Capture T in `verify`+ so T may be `void` in dart 2: `verify(m.void())` (#90)

Capture T in `verify`+ so T may be `void` in dart 2: `verify(m.void())`

While the semantics here are not yet fully decided, it is likely that
dart 2 will disallow passing void into dynamic. The most backwards
compatible solution here is to parameterize meta function types over T
so that the type `void` is inferred, *possibly* allowing the result of a
void method to be passed in.

This change unblocks us for incrementally rolling out the new void

It may in fact require a change to specify the `void` type directly,
rather than allowing `T` to be void for any `T`, but for now that's not
required and would hurt backwards compatibility since `void` didn't used
to be allowed there.

It also may not be allowed at all under any case, in which case we'd be
in for a fun ride.
3 files changed
tree: 26092cb4680ddcd89f4c27ffdd53df38c9d78cac
  1. .gitignore
  2. .travis.yml
  7. analysis_options.yaml
  8. lib/
  9. pubspec.yaml
  10. test/
  11. tool/

Mock library for Dart inspired by Mockito.

Pub Build Status

Current mock libraries suffer from specifying method names as strings, which cause a lot of problems:

  • Poor refactoring support: rename method and you need manually search/replace it's usage in when/verify clauses.
  • Poor support from IDE: no code-completion, no hints on argument types, can't jump to definition

Dart's mockito package fixes these issues - stubbing and verifying are first-class citizens.

Let's create mocks

import 'package:mockito/mockito.dart';

// Real class
class Cat {
  String sound() => "Meow";
  bool eatFood(String food, {bool hungry}) => true;
  int walk(List<String> places);
  void sleep() {}
  void hunt(String place, String prey) {}
  int lives = 9;

// Mock class
class MockCat extends Mock implements Cat {}

// mock creation
var cat = new MockCat();

Let's verify some behaviour!

//using mock object
//verify interaction

Once created, mock will remember all interactions. Then you can selectively verify whatever interaction you are interested in.

How about some stubbing?

// Unstubbed methods return null:
expect(cat.sound(), nullValue);

// Stubbing - before execution:
expect(cat.sound(), "Purr");

// You can call it again:
expect(cat.sound(), "Purr");

// Let's change the stub:
expect(cat.sound(), "Meow");

// You can stub getters:
expect(cat.lives, 9);

// You can stub a method to throw:
when(cat.lives).thenThrow(new RangeError('Boo'));
expect(() => cat.lives, throwsRangeError);

// We can calculate a response at call time:
var responses = ["Purr", "Meow"];
when(cat.sound()).thenAnswer(() => responses.removeAt(0));
expect(cat.sound(), "Purr");
expect(cat.sound(), "Meow");

By default, for all methods that return a value, mock returns null. Stubbing can be overridden: for example common stubbing can go to fixture setup but the test methods can override it. Please note that overridding stubbing is a potential code smell that points out too much stubbing. Once stubbed, the method will always return stubbed value regardless of how many times it is called. Last stubbing is more important, when you stubbed the same method with the same arguments many times. In other words: the order of stubbing matters, but it is meaningful rarely, e.g. when stubbing exactly the same method calls or sometimes when argument matchers are used, etc.

A quick word on async stubbing

Using thenReturn to return a Future or Stream will throw an ArgumentError. This is because it can lead to unexpected behaviors. For example:

  • If the method is stubbed in a different zone than the zone that consumes the Future, unexpected behavior could occur.
  • If the method is stubbed to return a failed Future or Stream and it doesn't get consumed in the same run loop, it might get consumed by the global exception handler instead of an exception handler the consumer applies.

Instead, use thenAnswer to stub methods that return a Future or Stream.

// BAD
    .thenReturn(new Future.value('Stub'));
    .thenReturn(new Stream.fromIterable(['Stub']));

    .thenAnswer((_) => new Future.value('Stub'));
    .thenAnswer((_) => new Stream.fromIterable(['Stub']));

If, for some reason, you desire the behavior of thenReturn, you can return a pre-defined instance.

// Use the above method unless you're sure you want to create the Future ahead
// of time.
final future = new Future.value('Stub');
when(mock.methodThatReturnsAFuture()).thenAnswer((_) => future);

Argument matchers

// You can use arguments itself:

// ... or collections:

// ... or matchers:

// ... or mix aguments with matchers:
when(cat.eatFood(argThat(startsWith("dry")), true).thenReturn(true);
expect(cat.eatFood("fish"), isTrue);
expect(cat.walk(["roof","tree"]), equals(2));
expect(cat.eatFood("dry food"), isFalse);
expect(cat.eatFood("dry food", hungry: true), isTrue);

// You can also verify using an argument matcher:

// You can verify setters:
cat.lives = 9;

If an argument other than an ArgMatcher (like any, anyNamed(), argThat, captureArg, etc.) is passed to a mock method, then the equals matcher is used for argument matching. If you need more strict matching consider use argThat(identical(arg)).

Verifying exact number of invocations / at least x / never


// Exact number of invocations:

// Or using matcher:

// Or never called:

Verification in order


Verification in order is flexible - you don't have to verify all interactions one-by-one but only those that you are interested in testing in order.

Making sure interaction(s) never happened on mock


Finding redundant invocations


Capturing arguments for further assertions

// Simple capture:
expect(verify(cat.eatFood(captureAny)).captured.single, "Fish");

// Capture multiple calls:
expect(verify(cat.eatFood(captureAny)).captured, ["Milk", "Fish"]);

// Conditional capture:
expect(verify(cat.eatFood(captureThat(startsWith("F")).captured, ["Fish"]);

Waiting for an interaction

// Waiting for a call:
await untilCalled(cat.chew()); //completes when cat.chew() is called

// Waiting for a call that has already happened:
await untilCalled(cat.eatFood(any)); //will complete immediately

Resetting mocks

// Clearing collected interactions:

// Resetting stubs and collected interactions:
expect(cat.eatFood("Fish"), false);


// Spy creation:
var cat = spy(new MockCat(), new Cat());

// Stubbing - before execution:

// Using mocked interaction:
expect(cat.sound(), "Purr");

// Using a real object:
expect(cat.lives, 9);


// Print all collected invocations of any mock methods of a list of mock objects:
logInvocations([catOne, catTwo]);

// Throw every time that a mock method is called without a stub being matched:

Strong mode compliance

Unfortunately, the use of the arg matchers in mock method calls (like cat.eatFood(any)) violates the Strong mode type system. Specifically, if the method signature of a mocked method has a parameter with a parameterized type (like List<int>), then passing any or argThat will result in a Strong mode warning:

[warning] Unsound implicit cast from dynamic to List<int>

In order to write Strong mode-compliant tests with Mockito, you might need to use typed, annotating it with a type parameter comment. Let's use a slightly different Cat class to show some examples:

class Cat {
  bool eatFood(List<String> foods, [List<String> mixins]) => true;
  int walk(List<String> places, {Map<String, String> gaits}) => 0;

class MockCat extends Mock implements Cat {}

var cat = new MockCat();

OK, what if we try to stub using any:


Let's analyze this code:

$ dartanalyzer --strong test/cat_test.dart
Analyzing [lib/cat_test.dart]...
[warning] Unsound implicit cast from dynamic to List<String> (test/cat_test.dart, line 12, col 20)
1 warning found.

This code is not Strong mode-compliant. Let's change it to use typed:

$ dartanalyzer --strong test/cat_test.dart
Analyzing [lib/cat_test.dart]...
No issues found

Great! A little ugly, but it works. Here are some more examples:

when(cat.eatFood(typed(any), typed(any))).thenReturn(true);

Named args require one more component: typed needs to know what named argument it is being passed into:

when(cat.walk(typed(any), gaits: typed(any, named: 'gaits')))

Note the named argument. Mockito should fail gracefully if you forget to name a typed call passed in as a named argument, or name the argument incorrectly.

One more note about the typed API: you cannot mix typed arguments with null arguments:

when(cat.eatFood(null, typed(any))).thenReturn(true); // Throws!
    typed(any))).thenReturn(true); // Works.

How it works

The basics of the Mock class are nothing special: It uses noSuchMethod to catch all method invocations, and returns the value that you have configured beforehand with when() calls.

The implementation of when() is a bit more tricky. Take this example:

// Unstubbed methods return null:
expect(cat.sound(), nullValue);

// Stubbing - before execution:

Since cat.sound() returns null, how can the when() call configure it?

It works, because when is not a function, but a top level getter that returns a function. Before returning the function, it sets a flag (_whenInProgress), so that all Mock objects know to return a “matcher” (internally _WhenCall) instead of the expected value. As soon as the function has been invoked _whenInProgress is set back to false and Mock objects behave as normal.

Be careful never to write when; (without the function call) anywhere. This would set _whenInProgress to true, and the next mock invocation will return an unexpected value.

The same goes for “chaining” mock objects in a test call. This will fail:

var mockUtils = new MockUtils();
var mockStringUtils = new MockStringUtils();

// Setting up mockUtils.stringUtils to return a mock StringUtils implementation

// Some tests

// Instead use this:

This fails, because verify sets an internal flag, so mock objects don't return their mocked values anymore but their matchers. So mockUtils.stringUtils will not return the mocked stringUtils object you put inside.

You can look at the when and Mock.noSuchMethod implementations to see how it‘s done. It’s very straightforward.

NOTE: This is not an official Google product