MCPcopy Create free account
hub / github.com/RUC-NLPIR/WebThinker / reliability_guard

Function reliability_guard

scripts/lcb_runner/evaluation/testing_util.py:678–761  ·  view source on GitHub ↗

This disables various destructive functions and prevents the generated code from interfering with the test (e.g. fork bomb, killing other processes, removing filesystem files, etc.) WARNING This function is NOT a security sandbox. Untrusted code, including, model- generated

(maximum_memory_bytes=None)

Source from the content-addressed store, hash-verified

676
677
678def reliability_guard(maximum_memory_bytes=None):
679 """
680 This disables various destructive functions and prevents the generated code
681 from interfering with the test (e.g. fork bomb, killing other processes,
682 removing filesystem files, etc.)
683 WARNING
684 This function is NOT a security sandbox. Untrusted code, including, model-
685 generated code, should not be blindly executed outside of one. See the
686 Codex paper for more information about OpenAI's code sandbox, and proceed
687 with caution.
688 """
689
690 if maximum_memory_bytes is not None:
691 import resource
692
693 resource.setrlimit(
694 resource.RLIMIT_AS, (maximum_memory_bytes, maximum_memory_bytes)
695 )
696 resource.setrlimit(
697 resource.RLIMIT_DATA, (maximum_memory_bytes, maximum_memory_bytes)
698 )
699 if not platform.uname().system == "Darwin":
700 resource.setrlimit(
701 resource.RLIMIT_STACK, (maximum_memory_bytes, maximum_memory_bytes)
702 )
703
704 faulthandler.disable()
705
706 import builtins
707
708 builtins.exit = None
709 builtins.quit = None
710
711 import os
712
713 os.environ["OMP_NUM_THREADS"] = "1"
714
715 os.kill = None
716 os.system = None
717 os.putenv = None
718 os.remove = None
719 os.removedirs = None
720 os.rmdir = None
721 os.fchdir = None
722 os.setuid = None
723 os.fork = None
724 os.forkpty = None
725 os.killpg = None
726 os.rename = None
727 os.renames = None
728 os.truncate = None
729 os.replace = None
730 os.unlink = None
731 os.fchmod = None
732 os.fchown = None
733 os.chmod = None
734 os.chown = None
735 os.chroot = None

Callers 1

run_testFunction · 0.70

Calls

no outgoing calls

Tested by

no test coverage detected