ctypes reimplementation of rshift for c_ulong

bit-shift, ctypes, operator-overloading, python

Solution

Since the `_ctypes._SimpleCData` type doesn't have the `Py_TPFLAGS_CHECKTYPES` flag, 2.x subclasses are treated as old-style numbers that use `__coerce__` in binary operations. See Objects/abstract.c for the calling scheme and the implementation in the function `binary_op1`.

For demonstration purposes this flag can be toggled on the type object, which you only need to define (vaguely with a lot of `void *`) up to the `tp_flags` field.

Hacking the `PyTypeObject`

from ctypes import *
import _ctypes

Py_TPFLAGS_CHECKTYPES = 1 << 4

class PyTypeObject(Structure):
    _fields_ = (('ob_refcnt', c_ssize_t),
                ('ob_type', c_void_p),
                ('ob_size', c_ssize_t),
                ('tp_name', c_char_p),
                ('tp_basicsize', c_ssize_t),
                ('tp_itemsize', c_ssize_t),
                ('tp_dealloc', c_void_p),
                ('tp_print', c_void_p),
                ('tp_getattr', c_void_p),
                ('tp_setattr', c_void_p),
                ('tp_compare', c_void_p),
                ('tp_repr', c_void_p),
                ('tp_as_number', c_void_p),
                ('tp_as_sequence', c_void_p),
                ('tp_as_mapping', c_void_p),
                ('tp_hash', c_void_p),
                ('tp_call', c_void_p),
                ('tp_str', c_void_p),
                ('tp_getattro', c_void_p),
                ('tp_setattro', c_void_p),
                ('tp_as_buffer', c_void_p),
                ('tp_flags', c_long))

Next, create an `unsigned long` subclass, and use the `from_address` factory to create a `PyTypeObject` for it. Get the address with built-in `id`, which is an implementation detail specific to CPython:

class c_ulong(_ctypes._SimpleCData):
    _type_ = "L"

    def __rshift__(self, other):
        print '__rshift__', self, other
        if hasattr(other, 'value'):
            other = other.value
        return c_ulong(self.value >> other)

c_ulong_type = PyTypeObject.from_address(id(c_ulong))

Demo

>>> a = c_ulong(16)
>>> b = c_ulong(2)

>>> a >> b
__rshift__ c_ulong(16L) c_ulong(2L)
c_ulong(4L)

>>> a >> 2
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
TypeError: unsupported operand type(s) for >>: 'c_ulong' and 'int'

The last step failed as expected. Now set the flag:

>>> c_ulong_type.tp_flags |= Py_TPFLAGS_CHECKTYPES

>>> a >> 2
__rshift__ c_ulong(16L) 2
c_ulong(4L)

Problem solved? But that's a hack. Try again with `__coerce__` implemented.

Implement `__coerce__`

class c_ulong(_ctypes._SimpleCData):
    _type_ = "L"

    def __rshift__(self, other):
        print '__rshift__', self, other
        if hasattr(other, 'value'):
            other = other.value
        return c_ulong(self.value >> other)

    def __coerce__(self, other):
        print '__coerce__', self, other
        try:
            return self, self.__class__(other)
        except TypeError:
            return NotImplemented

Demo

>>> a = c_ulong(16)
>>> b = c_ulong(2)

>>> a >> 2
__coerce__ c_ulong(16L) 2
__rshift__ c_ulong(16L) c_ulong(2L)
c_ulong(4L)

>>> 16 >> b
__coerce__ c_ulong(2L) 16
__rshift__ c_ulong(16L) c_ulong(2L)
c_ulong(4L)

Of course it fails if a `c_ulong` can't be created, such as for a `float`:

>>> a >> 2.0
__coerce__ c_ulong(16L) 2.0
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
TypeError: unsupported operand type(s) for >>: 'c_ulong' and 'float'

Problem

Original source