• TDD steps: Red (when we have failed tests) > Green (all tests are good) > Yellow (refactoring existing code, could imply some failed tests for a while, don’t add new behaviour here!)
  • Separate the three parts of test, should be easy to identify each one. Those parts are: Set, Execute and Check.
  • Always do the smaller step possible, that requires the minimum number of changes to get green.
  • Each new test should fail, don’t add a test that is green. That means that you are testing something already tested, and that also could mean that you have chosen an invalid order of tests or that you have added to much code in the class that we are testing. So you should do the minimum code, no more.
  • If you are adding a new test, and this requires to modify code that is gonna make fail other tests, comment the new test and apply the changes until old tests are good.
  • Unify similar tests when is possible, using tools like Parametizer in Java, DataProvider in PHP… we refactor also our tests.
  • Don’t abuse of assertTrue and assertFalse, don’t give a lot of information about which values are not good.
  • Don’t test a business rule if you don’t understand it, ask business first.
  • Only apply a refactor if that simplify our code, if it’s adding more complexity, wait as much as possible, probably some new business rule it will guide us.
  • If you have to test invalid cases, take the ones that just don’t follow one of the rules.
  • Do the refactor as fast as possible, when the complexity is bigger is much harder.
  • Use descriptive test names.
  • Each test just has to check one rule, not more. This implies having if it’s possible just one assert. If you have more ask yourself if you could split it in two tests. So each test should have only one reason to fail and if one assert fails is not gonna hide the others.
  • Test all known limits. Ex. quality decrease 1 each day, after expired date decrease 2. Important to test same expiration date and after. For that, we could use mutant testing.

Test Doubles



They behave as the real ones but taking some shortcut.


Objects that register their usage so we can check before (with expectations) or after (with spies) what have done.


Objects with predefined data.