[issue39295] usage of bitfields in ctypes structures changed between 3.7.5 and 3.7.6
Matthew Newville
report at bugs.python.org
Tue Jan 14 18:40:54 EST 2020
Matthew Newville <matt.newville at gmail.com> added the comment:
@eryksun Sorry for the imprecision -- I was mixing what we do on Linux and Windows. A minimum verifiable example for Linux/MacOS would be
import ctypes
class bitstruct(ctypes.Structure):
_fields_ = [('addr', ctypes.c_long),
('rbit', ctypes.c_uint, 1),
('wbit', ctypes.c_uint, 1)]
def handler(args):
print("handler: ", args.addr, args.rbit, args.wbit)
callback = ctypes.CFUNCTYPE(None, bitstruct)(handler)
This works with 3.7.5 but raises a TypeError with 3.7.6.
For Windows (or, well, 64-bit Windows, the only kind we bother to support), we find that we have to wrap the function and use a POINTER to the struct, so what we really use is more like
import os, functools
def make_callback(args, func):
""" make callback function"""
@functools.wraps(func)
def wrapped(arg, **kwargs):
if hasattr(arg, 'contents'):
return func(arg.contents, **kwargs)
return func(arg, **kwargs)
if os.name =='nt': # also check for 64-bit
cb = ctypes.CFUNCTYPE(None, ctypes.POINTER(args))(wrapped)
else:
cb = ctypes.CFUNCTYPE(None, bitstruct)(handler)
return cb
callback = make_callback(bitstruct, handler)
> ...
> This seems rights to me. There is no problem passing a pointer
> as a function parameter.
The problem here is that code that worked with 3.7.5 raises a TypeError with 3.7.6.
I don't know that the solution we came up with is actually the best approach. I've asked for such guidance a few times now. I don't know why using a pointer would be required for a structure containing a "u_int, 1", but not for other structures, but any guidance would be much appreciated.
----------
_______________________________________
Python tracker <report at bugs.python.org>
<https://bugs.python.org/issue39295>
_______________________________________
More information about the Python-bugs-list
mailing list