Predicting Bug Inducing Commits Using Commit-State Transition Analysis

Continuous software engineering promotes the use of a release-small, release-often deployment model, with a key focus on the shift-left approach, where quality is assessed as early as possible in the software development life-cycle. To effectively implement this shift-left strategy, developers need tools and techniques that can provide early warnings when a module may be entering a failure-prone state following certain edits. In this paper, we propose a fault-proneness detection framework that analyzes the evolution of quality and technical debt metrics as a system progresses from one commit to the next. This framework can be applied in real-time, as code is committed. Specifically, a metrics vector defines the state of each commit. Commit sequences then form a commit state transition model, which can be analyzed to identify transition paths most likely to lead to an imminent bug-inducing commit. Our approach has shown promising results when evaluated using the SonarQube Technical Debt open-source dataset.