Friday, July 21, 2006

Unit Testing with Mock Objects

Unit testing is a fundamental practice in Extreme Programming, but most non-trivial code is difficult to test in isolation. It is hard to avoid writing test suites that are complex, incomplete, and difficult to maintain and interpret. Using Mock Objects for unit testing improves both domain code and test suites. They allow unit tests to be written for everything, simplify test structure, and avoid polluting domain code with testing infrastructure.

You need to make sure that you test one feature at a time, and you want to be notified as soon as any problem occurs. Normal unit testing is hard because you are trying to test the code from outside.

There is a technique called Mock Objects in which we replace domain code with dummy
implementations that emulate real code. These Mock Objects are passed to the target domain code which they test from inside, also termed as Endo-Testing. This practice is similar to writing code stubs with two interesting differences: we test at a finer level of granularity than is usual, and we use our tests and stubs to drive the development of our production code.

Developing unit tests with Mock Objects leads to stronger tests and to better structure of both domain and test code. Unit tests written with Mock Objects have a regular format that gives the development team a common vocabulary. We believe that code should be written to make it easy to test, and have found that Mock Objects is a good technique to achieve this.

An essential aspect of unit testing is to test one feature at time; you need to know exactly what you are testing and where any problems are. Test code should communicate its intent as simply and clearly as possible. This can be difficult if a test has to set up domain state or the domain code causes side effects. Worse, the domain code might not even expose the features to allow you to set the state necessary for a test.

A Mock Object is a substitute implementation to emulate or instrument other domain code. It should be simpler than the real code, not duplicate its implementation, and allow you to set up private state to aid in testing. The emphasis in mock implementations is on absolute simplicity, rather than completeness. For example, a mock collection class might always return the same results from an index method, regardless of the actual parameters.

Mock Objects are not just stubs

As a technique, Mock Objects is very close to Server Stubs. The main concerns about using Server Stubs are: that stubs can be too hard to write, that the cost of developing and maintaining stubs can be too high, that dependencies between stubs can be cyclic, and that switching between stub and production code can be risky.

Why use Mock Objects?

An important aspect of Extreme Programming is not to commit to infrastructure before you have to. For example, we might wish to write functionality without committing to a particular database. Until a choice is made, we can write a mock class that provides the minimum behaviour that we would expect from our database. This means that we can continue writing the tests for our application code without waiting for a working database. The mock code also gives us an initial definition of the functionality we will require from the database.

Unit tests, as distinct from functional tests, should exercise a single piece of functionality. A unit test that depends on complex system state can be difficult to set up, especially as the rest of the system develops. Mock Objects avoid such problems by providing a lightweight emulation of the required system state. Furthermore, the setup of complex state is localised to one Mock Object instead of scattered throughout many unit tests.

Some unit tests need to test conditions that are very difficult to reproduce. For example, to test server failures we can write a Mock Object that implements the local proxy for the server.

Domain objects often fail some time after an error occurs, which is one reason that debugging can be so difficult. With tests that query the state of a domain object, all the assertions are made together after the domain code has executed. This makes it difficult to isolate the exact point at which a failure occurred. One of the authors experienced such problems during the development of a financial pricing library. The unit tests compared sets of results after each calculation had finished. Each failure required considerable tracing to isolate its cause, and it was difficult to test for intermediate values without breaking encapsulation.

Limitations of Mock Objects

As with any unit testing, there is always a risk that a Mock Object might contain errors, for example returning values in degrees rather than radians. Similarly, unit testing will not catch failures that arise from interactions between components. For example, the individual calculations for a complex mathematical formula might be within valid tolerances, and so pass their unit tests, but the cumulative errors might be unacceptable. This is why functional tests are still necessary, even with good unit tests. Extreme Programming reduces, but does not eliminate, such risks with practices such as Pair Programming and Continuous Integration.

Mock Objects reduce this risk further by the simplicity of their implementations.
In some cases it can be hard to create Mock Objects to represent types in a complex external library. The most difficult aspect is usually the discovery of values and structures for parameters that are passed into the domain code. In an event-based system, the object that represents an event might be the root of a graph of objects, all of which need mocking up for the domain code to work. This process can be costly and sometimes must be weighed against the benefit of having the unit tests. However, when only a small part of a library needs to be stubbed out, Mock Objects is a useful technique for doing so.
One important point that we have learned from trying to retrofit Mock Objects is that, in
statically typed languages, libraries must define their APIs in terms of interfaces rather than classes so that clients of the library can use such techniques.

1 comment:

Prasad said...

let know about the V-model,spiral and waterfall model in detail.