Reproduce “Semantic Fuzzing with Zest” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Rohan Padhye, Caroline Lemieux, Koushik Sen, Mike Papadakis, Yves Le Traon

Zest integrates coverage-guided fuzzing with parametric generators to efficiently explore the semantic analysis stages of programs processing structured inputs, significantly improving code coverage and accelerating the discovery of deep-seated semantic bugs. It addresses limitations of traditional fuzzing for complex inputs and enhances generator-based testing with feedback-directed exploration.

Key insights

  • Mutations on a low-level untyped bit sequence can deterministically produce high-level, syntactically valid structural changes in generator-produced inputs.
  • Explicitly guiding the fuzzer to explore paths reachable by *semantically valid* inputs, not just syntactically valid ones, is crucial for deep program exploration and bug finding.
  • Combining the guaranteed syntactic validity of generators with the feedback-directed search capabilities of fuzzing overcomes a fundamental limitation in testing complex software with structured inputs.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.