Data flow Subsumption and its Impact on Spectrum-based Fault Localization

Abstract

Debugging aims at finding and correcting software defects. To help the developer, fault localization techniques were developed using association metrics and code coverage data spectra to identify the most suspicious code snippets. They assist the developer by means of a ranking of the most suspicious spectra that guides the developer in his or her ”hunt” for defects. These techniques, when based on data flow spectra, use definition use associations (DUA) for ranking calculation. However, the coverage of given DUAs often guarantees the coverage of other DUAs, in a relationship between DUAs called subsumption. In practice, the subsumption relationship means that if a given DUA is covered, others are also guaranteed to be covered in certain conditions. Based on the subsumption property, this work presents an experiment in which fault localization effectiveness is assessed using only the spectra of the set of unconstrained DUAs, that is, the minimal set of DUAs that may guarantee coverage of all other DUAs of the software under test. For this experiment, we use a subset of programs of the Defects4J repository, data flow spectra, and the Ochiai association metric. Our results compare the rankings produced by the set of unconstrained DUAs against those produced by all DUAs for fault localization. They indicate that most of the faults reached by DUA spectra can be found by inspecting only the unconstrained DUAs.