[Pkg-fglrx-devel] Bug#695884: Re[2]: Bug#695884: amd potentially improper OpenCL dependencies

[Bob Bib]

Sun, 20 Jan 2013, 13:39 +01:00 from Vincent Danjean:
> Le 19/01/2013 23:27, Andreas Beckmann a écrit :
> > Adding Vincent to Cc, perhaps he has some ideas how to solve this
> > globally and properly ...
> > 
> > 
> > On 2013-01-19 22:57, Bob Bib wrote:
> >> Well, I've found one more minor dependencies defect:
> >> both 'amd-clinfo' & 'amd-opencl-icd' binary packages
> >> depend just on 'libopencl1' virtual package
> >> (provided by 'amd-libopencl1', 'nvidia-libopencl1' and
> 'ocl-icd-libopencl1').
> >>
> >>  Thus, I suggest replacing:
> >>
> >> Depends: libopencl1
> 
> Is a Depends on a virtual package really wrong ? Because
> this relation ship is really what we want here. The program
> depends on the ABI provided by any package providing
> libopencl1.
>
> >>  with following:
> >>
> >> Depends: amd-libopencl1 | libopencl1
> > Perhaps ACK, but maybe even
> > ocl-icd-libopencl1 | amd-libopencl1 | libopencl1
> 
> ocl-icd-libopencl1 beeing the only one in main, I think this
> would be better.
> However, I still fail to see the initial problem.

Similar to the NVIDIA's case, AMD's "fglrx-driver":
- it's doubtful that a free "ocl-icd-libopencl1"
can improve its freeness or buglessness;
- the current source package is already in "non-free",
so let it prefer its own native opencl library,
as "amd-opencl-dev" currently does:
Depends: amd-libopencl1 | libopencl1

> Bob, can you elaborate a little?
> 
> Regards,
> Vincent
> 
> >>  or, if you like, even with that:
> >>
> >> Depends: amd-libopencl1
> > NACK
> 
> NACK too. It is not what we want.
> 

OK, then we should also probably replace the change
made in experimental fglrx-driver/1:13.1-1
for "amd-clinfo" & "amd-libopencl1":
Recommends: amd-opencl-icd
 with following:
Recommends: amd-opencl-icd | opencl-icd