Artifact for ASE 2026 Submission#1047

1. The Benchmark Dataset Dataset- Performance Faults.csv: The core runtime-validated benchmark containing 2,479 bug-fix pairs extracted from six major Java repositories (HBase, Commons Pool, Spring, Guava, Glide, and Elasticsearch).2. Dataset Curation Pipeline (Stages 1–4): 01_mine_repositories.py and 01_mine_commits.py: Implements repository mining to identify candidate performance-relevant commits.02_extract_methods.py: A custom git-diff parser used to isolate before-fix and after-fix method versions.02_generate_jmh_harnesses.py: Automates the construction of JMH microbenchmark harnesses to execute code under controlled conditions.03_dynamic_validation.py: Executes the statistical validation protocol (Wilcoxon signed-rank test, $p < 0.05$) to confirm measurable runtime degradation.3. The FAULTEMBED Framework: 04_generate_embeddings.py: Generates structural vector representations using a frozen GraphCodeBERT backbone to capture data-flow graphs (DFGs).05_faultembed_classifier.py: The primary implementation of the retrieval-augmented classification pipeline (Algorithm 1), integrating FAISS similarity search and Random Forest classification.faultembed_ci.py: A lightweight implementation designed for deterministic screening within performance-aware CI/CD pipelines.4. Evaluation & Baselines 06_baselines.py: Scripts to execute and compare against the 17 baselines evaluated in the study, including rule-based tools (SonarQube, Infer) and LLM prompting paradigms.Technical Requirements: Python 3, Dependencies: transformers (for GraphCodeBERT), faiss-cpu, scikit-learn, and pandas.Java Development Kit (JDK): Required for running the JMH harnesses during the validation stage.1. 基准数据集 性能故障数据集.csv：这是经过运行时验证的核心基准数据集，包含从6个主流Java仓库（HBase、Commons Pool、Spring、Guava、Glide及Elasticsearch）中提取的2479个缺陷修复对。 2. 数据集构建流水线（第1-4阶段）： 01_mine_repositories.py与01_mine_commits.py：实现仓库挖掘流程，用于识别与性能相关的候选提交。 02_extract_methods.py：一款自定义Git差异解析器，用于分离缺陷修复前与修复后的方法版本。 02_generate_jmh_harnesses.py：自动构建Java微基准测试套件（Java Microbenchmark Harness，JMH），以在受控环境下执行代码。 03_dynamic_validation.py：执行统计验证流程（Wilcoxon符号秩检验，$p<0.05$），以确认存在可量化的运行时性能退化。 3. FAULTEMBED框架： 04_generate_embeddings.py：使用冻结的GraphCodeBERT主干网络生成结构向量表示，以捕获数据流图（Data-Flow Graph，DFG）。 05_faultembed_classifier.py：检索增强型分类流水线（算法1）的核心实现，集成FAISS（Facebook AI Similarity Search）相似度搜索与随机森林分类。 faultembed_ci.py：一款轻量级实现，专为性能感知型持续集成/持续交付（Continuous Integration/Continuous Delivery，CI/CD）流水线中的确定性筛查而设计。 4. 评估与基准测试 06_baselines.py：用于执行本研究中评估的17项基准测试并与之对比的脚本，涵盖基于规则的工具（SonarQube、Infer）以及大语言模型（Large Language Model，LLM）提示范式。 技术要求：Python 3；依赖库：transformers（用于GraphCodeBERT）、faiss-cpu、scikit-learn及pandas。 Java开发套件（Java Development Kit，JDK）：验证阶段运行JMH微基准测试套件时所需的依赖。