answer key questions about how the components actually behave on screen.

• Is Lightswitch.clicked() bound to anything such that the user can invoke it?

• Is the Lightswitch.message displayed?

• Can the user actually select the hero displayed by DashboardHeroComponent?

• Is the hero name displayed as expected (i.e, in uppercase)?

• Is the welcome message displayed by the template of WelcomeComponent?

These may not be troubling questions for the simple components illustrated above. But many components have complex interactions with the DOM elements described in their templates, causing HTML to appear and disappear as the component state changes.

To answer these kinds of questions, you have to create the DOM elements associated with the components, you must examine the DOM to confirm that component state displays properly at the appropriate times, and you must simulate user interaction with the screen to determine whether those interactions cause the component to behave as expected.

To write these kinds of test, you'll use additional features of the TestBed as well as other testing helpers.

CLI-generated tests

The CLI creates an initial test file for you by default when you ask it to generate a new component.

For example, the following CLI command generates a BannerComponent in the app/banner folder (with inline template and styles):

ng generate component banner --inline-template --inline-style --module app

It also generates an initial test file for the component, banner-external.component.spec.ts, that looks like this:

Reduce the setup

Only the last three lines of this file actually test the component and all they do is assert that Angular can create the component.

The rest of the file is boilerplate setup code anticipating more advanced tests that might become necessary if the component evolves into something substantial.

You'll learn about these advanced test features below. For now, you can radically reduce this test file to a more manageable size:

In this example, the metadata object passed to TestBed.configureTestingModule simply declares BannerComponent, the component to test.

There's no need to declare or import anything else. The default test module is pre-configured with something like the BrowserModule from @angular/platform-browser.

Later you'll call TestBed.configureTestingModule() with imports, providers, and more declarations to suit your testing needs. Optional override methods can further fine-tune aspects of the configuration.

{@a create-component}

createComponent()

After configuring TestBed, you call its createComponent() method.

TestBed.createComponent() creates an instance of the BannerComponent, adds a corresponding element to the test-runner DOM, and returns a ComponentFixture.

Do not re-configure TestBed after calling createComponent.

The createComponent method freezes the current TestBeddefinition, closing it to further configuration.

You cannot call any more TestBed configuration methods, not configureTestingModule(), nor get(), nor any of the override... methods. If you try, TestBed throws an error.

{@a component-fixture}

ComponentFixture

The ComponentFixture is a test harness for interacting with the created component and its corresponding element.

Access the component instance through the fixture and confirm it exists with a Jasmine expectation:

beforeEach()

You will add more tests as this component evolves. Rather than duplicate the TestBed configuration for each test, you refactor to pull the setup into a Jasmine beforeEach() and some supporting variables:

Now add a test that gets the component's element from fixture.nativeElement and looks for the expected text.

{@a native-element}

nativeElement

The value of ComponentFixture.nativeElement has the any type. Later you'll encounter the DebugElement.nativeElement and it too has the any type.

Angular can't know at compile time what kind of HTML element the nativeElement is or if it even is an HTML element. The app might be running on a non-browser platform, such as the server or a Web Worker, where the element may have a diminished API or not exist at all.

The tests in this guide are designed to run in a browser so a nativeElement value will always be an HTMLElement or one of its derived classes.

Knowing that it is an HTMLElement of some sort, you can use the standard HTML querySelector to dive deeper into the element tree.

Here's another test that calls HTMLElement.querySelector to get the paragraph element and look for the banner text:

{@a debug-element}

DebugElement

The Angular fixture provides the component's element directly through the fixture.nativeElement.

This is actually a convenience method, implemented as fixture.debugElement.nativeElement.

There's a good reason for this circuitous path to the element.

The properties of the nativeElement depend upon the runtime environment. You could be running these tests on a non-browser platform that doesn't have a DOM or whose DOM-emulation doesn't support the full HTMLElement API.

Angular relies on the DebugElement abstraction to work safely across all supported platforms. Instead of creating an HTML element tree, Angular creates a DebugElement tree that wraps the native elements for the runtime platform. The nativeElement property unwraps the DebugElement and returns the platform-specific element object.

Because the sample tests for this guide are designed to run only in a browser, a nativeElement in these tests is always an HTMLElement whose familiar methods and properties you can explore within a test.

Here's the previous test, re-implemented with fixture.debugElement.nativeElement:

The DebugElement has other methods and properties that are useful in tests, as you'll see elsewhere in this guide.

You import the DebugElement symbol from the Angular core library.

{@a by-css}

By.css()

Although the tests in this guide all run in the browser, some apps might run on a different platform at least some of the time.

For example, the component might render first on the server as part of a strategy to make the application launch faster on poorly connected devices. The server-side renderer might not support the full HTML element API. If it doesn't support querySelector, the previous test could fail.

The DebugElement offers query methods that work for all supported platforms. These query methods take a predicate function that returns true when a node in the DebugElement tree matches the selection criteria.

You create a predicate with the help of a By class imported from a library for the runtime platform. Here's the By import for the browser platform:

The following example re-implements the previous test with DebugElement.query() and the browser's By.css method.

Some noteworthy observations:

• The By.css() static method selects DebugElement nodes with a standard CSS selector.

• The query returns a DebugElement for the paragraph.

• You must unwrap that result to get the paragraph element.

When you're filtering by CSS selector and only testing properties of a browser's native element, the By.css approach may be overkill.

It's often easier and more clear to filter with a standard HTMLElement method such as querySelector() or querySelectorAll(), as you'll see in the next set of tests.

---

Component Test Scenarios

The following sections, comprising most of this guide, explore common component testing scenarios

Component binding

The current BannerComponent presents static title text in the HTML template.

After a few changes, the BannerComponent presents a dynamic title by binding to the component's title property like this.

Simple as this is, you decide to add a test to confirm that component actually displays the right content where you think it should.

Query for the <h1>

You'll write a sequence of tests that inspect the value of the <h1> element that wraps the title property interpolation binding.

You update the beforeEach to find that element with a standard HTML querySelector and assign it to the h1 variable.

{@a detect-changes}

createComponent() does not bind data

For your first test you'd like to see that the screen displays the default title. Your instinct is to write a test that immediately inspects the <h1> like this:

That test fails with the message:

expected '' to contain 'Test Tour of Heroes'.

Binding happens when Angular performs change detection.

In production, change detection kicks in automatically when Angular creates a component or the user enters a keystroke or an asynchronous activity (e.g., AJAX) completes.

The TestBed.createComponent does not trigger change detection. a fact confirmed in the revised test:

detectChanges()

You must tell the TestBed to perform data binding by calling fixture.detectChanges(). Only then does the <h1> have the expected title.

Delayed change detection is intentional and useful. It gives the tester an opportunity to inspect and change the state of the component before Angular initiates data binding and calls lifecycle hooks.

Here's another test that changes the component's title property before calling fixture.detectChanges().

{@a auto-detect-changes}

Automatic change detection

The BannerComponent tests frequently call detectChanges. Some testers prefer that the Angular test environment run change detection automatically.

That's possible by configuring the TestBed with the ComponentFixtureAutoDetect provider. First import it from the testing utility library:

Then add it to the providers array of the testing module configuration:

Here are three tests that illustrate how automatic change detection works.

The first test shows the benefit of automatic change detection.

The second and third test reveal an important limitation. The Angular testing environment does not know that the test changed the component's title. The ComponentFixtureAutoDetect service responds to asynchronous activities such as promise resolution, timers, and DOM events. But a direct, synchronous update of the component property is invisible. The test must call fixture.detectChanges() manually to trigger another cycle of change detection.

Rather than wonder when the test fixture will or won't perform change detection, the samples in this guide always call detectChanges() explicitly. There is no harm in calling detectChanges() more often than is strictly necessary.

---

Component with external files

The BannerComponent above is defined with an inline template and inline css, specified in the @Component.template and @Component.styles properties respectively.

Many components specify external templates and external css with the @Component.templateUrl and @Component.styleUrls properties respectively, as the following variant of BannerComponent does.

This syntax tells the Angular compiler to read the external files during component compilation.

That's not a problem when you run the CLI ng test command because it compiles the app before running the tests.

However, if you run the tests in a non-CLI environment, tests of this component may fail. For example, if you run the BannerComponent tests in a web coding environment such as plunker, you'll see a message like this one:

Error: This test module uses the component BannerComponent which is using a "templateUrl" or "styleUrls", but they were never compiled. Please call "TestBed.compileComponents" before your test.

You get this test failure message when the runtime environment compiles the source code during the tests themselves.

To correct the problem, call compileComponents() as explained below.

{@a component-with-dependency}

Component with a dependency

Components often have service dependencies.

The WelcomeComponent displays a welcome message to the logged in user. It knows who the user is based on a property of the injected UserService:

The WelcomeComponent has decision logic that interacts with the service, logic that makes this component worth testing. Here's the testing module configuration for the spec file, app/welcome/welcome.component.spec.ts:

This time, in addition to declaring the component-under-test, the configuration adds a UserService provider to the providers list. But not the real UserService.

{@a service-test-doubles}

Provide service test doubles

A component-under-test doesn't have to be injected with real services. In fact, it is usually better if they are test doubles (stubs, fakes, spies, or mocks). The purpose of the spec is to test the component, not the service, and real services can be trouble.

Injecting the real UserService could be a nightmare. The real service might ask the user for login credentials and attempt to reach an authentication server. These behaviors can be hard to intercept. It is far easier and safer to create and register a test double in place of the real UserService.

This particular test suite supplies a minimal mock of the UserService that satisfies the needs of the WelcomeComponent and its tests:

{@a get-injected-service}

Get injected services

The tests need access to the (stub) UserService injected into the WelcomeComponent.

Angular has a hierarchical injection system. There can be injectors at multiple levels, from the root injector created by the TestBed down through the component tree.

The safest way to get the injected service, the way that always works, is to get it from the injector of the component-under-test. The component injector is a property of the fixture's DebugElement.

{@a testbed-get}

TestBed.get()

You may also be able to get the service from the root injector via TestBed.get(). This is easier to remember and less verbose. But it only works when Angular injects the component with the service instance in the test's root injector.

In this test suite, the only provider of UserService is the root testing module, so it is safe to call TestBed.get() as follows:

For a use case in which TestBed.get() does not work, see the section Override a component's providers, which explains when and why you must get the service from the component's injector instead.

{@a service-from-injector}

Always get the service from an injector

Do not reference the userServiceStub object that's provided to the testing module in the body of your test. It does not work! The userService instance injected into the component is a completely different object, a clone of the provided userServiceStub.

{@a welcome-spec-setup}

Final setup and tests

Here's the complete beforeEach(), using TestBed.get():

And here are some tests:

The first is a sanity test; it confirms that the stubbed UserService is called and working.

The second parameter to the Jasmine matcher (e.g., 'expected name') is an optional failure label. If the expectation fails, Jasmine displays appends this label to the expectation failure message. In a spec with multiple expectations, it can help clarify what went wrong and which expectation failed.

The remaining tests confirm the logic of the component when the service returns different values. The second test validates the effect of changing the user name. The third test checks that the component displays the proper message when there is no logged-in user.

---

{@a component-with-async-service}

Component with async service

In this sample, the AboutComponent template hosts a TwainComponent. The TwainComponent displays Mark Twain quotes.

Note that value of the component's quote property passes through an AsyncPipe. That means the property returns either a Promise or an Observable.

In this example, the TwainComponent.getQuote() method tells you that the quote property returns an Observable.

The TwainComponent gets quotes from an injected TwainService. The component starts the returned Observable with a placeholder value ('...'), before the service can returns its first quote.

The catchError intercepts service errors, prepares an error message, and returns the placeholder value on the success channel. It must wait a tick to set the errorMessage in order to avoid updating that message twice in the same change detection cycle.

These are all features you'll want to test.

Testing with a spy

When testing a component, only the service's public API should matter. In general, tests themselves should not make calls to remote servers. They should emulate such calls. The setup in this app/twain/twain.component.spec.ts shows one way to do that:

{@a service-spy}

Focus on the spy.

The spy is designed such that any call to getQuote receives an Observable with a test quote. Unlike the real getQuote() method, this spy bypasses the server and returns a synchronous Observable whose value is available immediately.

You can write many useful tests with this spy, even though its Observable is synchronous.

{@a sync-tests}

Synchronous tests

A key advantage of a synchronous Observable is that you can often turn asynchronous processes into synchronous tests.

Because the spy result returns synchronously, the getQuote() method updates the message on screen immediately after the first change detection cycle during which Angular calls ngOnInit.

You're not so lucky when testing the error path. Although the service spy will return an error synchronously, the component method calls setTimeout(). The test must wait at least one full turn of the JavaScript engine before the value becomes available. The test must become asynchronous.

{@a fake-async}

Async test with fakeAsync()

The following test confirms the expected behavior when the service returns an ErrorObservable.

Note that the it() function receives an argument of the following form.

fakeAsync(() => { /* test body */ })`

The fakeAsync function enables a linear coding style by running the test body in a special fakeAsync test zone. The test body appears to be synchronous. There is no nested syntax (like a Promise.then()) to disrupt the flow of control.

{@a tick}

The tick() function

You do have to call tick() to advance the (virtual) clock.

Calling tick() simulates the passage of time until all pending asynchronous activities finish. In this case, it waits for the error handler's setTimeout();

The tick function is one of the Angular testing utilities that you import with TestBed. It's a companion to fakeAsync and you can only call it within a fakeAsync body.

Async observables

You might be satisfied with the test coverage of these tests.

But you might be troubled by the fact that the real service doesn't quite behave this way. The real service sends requests to a remote server. A server takes time to respond and the response certainly won't be available immediately as in the previous two tests.

Your tests will reflect the real world more faithfully if you return an asynchronous observable from the getQuote() spy like this.

Async observable helpers

The async observable was produced by an asyncData helper The asyncData helper is a utility function that you'll have to write yourself. Or you can copy this one from the sample code.

This helper's observable emits the data value in the next turn of the JavaScript engine.

RxJS defer() returns an observable. It takes a factory function that returns either a promise or an observable. When something subscribes to defer's observable, it adds the subscriber to a new observable created with that factory.

RxJS defer() transform the Promise.resolve() into a new observable that, like HttpClient, emits once and completes. Subscribers will be unsubscribed after they receive the data value.

There's a similar helper for producing an async error.

More async tests

Now that the getQuote() spy is returning async observables, most of your tests will have to be async as well.

Here's a fakeAsync() test that demonstrates the data flow you'd expect in the real world.

Notice that the quote element displays the placeholder value ('...') after ngOnInit(). The first quote hasn't arrived yet.

To flush the first quote from the observable, you call tick(). Then call detectChanges() to tell Angular to update the screen.

Then you can assert that the quote element displays the expected text.

{@a async}

Async test with async()

The fakeAsync() utility function has a few limitations. In particular, it won't work if the test body makes an XHR call.

XHR calls within a test are rare so you can generally stick with fakeAsync(). But if you ever do need to call XHR, you'll want to know about async().

The TestBed.compileComponents() method (see below) calls XHR to read external template and css files during "just-in-time" compilation. Write tests that call compileComponents() with the async() utility.

Here's the previous fakeAsync() test, re-written with the async() utility.

The async() utility hides some asynchronous boilerplate by arranging for the tester's code to run in a special async test zone. You don't have to pass Jasmine's done() into the test and call done() in promise or observable callbacks.

But the test's asynchronous nature is revealed by the call to fixture.whenStable(), which breaks the linear flow of control.

{@a when-stable}

whenStable

The test must wait for the getQuote() observable to emit the next quote. Instead of calling tick(), it calls fixture.whenStable().

The fixture.whenStable() returns a promise that resolves when the JavaScript engine's task queue becomes empty. In this example, the task queue becomes empty when the observable emits the first quote.

The test resumes within the promise callback, which calls detectChanges() to update the quote element with the expected text.

{@a jasmine-done}

Jasmine done()

While the async and fakeAsync functions greatly simplify Angular asynchronous testing, you can still fall back to the traditional technique and pass it a function that takes a done callback.

Now you are responsible for chaining promises, handling errors, and calling done() at the appropriate moments.

Writing test functions with done(), is more cumbersome than asyncand fakeAsync. But it is occasionally necessary. For example, you can't call async or fakeAsync when testing code that involves the intervalTimer() or the RxJS delay() operator.

Here are two mover versions of the previous test, written with done(). The first one subscribes to the Observable exposed to the template by the component's quote property.

The RxJS last() operator emits the observable's last value before completing, which will be the test quote. The subscribe callback calls detectChanges() to update the quote element with the test quote, in the same manner as the earlier tests.

In some tests, you're more interested in how an injected service method was called and what values it returned, than what appears on screen.

A service spy, such as the qetQuote() spy of the fake TwainService, can give you that information and make assertions about the state of the view.

---

{@a marble-testing}

Component marble tests

The previous TwainComponent tests simulated an asynchronous observable response from the TwainService with the asyncData and asyncError utilities.

These are short, simple functions that you can write yourself. Unfortunately, they're too simple for many common scenarios. An observable often emits multiple times, perhaps after a significant delay. A component may coordinate multiple observables with overlapping sequences of values and errors.

RxJS marble testing is a great way to test observable scenarios, both simple and complex. You've likely seen the marble diagrams that illustrate how observables work. Marble testing uses a similar marble language to specify the observable streams and expectations in your tests.

The following examples revisit two of the TwainComponent tests with marble testing.

Start by installing the jasmine-marbles npm package. Then import the symbols you need.

Here's the complete test for getting a quote:

Notice that the Jasmine test is synchronous. There's no fakeAsync(). Marble testing uses a test scheduler to simulate the passage of time in a synchronous test.

The beauty of marble testing is in the visual definition of the observable streams. This test defines a cold observable that waits three frames (---), emits a value (x), and completes (|). In the second argument you map the value marker (x) to the emitted value (testQuote).

The marble library constructs the corresponding observable, which the test sets as the getQuote spy's return value.

When you're ready to activate the marble observables, you tell the TestScheduler to flush its queue of prepared tasks like this.

This step serves a purpose analogous to tick() and whenStable() in the earlier fakeAsync() and async() examples. The balance of the test is the same as those examples.

Marble error testing

Here's the marble testing version of the getQuote() error test.

It's still an async test, calling fakeAsync() and tick(), because the component itself calls setTimeout() when processing errors.

Look at the marble observable definition.

This is a cold observable that waits three frames and then emits an error, The hash (#) indicates the timing of the error that is specified in the third argument. The second argument is null because the observable never emits a value.

Learn about marble testing

{@a marble-frame}

A marble frame is a virtual unit of testing time. Each symbol (-, x, |, #) marks the passing of one frame.

{@a cold-observable}

A cold observable doesn't produce values until you subscribe to it. Most of your application observables are cold. All HttpClient methods return cold observables.

A hot observable is already producing values before you subscribe to it. The Router.events observable, which reports router activity, is a hot observable.

RxJS marble testing is a rich subject, beyond the scope of this guide. Learn about it on the web, starting with the official documentation.

---

{@a component-with-input-output}

Component with inputs and outputs

A component with inputs and outputs typically appears inside the view template of a host component. The host uses a property binding to set the input property and an event binding to listen to events raised by the output property.

The testing goal is to verify that such bindings work as expected. The tests should set input values and listen for output events.

The DashboardHeroComponent is a tiny example of a component in this role. It displays an individual hero provided by the DashboardComponent. Clicking that hero tells the DashboardComponent that the user has selected the hero.

The DashboardHeroComponent is embedded in the DashboardComponent template like this:

The DashboardHeroComponent appears in an *ngFor repeater, which sets each component's hero input property to the looping value and listens for the component's selected event.

Here's the component's full definition:

{@a dashboard-hero-component}

While testing a component this simple has little intrinsic value, it's worth knowing how. You can use one of these approaches:

• Test it as used by DashboardComponent.

• Test it as a stand-alone component.

• Test it as used by a substitute for DashboardComponent.

A quick look at the DashboardComponent constructor discourages the first approach:

The DashboardComponent depends on the Angular router and the HeroService. You'd probably have to replace them both with test doubles, which is a lot of work. The router seems particularly challenging.

The discussion below covers testing components that require the router.

The immediate goal is to test the DashboardHeroComponent, not the DashboardComponent, so, try the second and third options.

{@a dashboard-standalone}

Test DashboardHeroComponent stand-alone

Here's the meat of the spec file setup.

Note how the setup code assigns a test hero (expectedHero) to the component's hero property, emulating the way the DashboardComponent would set it via the property binding in its repeater.

The following test verifies that the hero name is propagated to the template via a binding.

Because the template passes the hero name through the Angular UpperCasePipe, the test must match the element value with the upper-cased name.

This small test demonstrates how Angular tests can verify a component's visual representation—something not possible with component class tests—at low cost and without resorting to much slower and more complicated end-to-end tests.

Clicking

Clicking the hero should raise a selected event that the host component (DashboardComponent presumably) can hear:

The component's selected property returns an EventEmitter, which looks like an RxJS synchronous Observable to consumers. The test subscribes to it explicitly just as the host component does implicitly.

If the component behaves as expected, clicking the hero's element should tell the component's selected property to emit the hero object.

The test detects that event through its subscription to selected.

{@a trigger-event-handler}

triggerEventHandler

The heroDe in the previous test is a DebugElement that represents the hero <div>.

It has Angular properties and methods that abstract interaction with the native element. This test calls the DebugElement.triggerEventHandler with the "click" event name. The "click" event binding responds by calling DashboardHeroComponent.click().

The Angular DebugElement.triggerEventHandler can raise any data-bound event by its event name. The second parameter is the event object passed to the handler.

The test triggered a "click" event with a null event object.

The test assumes (correctly in this case) that the runtime event handler—the component's click() method—doesn't care about the event object.

Other handlers are less forgiving. For example, the RouterLink directive expects an object with a button property that identifies which mouse button (if any) was pressed during the click. The RouterLink directive throws an error if the event object is missing.

Click the element

The following test alternative calls the native element's own click() method, which is perfectly fine for this component.

{@a click-helper}

click() helper

Clicking a button, an anchor, or an arbitrary HTML element is a common test task.

Make that consistent and easy by encapsulating the click-triggering process in a helper such as the click() function below:

The first parameter is the element-to-click. If you wish, you can pass a custom event object as the second parameter. The default is a (partial) left-button mouse event object accepted by many handlers including the RouterLink directive.

The click() helper function is not one of the Angular testing utilities. It's a function defined in this guide's sample code. All of the sample tests use it. If you like it, add it to your own collection of helpers.

Here's the previous test, rewritten using the click helper.

---

{@a component-inside-test-host}

Component inside a test host

The previous tests played the role of the host DashboardComponent themselves. But does the DashboardHeroComponent work correctly when properly data-bound to a host component?

You could test with the actual DashboardComponent. But doing so could require a lot of setup, especially when its template features an *ngFor repeater, other components, layout HTML, additional bindings, a constructor that injects multiple services, and it starts interacting with those services right away.

Imagine the effort to disable these distractions, just to prove a point that can be made satisfactorily with a test host like this one:

This test host binds to DashboardHeroComponent as the DashboardComponent would but without the noise of the Router, the HeroService, or the *ngFor repeater.

The test host sets the component's hero input property with its test hero. It binds the component's selected event with its onSelected handler, which records the emitted hero in its selectedHero property.

Later, the tests will be able to easily check selectedHero to verify that the DashboardHeroComponent.selected event emitted the expected hero.

The setup for the test-host tests is similar to the setup for the stand-alone tests:

This testing module configuration shows three important differences:

1. It declares both the DashboardHeroComponent and the TestHostComponent.

2. It creates the TestHostComponent instead of the DashboardHeroComponent.

3. The TestHostComponent sets the DashboardHeroComponent.hero with a binding.

The createComponent returns a fixture that holds an instance of TestHostComponent instead of an instance of DashboardHeroComponent.

Creating the TestHostComponent has the side-effect of creating a DashboardHeroComponent because the latter appears within the template of the former. The query for the hero element (heroEl) still finds it in the test DOM, albeit at greater depth in the element tree than before.

The tests themselves are almost identical to the stand-alone version:

Only the selected event test differs. It confirms that the selected DashboardHeroComponent hero really does find its way up through the event binding to the host component.

---

{@a routing-component}

Routing component

A routing component is a component that tells the Router to navigate to another component. The DashboardComponent is a routing component because the user can navigate to the HeroDetailComponent by clicking on one of the hero buttons on the dashboard.

Routing is pretty complicated. Testing the DashboardComponent seemed daunting in part because it involves the Router, which it injects together with the HeroService.

Mocking the HeroService with a spy is a familiar story. But the Router has a complicated API and is entwined with other services and application preconditions. Might it be difficult to mock?

Fortunately, not in this case because the DashboardComponent isn't doing much with the Router

This is often the case with routing components. As a rule you test the component, not the router, and care only if the component navigates with the right address under the given conditions.

Providing a router spy for this component test suite happens to be as easy as providing a HeroService spy.

The following test clicks the displayed hero and confirms that Router.navigateByUrl is called with the expected url.

{@a routed-component-w-param}

Routed components

A routed component is the destination of a Router navigation. It can be trickier to test, especially when the route to the component includes parameters. The HeroDetailComponent is a routed component that is the destination of such a route.

When a user clicks a Dashboard hero, the DashboardComponent tells the Router to navigate to heroes/:id. The :id is a route parameter whose value is the id of the hero to edit.

The Router matches that URL to a route to the HeroDetailComponent. It creates an ActivatedRoute object with the routing information and injects it into a new instance of the HeroDetailComponent.

Here's the HeroDetailComponent constructor:

The HeroDetail component needs the id parameter so it can fetch the corresponding hero via the HeroDetailService.

The component has to get the id from the ActivatedRoute.paramMap property which is an Observable.

It can't just reference the id property of the ActivatedRoute.paramMap. The component has to subscribe to the ActivatedRoute.paramMap observable and be prepared for the id to change during its lifetime.

The Router guide covers ActivatedRoute.paramMap in more detail.

Tests can explore how the HeroDetailComponent responds to different id parameter values by manipulating the ActivatedRoute injected into the component's constructor.

You know how to spy on the Router and a data service.

You'll take a different approach with ActivatedRoute because

• paramMap returns an Observable that can emit more than one value during a test.

• You need the router helper function, convertToParamMap(), to create a ParamMap.

• Other routed components tests need a test double for ActivatedRoute.

These differences argue for a re-usable stub class.

ActivatedRouteStub

The following ActivatedRouteStub class serves as a test double for ActivatedRoute.

Consider placing such helpers in a testing folder sibling to the app folder. This sample puts ActivatedRouteStub in testing/activated-route-stub.ts.

Consider writing a more capable version of this stub class with the marble testing library.

{@a tests-w-test-double}

Testing with ActivatedRouteStub

Here's a test demonstrating the component's behavior when the observed id refers to an existing hero:

The createComponent() method and page object are discussed below. Rely on your intuition for now.

When the id cannot be found, the component should re-route to the HeroListComponent.

The test suite setup provided the same router spy described above which spies on the router without actually navigating.

This test expects the component to try to navigate to the HeroListComponent.

While this app doesn't have a route to the HeroDetailComponent that omits the id parameter, it might add such a route someday. The component should do something reasonable when there is no id.

In this implementation, the component should create and display a new hero. New heroes have id=0 and a blank name. This test confirms that the component behaves as expected:

---

Nested component tests

Component templates often have nested components, whose templates may contain more components.

The component tree can be very deep and, most of the time, the nested components play no role in testing the component at the top of the tree.

The AppComponent, for example, displays a navigation bar with anchors and their RouterLink directives.

While the AppComponent class is empty, you may want to write unit tests to confirm that the links are wired properly to the RouterLink directives, perhaps for the reasons explained below.

To validate the links, you don't need the Router to navigate and you don't need the <router-outlet> to mark where the Router inserts routed components.

The BannerComponent and WelcomeComponent (indicated by <app-banner> and <app-welcome>) are also irrelevant.

Yet any test that creates the AppComponent in the DOM will also create instances of these three components and, if you let that happen, you'll have to configure the TestBed to create them.

If you neglect to declare them, the Angular compiler won't recognize the <app-banner>, <app-welcome>, and <router-outlet> tags in the AppComponent template and will throw an error.

If you declare the real components, you'll also have to declare their nested components and provide for all services injected in any component in the tree.

That's too much effort just to answer a few simple questions about links.

This section describes two techniques for minimizing the setup. Use them, alone or in combination, to stay focused on the testing the primary component.

{@a stub-component}

Stubbing unneeded components

In the first technique, you create and declare stub versions of the components and directive that play little or no role in the tests.

The stub selectors match the selectors for the corresponding real components. But their templates and classes are empty.

Then declare them in the TestBed configuration next to the components, directives, and pipes that need to be real.

The AppComponent is the test subject, so of course you declare the real version.

The RouterLinkDirectiveStub, described later, is a test version of the real RouterLink that helps with the link tests.

The rest are stubs.

{@a no-errors-schema}

NO_ERRORS_SCHEMA

In the second approach, add NO_ERRORS_SCHEMA to the TestBed.schemas metadata.

The NO_ERRORS_SCHEMA tells the Angular compiler to ignore unrecognized elements and attributes.

The compiler will recognize the <app-root> element and the routerLink attribute because you declared a corresponding AppComponent and RouterLinkDirectiveStub in the TestBed configuration.

But the compiler won't throw an error when it encounters <app-banner>, <app-welcome>, or <router-outlet>. It simply renders them as empty tags and the browser ignores them.

You no longer need the stub components.

Use both techniques together

These are techniques for Shallow Component Testing , so-named because they reduce the visual surface of the component to just those elements in the component's template that matter for tests.

The NO_ERRORS_SCHEMA approach is the easier of the two but don't overuse it.

The NO_ERRORS_SCHEMA also prevents the compiler from telling you about the missing components and attributes that you omitted inadvertently or misspelled. You could waste hours chasing phantom bugs that the compiler would have caught in an instant.

The stub component approach has another advantage. While the stubs in this example were empty, you could give them stripped-down templates and classes if your tests need to interact with them in some way.

In practice you will combine the two techniques in the same setup, as seen in this example.

The Angular compiler creates the BannerComponentStub for the <app-banner> element and applies the RouterLinkStubDirective to the anchors with the routerLink attribute, but it ignores the <app-welcome> and <router-outlet> tags.

---

{@a routerlink}

Components with RouterLink

The real RouterLinkDirective is quite complicated and entangled with other components and directives of the RouterModule. It requires challenging setup to mock and use in tests.

The RouterLinkDirectiveStub in this sample code replaces the real directive with an alternative version designed to validate the kind of anchor tag wiring seen in the AppComponent template.

The URL bound to the [routerLink] attribute flows in to the directive's linkParams property.

The host metadata property wires the click event of the host element (the <a> anchor elements in AppComponent) to the stub directive's onClick method.

Clicking the anchor should trigger the onClick() method, which sets the stub's telltale navigatedTo property. Tests inspect navigatedTo to confirm that clicking the anchor set the expected route definition.

Whether the router is configured properly to navigate with that route definition is a question for a separate set of tests.

{@a by-directive}

{@a inject-directive}

By.directive and injected directives

A little more setup triggers the initial data binding and gets references to the navigation links:

Three points of special interest:

1. You can locate the anchor elements with an attached directive using By.directive.

2. The query returns DebugElement wrappers around the matching elements.

3. Each DebugElement exposes a dependency injector with the specific instance of the directive attached to that element.

The AppComponent links to validate are as follows:

{@a app-component-tests}

Here are some tests that confirm those links are wired to the routerLink directives as expected:

The "click" test in this example is misleading. It tests the RouterLinkDirectiveStub rather than the component. This is a common failing of directive stubs.

It has a legitimate purpose in this guide. It demonstrates how to find a RouterLink element, click it, and inspect a result, without engaging the full router machinery. This is a skill you may need to test a more sophisticated component, one that changes the display, re-calculates parameters, or re-arranges navigation options when the user clicks the link.

{@a why-stubbed-routerlink-tests}

What good are these tests?

Stubbed RouterLink tests can confirm that a component with links and an outlet is setup properly, that the component has the links it should have, and that they are all pointing in the expected direction. These tests do not concern whether the app will succeed in navigating to the target component when the user clicks a link.

Stubbing the RouterLink and RouterOutlet is the best option for such limited testing goals. Relying on the real router would make them brittle. They could fail for reasons unrelated to the component. For example, a navigation guard could prevent an unauthorized user from visiting the HeroListComponent. That's not the fault of the AppComponent and no change to that component could cure the failed test.

A different battery of tests can explore whether the application navigates as expected in the presence of conditions that influence guards such as whether the user is authenticated and authorized.

A future guide update will explain how to write such tests with the RouterTestingModule.

---

{@a page-object}

Use a page object

The HeroDetailComponent is a simple view with a title, two hero fields, and two buttons.

But there's plenty of template complexity even in this simple form.

Tests that exercise the component need ...

• to wait until a hero arrives before elements appear in the DOM.

• a reference to the title text.

• a reference to the name input box to inspect and set it.

• references to the two buttons so they can click them.

• spies for some of the component and router methods.

Even a small form such as this one can produce a mess of tortured conditional setup and CSS element selection.

Tame the complexity with a Page class that handles access to component properties and encapsulates the logic that sets them.

Here is such a Page class for the hero-detail.component.spec.ts

Now the important hooks for component manipulation and inspection are neatly organized and accessible from an instance of Page.

A createComponent method creates a page object and fills in the blanks once the hero arrives.

The HeroDetailComponent tests in an earlier section demonstrate how createComponent and page keep the tests short and on message. There are no distractions: no waiting for promises to resolve and no searching the DOM for element values to compare.

Here are a few more HeroDetailComponent tests to reinforce the point.

---

{@a compile-components}

Calling compileComponents()

You can ignore this section if you only run tests with the CLI ng test command because the CLI compiles the application before running the tests.

If you run tests in a non-CLI environment, the tests may fail with a message like this one:

Error: This test module uses the component BannerComponent which is using a "templateUrl" or "styleUrls", but they were never compiled. Please call "TestBed.compileComponents" before your test.

The root of the problem is at least one of the components involved in the test specifies an external template or CSS file as the following version of the BannerComponent does.

The test fails when the TestBed tries to create the component.

Recall that the app hasn't been compiled. So when you call createComponent(), the TestBed compiles implicitly.

That's not a problem when the source code is in memory. But the BannerComponent requires external files that the compile must read from the file system, an inherently asynchronous operation.

If the TestBed were allowed to continue, the tests would run and fail mysteriously before the compiler could finished.

The preemptive error message tells you to compile explicitly with compileComponents().

compileComponents() is async

You must call compileComponents() within an asynchronous test function.

If you neglect to make the test function async (e.g., forget to use async() as described below), you'll see this error message

Error: ViewDestroyedError: Attempt to use a destroyed view

A typical approach is to divide the setup logic into two separate beforeEach() functions:

1. An async beforeEach() that compiles the components

2. A synchronous beforeEach() that performs the remaining setup.

To follow this pattern, import the async() helper with the other testing symbols.

The async beforeEach

Write the first async beforeEach like this.

The async() helper function takes a parameterless function with the body of the setup.

The TestBed.configureTestingModule() method returns the TestBed class so you can chain calls to other TestBed static methods such as compileComponents().

In this example, the BannerComponent is the only component to compile. Other examples configure the testing module with multiple components and may import application modules that hold yet more components. Any of them could be require external files.

The TestBed.compileComponents method asynchronously compiles all components configured in the testing module.

Do not re-configure the TestBed after calling compileComponents().

Calling compileComponents() closes the current TestBed instance to further configuration. You cannot call any more TestBed configuration methods, not configureTestingModule() nor any of the override... methods. The TestBed throws an error if you try.

Make compileComponents() the last step before calling TestBed.createComponent().

The synchronous beforeEach

The second, synchronous beforeEach() contains the remaining setup steps, which include creating the component and querying for elements to inspect.

You can count on the test runner to wait for the first asynchronous beforeEach to finish before calling the second.

Consolidated setup

You can consolidate the two beforeEach() functions into a single, async beforeEach().

The compileComponents() method returns a promise so you can perform the synchronous setup tasks after compilation by moving the synchronous code into a then(...) callback.

compileComponents() is harmless

There's no harm in calling compileComponents() when it's not required.

The component test file generated by the CLI calls compileComponents() even though it is never required when running ng test.

The tests in this guide only call compileComponents when necessary.

---

{@a import-module}

Setup with module imports

Earlier component tests configured the testing module with a few declarations like this:

The DashboardComponent is simple. It needs no help. But more complex components often depend on other components, directives, pipes, and providers and these must be added to the testing module too.

Fortunately, the TestBed.configureTestingModule parameter parallels the metadata passed to the @NgModule decorator which means you can also specify providers and imports.

The HeroDetailComponent requires a lot of help despite its small size and simple construction. In addition to the support it receives from the default testing module CommonModule, it needs:

• NgModel and friends in the FormsModule to enable two-way data binding.

• The TitleCasePipe from the shared folder.

• Router services (which these tests are stubbing).

• Hero data access services (also stubbed).

One approach is to configure the testing module from the individual pieces as in this example:

Notice that the beforeEach() is asynchronous and calls TestBed.compileComponents because the HeroDetailComponent has an external template and css file.

As explained in Calling compileComponents() above, these tests could be run in a non-CLI environment where Angular would have to compile them in the browser.

Import a shared module

Because many app components need the FormsModule and the TitleCasePipe, the developer created a SharedModule to combine these and other frequently requested parts.

The test configuration can use the SharedModule too as seen in this alternative setup:

It's a bit tighter and smaller, with fewer import statements (not shown).

{@a feature-module-import}

Import a feature module

The HeroDetailComponent is part of the HeroModule Feature Module that aggregates more of the interdependent pieces including the SharedModule. Try a test configuration that imports the HeroModule like this one:

That's really crisp. Only the test doubles in the providers remain. Even the HeroDetailComponent declaration is gone.

In fact, if you try to declare it, Angular will throw an error because HeroDetailComponent is declared in both the HeroModule and the DynamicTestModule created by the TestBed.

Importing the component's feature module can be the easiest way to configure tests when there are many mutual dependencies within the module and the module is small, as feature modules tend to be.

---

{@a component-override}

Override component providers

The HeroDetailComponent provides its own HeroDetailService.

It's not possible to stub the component's HeroDetailService in the providers of the TestBed.configureTestingModule. Those are providers for the testing module, not the component. They prepare the dependency injector at the fixture level.

Angular creates the component with its own injector, which is a child of the fixture injector. It registers the component's providers (the HeroDetailService in this case) with the child injector.

A test cannot get to child injector services from the fixture injector. And TestBed.configureTestingModule can't configure them either.

Angular has been creating new instances of the real HeroDetailService all along!

These tests could fail or timeout if the HeroDetailService made its own XHR calls to a remote server. There might not be a remote server to call.

Fortunately, the HeroDetailService delegates responsibility for remote data access to an injected HeroService.

The previous test configuration replaces the real HeroService with a TestHeroService that intercepts server requests and fakes their responses.

What if you aren't so lucky. What if faking the HeroService is hard? What if HeroDetailService makes its own server requests?

The TestBed.overrideComponent method can replace the component's providers with easy-to-manage test doubles as seen in the following setup variation:

Notice that TestBed.configureTestingModule no longer provides a (fake) HeroService because it's not needed.

{@a override-component-method}

The overrideComponent method

Focus on the overrideComponent method.

It takes two arguments: the component type to override (HeroDetailComponent) and an override metadata object. The overide metadata object is a generic defined as follows:

type MetadataOverride = { add?: T; remove?: T; set?: T; };

A metadata override object can either add-and-remove elements in metadata properties or completely reset those properties. This example resets the component's providers metadata.

The type parameter, T, is the kind of metadata you'd pass to the @Component decorator:

selector?: string; template?: string; templateUrl?: string; providers?: any[]; ...

{@a spy-stub}

Provide a spy stub (HeroDetailServiceSpy)

This example completely replaces the component's providers array with a new array containing a HeroDetailServiceSpy.

The HeroDetailServiceSpy is a stubbed version of the real HeroDetailService that fakes all necessary features of that service. It neither injects nor delegates to the lower level HeroService so there's no need to provide a test double for that.

The related HeroDetailComponent tests will assert that methods of the HeroDetailService were called by spying on the service methods. Accordingly, the stub implements its methods as spies:

{@a override-tests}

The override tests

Now the tests can control the component's hero directly by manipulating the spy-stub's testHero and confirm that service methods were called.

{@a more-overrides}

More overrides

The TestBed.overrideComponent method can be called multiple times for the same or different components. The TestBed offers similar overrideDirective, overrideModule, and overridePipe methods for digging into and replacing parts of these other classes.

Explore the options and combinations on your own.

---

{@a attribute-directive}

Attribute Directive Testing

An attribute directive modifies the behavior of an element, component or another directive. Its name reflects the way the directive is applied: as an attribute on a host element.

The sample application's HighlightDirective sets the background color of an element based on either a data bound color or a default color (lightgray). It also sets a custom property of the element (customProperty) to true for no reason other than to show that it can.

It's used throughout the application, perhaps most simply in the AboutComponent:

Testing the specific use of the HighlightDirective within the AboutComponent requires only the techniques explored above (in particular the "Shallow test" approach).

However, testing a single use case is unlikely to explore the full range of a directive's capabilities. Finding and testing all components that use the directive is tedious, brittle, and almost as unlikely to afford full coverage.

Class-only tests might be helpful, but attribute directives like this one tend to manipulate the DOM. Isolated unit tests don't touch the DOM and, therefore, do not inspire confidence in the directive's efficacy.

A better solution is to create an artificial test component that demonstrates all ways to apply the directive.

The <input> case binds the HighlightDirective to the name of a color value in the input box. The initial value is the word "cyan" which should be the background color of the input box.

Here are some tests of this component:

A few techniques are noteworthy:

• The By.directive predicate is a great way to get the elements that have this directive when their element types are unknown.

• The :not pseudo-class in By.css('h2:not([highlight])') helps find <h2> elements that do not have the directive. By.css('*:not([highlight])') finds any element that does not have the directive.

• DebugElement.styles affords access to element styles even in the absence of a real browser, thanks to the DebugElement abstraction. But feel free to exploit the nativeElement when that seems easier or more clear than the abstraction.

• Angular adds a directive to the injector of the element to which it is applied. The test for the default color uses the injector of the second <h2> to get its HighlightDirective instance and its defaultColor.

• DebugElement.properties affords access to the artificial custom property that is set by the directive.

---

Pipe Testing

Pipes are easy to test without the Angular testing utilities.

A pipe class has one method, transform, that manipulates the input value into a transformed output value. The transform implementation rarely interacts with the DOM. Most pipes have no dependence on Angular other than the @Pipe metadata and an interface.

Consider a TitleCasePipe that capitalizes the first letter of each word. Here's a naive implementation with a regular expression.

Anything that uses a regular expression is worth testing thoroughly. Use simple Jasmine to explore the expected cases and the edge cases.

{@a write-tests}

Write DOM tests too

These are tests of the pipe in isolation. They can't tell if the TitleCasePipe is working properly as applied in the application components.

Consider adding component tests such as this one:

---

{@a test-debugging}

Test debugging

Debug specs in the browser in the same way that you debug an application.

1. Reveal the karma browser window (hidden earlier).

2. Click the DEBUG button; it opens a new browser tab and re-runs the tests.

3. Open the browser's “Developer Tools” (Ctrl-Shift-I on windows; Command-Option-I in OSX).

4. Pick the "sources" section.

5. Open the 1st.spec.ts test file (Control/Command-P, then start typing the name of the file).

6. Set a breakpoint in the test.

7. Refresh the browser, and it stops at the breakpoint.

---

{@a atu-apis}

Testing Utility APIs

This section takes inventory of the most useful Angular testing features and summarizes what they do.

The Angular testing utilities include the TestBed, the ComponentFixture, and a handful of functions that control the test environment. The TestBed and ComponentFixture classes are covered separately.

Here's a summary of the stand-alone functions, in order of likely utility:

<th>

  Function

</th>

<th>

  Description

  描述

</th>

<td style="vertical-align: top">

  <code>async</code>

</td>

<td>

  Runs the body of a test (`it`) or setup (`beforeEach`) function within a special _async test zone_.
  See [discussion above](#async).

</td>

<td style="vertical-align: top">

  <code>fakeAsync</code>

</td>

<td>

  Runs the body of a test (`it`) within a special _fakeAsync test zone_, enabling
  a linear control flow coding style. See [discussion above](#fake-async).

</td>

<td style="vertical-align: top">

  <code>tick</code>

</td>

<td>

  Simulates the passage of time and the completion of pending asynchronous activities
  by flushing both _timer_ and _micro-task_ queues within the _fakeAsync test zone_.

  <div class="l-sub-section">

  The curious, dedicated reader might enjoy this lengthy blog post,
  ["_Tasks, microtasks, queues and schedules_"](https://jakearchibald.com/2015/tasks-microtasks-queues-and-schedules/).

  </div>

  Accepts an optional argument that moves the virtual clock forward
  by the specified number of milliseconds,
  clearing asynchronous activities scheduled within that timeframe.
  See [discussion above](#tick).

</td>

<td style="vertical-align: top">

   <code>inject</code>

</td>

<td>

  Injects one or more services from the current `TestBed` injector into a test function.
  It cannot inject a service provided by the component itself.
  See discussion of the [debugElement.injector](#get-injected-services).

</td>

<td style="vertical-align: top">

  <code>discardPeriodicTasks</code>

</td>

<td>

  When a `fakeAsync` test ends with pending timer event _tasks_ (queued `setTimeOut` and `setInterval` callbacks),
  the test fails with a clear error message.

  In general, a test should end with no queued tasks.
  When pending timer tasks are expected, call `discardPeriodicTasks` to flush the _task_ queue
  and avoid the error.

</td>

<td style="vertical-align: top">

  <code>flushMicrotasks</code>

</td>

<td>

  When a `fakeAsync` test ends with pending _micro-tasks_ such as unresolved promises,
  the test fails with a clear error message.

  In general, a test should wait for micro-tasks to finish.
  When pending microtasks are expected, call `flushMicrotasks` to flush the  _micro-task_ queue
  and avoid the error.

</td>

<td style="vertical-align: top">

  <code>ComponentFixtureAutoDetect</code>

</td>

<td>

  A provider token for a service that turns on [automatic change detection](#automatic-change-detection).

</td>

<td style="vertical-align: top">

  <code>getTestBed</code>

</td>

<td>

  Gets the current instance of the `TestBed`.
  Usually unnecessary because the static class methods of the `TestBed` class are typically sufficient.
  The `TestBed` instance exposes a few rarely used members that are not available as
  static methods.

</td>

---

{@a testbed-class-summary}

TestBed class summary

The TestBed class is one of the principal Angular testing utilities. Its API is quite large and can be overwhelming until you've explored it, a little at a time. Read the early part of this guide first to get the basics before trying to absorb the full API.

The module definition passed to configureTestingModule is a subset of the @NgModule metadata properties.

type TestModuleMetadata = { providers?: any[]; declarations?: any[]; imports?: any[]; schemas?: Array<SchemaMetadata | any[]>; };

{@a metadata-override-object}

Each override method takes a MetadataOverride<T> where T is the kind of metadata appropriate to the method, that is, the parameter of an @NgModule, @Component, @Directive, or @Pipe.

type MetadataOverride = { add?: T; remove?: T; set?: T; };

{@a testbed-methods}

{@a testbed-api-summary}

The TestBed API consists of static class methods that either update or reference a global instance of theTestBed.

Internally, all static methods cover methods of the current runtime TestBed instance, which is also returned by the getTestBed() function.

Call TestBed methods within a beforeEach() to ensure a fresh start before each individual test.

Here are the most important static methods, in order of likely utility.

<th>

  Methods

</th>

<th>

  Description

  描述

</th>

<td style="vertical-align: top">

  <code>configureTestingModule</code>

</td>

<td>

  The testing shims (`karma-test-shim`, `browser-test-shim`)
  establish the [initial test environment](guide/testing) and a default testing module.
  The default testing module is configured with basic declaratives and some Angular service substitutes that every tester needs.

  Call `configureTestingModule` to refine the testing module configuration for a particular set of tests
  by adding and removing imports, declarations (of components, directives, and pipes), and providers.

</td>

<td style="vertical-align: top">

  <code>compileComponents</code>

</td>

<td>

  Compile the testing module asynchronously after you've finished configuring it.
  You **must** call this method if _any_ of the testing module components have a `templateUrl`
  or `styleUrls` because fetching component template and style files is necessarily asynchronous.
  See [above](#compile-components).

  After calling `compileComponents`, the `TestBed` configuration is frozen for the duration of the current spec.

</td>

<td style="vertical-align: top">

  <code>createComponent<T></code>

</td>

<td>

  Create an instance of a component of type `T` based on the current `TestBed` configuration.
  After calling `compileComponent`, the `TestBed` configuration is frozen for the duration of the current spec.

</td>

<td style="vertical-align: top">

  <code>overrideModule</code>

</td>

<td>

  Replace metadata for the given `NgModule`. Recall that modules can import other modules.
  The `overrideModule` method can reach deeply into the current testing module to
  modify one of these inner modules.

</td>

<td style="vertical-align: top">

  <code>overrideComponent</code>

</td>

<td>

  Replace metadata for the given component class, which could be nested deeply
  within an inner module.

</td>

<td style="vertical-align: top">

  <code>overrideDirective</code>

</td>

<td>

  Replace metadata for the given directive class, which could be nested deeply
  within an inner module.

</td>

<td style="vertical-align: top">

  <code>overridePipe</code>

</td>

<td>

  Replace metadata for the given pipe class, which could be nested deeply
  within an inner module.

</td>

<td style="vertical-align: top">

  {@a testbed-get}

  <code>get</code>

</td>

<td>

  Retrieve a service from the current `TestBed` injector.

  The `inject` function is often adequate for this purpose.
  But `inject` throws an error if it can't provide the service.

  What if the service is optional?

  The `TestBed.get()` method takes an optional second parameter,
  the object to return if Angular can't find the provider
  (`null` in this example):

  <code-example path="testing/src/app/demo/demo.testbed.spec.ts" region="testbed-get-w-null" title="app/demo/demo.testbed.spec.ts" linenums="false"></code-example>

  After calling `get`, the `TestBed` configuration is frozen for the duration of the current spec.

</td>

<td style="vertical-align: top">

  {@a testbed-initTestEnvironment}

  <code>initTestEnvironment</code>

</td>

<td>

  Initialize the testing environment for the entire test run.

  The testing shims (`karma-test-shim`, `browser-test-shim`) call it for you
  so there is rarely a reason for you to call it yourself.

  You may call this method _exactly once_. If you must change
  this default in the middle of your test run, call `resetTestEnvironment` first.

  Specify the Angular compiler factory, a `PlatformRef`, and a default Angular testing module.
  Alternatives for non-browser platforms are available in the general form
  `@angular/platform-<platform_name>/testing/<platform_name>`.

</td>

<td style="vertical-align: top">

  <code>resetTestEnvironment</code>

</td>

<td>

  Reset the initial test environment, including the default testing module.

</td>

A few of the TestBed instance methods are not covered by static TestBed class methods. These are rarely needed.

{@a component-fixture-api-summary}

The ComponentFixture

The TestBed.createComponent<T> creates an instance of the component T and returns a strongly typed ComponentFixture for that component.

The ComponentFixture properties and methods provide access to the component, its DOM representation, and aspects of its Angular environment.

{@a component-fixture-properties}

ComponentFixture properties

Here are the most important properties for testers, in order of likely utility.

<th>

  Properties

</th>

<th>

  Description

  描述

</th>

<td style="vertical-align: top">

  <code>componentInstance</code>

</td>

<td>

  The instance of the component class created by `TestBed.createComponent`.

</td>

<td style="vertical-align: top">

  <code>debugElement</code>

</td>

<td>

  The `DebugElement` associated with the root element of the component.

  The `debugElement` provides insight into the component and its DOM element during test and debugging.
  It's a critical property for testers. The most interesting members are covered [below](#debug-element-details).

</td>

<td style="vertical-align: top">

  <code>nativeElement</code>

</td>

<td>

  The native DOM element at the root of the component.

</td>

<td style="vertical-align: top">

  <code>changeDetectorRef</code>

</td>

<td>

  The `ChangeDetectorRef` for the component.

  The `ChangeDetectorRef` is most valuable when testing a
  component that has the `ChangeDetectionStrategy.OnPush` method
  or the component's change detection is under your programmatic control.

</td>

{@a component-fixture-methods}

ComponentFixture methods

The fixture methods cause Angular to perform certain tasks on the component tree. Call these method to trigger Angular behavior in response to simulated user action.

Here are the most useful methods for testers.

<th>

  Methods

</th>

<th>

  Description

  描述

</th>

<td style="vertical-align: top">

  <code>detectChanges</code>

</td>

<td>

  Trigger a change detection cycle for the component.

  Call it to initialize the component (it calls `ngOnInit`) and after your
  test code, change the component's data bound property values.
  Angular can't see that you've changed `personComponent.name` and won't update the `name`
  binding until you call `detectChanges`.

  Runs `checkNoChanges`afterwards to confirm that there are no circular updates unless
  called as `detectChanges(false)`;

</td>

<td style="vertical-align: top">

  <code>autoDetectChanges</code>

</td>

<td>

  Set this to `true` when you want the fixture to detect changes automatically.

  When autodetect is `true`, the test fixture calls `detectChanges` immediately
  after creating the component. Then it listens for pertinent zone events
  and calls `detectChanges` accordingly.
  When your test code modifies component property values directly,
  you probably still have to call `fixture.detectChanges` to trigger data binding updates.

  The default is `false`. Testers who prefer fine control over test behavior
  tend to keep it `false`.

</td>

<td style="vertical-align: top">

  <code>checkNoChanges</code>

</td>

<td>

  Do a change detection run to make sure there are no pending changes.
  Throws an exceptions if there are.

</td>

<td style="vertical-align: top">

  <code>isStable</code>

</td>

<td>

  If the fixture is currently _stable_, returns `true`.
  If there are async tasks that have not completed, returns `false`.

</td>

<td style="vertical-align: top">

  <code>whenStable</code>

</td>

<td>

  Returns a promise that resolves when the fixture is stable.

  To resume testing after completion of asynchronous activity or
  asynchronous change detection, hook that promise.
  See [above](#when-stable).

</td>

<td style="vertical-align: top">

  <code>destroy</code>

</td>

<td>

  Trigger component destruction.

</td>

{@a debug-element-details}

DebugElement

The DebugElement provides crucial insights into the component's DOM representation.

From the test root component's DebugElement returned by fixture.debugElement, you can walk (and query) the fixture's entire element and component subtrees.

Here are the most useful DebugElement members for testers, in approximate order of utility:

<th>

  Member

  成员

</th>

<th>

  Description

  描述

</th>

<td style="vertical-align: top">

  <code>nativeElement</code>

</td>

<td>

  The corresponding DOM element in the browser (null for WebWorkers).

</td>

<td style="vertical-align: top">

  <code>query</code>

</td>

<td>

  Calling `query(predicate: Predicate<DebugElement>)` returns the first `DebugElement`
  that matches the [predicate](#query-predicate) at any depth in the subtree.

</td>

<td style="vertical-align: top">

  <code>queryAll</code>

</td>

<td>

  Calling `queryAll(predicate: Predicate<DebugElement>)` returns all `DebugElements`
  that matches the [predicate](#query-predicate) at any depth in subtree.

</td>

<td style="vertical-align: top">

  <code>injector</code>

</td>

<td>

  The host dependency injector.
  For example, the root element's component instance injector.

</td>

<td style="vertical-align: top">

  <code>componentInstance</code>

</td>

<td>

  The element's own component instance, if it has one.

</td>

<td style="vertical-align: top">

  <code>context</code>

</td>

<td>

  An object that provides parent context for this element.
  Often an ancestor component instance that governs this element.

  When an element is repeated within `*ngFor`, the context is an `NgForRow` whose `$implicit`
  property is the value of the row instance value.
  For example, the `hero` in `*ngFor="let hero of heroes"`.

</td>

<td style="vertical-align: top">

  <code>children</code>

</td>

<td>

  The immediate `DebugElement` children. Walk the tree by descending through `children`.

  <div class="l-sub-section">

  `DebugElement` also has `childNodes`, a list of `DebugNode` objects.
  `DebugElement` derives from `DebugNode` objects and there are often
  more nodes than elements. Testers can usually ignore plain nodes.

  </div>

</td>

<td style="vertical-align: top">

  <code>parent</code>

</td>

<td>

  The `DebugElement` parent. Null if this is the root element.

</td>

<td style="vertical-align: top">

  <code>name</code>

</td>

<td>

  The element tag name, if it is an element.

</td>

<td style="vertical-align: top">

  <code>triggerEventHandler</code>

</td>

<td>

  Triggers the event by its name if there is a corresponding listener
  in the element's `listeners` collection.
  The second parameter is the _event object_ expected by the handler.
  See [above](#trigger-event-handler).

  If the event lacks a listener or there's some other problem,
  consider calling `nativeElement.dispatchEvent(eventObject)`.

</td>

<td style="vertical-align: top">

  <code>listeners</code>

</td>

<td>

  The callbacks attached to the component's `@Output` properties and/or the element's event properties.

</td>

<td style="vertical-align: top">

  <code>providerTokens</code>

</td>

<td>

  This component's injector lookup tokens.
  Includes the component itself plus the tokens that the component lists in its `providers` metadata.

</td>

<td style="vertical-align: top">

  <code>source</code>

</td>

<td>

  Where to find this element in the source component template.

</td>

<td style="vertical-align: top">

  <code>references</code>

</td>

<td>

  Dictionary of objects associated with template local variables (e.g. `#foo`),
  keyed by the local variable name.

</td>

{@a query-predicate}

The DebugElement.query(predicate) and DebugElement.queryAll(predicate) methods take a predicate that filters the source element's subtree for matching DebugElement.

The predicate is any method that takes a DebugElement and returns a truthy value. The following example finds all DebugElements with a reference to a template local variable named "content":

The Angular By class has three static methods for common predicates:

• By.all - return all elements.

• By.css(selector) - return elements with matching CSS selectors.

• By.directive(directive) - return elements that Angular matched to an instance of the directive class.

---

{@a faq}

Frequently Asked Questions

{@a q-spec-file-location}

Why put spec file next to the file it tests?

It's a good idea to put unit test spec files in the same folder as the application source code files that they test:

• Such tests are easy to find.

• You see at a glance if a part of your application lacks tests.

• Nearby tests can reveal how a part works in context.

• When you move the source (inevitable), you remember to move the test.

• When you rename the source file (inevitable), you remember to rename the test file.

---

{@a q-specs-in-test-folder}

When would I put specs in a test folder?

Application integration specs can test the interactions of multiple parts spread across folders and modules. They don't really belong to any part in particular, so they don't have a natural home next to any one file.

It's often better to create an appropriate folder for them in the tests directory.

Of course specs that test the test helpers belong in the test folder, next to their corresponding helper files.

{@a q-e2e}

Why not rely on E2E tests of DOM integration?

The component DOM tests describe in this guide often require extensive setup and advanced techniques where as the class-only test were comparatively simple.

Why not defer DOM integration tests to end-to-end (E2E) testing?

E2E tests are great for high-level validation of the entire system. But they can't give you the comprehensive test coverage that you'd expect from unit tests.

E2E tests are difficult to write and perform poorly compared to unit tests. They break easily, often due to changes or misbehavior far removed from the site of breakage.

E2E tests can't easily reveal how your components behave when things go wrong, such as missing or bad data, lost connectivity, and remote service failures.

E2E tests for apps that update a database, send an invoice, or charge a credit card require special tricks and back-doors to prevent accidental corruption of remote resources. It can even be hard to navigate to the component you want to test.

Because of these many obstacles, you should test DOM interaction with unit testing techniques as much as possible.