Karp: a language for NP reductions

In CS theory courses, NP reductions are a notorious source of pain for students and instructors alike. Invariably, students use pen and paper to write down reductions that łworkž in many but not all cases. When instructors observe that a student's reduction deviates from the expected one, they have to manually compute a counterexample that exposes the mistake. In other words, NP reductions are subtle yet, most of the time, unimplemented programs. And for a good reason: there exists no language tailored to NP reductions.

We introduce Karp, a language for programming and testing NP reductions. Karp combines an array of programming languages techniques: language-oriented programming and macros, solver-aided languages, property testing, higherorder contracts and gradual typing. To validate the correctness of Karp, we prove that its core is well-defined. To validate its pragmatics, we demonstrate that it is expressive and performant enough to handle a diverse set of reduction exercises from a popular algorithms textbook. Finally, we report the results from a preliminary user study with Karp.