Smart prediction for refactorings in the software test code

• Luana Martins UFBA
• Carla Bezerra UFC
• Heitor Costa UFLA
• Ivan Machado UFBA

Resumo

Test smells are bad practices to either design or implement a test code. Their presence may reduce the test code quality, harming the software testing activities, primarily from a maintenance perspective. Therefore, defining strategies and tools to handle test smells and improve the test code quality is necessary. State-of-the-art strategies encompass automated support mainly based on hard thresholds of rules, static and dynamic metrics to identify the test smells. Such thresholds are subjective to interpretation and may not consider the complexity of the software projects. Moreover, they are limited as they do not automate test refactoring but only count on developers’ expertise and intuition. In this context, a technique that uses historical implicit or tacit data to generate knowledge could assist the identification and refactoring of test smells. This study aims to establish a novel approach based on machine learning techniques to suggest developers refactoring strategies for test smells. As an expected result, we could understand the applicability of the machine learning techniques to handle test smells and a framework proposal that helps developers in decision-making regarding the refactoring of test smells.