Open classabbyamp opened 1 week ago
imo haivng linux-lts
follow the second latest LTS release is confusing, and don't see what value it really adds.
If someone wants to use an old kernel, they probably have a good reason for it, and probably will track that specific kernel package manually
I would propose:
linux-mainline
: no change (eg: 6.13)
linux
: remove
linux-lts
: follows latest LTS with nvidia and zfs support (eg: 6.6)
kernel.org offers mainline
, stable
and longterm
versions. maybe we can try and follow that to some extent? i'd find a bit surprising for the linux
package to track lts
and then have another linux-lts
package that tracks something even older.
mainline
should more or less follow upstream, like we are doing right now.
stable
can also follow upstream stable
, but maybe hold a version back, in case there are known issues.
nvidia
and zfs
users can then depend on linux-lts
and not hold back the other versions.
My original idea was to provide a lts package that rarely updated the major version so you don't need to fear breakage once a year. YMMV.
nvidia
andzfs
users can then depend onlinux-lts
and not hold back the other versions.
I take pride that our ZFS support is best in class. Shipping a linux
package that frequently fails to work with ZFS---especially when it is the default kernel for the distribution---is not acceptable. Likewise, a default kernel package that can't provide video to large numbers of users is no good.
As there is momentum behind keeping the linux
package the distribution default, we can't change that now. We are de facto pinning linux
at the latest LTS kernel, so let's codify that. In keeping with the spirit of @leahneukirchen's intention with linux-lts
, we could amend the proposal slightly: linux-lts
can point to a prior LTS kernel such that it is never the same series as linux
and is updated at most annually. Users looking for longer stability should just install a specific series and stop tracking the metas.
Since LTS releases appear at most annually, that sentence is redundant. But fine by me.
This would mean we update LTS to 6.12 in February when 6.13 is mainline? (+- 1 version)
The way I see it, there are two situations which need solving:
IMO, problem 1 should be solved by linux-mainline
, and problem 2 by linux-lts
. this leaves linux
redundant.
If keeping the linux
package is needed, then I propose killing linux-lts
instead, and pinning linux
to the most recent LTS kernel supported by zfs and nvidia
Personally, I don't see the value in having a meta package for the previous LTS release, which will always be at least a year old. What problem is that solving? Users who really need that stability should install a specific series and stop tracking the meta.
If keeping the
linux
package is needed, then I propose killinglinux-lts
instead, and pinninglinux
to the most recent LTS kernel supported by zfs and nvidia
Probably better would be to make linux
a meta package which depends on linux-lts
I don't know that either meta package is needed, but removing one is going to break workflows for people who use them to track new releases. Dropping -lts is probably the least surprising.
based on discussion on IRC, I've added linux-stable
, a package that follows what kernel.org calls stable, and adjusted the description of linux
.
In the long term, it would be nice to make linux
into a virtual package, which would allow for more flexible kernel series switching.
also, I started trying to make linux-headers
a virtual, and it didn't work while linux
/linux-headers
existed as a real package (and maybe another issue, i forget now): https://github.com/void-linux/void-packages/pull/43519
so there are maybe things that this would solve
The current kernel metapackage policy is:
linux-mainline
: follows the latest available kernellinux
: follows the latest available kernel that supports popular dkms modules, likezfs
andnvidia
linux-lts
: follows an LTS release and is behindlinux
In recent times, kernels have been released and gone EOL faster than
zfs
can release build fixes, andnvidia
in particular has had trouble with recent kernels, holding us back from bumpinglinux
to 6.10 or 6.11. In the recent past, we've hadlinux
point to a kernel that was already EOL (6.3 or 6.5), so we risked not getting backports for important fixes.I propose the following changes:
linux-mainline
: no changelinux
: follows the latest LTS releaselinux-lts
: follows the second-latest LTS releaseThis should help us keep
linux
on a supported kernel and provide a more reliable indicator for when we can update the default kernel series, I think.@void-linux/pkg-committers, thoughts?