Collaborative Research: SHF: Medium: Efficient and Trustworthy Proof Engineering

在验证助手(如CoQ)中对软件进行正式验证可以确定软件的正确性，防止可能导致重大经济损失甚至生命损失的软件错误。不幸的是，验证助手目前不能很好地适应大规模的软件开发，并且在开发时间和专业知识方面都很昂贵。该项目的目标是通过简化大型验证项目的开发和维护的技术来提高证明工程师(即证明助手的用户)的生产率，并增加证明工程师通常使用的工具链中的可信度。该项目的新颖性包括用于证据构建、提取和维护的基于学习和分析的方法，以及用于建立证据助手可信度的测试技术。该项目的影响是提高了生产率和软件质量。该项目开发了一些技术，帮助证明工程师(1)通过学习和执行约定、自动定位相关引理和合成通用不变量来构建证明；(2)通过检查假设违规的运行时监视以及对生成逻辑规范的可执行变体的新支持，来增强从验证的构件中提取可执行代码；以及(3)通过检测脆弱的验证脚本以及学习常见的转换来促进大型验证库的维护。此外，为了增加对证据工程工具链的信任，调查人员开发了针对证据辅助核心组件的测试技术。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

Formal verification of software in a proof assistant (such as Coq) can establish the correctness of software, preventing software bugs that could otherwise lead to significant financial losses or even loss of life. Unfortunately, proof assistants are not currently well adapted to large-scale software development and are expensive to use in terms of both development time and expertise. The goal of this project is to increase productivity of proof engineers (i.e., users of proof assistants) via techniques that simplify development and maintenance of large verification projects, as well as to increase trustworthiness in the toolchain commonly used by proof engineers. The project's novelties include learning-based and analytical approaches for proof construction, extraction, and maintenance, as well as testing techniques for establishing the trustworthiness of proof assistants. The project's impacts are increased productivity and increased software quality.This project develops techniques that help proof engineers (1) construct proofs by learning and enforcing conventions, automatically locating relevant lemmas, and synthesizing generalized invariants; (2) augment the extraction of executable code from verified artifacts with runtime monitoring for checking assumption violations and with novel support for generating executable variants of logical specifications; and (3) facilitate the maintenance of large proof repositories by detecting brittle proof scripts, as well as learning common transformations. Furthermore, to increase trust in the proof engineering toolchain, the investigators develop testing techniques that target the core components of proof assistants.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.