reworked how installations behave when APT mirror connectivity test fails:
for automated installations and interactive desktop installations, the default policy was changed from “abort” to “offline-install”. This means that we now revert automatically to a semi-offline installation (i.e., we only fetch packages from the pool, not from the archive).
for interactive server installations, the user must now decide between going back (and edit the settings) or proceeding with a semi-offline installation. The “Continue anyway” option is no longer offered, after the connectivity test fails. This was approved as a UI freeze exception. Bug #2059898 “[UIFe] Subiquity fails to download packages after ...” : Bugs : subiquity
spent time investigating slow subiquity startup in VMs when run by the desktop installer. Nothing obvious came up from the investigation.
brainstormed on how to properly handle already mounted partitions in the installer.
did some non regression testing with subiquity on focal and jammy.
In review and waiting for TIOBE to setup the project.
Misc.
Biannual check-in
Went down the ubuntu core/snaps/gadget/etc. rabbit hole. This is severely lacking documentation but this is identified and will hopefully improve in the future.
The specification is now approved! TLS 1.0 and 1.1 is recognized as deprecated in Ubuntu too and while some softare stil enable these by default, this is considered a bug rather than a feature.
libtracefs
I got the package in a better shape, especially its testsuite which was superficial, and fixed/skipped/ignored half a dozen different issues in CI. The package has migrated and is in now in main. Thanks @upils , @slyon and @paelzer !
Prepared 3.8.5-2ubuntu1 which merges from debian and uses the upstream fix for the regression observed in 3.8.5 (no configuration file led to the wrong default configuration for a new setting). This is a late update in order to have as many fixes as possible for this core package, including security ones, and also future-proofing for the Ubuntu LTS as gnutls finally offers a way to disable RSAES-PKCS1-v1_5 which has been deprecated for decades and has been considered legacy for years. Thanks @vpa19771 for the review.
Like with gnutls28, I’m looking at using the most recent version possible for openssl. At the moment, 3.0.13 is already in the archive but there are fixes in the openssl-3.0 git branch which I’ve picked up in order to create 3.0.13+20240415.45c2a8-0ubuntu1.
Merge request but I’m re-spinning it for a couple minor changes
Work on automation to trigger tests in a PPA after the builds have been published; it’s very basic code but very convenient; thanks @andersson123 for the API key and @bryce for ppa-dev-tools
We still need to respin cd-boot-images-riscv64 (LP: #2062166) Sponsor needed
U-Boot
In upstream U-Boot the loading of device-trees from directory /dtb in ESP currently is not working for the UEFI boot manager boot method. I am working on a fix.
Milk-V Mars CM
Patches for support in U-Boot sent upstream. Currently under upstream review.
Removed nvidia hack from initramfs-tools (LP: #2060032) as the underlying issue was fixed (LP: #1970069)
Requested removal of mtd-utils on armhf architecture (tests failing on armhf, and the consequences of failure here are too severe to ignore) (LP: #2061990)
Enabled setting the embedded switch mode on an SR-IOV interface when it doesn’t have virtual functions, I did some refactoring and lots of testing along the way. PR#454
Imported my random Netplan YAML generator to our tests and enabled it in our CI. It will generate thousands of (controlled) random Netplan YAMLs and call the parser against them to try to detect problems such as memory leaks and crashes PR#453
Fixed a couple of new issues found with the config generator PR#451PR#450
Helped with lots of testing with the new networkd wait-online strategy that Lukas is working on PR#456
Currently trying to move forward with the parser flags work to ignore parser errors.
Distro
Tried for a while to get swiftlang building on Noble. After a few patches I’ve got it working on amd64 but the swift compiler is crashing on arm64. The culprit seems to be the vendored LLVM 13. I added the details in the LP ticket. LP#2060818 and PPA
Discourse was down last week and I never went and pasted my status there after the fact, so here’s two weeks worth of status report:
Subiquity
LP: #2060547 - An expected autoinstall failure case will still allow the installer to progress significantly and then abruptly hang, which obscures the previously emitted error message. A terrible UX problem. I have opened PR: #1965 to address the issue but since the fix for this is a little involved, we are going to wait until after release to avoid potentially introducing more severe regressions in user creation.
LP: #2060353- I investigated a bug that causes the installer to hang when installing nvidia-driver-550. Initially I was unable to reproduce the bug, but it turns out that was a separate (weird and interesting) problem with my test package. I opened PR #1969 (draft) which would solve the solution in the short term if needed, but this will naturally be taken care of as we move towards supporting MOK enrollment.
PR: #1964 - Adds a pre-commit hook to Subiquity to check documentation spelling. I consistently forgot to do this and had to wait for CI to yell at me, so I decided to quickly write this up. Thanks @dbungert for later improvements.
Defining some improvements we want to make to Subiquity’s documentation (more tutorials please and thank you @rkratky )
Drafting release notes
Subiquity bug triage this week, with an increased focus this week to keep on top of bugs before the RC build
LP: #2062085 - I opened a bug to track a known issue where Subiquity will happily let you put /boot on a FAT-based filesystem but crashes when configuring the kernel since FAT doesn’t support symlinks.
LP: #2061042 - I opened this bug to track an issue where a particular autoinstall configuration for installing reset partitions can crash the installer. Draft PR: #1972 opened to fix the problem, but since it’s not critical I am deferring the fix until after the release.
Distro
I was on +1 maintenance the week of April 2nd, where I focused on xz-utils related efforts. I also spent some time tacking down reasons for desktop meta-package install-ability issues. You can read my report here. I had some follow up last week on related bugs I opened, thank you @kanashiro and @ginggs for the sponsorship!
After receiving multiple private bug reports of failed online installs due to a particular out-of-sync mirror, I reached out to them and they are currently working on updating their mirror.