AromaLIA: a language-independent approach to detect test smells

Abstract

Test smells are indicators of poor practices or suboptimal decisions made during test-code development, which can undermine its maintainability and evolution. This topic has gained increasing relevance in recent years, and numerous studies have proposed techniques to detect test smells. However, most existing solutions support only a very limited set of programming languages, often just one. As a result, applying test smell detection to a new language typically requires developing an entirely new solution, even when similar problems have already been addressed elsewhere. In this context, the objective of this work is to propose a languageindependent approach for detecting test smells. To understand how this topic has been explored in the literature, we conducted a Systematic Mapping Study (SMS) analyzing 117 studies published up to August 2025. From this review, we identified, among other aspects, the most frequently discussed test smells across multiple programming languages and those considered most critical, that is, the ones that require the greatest attention from the community. Based on these findings, we developed our approach and evaluated its effectiveness in C#, Java, JavaScript, TypeScript, and Python across 10 different test smells, comparing it with existing language-specific techniques. The results were promissing: our solution demonstrated superior effectiveness, achieving 97% precision, 96% recall, and a 97% F1-score. To support practical adoption, we also implemented and released a test smell detection tool built upon our languageindependent methodology. The outcomes of this work have the potential to greatly simplify the incorporation of new programming languages into test smell detection workflows, enabling the reuse of prior efforts across diverse ecosystems. This benefits both industry practitioners seeking high-quality testing tools and researchers interested in studying test smells in additional languages.