We distributed the set among the Archive admins and start with two each
of those that are dependencies of later builds. I started building and
evaluating cuda-cudart and cuda-cupti.

Start with cuda-cudart

# License

All good AFAICS

# Package namespace

There are some close misses like:
nvidia-cuda-toolkit/resolute 12.4.131~12.4.1-8 amd64   NVIDIA CUDA development 
toolkit
nvidia-cuda-toolkit-doc/resolute 12.4.1-8 all   NVIDIA CUDA and OpenCL 
documentation
nvidia-cuda-toolkit-gcc/resolute 12.4.1-8 amd64   NVIDIA CUDA development 
toolkit (GCC compatibility)

No conflict to block you, but I assume those might conflict one day.
I know from the tracking document in section "nvidia-cuda-toolkit retirement" 
that you are after that - so that is even less of a blocker. Timing is extra 
tight, but ok for the path we are on.

# Binary namespace

We all have discussed enough about the /usr/lib - it has been accepted
for now by the TB with the order to work with Nvidia to get that away in
the future. So I first thought the following warning would be non
silenced about just that - but it seems to be more on inspection:

W: cuda-cudart-13-1 source: binaries-have-file-conflict cuda-cudart-13-1 
cuda-cudart-dev-13-1 usr/local/cuda-13.1/lib64
W: cuda-cudart-13-1 source: binaries-have-file-conflict cuda-cudart-13-1 
cuda-driver-dev-13-1 usr/local/cuda-13.1/lib64
W: cuda-cudart-13-1 source: binaries-have-file-conflict cuda-cudart-dev-13-1 
cuda-driver-dev-13-1 usr/local/cuda-13.1/lib64

But looking into a build that I did I see each of them have

 832 lrwxrwxrwx root/root         0 2026-02-06 22:17
./usr/local/cuda-13.1/lib64 -> targets/x86_64-linux/lib

Which is at least unexpected, does that not make all of them conflicting 
installs?
They are not packages that have a "Breaks" or "Conflicts" statement.

I test installed this from the PPA and it gladly works these days.
I'm sure if those would not be symlinks it would break, but being symlinks 
makes this ok.

Seen in lintian:

W: cuda-cudart-13-1 source: binaries-have-file-conflict cuda-cudart-13-1 
cuda-cudart-dev-13-1 usr/local/cuda-13.1/lib64
W: cuda-cudart-13-1 source: binaries-have-file-conflict cuda-cudart-13-1 
cuda-driver-dev-13-1 usr/local/cuda-13.1/lib64
W: cuda-cudart-13-1 source: binaries-have-file-conflict cuda-cudart-dev-13-1 
cuda-driver-dev-13-1 usr/local/cuda-13.1/lib64

But it works and after install I see:

dpkg -S /usr/local/cuda-13.1/lib64
cuda-driver-dev-13-1, cuda-cudart-dev-13-1, cuda-culibos-dev-13-1, 
cuda-cccl-13-1, cuda-cudart-13-1: /usr/local/cuda-13.1/lib64

Thereby this is negligible.

If you want to resolve it down the road you could be inspired by how
this is handled usually when it is a real file. In that case you often
have some "foo-common" package that brings the thing everyone need and
every package needing it depends on "foo-common".

If you decide against resolving consider a lintian override with a
comment why the decision was taken.

# Misc

I can't resist to hate binary-only content, yes I know you target multiverse 
because you are aware of that. So I can't stop you but let me express my 
distaste of it, whenever you can have a chance to influence that it becomes a 
source upload - do it!
Making them better serviceable even if we are left alone and inspectable and 
... $default-argument :-/
I tried to convince myself to at least appreciate the faster build as it does 
not build.

-- 
You received this bug notification because you are a member of Ubuntu
Bugs, which is subscribed to Ubuntu.
https://bugs.launchpad.net/bugs/2141744

Title:
  [needs-packaging] cuda-13-1

To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+bug/2141744/+subscriptions


-- 
ubuntu-bugs mailing list
[email protected]
https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs

Reply via email to