Co-evolution analysis of production and test code by learning association rules of changes

Abstract

Many modern software systems come with automated tests. While these tests help to maintain code quality by providing early feedback after modifications, they also need to be maintained. In this paper, we replicate a recent pattern mining experiment to find patterns on how production and test code co-evolve over time. Understanding co-evolution patterns may directly affect the quality of tests and thus the quality of the whole system. The analysis takes into account fine grained changes in both types of code. Since the full list of fine grained changes cannot be perceived, association rules are learned from the history to extract co-change patterns. We analyzed the occurrence of 6 patterns throughout almost 2500 versions of a Java system and found that patterns are present, but supported by weaker links than in previously reported. Hence we experimented with weighting methods and investigated the composition of commits.