PyObject_IsInstance() is used incorrectly for testing if Python object is an instance of specified builtin type before direct access to internals of object. This is not correct, because PyObject_IsInstance() checks the __class__ attribute that can be modified and even can be dynamic property. Correct way is to check static type. Proposed patch replaces PyObject_IsInstance() with PyObject_TypeCheck() if this is appropriate. See also similar issues and .
Is there any chance that these changes will break working code, or is it the case that if the current check passes incorrectly one will always get a segfauilt or other error?
> is it the case that if the current check passes incorrectly > one will always get a segfauilt or other error? Yes, that is the case. All four of these checks precede a reference to an structure member that depends on being an exact type or subtype. So, yes they are all necessary to prevent segfaults or other undefined behavior.
Yes, it is the case that if the current check passes incorrectly one will always get a segfauilt or other error. Added tests for types.SimpleNamespace and sqlite3.Cursor. It is not easy to reproduce a bug for StopIterator (not sure it is reproducible), but the code looks definitely erroneous in any case.
