TDD is a process that relies on the repetition of a very short development cycle. It is based on the test-first concept of extreme programming (XP) that encourages simple design with a high-level of confidence. The procedure that drives this cycle is called Red-Green-Refactor.
The procedure itself is simple and it consists of a few steps that are repeated over and over again:
- Write a test
- Run all tests
- Write the implementation code
- Run all tests
- Refactor
- Run all tests
Since a test is written before the actual implementation, it is supposed to fail. If it doesn't, the test is wrong. It describes something that already exists or it was written incorrectly. Being in the green state while writing tests is a sign of a false positive. Tests like these should be removed or refactored.
While writing tests, we are in the red state. When the implementation of a test is finished, all tests should pass and then we will be in the green state.
If the last test failed, the implementation is wrong and should be corrected. Either the test we just finished is incorrect or the implementation of that test did not meet the specification we had set. If any but the last test failed, we broke something and changes should be reverted.
When this happens, the natural reaction is to spend as much time as needed to fix the code so that all tests are passing. However, this is wrong. If a fix is not done in a matter of minutes, the best thing to do is to revert the changes. After all, everything worked not long ago. An implementation that broke something is obviously wrong, so why not go back to where we started and think again about the correct way to implement the test? That way, we wasted minutes on a wrong implementation instead of wasting much more time to correct something that was not done right in the first place. Existing test coverage (excluding the implementation of the last test) should be sacred. We change the existing code through intentional refactoring, not as a way to fix recently written code.
Do not make the implementation of the last test final, but provide just enough code for this test to pass.
Write the code in any way you want, but do it fast. Once everything is green, we have confidence that there is a safety net in the form of tests. From this moment on, we can proceed to refactor the code. This means that we are making the code better and more optimal without introducing new features. While refactoring is in place, all tests should be passing all the time.
If, while refactoring, one of the tests failed, refactoring broke an existing functionality and, as before, changes should be reverted. Not only that, at this stage we are not changing any features, but we are also not introducing any new tests. All we're doing is making the code better while continuously running all tests to make sure that nothing got broken. At the same time, we're proving code correctness and cutting down on future maintenance costs.
Once refactoring is finished, the process is repeated. It's an endless loop of a very short cycle.