I just spent the last 30 minutes debugging a mock object. I was running unit tests through the debugger and inspecting the objects I thought the code was calling. To my surprise this well-defined well-used object was returning an un-expected result in the tests. Only later did I realize the test wasn’t calling the real object. It was calling a mock, and the test writer had incorrectly defined the mock result.

This illustrates why I think isolation testing with mocks can lead to a misplaced sense of code coverage. Lets look at some code so I can explain.

Consider the following

RequestContext target = mock(RequestContext.class);
when(target.getParameterNames().thenReturn( new String[]{ 'marker' } ));
when(target.getBaseUri()).thenReturn(new IRI('http://localhost:8080/atom'));

Since we don’t have access to a real RequestContext() it makes sense to mock this out. Its only when we actually deploy our application do we realize the subtle errors here.

The actual call to target.getBaseUri() returns http://localhost:8080/atom/ instead of http://localhost:8080/atom In addition, this bit of code target.getTargetPath() returns '/foo/bar?marker=1' instead of '/foo/bar'

Both of these issue are impossible to detect without actually running your code with a real request that returns a real RequestContext().

This illustrates the biggest downside to using mocks for testing. When you isolate a test with mocks your almost guaranteed your classes are going to pass the unit tests because you are in complete control of the inputs your class is subjected too. In a real world application this is simply not true.

One of the values of good testing is to validate the class works within a certain context. By separating your classes from the context they run in; you completely negate a huge benefit to code testing.

The need for functional tests

In contrast to isolated unit tests, functional tests exercise your code with a real world context.

Functional tests have the following benefits

  1. Broader strokes of code coverage
  2. Tracing errors with a debugger can be much faster than sorting through unit tests guessing what part of the system is acting up, and tweaking unit tests
  3. Users can submit inputs that cause the error, which can and should be easily turned into a functional test.
  4. Confidence your application delivers the service it was designed to provide. The user — and by extension you — don’t care if all the classes work as expected in an isolated environment. They do care if your application provides the functionality promised.

Don’t get me wrong I’m not anti mock. I just don’t subscribe to the notation that EVERY class must have an isolation test. Using a mock; like any good tool — at the right time and for the right purpose is important. Many of the classes we write are better served in a functional test. We still get the same code coverage, and the test is less fragile.