Hi Janick, On Mon, 06 Sep 2010 11:11:44 +0200, Janick Martinez Esturo <[email protected]> wrote: > -----BEGIN PGP SIGNED MESSAGE----- > Hash: SHA1 > > As the numpy package does officially support python 3 since > last weeks release 1.5.0 I was wondering if there are any > plans to move the pycuda / pyopencl and meshpy packages > to support python 3 as well?
Yes, definitely. It won't happen overnight, but now seems to be the
right time to get this started. Our main dependencies, Boost.Python and
numpy, are available for Py3.
> I know that this would result
> in a lot of work since the buffer interface changed
> in the Python C Api.
Not really. Py{CUDA,OpenCL} consume the buffer interface only in a few
places that could easily be #ifdef'd.
> However, perhaps it will pay out in the long run for the future?
No question--Py3 is the future, but I would prefer 2 to be supported (by
Py{CUDA,OpenCL}) for a while to come.
> I'm asking since I'm starting
> a new project that is supposed to use GPU computing heavily
> and I don't want to rely on python 3 since no new features
> will be integrated in python 2 anymore. Perhaps I can help
> with the transition, I think a common code base which
> can automatically be transformed by 2to3 in the way the
> numpy package is doing it could be an effective way to
> provide both versions.
I agree, let's try 2to3. I've just built myself a Python3 test
environment, but I'm stuck with some numpy build problem...
Andreas
pgp9mugxoNpvn.pgp
Description: PGP signature
_______________________________________________ PyCUDA mailing list [email protected] http://lists.tiker.net/listinfo/pycuda
