Making Software Verification Tools Really Work


We discuss problems and barriers which stand in the way of producing verification tools that are robust, scalable and integrated in the software development cycle. Our analysis is that these barriers span a spectrum from theoretical, through practical and even logistical issues. Theoretical issues are the inherent complexity of program verification and the absence of a common, accepted semantic model in tools. Practical hurdles include the challenges arising from real-world systems features, such as floating-point arithmetic and weak memory. Logistical obstacles we identify are the lack of standard benchmarks to drive tool quality and efficiency, and the difficulty for academic research institutions of allocating resources to tool development. We propose simple measures which we, as a community, could adopt to make the design of serious verification tools easier and more credible. Our long-term vision is for the community to produce tools that are indispensable for a developer but so seamlessly integrated into a development environment, as to be invisible.