Property-based testing (PBT), widely used in functional languages and interactive theorem provers, works by randomly generating many inputs to a system under test. While PBT has also seen some use in low-level languages like C, users in this setting must craft all their own generators by hand, rather than letting the tool synthesize most generators automatically from types or logical specifications. For low-level code with complex memory ownership patterns, writing such generators can waste significant amounts of time. CN, a specification and verification framework for C, features a streamlined presentation of separation logic that is specially tuned to present only “easy” logical problems to an underlying constraint solver. Prior work on the Fulminate testing framework has shown that CN’s streamlined specifications can also be checked effectively at run time, providing an oracle for testing whether a memory state satisfies a pre-or postcondition. We show that the restricted syntax of CN is also a good basis for deriving generators for random inputs satisfying separation-logic preconditions. We formalize the semantics for a DSL describing these generators, as well as optimizations that reorder when values are generated and propagate arithmetic constraints. Using this DSL, we implement a property-based testing tool, Bennet, that generates and runs random tests for C functions annotated with CN specifications. We evaluate Bennet on a corpus of programs with CN specifications and show that it can efficiently generate bug-revealing inputs for heap-manipulating programs with complex preconditions.

@article{paperbot2902,
  title = {Bennet: Randomized Specification Testing for Heap-Manipulating Programs},
  author = {Zain K Aamer and Benjamin C. Pierce},
  journal = {Proceedings of the ACM on Programming Languages},
  volume = {9},
  number = {OOPSLA2},
  year = {2025},
  doi = {10.1145/3764115}
}