paperbot · PL 论文追踪

全部论文 待读 收藏

P2 Defects4REST: A Benchmark of Real-World Defects to Enable Controlled Testing and Debugging Studies for REST APIs

Defects4REST:面向 REST API 受控测试与调试研究的真实缺陷基准
ICSE Proceedings of the 2026 IEEE/ACM 48th International Conference on Software Engineering2026匹配度 2/10 · 该论文是 REST API 测试与调试的真实缺陷基准,属于软件测试/调试方向;与研究画像关注的静态分析、Agentic/LLM 程序、浮点程序分析、贝叶斯程序分析等核心方向没有直接交集。其缺陷基准构建思路虽与“重要场景缺乏 benchmark”有一定共鸣,但任务领域和技术路线均偏离主线。引用 2 inbox
Rahil Mehta, Pushpak Katkhede, Manish Motwani
crossref: 2026-09-12

AI 速览

一句话:构建首个从 GitHub 挖掘的 REST API 真实缺陷基准,含 607 个缺陷、分类体系与缺陷/修复版本,并揭示现有测试工具仅能检测 10% 的缺陷。

现有 REST API 测试与调试研究常任意选取 API 进行评估,缺乏对真实项目中可自动检测或修复缺陷的系统认识。本文提出 Defects4REST——首个公开的 REST API 项目缺陷基准,通过聚类、主题建模与人工验证的半自动化流程,从开源 GitHub 仓库中挖掘出 50 个项目的 607 个缺陷,并归纳为 6 类、13 个子类的分类体系。每个缺陷关联 issue 报告、开发者修改后的补丁文件与提交信息,并提供便捷的缺陷版本与修复版本访问及工件分析能力。作者用 11 个项目的 30 个缺陷样本评估了四款最先进的 REST API 测试工具,发现仅检测出 10%(3/30)的真实缺陷,原因包括语义验证、状态化测试、基础设施感知测试等九项根本性局限。基准与 issue 挖掘框架均已开源,以支持 REST API 测试与调试技术的可复现研究。

问题
REST API 测试与调试研究缺少真实世界缺陷基准,现有评估依赖任意选取的 API,无法反映可自动检测或修复的真实缺陷分布与特征。
方法
通过聚类、主题建模与人工验证相结合的半自动化流程,从 GitHub 开源仓库中挖掘 REST API 缺陷,构建带分类体系、issue 报告、补丁与提交信息的缺陷基准,并提供缺陷/修复版本访问。
评估
在来自 11 个项目的 30 个缺陷样本上评估四款最先进的 REST API 测试工具,仅检测出 10%(3/30)的缺陷,并归纳出语义验证、状态化测试、基础设施感知测试等九项局限。
原文摘要(Abstract)

Existing research on REST API testing and debugging often relies on arbitrarily selected APIs for evaluation, leaving a critical gap in our understanding of real world defects in REST API based projects, particularly those that can be automatically detected or repaired. To address this gap, we present Defects4REST, the first public benchmark of defects in REST API projects, systematically mined from open source GitHub repositories. Using a semi-automated process that combines clustering, topic modeling, and manual validation, we build a dataset of 607 REST API defects from 50 projects and organize them into a taxonomy of 6 defect types and 13 sub-types. Each defect is linked to an issue report, the developer-modified versions of the patched files, and the corresponding commit messages. Defects4REST features easy access to the buggy and patched versions of projects for each defect and enables analysis of the associated artifacts. To evaluate current automated testing capabilities, we assess four state-of-the-art REST API testing tools on a sample of 30 defects from 11 projects. Our evaluation reveals that tools detect only 10% (3/30) of real-world defects due to nine fundamental limitations including semantic validation, stateful testing, and infrastructure-aware testing. We release both the benchmark and the issue mining framework to support reproducible research in REST APIs, enabling the development and evaluation of REST API testing and debugging techniques.

链接与引用

DOI 原文 · PDF(开放获取) ·

BibTeX
@inproceedings{paperbot11716,
  title = {Defects4REST: A Benchmark of Real-World Defects to Enable Controlled Testing and Debugging Studies for REST APIs},
  author = {Rahil Mehta and Pushpak Katkhede and Manish Motwani},
  booktitle = {ICSE},
  year = {2026},
  doi = {10.1145/3744916.3787787}
}