Hello Guix,

some updates related to the release. I think it is important to inform
about this especially since the release candidate is out and hopefully
the release will be out soon as well. I am not sure what date the
release will be, but I am pretty sure I/we won't manage to do it
tommorrow, being the first release target, see
<https://codeberg.org/guix/release-planning/wiki/release-1.5.0-project>

During the week the RC1 is out, couple of issues have been discovered.
Some of them already do have a solution but some don't. Not all of these
issues are completely critical, especially since some of the issues also
presented in 1.4.0 and it seems there aren't really reports of them. But
mostly, it all looks fine.

We have a milestone, so up to date state can always be found there:
<https://codeberg.org/guix/guix/milestone/26686>.

I tried to order the issues by priority. I expect the last three issues
to not be fixed in time for the release and I don't think they matter
that much.

- AArch64 boot issues

  It has been established based on previous e-mails to the mailing list
  that the AArch64 system artifacts are by no means satisfactory. They
  still do not boot on hardware that would definitely be a candidate for
  installation (ie. ThinkPad Snapdragon). Also, on headless systems,
  there is no shell started on the serial port, making installation
  possible only on systems with a display.

  I have already prepared to PRs that should hopefully resolve these
  issues. In case you're an owner of an UEFI AArch64 device, it would be
  helpful if you tried them out (pre-built images are published in the
  PRs). Ideally, the image should be bootable both from USB drives and
  SD cards (though these might be trickier and I expect there will be
  issues picked up over the coming months to make the system
  artifacts better for next release, bootable on more platforms).

  <https://codeberg.org/guix/guix/pulls/5320>,
  <https://codeberg.org/guix/guix/pulls/5346>,
  <https://codeberg.org/guix/guix/pulls/5359>.

- Access to ftp.gnu.org

  Apart from technical problems, there is also still a pending problem
  of access to {alpha,ftp}.gnu.org.

  Unfortunately it's unclear how to proceed with publishing of the
  artifacts. I have requested access, but received no reply from Guix
  maintainers. Efraim wanted to request access, but I haven't heard from
  him in a long time.

- Cannot log in to Gnome through SDDM 
<https://codeberg.org/guix/guix/issues/3629>

  This is a problem on AArch64, where SDDM is the default in
  %desktop-services. That means that users who use the installer to
  install Gnome, will end up with a system where they cannot log in to
  Gnome.

  It seems GDM is supported on AArch64 as well, so one workaround would
  be to change the default on AArch64 to GDM. However, this change seems
  a bit drastic to existing users who might get confused by it, it is
  a breaking change in the end.

- NEWS <https://codeberg.org/guix/guix/issues/5165>

  There has been many changes since 1.4.0, but not many of them are
  documented in the NEWS. Changes worth mentioning have to be gathered
  and added to the file afterwards. In case you remember significant
  changes during the three years, do share them so they are not missed.

- Cancellation of Guix System installation leads to 'corrupted' database
  <https://codeberg.org/guix/guix/issues/5324>

  When in the graphical installer, at the step that performs the system
  initialization, cancellation by C-c will keep the sqlite database in a
  wrong state, leading to `No such file or directory` on .drv files on
  subsequent runs of the installation.

  This can also be an issue in the manual installation, after stopping
  the cow-store service. As long as the cow-service is not stopped, it
  should all be fine.

- QEMU not building on some machines on AArch64 
<https://codeberg.org/guix/guix/issues/2447>

  This is a problem for us, since it affects machines on
  ci.guix.gnu.org. For this release, there is an effort to build the
  artifacts through CI. For the system images, qemu-minimal is
  necessary to generate the images.

  A workaround here could be reintroducing a patch that disabled
  migration-test. It has been removed, because on x86_64 there are no
  longer any issues. And neither on some AArch64 machines, but seems
  that not all.

  Still, even after this is fixed, it takes CI quite a large amount of
  time to build everything for AArch64. So while we've got the tarball
  few days after the build was handed to CI, the iso and qcow2 probably
  still could not have been biult (even if it wasn't for the QEMU
  build issues). This has been known beforehand since the AArch64 builds
  are behind.

- Plasma not building on AArch64

  This is fixed already, but in RC1, choosing plasma in the installer
  will lead to long build process only to meet with a failure.

  See <https://codeberg.org/guix/guix/pulls/4846> in case you're
  interested in details.

Issues that I think we can leave if there is no time to fix them,
follow. I will definitely discuss them with rest of the release-team to
confirm they also don't think they're critical, though.

- guix system switch-generation picks wrong bootloader on new system
  <https://codeberg.org/guix/guix/issues/5109>

  This is visible mainly on AArch64, where grub-pc is not supported and
  switch-generation is sometimes impossible, like from generation two to
  first one, failing on bootloader installation.

- Shepherd (and the Guix Installer) hangs when date changes by more than
  a ~day <https://codeberg.org/guix/guix/issues/5115>

  This has been an issue also in 1.4.0 and I couldn't find any opened
  issues related to the installation itself. It can happen for example
  on systems that do not have hardware clock or if the hardware clock
  has been reset due to discharged battery. The date has to be set then,
  because otherwise sites do not load due to certificates not being
  valid. This is still done manually, though. The installer does not
  ship with NTP. So the user could be somewhat aware of this being
  caused by the date change they just did.

  It is unlikely this will be fixed in 1.5.0, I think. The issue is in
  Guile Fibers and it has been known for more than two years. While
  lately there has been activity in the issue, there is still no
  definitive fix published.

- Generating source tarball with Guix and through CI
  <https://codeberg.org/guix/guix/issues/4769>

  As part of the effort of getting the artifacts generated through CI,
  it would be great if the source was also generated by it. There is a
  `dist-package` procedure in Guix. However, it has problems, currently
  it leaves in store paths, through the various patch phases. Moreover
  these patches are necessary for running `make`, so it's not as easy to
  get rid of them. We cannot tolerate having local store paths in the
  source tarball.

  I have started work on this
  <https://codeberg.org/guix/guix/pulls/4978>, but I was unable to
  finish it and had other more pressing issues to look at.

  It thus seems that for this release, we will be left with generation
  of the source manually through the Makefile's `dist` target, like was
  also the case with previous releases. Well still, it's a huge step
  forward in the other artifacts that are generated through CI.

Phew, sorry for such a long e-mail. In case you are interested,
definitely feel free to help with any of the mentioned issues, it wil be
welcome. :) Let's align in the associated issues and/or PRs for the issue.

Rutherther

Reply via email to