TESTEVO-BENCH
收藏资源简介:
TESTEVO-BENCH是由滑铁卢大学与谷歌联合创建的一个可执行、动态更新的基准测试数据集,专注于测试与代码协同演化的研究。该数据集包含1,255个任务,涵盖测试生成与测试更新两个核心赛道,数据来源于152个开源Java项目的59,950条候选协同演化记录,并经过严格的构建、执行与语义验证流程筛选而成。其创建过程通过三阶段自动化流水线实现,包括从提交历史中挖掘候选对、基于执行的清洗以及任务构建,确保每个任务都具备可重现的执行环境与高质量的语义关联。该数据集旨在评估人工智能代理在软件工程中理解代码变化并相应生成或更新测试的能力,为解决测试自动化中的语义对齐与执行可靠性问题提供标准化的评估平台。
TESTEVO-BENCH is an executable and dynamically updatable benchmark dataset jointly created by the University of Waterloo and Google, focusing on research into the co-evolution of tests and code. The dataset comprises 1,255 tasks spanning two core tracks: test generation and test update. It is sourced from 59,950 candidate co-evolution records across 152 open-source Java projects, and has been filtered through rigorous construction, execution, and semantic validation processes. Its creation follows a three-stage automated pipeline, including mining candidate pairs from commit histories, execution-based cleaning, and task construction, ensuring that each task has a reproducible execution environment and high-quality semantic associations. This dataset aims to evaluate the ability of AI Agents to understand code changes and generate or update tests accordingly in software engineering, providing a standardized evaluation platform for addressing the issues of semantic alignment and execution reliability in test automation.




