[Python-Dev] [docs] undocumented argtypes magic in ctypes?

[Eli Bendersky]

On Wed, Mar 6, 2013 at 8:33 AM, Andrew Svetlov <andrew.svetlov at gmail.com>wrote:

> Looks like bug for me.
>

ctypes seems to auto-convert arguments when argtypes is specified. This
fact is documented. However, I'm not sure whether this auto-conversion is
advanced enough to apply byref. Because otherwise, DIRENT is certainly not
convertible to DIRENT_p

Eli




>
> On Tue, Mar 5, 2013 at 4:26 PM, Eli Bendersky <eliben at gmail.com> wrote:
> > Hello,
> >
> > While playing with ctypes a bit, I noticed a feature that doesn't appear
> to
> > be documented. Suppose I import the readdir_r function (assuming DIRENT
> is a
> > correctly declared ctypes.Structure):
> >
> > DIR_p = c_void_p
> > DIRENT_p = POINTER(DIRENT)
> > DIRENT_pp = POINTER(DIRENT_p)
> >
> > readdir_r = lib.readdir_r
> > readdir_r.argtypes = [DIR_p, DIRENT_p, DIRENT_pp]
> > readdir_r.restype = c_int
> >
> > It seems that I can then call it as follows:
> >
> > dirent = DIRENT()
> > result = DIRENT_p()
> >
> > readdir_r(dir_fd, dirent, result)
> >
> > Note that while readdir_r takes DIRENT_p and DIRENT_pp as its second and
> > third args, I pass in just DIRENT and DIRENT_p, accordingly. What I
> should
> > have done is use byref() on both, but ctypes seems to have some magic
> > applied when argtypes declares pointer types. If I use byref, it still
> > works. However, if I keep the same call and comment out the argtypes
> > declaration, I get a segfault.
> >
> > So, is it a feature that should be documented, explicitly discouraged or
> is
> > it a bug?
> >
> > Eli
> >
> >
> > _______________________________________________
> > docs mailing list
> > docs at python.org
> > http://mail.python.org/mailman/listinfo/docs
> >
>
>
>
> --
> Thanks,
> Andrew Svetlov
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.python.org/pipermail/python-dev/attachments/20130306/d87ac147/attachment.html>