May 15, 2013
Legacy Code Without Automated Tests Is Not An Excuse For Less RigourWhile I'm on the subject of bad ideas when refactoring legacy code, I feel I should draw attention to what appears to be a common misunderstanding - even among us experts.
I watch a lot of screencasts where folk demonstrate how they would refactor legacy code. Typically, they start by stating bluntly that you shouldn't refactor code without automated tests.
Then they go on to do exactly that so that they can write their first automated unit test in order to make the code initially testable - usually to introduce some kind of dependency injection.
Some excuse themselves from the need to re-test while they do these initial refactorings because they were using automated refactoring tools. I'm not quite sure how this urban myth got started, but let me burst that bubble right here.
At the time of writing, no refactoring tool is that reliable. Even if you're expert at using the tool, and selecting all the right options for more complex refactorings , which most of us aren't, every once in a while the tool screws up our code. And when I say "once in a while", I mean regularly.
I've learned from the school of Hard Knocks to re-test my code even after using the simplest automated refactorings. Even if those tests run slowly. Even if I have to follow manual test scripts and click the buttons myself.
The reason we fear legacy code is because it is difficult to change. It's difficult to change because it's easy to break. The time to be giving ourselves a hall pass to excuse ourselves from regression testing is not right at the start, when the code is probably at ots most brittle.
In these early stages, when our priority is probably getting fast-running automated tests (i.e., unit tests) in place to enable the kind of architectural refactoring we want to do, we must approach the code with utmost care. That means we need to apply the greatest rigour.
Posted 2 weeks, 6 days ago on May 15, 2013