Notes from the Graz Plasma sprint

A few days ago I returned home from a wonderful Plasma sprint in Graz, Austria. Between COVID-19 and there being no Plasma sprint last year in favor of the Goals mega-sprint, this was actually only my my third in-person Plasma sprint! So I was very excited to attend. There’s much to talk about!

Sprints are often said to come in two flavors: “talking” sprints, which are mostly about discussing big topics; and “working” sprints, where folks write lots of code. This one was a bit of both — a good mix, I think.

For my part, I wanted to talk about some big topics and see if we could get them unblocked. And talk about them we did! There were many more discussions besides these, but here’s a summary of the ones I led:

Plasma LTS

It’s no secret that our Plasma LTS (“Long-Term Support”) product isn’t great. It really only means we backport bug-fixes for longer than usual — usually without even testing them, since no Plasma developers enjoy living on or testing old branches. And there’s no corresponding LTS product for Frameworks or Gear apps, leaving a lot of holes in the LTS umbrella. Then there’s the fact that “LTS” means different things to different people; many have an expansive definition of the term that gives them expectations of stability that are impossible to meet.

Our conclusion was that the fairly limited nature of the product isn’t meeting anyone’s expectations, so we decided to not continue it. Instead, we’ll lengthen the effective support period of normal Plasma releases a bit by adding on an extra bug-fix release, taking us from five to six.

We also revisited the topic of reducing from three to two Plasma feature releases per year, with a much longer bug-fix release schedule. It would effectively make every Plasma version a sort of mini-LTS, and we’d also try to align them with the twice-yearly release schedules of Kubuntu and Fedora.

For some background, last Akademy we decided to postpone making this schedule change until all the KDE items on the “Wayland known significant issues” wiki page are addressed. During the sprint, we took another look and found the list much shorter than it was last year, with most remaining items in progress or nearing completion! So we agreed to revisit the topic again around this year’s Akademy in about 4 months (reminder to submit a talk!).

I hope that by then we’ve either got everything done, or can consider it close enough that we can pull the trigger on the new schedule anyway — the latter outcome being what we did for the Wayland-by-default rollout.

However, the concept of “Long-Term Support” doesn’t go away just because we’re not giving that label to any of our software releases anymore. Really, it was always a label applied by distros anyway — the distros doing the hard work of building an LTS final product out of myriad software components that were never themselves declared LTS by their own developers. It’s a lot of work.

So we decided to strengthen our messaging that users of KDE software on LTS distros should be reporting issues to their distro, and not to KDE. An LTS software stack is complex and requires a lot of engineering effort to stabilize; the most appropriate people to triage issues on LTS distros are the engineers putting them together. This will free up time among KDE’s bug triagers and developers to focus on current issues they can reproduce and fix, rather than wasting time on issues that can’t be reproduced due to a hugely different software stack, or that were fixed months or years ago yet reported to us anyway due to many users’ unfamiliarity with software release schedules and bug reporting.

3rd-party content and theming

We’ve had some difficulty with the UX for how users get 3rd-party content, and what it does to their system once they’ve gotten it. Many folks will remember the issue last year when a defective 3rd-party Global Theme caused user data loss. It was Very Bad.

The issue here is that QML-based theming is just inherently dangerous because QML is code; there’s not really a way to make QML theming safe, so we’re working on moving away from it. David Edmundson wrote about this recently, too.

So far we’ve already removed QML-themability from the lock screen (a component that’s very sensitive to stability and security needs!), and during this sprint, we hit the OSDs too. We also made plans to remove QML-themability from splash screens and login screen themes, and instead you’ll simply be able to choose custom images.

However, some things will always have to contain QML code, like 3rd-party widgets and wallpaper plugins. For these, we devised a plan to sandbox them so they can’t blow up the world if they misbehave. This will also make Global Themes safe, since Global Themes can include them. I wasn’t able to follow all the details of the proposed sandboxing system, so others will have to fill in the blanks with their own blog posts!

Finally, we talked about the distribution channel of https://store.kde.org, exposed via the “Get new [thing]” dialogs sprinkled throughout System Settings and KDE apps. What some might not know is that this distribution channel is not actually owned by KDE; it’s simply a themed front-end to https://pling.com. And that’s one of the issues: people think this is owned by KDE, and it’s not! Some other concerns included the lack of approval-required moderation for new content with stability or security implications; lack of automatic linting for content to make sure it’s valid; inability to specify a minimum Plasma version for QML-based content; and the place being sadly flooded with low-effort AI-created junk. We also talked about some UX issues in these dialogs and in Discover, and how we can address them.

We brainstormed what our ideal 3rd-party content distribution mechanism would look like if we were creating one from scratch, and the degree to which our current UX does and doesn’t approach it. I’ll be reaching out to the folks behind Pling to see if we can work on any improvements there so we can make reality converge more with our desires!

Activities

Activities has been in a weird place for a long time now. It’s a feature that’s somewhat difficult to explain in an elevator pitch, and with a more limited scope than you might expect. We all pretty much agreed that it’s not ideal, and not as useful as it could be.

So we brainstormed many alternatives, taking into account feedback and experiences from people at the sprint who currently do use Activities, or would like to if it met their needs.

Something that came up over and over again was the desire to use certain apps differently in different Activities. For example in your “Home” Activity, you could have your email client set up to only show your home email accounts, whereas in your “Work” activity, you could have the same app set up with only the work email accounts, or with both. But it would be the same email client app in each Activity, just configured differently!

This functionality right now needs to be provided by each app implementing its own “profiles” or “sessions” feature — and of course, most don’t. So an idea that stuck was for us to make this into a system service that can basically bolt the functionality of multiple profiles/sessions onto any app! This would be easiest for containerized apps that already have their own separate location for configuration data, so this is where we’ll start. But it’s possible we’ll also be able to open it up to un-containerized traditionally-packaged apps too, using Firejail or another similar technology.

We thought this feature would be useful even outside of Activities, so our conclusion thus far has been to build it first! After it’s in production and the kinks have been worked out, it would then become the basis for the Activities system’s new scope: “configure and use your apps and virtual desktops differently in each Activity.”

There are no timelines for any of this yet; it’s still in the “turn a discussion into a plan” phase.

Telemetry

There was broad agreement that the status quo is not ideal: we collect very little data from people who have opted in (because it’s off by default), and we don’t have a path towards changing what data we collect in response to newly-discovered questions we’d like answered.

For example, let’s say we want to remove a very niche KWin effect that we think probably nobody’s using. Right now, we have no way of knowing how many people have it turned on, and of them, how many people are actually using it — let alone reaching out to ask them why! So we have to just go by gut feelings when we make that decision, or get spooked and not do it at all.

So we decided to change the way we collect telemetry to be more like the common and successful Steam Hardware Survey. Our telemetry UX will be the same: people will see a dialog window asking them to participate in the survey, and in there, they’ll see what data will be collected. They’ll have the opportunity to say yes, say no, or turn it off forever. And for each survey, we can tailor the data collected to what we actually want to know! So we could have it collect information about that KWin effect, and it could even prompt the people using it to write a little bit of text explaining why.

We also discussed how the current place to view collected data is not ideal. Right now there’s a GUI app which is slightly broken, and a web page you have to type raw SQL queries into to see anything besides the default visualizations. Not ideal! We brainstormed a better web UX so we can make better use of the data we do currently collect. We also agreed that we want to make the aggregated data public, just like Valve does with the results of the Steam Hardware Survey.

And more

In addition, I got in a lot of hacking time between discussion sessions, which felt useful. Being able to sit down with other people unblocked some work, both mine and theirs, as a result of some productive face-to-face conversations! And I got to see a lot of old friends again, and meet a few in person for the first time. The city of Graz was lovely too — such a sensible and humane-feeling place.

Thank you to TU Graz for hosting us for this sprint, to Techpaladin Software for sponsoring my travel and lodging costs, and to Harald Sitter for organizing everything!