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

[Russ Allbery]

Graham Inggs <graham.inggs at gmail.com> writes:
> On 16 July 2013 11:37, Andreas Beckmann <anbe at debian.org> wrote:

>> Yes:
>>
>>    dh_shlibdeps -a
>> dpkg-shlibdeps: error: couldn't find library libcuda.so.1 needed by
>> debian/nvidia-profiler/usr/bin/nvprof (ELF format: 'elf64-x86-64';
>> RPATH: '')

> Thanks.  I wonder if we can override this.  My thinking is there's no
> compiling or linking happening when nvidia-cuda-toolkit is built, files
> are just unpacked and put in the right places, unless I missed
> something.

The build system needs access to the shlibs file for the library.  If we
copied that information somewhere else, we'd create the problem where the
two copies could get out of sync, which would create some weird bugs.
It's generally best practice to install all of the libraries used by
non-free binary software during the build so that the normal tools can
generate proper package dependencies using the same techniques that are
used for normal packages.

-- 
Russ Allbery (rra at debian.org)               <http://www.eyrie.org/~eagle/>