castironpi at gmail.com
Sun Aug 24 16:26:46 CEST 2008
On Aug 23, 7:11 pm, castironpi <castiro... at gmail.com> wrote:
> On Aug 23, 6:43 pm, marek.ro... at wp.pl wrote:
> > castironpi napisa³(a):
> > > Is there a way to initialize a ctypes Structure to point to an offset
> > > into a buffer? I don't know if the way I'm doing it is supported.
> > There is a high probability you're abusing ctypes too much, but it's
> > possible. The following seems to work:
> > from ctypes import *
> > class S(Structure):
> > _fields_ = [('x', c_uint), ('y', c_int)]
> > rawdata = create_string_buffer('\xEE\xFF\x78\x56\x34\x12\xFF\xFF\xFF
> > \xFF\xAA')
> > # Try to make a structure s of type S which takes its data from
> > rawdata
> > # buffer, starting at index 2
> > s = cast(c_void_p(addressof(rawdata)+2), POINTER(S)).contents
> > print hex(s.x), s.y # Should be 12345678h and -1
> Output is 0x12345678L -1, as you state. I understand that '\xEE\xFF'
> is skipped with addressof(rawdata)+ 2, which makes +2 an offset into
> the buffer.
> At this point, I'd say the use of 'cast' is dubious, but possible to
> support. My problem comes in, in that the buffer I have comes from a
> non-ctypes source. It is a, <drumroll please> mmap.
> My goals in exploring this are persistence and IPC, which are
> certainly not abusing Python too much. 'ctypes' may not be right for
> the job though. The solution I looked at got even worse than 'from
> _ctypes import _cast_addr'. I want a supported way to do it.
The solution could be as simple as returning a ctypes pointer from an
mmap method. Does this require a PEP, or ought a patch to do it?
More information about the Python-list