more virtual packages for providing libcuda.so.1 and libOpenCL.so.1

[Graham Inggs]

Hi Alberto / Andreas

On 09/09/2013 14:03, Alberto Milone wrote:
> On 29/08/13 10:42, Graham Inggs wrote:
>> One solution is to separate libopencl1 out of the Nvidia and AMD
>> drivers, however, since the open-source ocl-icd-libopencl1 is able to
>> load all of the available ICDs, could we simply remove libOpenCL.so from
>> the Nvidia and AMD drivers and no longer provide libopencl1?
>> We must keep the Nvidia and AMD ICDs and continue to provide opencl-icd
>> though.
>>
>> Regards
>> Graham
> Hi Graham,
>
> Before I commit to anything, I'd like to discuss this with NVIDIA.
>
> Cheers,
>

@Alberto, don't do anything just yet.
Since nvidia-cuda-toolkit 5.0.35-7 landed in Saucy this weekend and it 
is no longer co-installable with nvidia-319 and nvidia-319-updates.

@Andreas, I have some questions about some of the changes in -7:

- ${package:nvidia-libopencl1} | libopencl1,
+ ${package:nvidia-libopencl1},

Does this mean nvidia-cuda-toolkit can only be used with Nvidia's OpenCL 
loader?
What happens if someone installs nvidia-cuda-toolkit alongside 
python-pythoncl, won't they end up with a "version OPENCL_1.2 not 
defined in file libOpenCL.so.1 with link time reference" error as in LP: 
#1174205?

  package_nvidia-libopencl1 \
-    = nvidia-libopencl1 \
+    = nvidia-libopencl1 (>= $${nvidia:MinVersion}) \
      | nvidia-current

We recently started providing libopencl1 in the Ubuntu Nvidia drivers, 
but it looks like we should be providing nvidia-libopencl1 instead, 
although that won't satisfy the versioned dependency.

@Alberto, would you please reconsider splitting nvidia-libopencl1 and 
libcuda1 into separate packages?

Regards
Graham