Hi everyone, below you will find the updates from the Desktop team from the last week.
If you’re interested in discussing a topic please start a thread in the Desktop area of the Community Hub (this site).
We also have our weekly meeting on IRC. We meet on Tuesday at 13:30 UTC in #ubuntu-desktop on Freenode. There will be an “Any Other Business” section at the end where you are welcome to raise topics. These topics might be discussed during the meeting, or afterwards depending on the time, depth of conversation, topic and so on.
Last week’s notes are here: https://community.ubuntu.com/t/monday-14th-january-2019/9344
Next week’s notes are here: https://community.ubuntu.com/t/monday-28th-january-2019/9524
Mutter and Friends
All that involves packaging listed here needs review, so feel free to sponsor
Gnome Shell performance (stutter | latency | CPU):
Gnome Shell other:
- I mostly worked on autopkgtest this week
- The queues got to over 4000 per arch at once point
- The controller started dying under the IO load. What happens is when a test needs to make a new testbed, the state of the test directory is copied to the controller and then restored back once the new one is provisioned. When this is happening for many large tests in parallel (e.g. libreoffice, KDE), this causes a lot of IO on the host. To cope with this I reduced our capacity by about 90%, so that there were fewer concurrent tests running. Once the big ones were out of the way I was able to scale back up. This required some close watching of the machine.
- I also started a thread about some (IMO) unreasonable tests.
- i386 is failing to reboot sometimes - investigated that a bit and passed on to the kernel team
- amd64 was broken in disco-proposed yesterday due to a klibc upload. Took a while to get to the point of being able to identify that (we weren’t reporting the failures properly to the web UI / proposed-migration).
and now there’s some weird problem with tests that build their packages, which is breaking all systemd-upstream (PR) tests. I took a bit of a look yesterday and will need to continue with that. Don’t understand it atm.
- Asked debian-gnome if we can update to 3.34 in experimental, nobody said no so will be doing that.
- Some reviews of mutter/gnome-shell stuff for Marco
- Updated glib to 2.59, seems that release is slightly buggy so we removed it from disco-proposed for now (e.g. GNOME/glib#1659 which I’ll keep looking at next after autopkgtest).
I don’t have any rls bugs at the minute but we need to get dnsmasq to migrate.
xdg-desktop-portal OpenURI.OpenFile support:
- We discovered that launching applications associated with local files via the glib API was not functioning correctly for snap applications. This seems to be a result of a security improvement covered by xdg-desktop-portal issue #167 where the corresponding glib change never got merged.
- I filed glib MR !585 to fix this, which I can hopefully get merged this week.
xdg-desktop-portal Inhibit support:
- The inhibit APIs in GTK (i.e. block screensaver, logout, suspend, etc) also didn’t seem to function with snap confinement: rather than using the portal API it was trying to issue the
org.gnome.SessionManager D-Bus API instead.
- It turns out that this had been fixed in the 3.24 branch. It looks like it still needs to be able to call the
StartServiceByName bus API, but that can probably be avoided with some more code changes. Either way, it means we need to package a newer GTK with snaps that found in 18.04 (GTK 3.22).
gtk-common-themes Yaru/communitheme support:
- I started putting together a branch to move some of the logic from gtk-common-themes MR !8 into a Yaru build option to make it a bit easier to maintain. Once that’s sorted, it should be easier to set up the CI trigger to build gtk-common-themes snaps with custom Yaru branches to fix the frankenstein snaps currently pushed to edge.