shubhamugare/infer-pulse-eval

Infer Pulse静态分析评估数据集包含523个从Meta的Infer静态分析器测试套件中提取的C语言函数，主要用于评估大型语言模型在静态分析任务上的表现，尤其是针对C代码的内存安全漏洞检测。数据集中的所有示例都提供了测试划分，并带有地面真实错误的注释。数据集包括51个独特的源文件，5种错误类别，并采用智能匿名化策略以保留语义意义。每个示例包含唯一的标识符、原始和匿名化的函数名称、完整的C函数代码、上下文信息、是否含有错误的布尔值、错误类型、错误行偏移量、原始文件中的绝对行号、错误严重性等级、错误跟踪信息和主要错误类别。

The Infer Pulse Static Analysis Evaluation Dataset contains 523 C functions extracted from Metas Infer static analyzer test suite, designed for evaluating Large Language Models (LLMs) on static analysis tasks, particularly memory safety bug detection in C code. The dataset is provided in a test split with ground truth bug annotations. It includes 51 unique source files, five bug categories, and uses smart anonymization to preserve semantic meaning. Each example in the dataset features a unique identifier, original and anonymized function names, complete C function code, context information, a boolean indicating the presence of bugs, bug types, line offsets, absolute line numbers in the original file, bug severities, bug traces, and the primary bug category.