BEARS-BENCHMARK

BEARS-BENCHMARK是一个包含251个可重现错误的Java错误基准，用于自动程序修复研究。该数据集由巴西联邦大学的研究人员创建，旨在通过使用持续集成（CI）构建状态来识别错误和修补程序版本候选，从而收集和存储错误。数据集中的错误来自72个不同的项目，涵盖了多种领域和项目规模，提供了广泛的项目多样性。BEARS-BENCHMARK的设计使得研究社区可以通过自动创建带有错误的分支来轻松扩展，这些分支可以用于向BEARS-BENCHMARK仓库提交拉取请求。该数据集的目标是简化实验重现、确保结果的普遍性、允许直接比较竞争方法，并促进研究领域随时间的进展。

BEARS-BENCHMARK is a Java bug benchmark containing 251 reproducible bugs, tailored for automated program repair research. This dataset was developed by researchers from the Federal University of Brazil, with the goal of collecting and curating bugs by leveraging continuous integration (CI) build statuses to identify erroneous code and candidate patch versions. The bugs in this dataset originate from 72 distinct projects spanning multiple domains and project scales, providing extensive project diversity. BEARS-BENCHMARK is engineered to allow the research community to easily extend the benchmark by automatically creating buggy branches, which can be submitted as pull requests to the BEARS-BENCHMARK repository. The core goals of this dataset are to simplify experimental reproducibility, ensure the generalizability of research results, enable direct comparison of competing approaches, and facilitate the advancement of the automated program repair research field over time.