paperbot · PL 论文追踪

全部论文 待读 收藏

Don’t Retest Everything: Auditable Regression Scoping for Safety-Critical Systems

不要重测一切:安全关键系统的可审计回归范围界定
ASE Proceedings of the 41st IEEE/ACM International Conference on Automated Software Engineering2026引用 1 inbox
Francesco Basciani, Daniele Masti, Patrizio Pelliccione, Alberto Petrucci, Franco Raimondi, Emerson Sales, Paolo Serri
crossref: 2026-10-09

AI 速览

一句话:提出面向安全关键系统的可审计回归范围界定框架 CoCoS,并给出工业验证与复制材料。

论文针对安全关键系统中回归测试范围难以界定、全量重测成本高且需可审计的问题,提出 CoCoS 框架,用于确定变更后需重测的函数并支持审计。复制包给出工业验证流程、CoCoS 源码与配置脚本,可在 cocos-only 与 diffsitter 模式下复现实验;还包含匿名专家画像、场景级定量结果、变更函数与需重测函数统计、从业者反馈及反馈到工具改进的追踪。通过专家访谈与焦点小组会议记录,验证框架在实际工业环境中的适用性、可审计性与可复现性。

问题
安全关键系统中回归测试若全量重测成本过高,但缩减测试范围又必须保证安全性与可审计性。
方法
提出 CoCoS 回归范围界定与审计框架,并提供复制包、配置脚本及在 cocos-only 与 diffsitter 模式下的实验流程。
评估
通过工业验证、匿名专家画像、场景级定量结果、定性反馈和会议记录评估,并追踪从业者反馈到工具改进。
原文摘要(Abstract)

Replication Package: "Don’t Retest Everything: Auditable Regression Scoping for Safety-Critical Systems" CoCoS Validation Materials This package contains validation materials for the CoCoS study. Participant and organization identifiers are anonymized. Technical repository and function identifiers used in the quantitative validation tables are preserved where they do not identify the authors. The material documents: the industrial validation process; the anonymized profiles of the experts involved; scenario-level quantitative results; qualitative feedback and resulting refinements; interview/focus-group notes and normalized English meeting minutes for the documented design and validation meetings. Contents Logs (logs.zip): Raw experimental data, test logs, and execution traces used in the analysis. CoCoS Repository (CoCoS-repo.zip): Source code and tools for the CoCoS framework, including scripts for regression scoping and auditing. Configuration scripts (CoCoS-configuration-scripts.zip): The source script used to run CoCoS in "cocos-only" and "diffsitter" mode validation_process.md: overview of the validation process and meeting-based evaluation method. industrial_expert_profiles.md / industrial_expert_profiles.csv: anonymized expert profiles. development_environment.md: summary of the development environment and toolchain used during validation. scenario_quantitative_results.md / scenario_quantitative_results.csv: quantitative results and codebase statistics. scenario_function_results.csv: changed functions and functions requiring retest in the reported scenarios. practitioner_feedback_summary.md: concise qualitative feedback summary. feedback_to_refinements.csv: traceability between practitioner feedback and tool refinements. meeting_index.csv: index of documented meetings. meeting_minutes.zip: interview/focus-group notes and normalized English minutes for each documented meeting. translation_and_anonymization_note.md: note on translation, normalization, and anonymization. Scope The package does not include proprietary source code, raw logs, patches, internal paths, organization names, or participant identities. The meeting minutes are not verbatim transcripts. They are normalized English minutes derived from the original notes, with sensitive details removed. Setup Instructions Unzip the Archives: unzip logs.zip unzip CoCoS-repo.zip unzip CoCoS-configuration-scripts.zip Replicating Results: Follow the instructions in CoCoS-repo/README.md to run the regression scoping experiments. Use the provided logs to validate or reproduce the analysis. If you want to use the script place them in the same directory where CoCoS-repo has been extracted.

链接与引用

DOI 原文 · PDF(开放获取) ·

BibTeX
@inproceedings{paperbot12575,
  title = {Don’t Retest Everything: Auditable Regression Scoping for Safety-Critical Systems},
  author = {Francesco Basciani and Daniele Masti and Patrizio Pelliccione and Alberto Petrucci and Franco Raimondi and Emerson Sales and Paolo Serri},
  booktitle = {ASE},
  year = {2026},
  doi = {10.1145/3832783.3837416}
}