Linear types for large-scale systems verification

Reasoning about memory aliasing and mutation in software verification is a hard problem. This especially true for systems using SMT-based automated theorem provers. Memory reasoning SMT typically requires nontrivial amount of manual effort to specify heap invariants, as well extensive alias from the solver. In this paper, we present hybrid approach that combines linear types with reasoning. We integrate into Dafny, language an backend, show two approaches complement each other. By separating conditions, reduce solving time. At same time, expressiveness queries extends flexibility type system. particular, it allows our system easily correctly mix nonlinear data novel ways, encapsulating inside vice-versa. formalize core extensions, prove soundness, provide algorithms checking. evaluate by converting implementation verified storage (about 24K lines code proof) written use extended Dafny. The resulting uses 91% remaining 9%. converted has 28% fewer proofs 30% shorter time overall. discuss development overhead original due highlight improved developer experience when types.