Python pickle `load_reduce` function doesn't check if `args` is a tuple

[Legoclones]

Bug report

Bug description:

When load_reduce() is called in Python pickle, the args variable is pulled off the stack and passed into the functions using the * operator. This syntax is type-permissive, meaning the type can be any iterable that will return the needed arguments.

cpython/Lib/pickle.py

Lines 1723 to 1727 in 29acc08

	def load_reduce(self):
	stack = self.stack
	args = stack.pop()
	func = stack[-1]
	stack[-1] = func(*args)

However, in C _pickle, argtup is pulled off the stack and passed into PyObject_CallObject(), which checks if the args parameter is specifically a PyTuple().

cpython/Modules/_pickle.c

Line 6964 in 29acc08

obj = PyObject_CallObject(callable, argtup);

cpython/Objects/call.c

Lines 460 to 473 in 29acc08

	PyObject_CallObject(PyObject *callable, PyObject *args)
	{
	PyThreadState *tstate = _PyThreadState_GET();
	assert(!_PyErr_Occurred(tstate));
	if (args == NULL) {
	return _PyObject_CallNoArgsTstate(tstate, callable);
	}
	if (!PyTuple_Check(args)) {
	_PyErr_SetString(tstate, PyExc_TypeError,
	"argument list must be a tuple");
	return NULL;
	}
	return _PyObject_Call(tstate, callable, args, NULL);
	}

This means that any non-tuple iterator being used as the arguments passed into the function will cause C _pickle to throw an error, but Python pickle deserialization to unfold just fine.

payload:      b'cbuiltins\nprint\n}R.'

pickle:
None
_pickle.c:    FAILURE argument list must be a tuple
pickletools:
    0: c    GLOBAL     'builtins print'
   16: }    EMPTY_DICT
   17: R    REDUCE
   18: .    STOP
highest protocol among opcodes = 1

I think the easiest way to remedy the discrepancy is to explicitly type check the args parameter in pickle.py's load_reduce() function.

CPython versions tested on:

CPython main branch

Operating systems tested on:

Linux

Linked PRs

• gh-144412: Require tuple args in pickle REDUCE #144419

[serhiy-storchaka]

There is a similar lack of checks in load_newobj() and load_newobj_ex(). I am not sure this is a bug.