import atheris with atheris.instrument_imports(): from math import isnan import sys import warnings import tomli_w import tomli # Disable any caching used so that the same lines of code run # on a given input consistently. tomli._re.cached_tz = tomli._re.cached_tz.__wrapped__ # Suppress all warnings. warnings.simplefilter("ignore") def test_one_input(input_bytes: bytes) -> None: # We need a Unicode string, not bytes fdp = atheris.FuzzedDataProvider(input_bytes) data = fdp.ConsumeUnicode(sys.maxsize) try: toml_obj = tomli.loads(data) except (tomli.TOMLDecodeError, RecursionError): return except BaseException: print_err(data) raise try: recovered_data = tomli_w.dumps(toml_obj) except RecursionError: return except BaseException: print_err(data) raise roundtripped_obj = tomli.loads(recovered_data) normalize_toml_obj(roundtripped_obj) normalize_toml_obj(toml_obj) if roundtripped_obj != toml_obj: sys.stderr.write( f"Original dict:\n{toml_obj}\nRoundtripped dict:\n{roundtripped_obj}\n" ) sys.stderr.flush() print_err(data) raise Exception("Dicts not equal after roundtrip") def print_err(data): codepoints = [hex(ord(x)) for x in data] sys.stderr.write(f"Input was {type(data)}:\n{data}\nCodepoints:\n{codepoints}\n") sys.stderr.flush() def normalize_toml_obj(toml_obj: dict) -> None: """Make NaNs equal when compared (without using recursion).""" to_process = [toml_obj] while to_process: cont = to_process.pop() for k, v in cont.items() if isinstance(cont, dict) else enumerate(cont): if isinstance(v, float) and isnan(v): cont[k] = "nan" elif isinstance(v, (dict, list)): to_process.append(v) def main(): # For possible options, see https://llvm.org/docs/LibFuzzer.html#options fuzzer_options = sys.argv atheris.Setup(fuzzer_options, test_one_input) atheris.Fuzz() if __name__ == "__main__": main()