Home // VALID 2024, The Sixteenth International Conference on Advances in System Testing and Validation Lifecycle // View article
Authors:
Carlos Galindo
Manuel Gregorio
Josep Silva
Keywords: EvoSuite; automated test unit generation
Abstract:
Unit testing is crucial for ensuring software quality and reliability. Although recent advancements in artificial intelligence, particularly Large Language Models (LLMs), offer promise for automating unit test generation, they often struggle with compilation due to an insufficient understanding of specific code rules and execution errors, primarily caused by incorrect assertions. This paper focuses on EvoSuite, a leading state-of-the-art Search-Based Software Testing (SBST) tool that originated in academic research and has proven to be a more reliable alternative for generating unit tests, particularly in Java. EvoSuite excels by directly targeting code coverage and optimizing test generation based on actual program behavior, overcoming many challenges LLMs face. We share our experiences and challenges with EvoSuite across various projects, which have provided valuable insights for its subsequent application in ASys, a system for automatically evaluating Java code. The study explores challenges such as generating tests for overloaded methods and running tests across different environments. We also discuss solutions for these challenges, including method-specific test generation strategies and ensuring test execution compatibility. Our findings highlight the limitations and potential improvements for EvoSuite, offering valuable insights for developers and researchers aiming to enhance automated unit test generation in their projects.
Pages: 15 to 21
Copyright: Copyright (c) IARIA, 2024
Publication date: September 29, 2024
Published in: conference
ISSN: 2308-4316
ISBN: 978-1-68558-199-2
Location: Venice, Italy
Dates: from September 29, 2024 to October 3, 2024