On June 3, 2020, TechTalk hosted a meetup at which I spoke about managing the various kinds of uncertainty that we routinely encounter on projects that involve legacy code. I presented a handful of ideas for how we might improve our practices related to testing, design, planning, and collaboration. These ideas and practices help us with general software project work, but they help us even more when working with legacy code, since legacy code tends to add significant uncertainty and pressure to every bit of our work. Fortunately, we can build our skill while doing everyday work away from legacy code, then exploit that extra skill when we work with legacy code. Here are some questions that came up during this session and some answers to those questions.
You’ll find the remaining articles in this series here as they are released.
The Risks Related to Refactoring Without Tests
What we should say to project planners who are afraid to let us do refactoring without tests, because some folks in our team are not very good at refactoring and make mistakes? How to convince them it can work for some good programmers?
First, I recognize that if I were the project planner, then I would worry about this, too! I probably don’t know how to judge the refactoring skill of the programmers in the group, so I wouldn’t know whom to trust to refactor without tests. Moreover, I probably can’t calculate the risk associated with refactoring without tests, so I wouldn’t know when to trust anyone to refactor without tests, even if I feel confident in their skill. Once I have thought about these things, it becomes easier to formulate a strategy, because I can ask myself what would make me feel better in this situation? I encourage you to ask yourself this question and write down a few ways that you believe you could increase your confidence from the point of view of the project planner. I can provide a few general ideas here.
I encourage you to build trust by telling the project planner that you are aware of the risks, that you care about protecting the profit stream of the code base, and that you are prepared to discuss the details with them. It often helps a lot simply to show them that you and they are working together to solve this problem and not that you are doing what helps you while creating problems for them.
I would ask the project planners what specifically they are worried about, then matching my strategies to their worries. For example, microcommitting provides one way to manage the risk of refactoring without tests, because it reduces the cost of recovering from a mistake. At the same time, if the project planner worries about different risks than the ones I have thought about, then my strategies might not make them feel any more secure! If I know more about which risks affect them more or concern them more, then I can focus my risk-management work on those points, which also helps to build trust.
I would emphasize that we do not intend to do this as a primary strategy forever. We don’t feel comfortable doing it, either! Even so, we must make progress somehow. We refactor without tests because it would be even more expensive to add “enough” tests than to recover from our mistakes. Of course, we have to be willing to explain our judgment here and we have to be prepared that we are wrong in that judgment! I am always prepared to take suggestions from anyone who has better ideas, but outside of that, they hired me to do good work and make sound decisions, so if they don’t trust me, then I must try to earn their trust or they should give my job to someone that they trust more. I don’t mean this last part as a threat, but merely as a reminder that if they hire me to do the job, but they never trust me, then they should hire someone else!
How about pair-refactoring?
I love it! Refactoring legacy code is often difficult and tiring work, so pair-refactoring fits well even in places where “ordinary” pair programing might not be needed. Refactoring legacy code often alternates periods of difficulty understanding what to do next with long periods of tedious work. Working in pairs significantly increases the profit from both of those kinds of tasks.
You also need this refactoring-without-tests skill, to effectively refactor your tests!
Maybe! I don’t say you need it, but it would probably help you. Your production code helps you to refactor your tests: if you change your tests and they now expect the wrong behavior, then your production code will fail that test for “the right reasons”. It doesn’t provide perfect coverage, but it helps more than you might expect. In that way, the production code helps to test the tests.
Moreover, tests tend to have simpler design than the production code. This means that we might never need to refactor tests in certain ways that feel common when we refactor production code. I almost always write tests with a cyclomatic complexity of 1 (no branching), so the risk when refactoring tests tends to be much lower than when refactoring legacy code. This makes refactoring tests generally safer.