Comparing spectrum based fault localisation against test-to-code traceability links

The recent shift towards automated software tests stimulated research interest in fault localisation. Fault localisation addresses the question which program elements need to be fixed to repair a failing test. The current state of the art in that field is named spectrum based fault localisation, which relies on dynamic coverage information from both failing and passing test cases to pinpoint the faulty program elements. This is in sharp contrast with the naive approach which extracts traceability links between the test code and the program elements under test and enumerates those until the faulty element is found. In this paper we ask ourselves the question whether the state-of-the-art approach (spectrum based fault localisation) is so much better than the naive approach (test-to-code traceability). We demonstrate on 178 defects from three representative projects in the recent Defects4J dataset that spectrum based fault localisation does not perform better than test-to-code traceability. This implies that future improvements in spectrum based fault localisation should also be compared against naive approaches, such as test-to-code traceability.

Publication year:2018

Keywords:P1 Proceeding

BOF-keylabel:yes

Authors from:Government

Accessibility:Open