+1 Maintenance Report

Athos Ribeiro athos.ribeiro at canonical.com
Fri Jun 7 21:03:35 UTC 2024


I did +1 maintenance from 2024-06-03 to 2024-06-07.

I start the week by running the ubuntu-archive-tools find-proposed-cluster
script. There was nothing relevant there at this point of the cycle.

Then I started looking at individual packages, no hard rules, but I was
trying to focus on the bottom half of the list. Coincidently, the first
2 packages I looked at had infrastructure related failures. I then
proceeded to run the archive tools retry script:

./retry-autopkgtest-regressions --log-regex='unexpected eof from the testbed'

I sticked to this regex for the 5 days I was working on +1 maintainance and
found dozens of (non-duplicated) occurrences each day.

On Thursday and Friday, mirespace was shadowing me during my +1 shifts as our
timezone differences allowed (I suppose she will reply to this thread with her
findings later).

Below are comments on individual packages I worked on throghout the week (not
including test retries).

## qiime and q2-* packages
   qiime was blocking several q2-* packages. They all had the same python 3.12
   compatibility issue. I add a short patch to each of them and forwarded them
   all to Debian as well. Below are the packages cleared (or in their migration
   process) from the migration list. If any of those are still in the excuses
   page, they are most likely waiting to be accepted from the new queue or
   waiting on a dependency which is pending in new.
   quiime, q2-phylogeny, q2-types, q2-feature-table, q2-dada2, q2cli, q2-taxa,
   q2-quality-filter, q2-metadata, q2-fragment-insertion, q2-feature-classifier,
   q2-diversity-lib, q2-demux, q2-cutadapt, q2-alignment, q2-quality-control,
   q2-emperor, and q2-sample-classifier

## ruby-toml and ruby-parslet
   This is related to https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=1070989
   Here, we applied an upstream fix to ruby-toml and re-triggered ruby-parslet
   tests. Both packages migrated.

## wannier90
   This also had python 3.12 compatibility issues. We uploaded a fix and
   forwarded it to Debian.

## python-awkward
   We sent a fix to the debian maintainers so it would no longer use internet
   connections during the package build. The package is still needing some
   fixes, so I did not upload it to Ubuntu (it would most likely migrate, I am
   unsure if we want it to migrate as is).

## edflib
   Upstream does not support big endian architectures. I applied a patch
   available in Debian's salsa to avoid building to s390x and asked the
   ubuntu-archive team to remove it from s390x. This was done and the package
   migrated.

## node-yarnpkg
   This was an interesting one. The package initially FTBFS due to issues with
   "-Bsymbolic-functions". However, removing the flag from the build got us into
   a different FTBFS issue also affecting Debian
   (https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=1067283). This happens
   because the off_t type is being configured to be 72-bit long by cmake. After
   some investigation and a pair debugging session with sergiodj (thanks,
   Sergio), we found out that the build of an embedded library injects a hook
   into cmake to call "node" (JS) with the "--experimental-wasm-threads" option.
   This option has been removed from node 20 and node now exits with code 9 on
   errors. This error code ends up being picked up by the embeded library cmake
   checks which ends up being set as the size of off_t (9 bytes). I reported
   this to Debian and filed LP: #2068769

## pytorch
   pytorch FTBFS with llvm-18.  This was already reported last week.  there are
   several other packages blocked on this. I found some more upstream patches
   related to the issue but we are still getting (different) build failures
   there. See LP: #2067720.
   These are the packages directly blocked by pytorch: tabnet,
   pytorch-geometric, baler, open3d, pytorch-ignite, pytorch-cuda,
   pytorch-sparse, pytorch-cluster, pytorch-scatter, pytorch-text,
   pytorch-vision, and skorch

## python-lsp-ruff
   missing python3-ruff (provided by ruff), which was added to the
   sync/blocklist due to missing rust-clearscreen. This is now in the archive
   and we should sync ruff back in. I tried sync'ing it with

   $ syncpackage -r oracular-proposed -d unstable -v --force ruff
   and
   $ copy-package -b -s noble --to-suite oracular-proposed -y -e 0.0.291+dfsg1-3 ruff

   Both commands reported successful status, but it seems I have no permissions
   to copy packages in that exclusion list (?).

-- 
Athos Ribeiro



More information about the ubuntu-devel mailing list