On 8 August 2014 19:43, Vincent Danjean <vdanjean...@free.fr> wrote:
> For what I remember, both these packages have the libOpenCL.so file
> so they are buggy (should declare a conflict directly or indirectly)
libOpenCL.so is shipped by amd-opencl-dev, nvidia-opencl-dev and
ocl-icd-libopencl1.
However, the conflicts relationship does not need to be in all three
packages, simply having it in ocl-icd-libopencl1 is enough.
(amd-opencl-dev, nvidia-opencl-dev and ocl-icd-opencl-dev already
conflict with each other because they all provide the opencl-dev
virtual package)
>> Ignoring request to alter fixed versions of bug #755513 to the same values
>> previously set
>> Bug #755513 [ocl-icd-libopencl1] nvidia-opencl-dev: binary conflict with
>> ocl-icd-libopencl1
>> Marked as found in versions ocl-icd/2.1.0-1.
>
> There is nothing new in this package related to this problem.
libOpenCL.so has been in nvidia-opencl-dev since August 2013 and in
amd-opencl-dev since October 2013 (see changelog excerpt below).
The next upload of ocl-icd after that date should have had the
conflicts on amd-app changed to amd-opencl-dev and the conflicts on
nvidia-libopencl1-dev changed to nvidia-opencl-dev.
> So, please, explain your reassign or revert it back.
This bug was preventing nvidia-cuda-toolkit from migrating to testing
and I believe there is an upload to pycuda (which depends on
nvidia-cuda-toolkit) coming soon.
Considering that by shipping libOpenCL.so, ocl-icd-libopencl1 is the
package breaking policy, and that you indicated libOpenCL.so might
move to the ocl-icd-opencl-dev package, I believe it is simpler for
this bug to be fixed in ocl-icd-libopencl1 alone without affecting
amd-opencl-dev and nvidia-opencl-dev as well.
fglrx-driver (1:13.4-4) unstable; urgency=low
* Move the libOpenCL.so symlink from amd-libopencl1 to amd-opencl-dev and
adjust Depends/Breaks/Replaces accordingly.
-- Andreas Beckmann <a...@debian.org> Thu, 17 Oct 2013 22:18:40 +0200
--
To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org
with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
