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

Andreas Beckmann anbe at debian.org
Thu Jul 4 15:01:12 UTC 2013


On 2013-06-28 13:35, Graham Inggs wrote:
> On 28 June 2013 13:23, Alberto Milone <alberto.milone at canonical.com> wrote:
> 
>>> They have a real package named libcuda1, and a real package named
>>> nvidia-libcopencl1 which provides the libopencl1 virtual package.
>>
>> Ok, let's stick with the original plan then (as we won't provide real
>> packages in Ubuntu)
>>
> 
> I think Andreas meant there is already a libopencl1 virtual package, there
> is no need to use a new one.
> Maybe he can clarify for us before you go ahead.

How does Ubuntu handle the (non-)coinstallability of nvidia-current,
nvidia-304, nvidia-310, ...?

For CUDA, you should have a
  Provides: libcuda-5.0-1
in those packages, but probably no Conflicts/Breaks/Replaces.

And soon you need to provide additionally
  Provides: libcuda-5.5-1

This is because there are no versioned Provides.


nvidia-cuda-toolkit will get
  Build-Depends: libcuda1 (>= 3XX) | libcuda-5.0-1
and the current beta will get
  Build-Depends: libcuda1 (>= 3YY) | libcuda-5.5-1
(and the Depends in libcudart* will be similar).


For OpenCL it depends on whether we need to distinguish versions.
If we can go with only one version for now, I would suggest to use the
already existing libopencl1 virtual package.

If we need specifically 1.2 or later, use libopencl-1.2-1 as the virtual
package name.


Andreas



More information about the pkg-nvidia-devel mailing list