[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Needing branches for releases ?



On 2025-07-07 22:35, Rix, Tom wrote:
> I am interested in expanding my POC to cover more of the stack and have some branching to try out some release and/or ci workflows.

The Debian Janitor [1, 2] is a tool that can automate away a lot of
packaging tasks, especially the lintian-related ones. It is used
extensively by some other teams.

One of its features is the ability to auto-update a package from git,
using routine-update [4]. This works fine for some of the easier
packages, eg: Python modules. Though I haven't tried it, I would expect
this to just work at least for patch releases.

> It is very challenging to have a very fast upstream and a lot of packages. Patch releases happen about every 4-6 weeks.  This takes us in fedora about 2 weeks but this leans on a lot of fedora processes that I am not sure have a Debian equivalent.  If there is a some better/different way folks aways wanted to try out, I would be willing to give those a try otherwise I'll do the best I can to reduce the manual overhead per package.

I guess it depends on how much overhead is involved. For a patch
release, I would expect it to be very low. In a minor release, I would
expect some patches to be dropped (those applied upstream), and possibly
some larger tooling changes (changing the docs build process). No idea
what triggers a major update.

Looking back at the libraries I updated to 6.4, the biggest pain points were
   (1) Updating our custom downstream patches, eg: for changing file
       locations
   (2) Discovering/dealing with new features
   (3) SOVER bumps, though this is mostly just a chore
   (4) The change to the documentation build, though once I had this
       figured out, it was also just a 5-min chore

I think that if we had tracked upstream more closely rather than letting
the versions drift so far back, all of this would have been much
simpler. This applies to rocblas especially, the 5.5 -> 6.4 update is
substantial and I'm glad that Christian Bayle had already done much of
the needed work for 6.1.

Having said that, even with all of the above and the substantial version
delta, the library updates so far didn't really take that much time, and
I might finish the other libraries during DebConf (though all I myself
needed has now been updated, so no promises there).

The bigger problem, I think, are keeping hipcc and other core components
updated. I don't have a feel for that yet.

Best,
Christian

[1]: https://janitor.debian.net/, down at the moment
[2]: https://wiki.debian.org/Janitor
[3]: https://packages.debian.org/sid/routine-update
[4]: https://packages.debian.org/sid/routine-update


Reply to: