gh-138479: Ensure that __typing_subst__ returns a tuple
#138482
Conversation
ZeroIntensity
requested review from
JelleZijlstra and
AlexWaygood
as code owners
ZeroIntensity
added
needs backport to 3.13
| @@ -525,6 +525,15 @@ _Py_subs_parameters(PyObject *self, PyObject *args, PyObject *parameters, PyObje | |||
| return NULL; | |||
| } | |||
| if (unpack) { | |||
| if (!PyTuple_Check(arg)) { | |||
There was a problem hiding this comment.
Do we care about tuple subclasses here? If so, do we want to test this? Or use PyTuple_CheckExact?
There was a problem hiding this comment.
I think it would be fine to do either. It would technically be a breaking change to disallow tuple subclasses here, but I doubt anyone is actually doing that in practice.
There was a problem hiding this comment.
It's an internal undocumented API so I wouldn't feel terrible about disallowing tuple subclasses, but we could go either way.
There was a problem hiding this comment.
Let's just use PyTuple_Check for the time being. There's no drawback to keeping subclass support here, so we might as well keep it.
| "expected a tuple, not %T", arg); | ||
| Py_DECREF(arg); | ||
| return NULL; | ||
| } |
There was a problem hiding this comment.
Can't comment there but are we leaking newargs on line 543 below?
There was a problem hiding this comment.
Looks like it. Do you want me to fix that in this PR?
Objects/genericaliasobject.c
Outdated
| jarg = tuple_extend(&newargs, jarg, | ||
| &PyTuple_GET_ITEM(arg, 0), PyTuple_GET_SIZE(arg)); | ||
| Py_DECREF(arg); | ||
| if (jarg < 0) { | ||
| Py_DECREF(item); | ||
| Py_XDECREF(tuple_args); | ||
| Py_DECREF(newargs); |
There was a problem hiding this comment.
Ah, tuple_extend steals the reference on failure. I added a comment here instead. Thanks for clearing this up.
Objects/genericaliasobject.c
Outdated
| Py_DECREF(item); | ||
| Py_XDECREF(tuple_args); | ||
| PyErr_Format(PyExc_TypeError, | ||
| "expected a tuple, not %T", arg); |
There was a problem hiding this comment.
This error message is not helpful. Please add what method of what type has been called (this was method __typing_subst__ of original argument PyTuple_GET_ITEM(args, iarg)).
There was a problem hiding this comment.
Done, it's now expected __typing_subst__ of %T objects to return a tuple, not %T. Is that good?
Co-authored-by: Serhiy Storchaka <[email protected]>
Also fix incorrect DECREF.
Raise an exception if
__typing_subst__returns a non-tuple object.The test case is a little messy because it's based on fuzzer-generated code, so I can come up with my own repro if people aren't happy with using it.