test 0.12.6

test provides a standard way of writing and running tests in Dart.

Writing Tests

Tests are specified using the top-level test() function, and test assertions are made using expect():

import "package:test/test.dart";

void main() {
  test("String.split() splits the string on the delimiter", () {
    var string = "foo,bar,baz";
    expect(string.split(","), equals(["foo", "bar", "baz"]));

  test("String.trim() removes surrounding whitespace", () {
    var string = "  foo ";
    expect(string.trim(), equals("foo"));

Tests can be grouped together using the group() function. Each group's description is added to the beginning of its test's descriptions.

import "package:test/test.dart";

void main() {
  group("String", () {
    test(".split() splits the string on the delimiter", () {
      var string = "foo,bar,baz";
      expect(string.split(","), equals(["foo", "bar", "baz"]));

    test(".trim() removes surrounding whitespace", () {
      var string = "  foo ";
      expect(string.trim(), equals("foo"));

  group("int", () {
    test(".remainder() returns the remainder of division", () {
      expect(11.remainder(3), equals(2));

    test(".toRadixString() returns a hex string", () {
      expect(11.toRadixString(16), equals("b"));

Any matchers from the matcher package can be used with expect() to do complex validations:

import "package:test/test.dart";

void main() {
  test(".split() splits the string on the delimiter", () {
    expect("foo,bar,baz", allOf([

You can use the setUp() and tearDown() functions to share code between tests. The setUp() callback will run before every test in a group or test suite, and tearDown() will run after. tearDown() will run even if a test fails, to ensure that it has a chance to clean up after itself.

import "package:test/test.dart";

void main() {
  var server;
  var url;
  setUp(() async {
    server = await HttpServer.bind('localhost', 0);
    url = Uri.parse("http://${server.address.host}:${server.port}");

  tearDown(() async {
    await server.close(force: true);
    server = null;
    url = null;

  // ...

Running Tests

A single test file can be run just using pub run test:test path/to/test.dart (on Dart 1.10, this can be shortened to pub run test path/to/test.dart).

Single file being run via pub run

Many tests can be run at a time using pub run test:test path/to/dir.

Directory being run via

It's also possible to run a test on the Dart VM only by invoking it using dart path/to/test.dart, but this doesn't load the full test runner and will be missing some features.

The test runner considers any file that ends with _test.dart to be a test file. If you don't pass any paths, it will run all the test files in your test/ directory, making it easy to test your entire application at once.

By default, tests are run in the Dart VM, but you can run them in the browser as well by passing pub run test:test -p chrome path/to/test.dart. test will take care of starting the browser and loading the tests, and all the results will be reported on the command line just like for VM tests. In fact, you can even run tests on both platforms with a single command: pub run test:test -p "chrome,vm" path/to/test.dart.

Restricting Tests to Certain Platforms

Some test files only make sense to run on particular platforms. They may use dart:html or dart:io, they might test Windows' particular filesystem behavior, or they might use a feature that's only available in Chrome. The @TestOn annotation makes it easy to declare exactly which platforms a test file should run on. Just put it at the top of your file, before any library or import declarations:


import "dart:io";

import "package:test/test.dart";

void main() {
  // ...

The string you pass to @TestOn is what's called a "platform selector", and it specifies exactly which platforms a test can run on. It can be as simple as the name of a platform, or a more complex Dart-like boolean expression involving these platform names.

Platform Selector Syntax

Platform selectors can contain identifiers, parentheses, and operators. When loading a test, each identifier is set to true or false based on the current platform, and the test is only loaded if the platform selector returns true. The operators ||, &&, !, and ? : all work just like they do in Dart. The valid identifiers are:

  • vm: Whether the test is running on the command-line Dart VM.

  • dartium: Whether the test is running on Dartium.

  • content-shell: Whether the test is running on the headless Dartium content shell.

  • chrome: Whether the test is running on Google Chrome.

  • phantomjs: Whether the test is running on PhantomJS.

  • firefox: Whether the test is running on Mozilla Firefox.

  • safari: Whether the test is running on Apple Safari.

  • ie: Whether the test is running on Microsoft Internet Explorer.

  • dart-vm: Whether the test is running on the Dart VM in any context, including Dartium. It's identical to !js.

  • browser: Whether the test is running in any browser.

  • js: Whether the test has been compiled to JS. This is identical to !dart-vm.

  • blink: Whether the test is running in a browser that uses the Blink rendering engine.

  • windows: Whether the test is running on Windows. If vm is false, this will be false as well.

  • mac-os: Whether the test is running on Mac OS. If vm is false, this will be false as well.

  • linux: Whether the test is running on Linux. If vm is false, this will be false as well.

  • android: Whether the test is running on Android. If vm is false, this will be false as well, which means that this won't be true if the test is running on an Android browser.

  • posix: Whether the test is running on a POSIX operating system. This is equivalent to !windows.

For example, if you wanted to run a test on every browser but Chrome, you would write @TestOn("browser && !chrome").

Running Tests on Dartium

Tests can be run on Dartium by passing the -p dartium flag. If you're using the Dart Editor, the test runner will be able to find Dartium automatically. On Mac OS, you can also install it using Homebrew. Otherwise, make sure there's an executable called dartium (on Mac OS or Linux) or dartium.exe (on Windows) on your system path.

Similarly, tests can be run on the headless Dartium content shell by passing -p content-shell. The content shell is installed along with Dartium when using Homebrew. Otherwise, you can downloaded it manually from this page; if you do, make sure the executable named content_shell (on Mac OS or Linux) or content_shell.exe (on Windows) is on your system path.

In the future, there will be a more explicit way to configure the location of both the Dartium and content shell executables.

Asynchronous Tests

Tests written with async/await will work automatically. The test runner won't consider the test finished until the returned Future completes.

import "dart:async";

import "package:test/test.dart";

void main() {
  test("new Future.value() returns the value", () async {
    var value = await new Future.value(10);
    expect(value, equals(10));

There are also a number of useful functions and matchers for more advanced asynchrony. The completion() matcher can be used to test Futures; it ensures that the test doesn't finish until the Future completes, and runs a matcher against that Future's value.

import "dart:async";

import "package:test/test.dart";

void main() {
  test("new Future.value() returns the value", () {
    expect(new Future.value(10), completion(equals(10)));

The throwsA() matcher and the various throwsExceptionType matchers work with both synchronous callbacks and asynchronous Futures. They ensure that a particular type of exception is thrown:

import "dart:async";

import "package:test/test.dart";

void main() {
  test("new Future.error() throws the error", () {
    expect(new Future.error("oh no"), throwsA(equals("oh no")));
    expect(new Future.error(new StateError("bad state")), throwsStateError);

The expectAsync() function wraps another function and has two jobs. First, it asserts that the wrapped function is called a certain number of times, and will cause the test to fail if it's called too often; second, it keeps the test from finishing until the function is called the requisite number of times.

import "dart:async";

import "package:test/test.dart";

void main() {
  test("Stream.fromIterable() emits the values in the iterable", () {
    var stream = new Stream.fromIterable([1, 2, 3]);

    stream.listen(expectAsync((number) {
      expect(number, inInclusiveRange(1, 3));
    }, count: 3));

Running Tests with Custom HTML

By default, the test runner will generate its own empty HTML file for browser tests. However, tests that need custom HTML can create their own files. These files have three requirements:

  • They must have the same name as the test, with .dart replaced by .html.

  • They must contain a link tag with rel="x-dart-test" and an href attribute pointing to the test script.

  • They must contain <script src="packages/test/dart.js"></script>.

For example, if you had a test called custom_html_test.dart, you might write the following HTML file:

<!doctype html>
<!-- custom_html_test.html -->
    <title>Custom HTML Test</title>
    <link rel="x-dart-test" href="custom_html_test.dart">
    <script src="packages/test/dart.js"></script>
    // ...

Configuring Tests

Skipping Tests

If a test, group, or entire suite isn't working yet and you just want it to stop complaining, you can mark it as "skipped". The test or tests won't be run, and, if you supply a reason why, that reason will be printed. In general, skipping tests indicates that they should run but is temporarily not working. If they're is fundamentally incompatible with a platform, @TestOn/testOn should be used instead.

To skip a test suite, put a @Skip annotation at the top of the file:

@Skip("currently failing (see issue 1234)")

import "package:test/test.dart";

void main() {
  // ...

The string you pass should describe why the test is skipped. You don't have to include it, but it's a good idea to document why the test isn't running.

Groups and individual tests can be skipped by passing the skip parameter. This can be either true or a String describing why the test is skipped. For example:

import "package:test/test.dart";

void main() {
  group("complicated algorithm tests", () {
    // ...
  }, skip: "the algorithm isn't quite right");

  test("error-checking test", () {
    // ...
  }, skip: "TODO: add error-checking.");


By default, tests will time out after 30 seconds of inactivity. However, this can be configured on a per-test, -group, or -suite basis. To change the timeout for a test suite, put a @Timeout annotation at the top of the file:

@Timeout(const Duration(seconds: 45))

import "package:test/test.dart";

void main() {
  // ...

In addition to setting an absolute timeout, you can set the timeout relative to the default using @Timeout.factor. For example, @Timeout.factor(1.5) will set the timeout to one and a half times as long as the default—45 seconds.

Timeouts can be set for tests and groups using the timeout parameter. This parameter takes a Timeout object just like the annotation. For example:

import "package:test/test.dart";

void main() {
  group("slow tests", () {
    // ...

    test("even slower test", () {
      // ...
    }, timeout: new Timeout.factor(2))
  }, timeout: new Timeout(new Duration(minutes: 1)));

Nested timeouts apply in order from outermost to innermost. That means that "even slower test" will take two minutes to time out, since it multiplies the group's timeout by 2.

Platform-Specific Configuration

Sometimes a test may need to be configured differently for different platforms. Windows might run your code slower than other platforms, or your DOM manipulation might not work right on Safari yet. For these cases, you can use the @OnPlatform annotation and the onPlatform named parameter to test() and group(). For example:

@OnPlatform(const {
  // Give Windows some extra wiggle-room before timing out.
  "windows": const Timeout.factor(2)

import "package:test/test.dart";

void main() {
  test("do a thing", () {
    // ...
  }, onPlatform: {
    "safari": new Skip("Safari is currently broken (see #1234)")

Both the annotation and the parameter take a map. The map's keys are platform selectors which describe the platforms for which the specialized configuration applies. Its values are instances of some of the same annotation classes that can be used for a suite: Skip and Timeout. A value can also be a list of these values.

If multiple platforms match, the configuration is applied in order from first to last, just as they would in nested groups. This means that for configuration like duration-based timeouts, the last matching value wins.

Testing With barback

Packages using the barback transformer system may need to test code that's created or modified using transformers. The test runner handles this using the --pub-serve option, which tells it to load the test code from a pub serve instance rather than from the filesystem.

Before using the --pub-serve option, add the test/pub_serve transformer to your pubspec.yaml. This transformer adds the necessary bootstrapping code that allows the test runner to load your tests properly:

- test/pub_serve:
    $include: test/**_test{.*,}.dart

Note that if you're using the test runner along with polymer, you have to make sure that the test/pub_serve transformer comes after the polymer transformer:

- polymer
- test/pub_serve:
    $include: test/**_test{.*,}.dart

Then, start up pub serve. Make sure to pay attention to which port it's using to serve your test/ directory:

$ pub serve
Loading source assets...
Loading test/pub_serve transformers...
Serving my_app web on http://localhost:8080
Serving my_app test on http://localhost:8081
Build completed successfully

In this case, the port is 8081. In another terminal, pass this port to --pub-serve and otherwise invoke pub run test:test as normal:

$ pub run test:test --pub-serve=8081 -p chrome
"pub serve" is compiling test/my_app_test.dart...
"pub serve" is compiling test/utils_test.dart...
00:00 +42: All tests passed!

Further Reading

Check out the API docs for detailed information about all the functions available to tests.


  • Add a machine-readable JSON reporter. For details, see the protocol documentation.

  • Skipped groups now properly print skip messages.


  • Declare compatibility with Dart 1.14 and 1.15.


  • Fixed a deadlock bug when using setUpAll() and tearDownAll().


  • Add setUpAll() and tearDownAll() methods that run callbacks before and after all tests in a group or suite. Note that these methods are for special cases and should be avoided—they make it very easy to accidentally introduce dependencies between tests. Use setUp() and tearDown() instead if possible.

  • Allow setUp() and tearDown() to be called multiple times within the same group.

  • When a tearDown() callback runs after a signal has been caught, it can now schedule out-of-band asynchronous callbacks normally rather than having them throw exceptions.

  • Don't show package warnings when compiling tests with dart2js. This was accidentally enabled in 0.12.2, but was never intended.


  • If a tearDown() callback throws an error, outer tearDown() callbacks are still executed.


  • Don't compile tests to JavaScript when running via pub serve on Dartium or content shell.


  • Support http_parser 1.0.0.


  • Fix a broken link in the README.


  • Internal changes only.


  • Widen the Dart SDK constraint to include 1.13.0.


  • Make source maps work properly in the browser when not using --pub-serve.


  • Fix a memory leak when running many browser tests where old test suites failed to be unloaded when they were supposed to.


  • Require Dart SDK >= 1.11.0 and shelf >= 0.6.0, allowing test to remove various hacks and workarounds.


  • Add a --pause-after-load flag that pauses the test runner after each suite is loaded so that breakpoints and other debugging annotations can be added. Currently this is only supported on browsers.

  • Add a Timeout.none value indicating that a test should never time out.

  • The dart-vm platform selector variable is now true for Dartium and content shell.

  • The compact reporter no longer prints status lines that only update the clock if they would get in the way of messages or errors from a test.

  • The expanded reporter no longer double-prints the descriptions of skipped tests.


  • Widen the constraint on analyzer to include 0.26.0.


  • Fix an uncaught error that could crop up when killing the test runner process at the wrong time.


  • Add a missing dependency on the collection package.


This version was unpublished due to issue 287.

  • Properly report load errors caused by failing to start browsers.

  • Substantially increase browser timeouts. These timeouts are the cause of a lot of flakiness, and now that they don't block test running there's less harm in making them longer.


This version was unpublished due to issue 287.

  • Fix a crash when skipping tests because their platforms don't match.


This version was unpublished due to issue 287.

  • The compact reporter will update the timer every second, rather than only updating it occasionally.

  • The compact reporter will now print the full, untruncated test name before any errors or prints emitted by a test.

  • The expanded reporter will now always print the full, untruncated test name.


This version was unpublished due to issue 287.

  • Limit the number of test suites loaded at once. This helps ensure that the test runner won't run out of memory when running many test suites that each load a large amount of code.


This version was unpublished due to issue 287.

  • Improve the display of syntax errors in VM tests.

  • Work around a Firefox bug. Computed styles now work in tests on Firefox.

  • Fix a bug where VM tests would be loaded from the wrong URLs on Windows (or in special circumstances on other operating systems).


  • Fix a bug that caused the test runner to crash on Windows because symlink resolution failed.


  • If a future matched against the completes or completion() matcher throws an error, that error is printed directly rather than being wrapped in a string. This allows such errors to be captured using the Zone API and improves formatting.

  • Improve support for Polymer tests. This fixes a flaky time-out error and adds support for Dartifying JavaScript stack traces when running Polymer tests via pub serve.

  • In order to be more extensible, all exception handling within tests now uses the Zone API.

  • Add a heartbeat to reset a test's timeout whenever the test interacts with the test infrastructure.

  • expect(), expectAsync(), and expectAsyncUntil() throw more useful errors if called outside a test body.


  • Convert JavaScript stack traces into Dart stack traces using source maps. This can be disabled with the new --js-trace flag.

  • Improve the browser test suite timeout logic to avoid timeouts when running many browser suites at once.


  • Add a --verbose-trace flag to include core library frames in stack traces.


Test Runner

0.12.0 adds support for a test runner, which can be run via pub run test:test (or pub run test in Dart 1.10). By default it runs all files recursively in the test/ directory that end in _test.dart and aren't in a packages/ directory.

The test runner supports running tests on the Dart VM and many different browsers. Test files can use the @TestOn annotation to declare which platforms they support. For more information on this and many more new features, see the README.

Removed and Changed APIs

As part of moving to a runner-based model, most test configuration is moving out of the test file and into the runner. As such, many ancillary APIs have been removed. These APIs include skip_ and solo_ functions, Configuration and all its subclasses, TestCase, TestFunction, testConfiguration, formatStacks, filterStacks, groupSep, logMessage, testCases, BREATH_INTERVAL, currentTestCase, PASS, FAIL, ERROR, filterTests, runTests, ensureInitialized, setSoloTest, enableTest, disableTest, and withTestEnvironment.

FailureHandler, DefaultFailureHandler, configureExpectFailureHandler, and getOrCreateExpectFailureHandler which used to be exported from the matcher package have also been removed. They existed to enable integration between test and matcher that has been streamlined.

A number of APIs from matcher have been into test, including: completes, completion, ErrorFormatter, expect,fail, prints, TestFailure, Throws, and all of the throws methods. Some of these have changed slightly:

  • expect no longer has a named failureHandler argument.

  • expect added an optional formatter argument.

  • completion argument id renamed to description.


  • Internal code cleanups and documentation improvements.


  • Bumped the version constraint for matcher.


  • Bump the version constraint for matcher.


  • Narrow the constraint on matcher to ensure that new features are reflected in unittest's version.


  • Prints a warning instead of throwing an error when setting the test configuration after it has already been set. The first configuration is always used.


  • Fix bug in withTestEnvironment where test cases were not reinitialized if called multiple times.


  • Add reason named argument to expectAsync and expectAsyncUntil, which has the same definition as expect's reason argument.

  • Added support for private test environments.


  • Refactored package tests.


  • Release test functions after each test is run.



  • Updated maximum matcher version.


  • Removed unused files from tests and standardized remaining test file names.


  • Widen the version constraint for stack_trace.


  • Deprecated methods have been removed: expectAsync0, expectAsync1, and expectAsync2 - use expectAsync instead expectAsyncUntil0, expectAsyncUntil1, and expectAsyncUntil2 - use expectAsyncUntil instead guardAsync - no longer needed protectAsync0, protectAsync1, and protectAsync2 - no longer needed

  • matcher.dart and mirror_matchers.dart have been removed. They are now in the matcher package.

  • mock.dart has been removed. It is now in the mock package.


  • Fixed deprecation message for mock.


  • Moved to triple-slash for all doc comments.


  • DEPRECATED matcher.dart and mirror_matchers.dart are now in the matcher package. mock.dart is now in the mock package.

  • equals now allows a nested matcher as an expected list element or map value when doing deep matching.

  • expectAsync and expectAsyncUntil now support up to 6 positional arguments and correctly handle functions with optional positional arguments with default values.


  • Each test is run in a separate Zone. This ensures that any exceptions that occur is async operations are reported back to the source test case.

  • DEPRECATED guardAsync, protectAsync0, protectAsync1, and protectAsync2 * Running each test in a Zone addresses the need for these methods.

  • NEW! expectAsync replaces the now deprecated expectAsync0, expectAsync1 and expectAsync2

  • NEW! expectAsyncUntil replaces the now deprecated expectAsyncUntil0, expectAsyncUntil1 and expectAsyncUntil2

  • TestCase: Removed properties: setUp, tearDown, testFunction enabled is now get-only * Removed methods: pass, fail, error

  • interactive_html_config.dart has been removed.
  • runTests, tearDown, setUp, test, group, solo_test, and solo_group now throw a StateError if called while tests are running.

  • rerunTests has been removed.

1. Depend on it

Add this to your package's pubspec.yaml file:

  test: "^0.12.6"

2. Install it

You can install packages from the command line:

$ pub get

Alternatively, your editor might support pub. Check the docs for your editor to learn more.

3. Import it

Now in your Dart code, you can use:

import 'package:test/test.dart';


A library for writing dart unit tests.


Email misc@dartlang.org Dart Team