xtekky/gpt4free · error · ImportError

Failed to load workspace module '{name}' ({source_path}):\n{

Error message

Failed to load workspace module '{name}' ({source_path}):\n{traceback.format_exc()}

What it means

Thrown by the .pa.py sandbox loader when compile() succeeded but exec() of the module body raised any exception — e.g. NameError, ImportError from a blocked module, ZeroDivisionError at import time, or a raised error inside a decorator. The whole traceback is embedded in the ImportError message. The loader deliberately raises rather than swallowing, so a half-initialized module never enters sys.modules.

Source

Thrown at g4f/mcp/pa_provider.py:368

    module_globals["__package__"] = module.__package__
    module.__dict__.update(module_globals)

    try:
        compiled = compile(code, str(source_path), "exec")
    except SyntaxError:
        raise ImportError(
            f"Syntax error in workspace module '{name}' "
            f"({source_path}):\n{traceback.format_exc()}"
        )

    # Execute in the current thread (no timeout — module loading is expected
    # to be fast and we need the module object synchronously).
    prev_depth = sys.getrecursionlimit()
    sys.setrecursionlimit(MAX_RECURSION_DEPTH)
    try:
        exec(compiled, module.__dict__, module.__dict__)  # noqa: S102
    except Exception:
        raise ImportError(
            f"Failed to load workspace module '{name}' "
            f"({source_path}):\n{traceback.format_exc()}"
        )
    finally:
        sys.setrecursionlimit(prev_depth)

    sys.modules[name] = module
    return module


# ---------------------------------------------------------------------------
# Restricted os shim
# ---------------------------------------------------------------------------


def _make_restricted_os() -> types.ModuleType:
    """Return a restricted ``os`` module that only exposes safe, read-only
    attributes (``urandom``, ``name``, ``sep``, ``linesep``, ``altsep``,

View on GitHub (pinned to 973504e177)

Solutions

  1. Read the embedded traceback to find the failing line; move import-time logic into a function invoked later.
  2. Replace blocked imports with allowed ones (only SAFE_MODULES and the allowed g4f subpaths can be imported).
  3. Wrap risky module-level statements in try/except inside the sandbox module if failure is expected.
  4. Re-test locally with the same restricted import policy to reproduce before re-uploading.

Example fix

// before (module.pa.py)
import requests  # blocklisted -> exec fails -> ImportError
DATA = requests.get('https://x').text

// after
ALLOWED = True
def fetch():
    raise NotImplementedError('use the provided tools instead')
Defensive patterns

Strategy: try-catch

Validate before calling

def scan_for_blocked_imports(source: str, allowed: set) -> list:
    import ast
    tree = ast.parse(source)
    blocked = []
    for node in ast.walk(tree):
        if isinstance(node, ast.Import):
            blocked += [a.name for a in node.names if a.name.split('.')[0] not in allowed]
        elif isinstance(node, ast.ImportFrom) and node.level == 0 and node.module:
            if node.module.split('.')[0] not in allowed:
                blocked.append(node.module)
    return blocked

Try / catch

try:
    module = load_workspace_module(name, source_path)
except ImportError as e:
    if "Failed to load workspace module" in str(e):
        logger.error("sandbox module crashed at exec:\n%s", e)
        return None
    raise

Prevention

When it happens

Trigger: Module-level code that imports a blocklisted module (import requests inside the sandbox), references an undefined global, performs network/file work at import time that fails, or a subclass whose decorator executes immediately.

Common situations: Tool modules doing heavy work at top level instead of inside functions; imports of modules not in SAFE_MODULES; code assuming internet access during load in a restricted environment; a dependency on g4f internals not in _ALLOWED_G4F_SUBPATHS.

Related errors

AI-assisted analysis of xtekky/gpt4free@973504e177 (2026-08-14). Data as JSON: /api/errors/1eff0c2c4472386c. Report an issue: GitHub.