When Testing Meets Refactoring: Catalogue, Detection, and Recommendation
TL;DR
This paper presents a unified catalogue of test refactoring patterns derived from a mixed-method approach involving repository mining, developer surveys, and analysis of StackOverflow discussions, to establish a rigorous benchmark for evaluating future automated recommendation systems, including those driven by Large Language Models.
Abstract
Automated software testing is critical for modern software development, yet test code itself requires maintenance and refactoring to prevent decay. While theoretical test-specific code smells are well-documented, the practical automation of test refactoring—both detection and recommendation—remains incipient, leading to fragmented knowledge and inadequate tooling. This paper bridges this gap by presenting a unified catalogue of test refactoring patterns derived from a mixed-method approach involving repository mining, developer surveys, and analysis of StackOverflow discussions. To operationalize this catalogue, we extend RefactoringMiner with test-aware detection capabilities and introduce PatternTracker, a tool for identifying and tracking the evolution of test patterns across version history. Leveraging this infrastructure, we analyzed 1.7 million Java projects to generate a high-confidence ground-truth dataset of test refactorings. Our results validate the prevalence of these patterns in open-source projects and establish a rigorous benchmark for evaluating future automated recommendation systems, including those driven by Large Language Models.