Evaluating the Likelihood of Bug Inducing Commits Using Metrics Trend Analysis
Abstract
Continuous software engineering principles advocate a release-small, release-often process model, where new functionality is added to a system, in small increments and very frequently. In such a process model, every time a change is introduced it is important to identify as early as possible, whether the system has entered a state where faults are more likely to occur. In this paper, we present a method that is based on process, quality, and source code metrics to evaluate the likelihood that an imminent bug-inducing commit is highly probable. More specifically, the method analyzes the correlations and the rate of change of selected metrics. The findings from the technical debt dataset extracted data from SonarQube indicate that before bug-inducing commits, metrics that otherwise are not correlated, suddenly exhibit a high correlation or a high rate of change. This metric behavior can then be used for assessing an impending bug-inducing commit. The technique is programing language agnostic, based on metrics that can be extracted without the use of specialized parsers and can be applied to forewarn developers that a file, or a collection of files, has entered a state where faults are highly probable.