Research Report: Not All Specifications Are Created Equal

No ratings

Presented at LangSec Workshop 2024 by

Meng Xu

Software developers who are newly introduced to formal verification often have subtle but distinct interpretations on the roles of specifications (specs). Unsurprisingly, these different interpretations affect the types of specs developers tend to write, which, without coordination, can lead to fragmented assurance guarantee and ultimately impairs the overall effectiveness of the entire formal verification effort. This paper is an experience report on rolling out a formal verification system in a smart contract setting (where correctness matters financially). In this process, we discover three popular views about what “a spec is operationally” among experienced industrial developers yet novice in formal methods: 1) specs are contracts between implementation and functionalities conveyed to end users; 2) specs are extensions to the type system; and 3) specs are definitions of high-level state machines. While which interpretation is closer to the real purpose(s) of specs is a judgement call, one important view is missing: some specs are abstracting specs that lock in requirements or intention (hence the more the better), some specs are proof assistance that facilitates the refinement proof between implementation and abstracting specs and hence, should be written on an as-need basis; while other specs are denotational specs and do not add assurance from a formal methods perspective. Absence this distinction, it is easy to fall into a trap in formal verification where specs accumulate but little assurance is added holistically.