Aslak and I were brought on to a project with the task of refactoring a legacy codebase. A complex application with approximately 30 thousands lines of code, written over a 2 year timeframe by many programmers. The code worked as intended and was actually a success compared to all previous efforts to accomplish the same sort of functionality. The team is somewhat agile and uses CruiseControl for automated builds.
We had some clear directives from the client:
At first glance, the code didn't look too badly written and Clover was reporting over 60% coverage from approximately 1100 Junit tests. Hmm... Diving into the test suite usually seems like a good starting point to figure out where to start.
To my dismay, there were some serious issues with the way that the unit tests were written. I should say so-called unit tests, cause there were some major problems to address:
The majority of those tests weren't actually unit tests at all. Sure, they all extended JUnit TestCase, but most of them did some sort of file I/O, many included calls to Thread.sleep(), and during their execution there would be about 20-30 AWT windows popping up and flashing stuff in my face. As a matter of fact, they were system tests masquerading as unit test.