The Effect of Testability on Fault Proneness: A Case Study of the Apache HTTP Server

Numerous studies have identified measures that relate to the fault-proneness of software components. An issue practitioners face in implementing these measures is that the measures tend to provide predictions at a very high level, for instance the per-module level, so it is difficult to provide specific recommendations based on those predictions. We examine a more specific measure, called software testability, based on work in test case generation. We discuss how it could be used to make more specific code improvement recommendations at the line-of-code level. In our experiment, we compare the testability of fault prone lines with unchanged lines. We apply the experiment to Apache HTTP Server and find that developers more readily identify faults in highly testable code. We then compare testability as a fault proneness predictor to McCabe´s cyclomatic complexity and find testability has higher recall.