Zero-cost meta-programmed stateful functors in F*
share
Writing code is hard; proving it correct is even harder. As the scale of verified software projects reaches new heights, the problem of efficiently verifying large amounts of software becomes more and more salient. Nowhere is this issue more evident than in the context of verified cryptographic libraries. To achieve feature-parity and be competitive with unverified cryptographic libraries, a very large number of algorithms and APIs need to be verified. However, the task is oftentimes repetitive, and factoring out commonality between algorithms is fraught with difficulties, requiring until now a significant amount of manual effort. This paper shows how a judicious combination of known functional programming techniques leads to an order-of-magnitude improvement in the amount of verified code produced by the popular HACL* cryptographic library, without compromising performance. We review three techniques that build upon each other, in order of increasing sophistication. First, we use dependent types to crisply capture the specification and state machine of a block algorithm, a cryptographic notion that was until now only informally and imprecisely specified. Next, we rely on partial evaluation to author a higher-order, stateful functor that transforms any unsafe block API into a safe counterpart. Finally, we rely on elaborator reflection to automate the very process of authoring a functor, using a code-rewriting tactic. This culminates in a style akin to templatized C++ code, but relying on a userland tactic and partial evaluation, rather than built-in compiler support.
READ FULL TEXT