TDD Takes Years

Unit testing is a practice that takes years to sink in. For many the first experience with the green/red bar is interesting, but not life altering. Maybe it was just a quick demo. They go back to the normal debugging patterns in the IDE or with printing output to the terminal. They try testing for a bit and find battling with legacy code just doesn’t feel like its worth the effort.

How do you effect change here? After experience with several groups I’ve found a gradual approach to work better than total immersion:

  • Exposing the team to the concepts, especially if they haven’t really seen a true TDD session.
  • Explaining the benefits to them as developers with reduction in defects, better code that’s easier to refactor, and knowing that when you get asked to work on a legacy codebase with a decent test harness that you won’t have to cross your fingers that the new feature you added didn’t break some other code.
  • Continuing the conversation through one on ones and other means of how testing is going and where they’re having trouble writing tests or how writing the tests first might work a little better, or even starting with just some larger integration tests. With one team I asked the question of all the developers every week, ‘How many tests did you write this week?’
  • Celebrating as code coverage metrics start climbing and pointing back at the reduction in defects.

While I often hoped the process of becoming real TDD style developers would take people a few months, it often took years. I never saw overnight success with adoption, but after time I started to see ‘aha’ moments among many of the developers:

  • A developer who was dedicated to testing let about 20 unit tests start failing because he’d made some changes to the domain model. He knew he needed to get back to them, but had prioritized some other work. As the tech lead on the project who I’d worked with for years saw the failure email from the CI server he rushed over right away and got the developer refocused on fixing the tests first.
  • Another developer finished off a recent project release and had zero real defects. The code coverage was well over 80% and among the best tested code in the organization. I’d started introducing testing to that developer over 6 years ago, but now he feels bad not writing tests.

As a manager and a person I treasure these stories as I hear them even if it they were years in the making. They’re more important than any kudos on an annual review or a big end of year bonus.