SHF: Medium: Collab Research: Synthesizing Verified Analyzers for Critical Software

一个完整的软件系统的可靠性取决于用来构建它的每个工具的可靠性。这些工具中包括程序分析器，它是用于验证不存在特定类型的错误（如不安全的内存访问）的自动化工具。虽然编译器用于程序优化，软件开发人员用于消除软件缺陷，但程序分析器本身并没有得到验证：它们的可靠性在很大程度上是假定的，并且在当前实践中，它们驻留在软件的可信计算基础上。该项目开发(a)直接从程序分析器的规格中合成程序分析器的基础理论；(b)程序分析器的实际实现；(c)通过正式方法、软件开发和经验案例研究的混合，对基础技术和实现进行严格的评估。这些结果的基础是在实践中广泛采用这些工具的潜力，从而更普遍地提高软件的可靠性。该项目的技术和工具将使声音程序分析器在证明助手中以交互式、大部分自动化的方式演绎合成，并使用带有伽罗瓦连接的抽象解释的计算框架。研究人员首先通过比较现有的工具来评估这种方法：Fiat，一个在定理证明器Coq中用于半自动演绎合成的现有工具，但不支持伽罗瓦连接，以及建设性伽罗瓦连接，一个在Agda语言中嵌入伽罗瓦连接的现有框架，但不支持自动化。研究人员将这些结果与现有的“按结构正确”程序分析器的纸上推导结果以及现有的没有使用抽象解释框架推导的信息流分析器进行了比较。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

The reliability of a complete software system hinges on the reliability of each tool used to construct it. Among these tools are program analyzers which are automated tools for verifying the absence of specific classes of errors such as unsafe memory accesses. While used both for program optimization by compilers, and for eliminating software defects by software developers, program analyzers by themselves are not verified: their reliability is largely assumed and, in current practice, they inhabit a software's trusted computing base. This project develops (a) foundational theories for synthesizing program analyzers directly from their specifications; (b) practical implementations of program analyzers; and (c) rigorous evaluations of both foundational techniques as well as implementations via a mixture of formal methods, software development, and empirical case studies. Underlying these results is the potential for widespread adoption of these tools in practice thus leading to higher reliability of software more generally.The project's techniques and tools will enable the deductive synthesis of sound program analysers in proof assistants in an interactive, mostly-automated style, and using the calculational framework of abstract interpretation with Galois connections. The investigators evaluate this approach by first comparing to existing tools: Fiat, an existing tool for semi-automated deductive synthesis in the theorem prover Coq but which does not support Galois connections, and Constructive Galois Connections, an existing framework for embedding Galois connections in Agda language but which does not support automation. The investigators compare these results with existing on-paper derivations of correct-by-construction program analyzers, as well as existing information flow analyzers which were not derived using the abstract interpretation framework.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.