Give an Inch and Take a Mile? Effects of Adding Reliable Knowledge to Heuristic Feature Tracing

Tracing features to software artifacts is a crucial yet challenging activity for developers of variability-intensive software projects. Developers can provide feature traces either proactively in a manual and rarely semi-automated way or recover them retroactively where automated approaches mainly rely on heuristics. While proactive tracing promises high reliability as developers know which features they realize when working on them, the task is cumbersome and without immediate benefit. Conversely, automated retroactive tracing offers high automation by employing heuristics but remains unreliable and dependent on the quality of the heuristic. To exploit the benefits of proactive and retroactive tracing while mitigating their drawbacks, this paper examines how providing a minimal seed of accurate feature traces proactively (give an inch) can boost the accuracy of automated, heuristic-based retroactive tracing (take a mile). We examine how comparison-based feature location, as one representative of retroactive feature tracing, can benefit from increasing amounts of proactively provided feature mappings. For retroactive comparison-based feature tracing, we find not only that increasing amounts of proactive information can boost the overall accuracy of the tracing but also that the number of variants available for comparison affects the effectiveness of the combined tracing. As a result, our work lays the foundations to optimize the accuracy of retroactive feature tracing techniques with pinpointed proactive knowledge exploitation.