crossfuzz 0.0.1

Python harness for the cross_fuzz differential fuzzer

crossfuzz

Python harness for cross_fuzz, a coverage-guided differential fuzzer across C, C++, Go, Java, JavaScript/TypeScript, Python, and Rust. The coordinator sends the same generated input to multiple implementations of the same function, collects coverage from every target, and flags any divergence in outputs.

This package is the Python side: it handles the shared-memory IPC, pipe protocol, and branch-arc coverage collection so you only have to write the target function.

Install

pip install crossfuzz

You also need the crossfuzz coordinator binary — see the main repo for build/install instructions.

Quick start

Write a target script that calls crossfuzz.fuzz(your_function):

import crossfuzz

def my_target(data: bytes) -> bytes:
    # Your code under test. Raise an exception to signal an error.
    return data

crossfuzz.fuzz(my_target)

Point the coordinator at it from a crossfuzz.toml config:

[[target]]
name = "python_impl"
language = "python"
binary = "python3"
args = ["my_target.py"]

Then run the campaign with crossfuzz run crossfuzz.toml.

API

• crossfuzz.fuzz(target, settings=None) — persistent fuzz loop. target(data: bytes) -> bytes.
• crossfuzz.filter(target, settings=None) — persistent filter loop. target(data: bytes) -> tuple[bytes, bool].
• crossfuzz.compare(target, settings=None) — custom comparator loop. target(input: bytes, names: list[str], outputs: list[bytes]) -> str.
• crossfuzz.Settings(instrument=True, warmup=0, transform=False) — configuration. Set instrument=False for thin HTTP-trigger harnesses where coverage comes from an instrumented server process.

Platform

POSIX only — the harness uses mmap and inherited file descriptors 3/4 to talk to the coordinator. Windows is not supported.

License

MIT