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

Transition to usrmerge in full swing [was: Re: transition to usrmerge to start around 2022-09-15 (next Thursday)]

On Sat, 2022-09-17 at 02:30 +0100, Luca Boccassi wrote:
> On Thu, 2022-09-15 at 22:57 +0100, Luca Boccassi wrote:
> > On Sat, 2022-09-10 at 15:37 +0200, Ansgar wrote:
> > > Hi,
> > > 
> > > the transition to usrmerge as described in [1] is planned to
> > > start
> > > around 2022-09-15 (next Thursday).
> > > 
> > > init-system-helpers 1.65~exp1 in experimental adds the new
> > > dependency
> > > on
> > > "usrmerge | usr-is-merged" and will be uploaded to unstable to
> > > start
> > > the
> > > transition. Feel free to test and report any issues.
> > > 
> > > Recent versions of debootstrap[2] will setup the usr-is-merged
> > > package to
> > > avoid installing additional dependencies required by usrmerge.
> > > The
> > > usr-is-merged package can also be manually installed in existing
> > > systems
> > > for the same reason.
> > > 
> > > Debian's buildds will continue to use the legacy filesystem
> > > layout
> > > for
> > > Debian 12 (bookworm).
> > > 
> > > We will send an announcement to debian-devel-announce@ once the
> > > upload
> > > to unstable happens.
> > > 
> > > Ansgar
> > > 
> > >   [1]: https://lists.debian.org/debian-ctte/2022/09/msg00005.html
> > >   [2]: debootstrap 1.0.114+deb10u1, 1.0.123+deb11u1, 1.0.127
> > 
> > Quick update: three minor issues where found, two with i-s-h itself
> > (one solved in experimental just now about test deps
> > uninstallability
> > on some ports and one piuparts seemingly false positive about
> > /etc/shells that I'll fix tomorrow), and one in
> > usrmerge+nspawn+arm64
> > [0]. I have just come back home from LPC so did not have much time
> > today, will have a look at the latter two tomorrow and then upload
> > to
> > unstable once the situation is clearer.
> > 
> > [0] https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=1019575
> All 3 issues are solved or pending:
> 1) init-system-helpers build issue on some ports architectures has
> been
> fixed (new test dependency fakeroot is not available everywhere, made
> optional)
> 2) salsa CI issue with piuparts job image not being built by
> debootstrap/mmdebstrap and thus not installing usrmerge/usr-is-
> merged,
> causing a false positive when usrmerge is pulled in by the package-
> under-test thus falsely attributing the /etc/shells change to it,
> fixed
> by installing usrmerge in the pre-built image, MR pending:
> https://salsa.debian.org/salsa-ci-team/pipeline/-/merge_requests/372
> 3) systemd-nspawn on aarch64 merged systems creates a /lib64 ->
> /usr/lib/aarch64-linux-gnu symlink which is not created by
> debootstrap/mmdebstrap, so the usr-is-merged preinst check doesn't
> expect it and erroneously fails. Fixed with NMU to just enforce that
> the arch-specific libdirs are symlinks to /usr/lib* instead of
> exactly
> the same dirname under /usr, as this is not the right place to care
> about that detail, the only important thing in this context is the
> target being under /usr.
> Unless any new issue pops up, I'll upload i-s-h to unstable to start
> the transition tomorrow evening.

It's been almost two weeks now, so another (and probably final) status
update, as init-system-helpers was uploaded and has migrated to
bookworm and everything seems to be going very smoothly, without any
major issue reported. A few minor issues and corner cases were reported
and fixed in various places, with contributions from multiple people.
In usrmerge itself:

- https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=1020312
- https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=1019985
- https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=1019506
- https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=1019575
- https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=1020549
- https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=926699

Some of these are fixed in unstable, with the new version migrating in
a couple of days.

A few issues in other packages were found and fixed too - these are
pre-existing bugs that were never found before:

- https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=1020426
- https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=1020783

Note that the second one means cdebootstrap currently doesn't work
(usual multiarch dependency resolution issue, with a custom parser not
recognizing the :any suffix and failing), a patch is available and a MR
is open.

A new version of debootstrap has also just migrated - I made a mistake
with the July update, and failed to notice that both usr-is-merged and
usrmerge are installed, which means unnecessary dependencies are pulled
in by default. The new version fixes that. It will also be available in
backports on Monday, and once it has marinated enough I intend to push
to to proposed-updates too.

There are 3 issues with a pending or incomplete solutions:

- a MR is pending on glibc to create the biarch symlinks on the
multiarch libc6 packages too (same fix is already available for biarch
- usrmerge needs some mildly invasive changes for hurd-i386 which we
are a bit wary of rolling out to other architectures to avoid
increasing risk, so for now it is handled via a specific upload to
debian-ports unreleased (thanks Samuel!)
- WSL images were still built and published unmerged, and that has now
been fixed, the maintainer uploaded a merged image to the MS store. Due
to filesystem weirdness in WSL 1.0 (it's fine in 2.0), converting
doesn't seem to work (note that everything should switch to WSL 2.0
anyway these days), there's an upstream issue about it, but there's
nothing we can do on our side

Final note, since it has been reported a few times: when using a
container manager (like Podman) that stacks multiple images with
OverlayFS, note that the conversion cannot happen, as you cannot modify
the lower layer - this is just how OverlayFS works. The base layer
needs to be converted (or rebuilt) individually. There's nothing we can
do about this, apart from detecting and aborting (without touching
anything) with a clear message.

Huge thanks to everybody who contributed fixes and/or good quality and
actionable bug reports!

Kind regards,
Luca Boccassi

Attachment: signature.asc
Description: This is a digitally signed message part

Reply to: