Dynamic feature traces: finding features in unfamiliar code

This paper introduces an automated technique for feature location: helping developers map features to relevant source code. Like several other automated feature location techniques, ours is based on execution-trace analysis. We hypothesize that these techniques, which rely on making binary judgments about a code element's relevance to a feature, are overly sensitive to the quality of the input. The main contribution of this paper is to provide a more robust alternative, whose most distinguishing characteristic is that it employs ranking heuristics to determine a code element's relevance to a feature. We believe that our technique is less sensitive with respect to the quality of the input and we claim that it is more effective when used by developers unfamiliar with the target system. We validate our claim by applying our technique to three systems with comprehensive test suites. A developer unfamiliar with the target system spent a limited amount of effort preparing the test suite for analysis. Our results show that under these circumstances our ranking-based technique compares favorably to a technique based on binary judgements.