Received from Andreas Kloeckner on Sun, Sep 05, 2010 at 04:21:03PM EDT:
On Fri, 20 Aug 2010 13:34:59 -0400, Lev Givon
> Given that the latest revision of PyCUDA apparently needs to remove
> setuptools before downloading and installing distribute, is it
> effectively no longer possible to build/install PyCUDA locally (i.e.,
> within a user directory) on a Linux system? If I run configure.py as a
> non-root user, setuptools obviously can't be removed if it is
> installed globally.
Humph. That could be an issue.
Ok, suppose you have setuptools globally, and you install a
distribute-based virtualenv (virtualenv --distribute), what errors do
If I create a distribute-based virtualenv named CUDA in my home
directory (using virtualenv 1.4.9), activate it, and then install the
latest revision of pycuda using the installation root ~/CUDA after
specifying the configure options
./configure.py --cudadrv-lib-dir=/usr/lib/nvidia-current,/usr/lib64/nvidia-current \
(which is what I had been previously using to configure pycuda in the
past) the installation completes successfully.
Once you let me know, I'm planning on talking to the distribute
guys about what their suggested path of action is in a case like this.
Sorry for the trouble,