End of Line blog

Thoughts on software development, by Adam Ruka

Testing with Doubles, or why Mocks are Stupid – Part 2

Part 1 | Part 3 | Part 4

In Part 1, we saw the definition of the various kinds of Test Doubles. What was not covered, however, were the guidelines on when to prefer using one over the other (or, alternatively, why do we actually need more than one kind?). Part 2 and 3 deal with this topic.

Object methods classification

But before we can answer that question, we need to introduce a classification of methods in object-oriented programming. All methods can be characterised on two axises:

  1. Depending on the direction, methods can be either incoming or outgoing. When somebody calls a method on your object, it’s considered incoming; when your object calls methods on it’s collaborators, they’re outgoing. This classification obviously depends on the perspective of the object you’re looking from – one object’s outgoing message is another object’s incoming one. As we’re talking about tests, we usually consider this classification from the point of view of the class under test – so the behaviors we want to test are incoming, while the calls made to dependencies are outgoing.
  2. A method can be either a Query or a Command. Queries don’t affect the receiver’s state, and are executed for their return value; Commands tell the object to do something (and thus change it’s internal state), and are executed for their side-effects. Command methods may or may not also return a value to the caller.

Obviously, as this is an article about Test Doubles, our class under test has some inconvenient dependencies that we need to deal with. So, we have two basic cases: testing incoming Queries with an outgoing Query dependency, and testing incoming Commands with an outgoing Command dependency.

Testing an incoming Query method with outgoing Query dependencies

Imagine you’re creating an object which is supposed to translate some text from one language to another. To make the translation, you have an external service:

public interface DictionaryService {
    Map<String, String> lookupWords(Set<String> words);

This service returns a mapping from each word in the Set given as the argument to it’s translation – if a translation couldn’t be found, the word itself is used as the value. Your object is supposed to return a String with each word from the original text substituted by it’s translation (OK, so this probably won’t put Google Translate out of business, but bear with me – it’s just an example). The implementation might look something like this:

public class Translator {
    private final DictionaryService dictionaryService;

    public Translator(DictionaryService dictionaryService) {
        this.dictionaryService = dictionaryService;

    public String translate(String text) {
        String[] words = text.split("\\s+");
        Set<String> wordBag = Sets.newHashSet(words);
        Map<String, String> translations = dictionaryService.lookupWords(wordBag);

        return Stream.of(words).map(translations::get).collect(Collectors.joining(" "));

We want to unit-test this piece of code. Let’s see how each Test Double fares in this.

Attempt #1 – Mock

A test using a Mock would looke something like this:

public void test_with_mock() throws Exception {
    DictionaryService dictionaryService = mock(DictionaryService.class);
    Translator translator = new Translator(dictionaryService);

    translator.translate("A dog chases a cat");

    verify(dictionaryService).lookupWords(Sets.newHashSet("A", "dog", "chases", "a", "cat"));

This test passes, but it has some pretty obvious flaws:

So a Mock doesn’t really work here. How would a test using a Stub look like?

Attempt #2 – Stub
public void test_with_stub() throws Exception {
    DictionaryService dictionaryService = mock(DictionaryService.class);
            "A", "A", "a", "a", "dog", "god", "chases", "sesahc", "cat", "tac"));

    Translator translator = new Translator(dictionaryService);
    String result = translator.translate("A dog chases a cat");

    assertThat(result).isEqualTo("A god sesahc a tac");

This test is a lot better. A bigger part of the code is covered (including verifying the Query result), and the test is less brittle – it can actually help us refactor the production code without breaking.

However, while better, this test is still far from perfect:

Is there a way to have your mocking cake and eat it too?

Attempt #3 – Fake

While yes, there is:

public void test_with_fake() throws Exception {
    DictionaryService dictionaryService = new DictionaryService() {
        public Map<String, String> lookupWords(Set<String> argument) {
            return argument.stream().collect(Collectors.toMap(s -> s,
                    s -> new StringBuilder(s).reverse().toString()));
    Translator translator = new Translator(dictionaryService);
    String result = translator.translate("A dog chases a cat");

    assertThat(result).isEqualTo("A god sesahc a tac");

This Fake DictionaryService does the translation by simply reversing it’s input words, similarly to the Stub. The difference is that it doesn’t just blindly return the same Map every time it’s called – this code is actually a correct implementation of the DictionaryService (at least from a purely API standpoint – it’s obviously pretty poor when judging the quality of the translation!). And because of that, this test will not be brittle – it will adapt and stay passing as we correctly refactor the production code, and become red when we make a mistake in that refactoring. This is the kind of test that we want!

And so, my recommendation is:

For testing incoming Query methods, assert on the results of the Query, and avoid Mocks for outgoing dependencies, preferring Stubs (or Fakes) instead.

End of Part 2

That’s it for Part 2. In Part 3 , we’ll tackle testing an incoming Command method with outgoing Command dependencies.

Part 1 | Part 3 | Part 4