Planet GNOME

WPE WebKit and WebKitGTK 2.54 have been released with a bunch of improvements and new APIs as usual, but there’s one point that kept the Igalia WebKit graphics team busy for the whole cycle: the new Skia-based compositor. The replacement of Cairo with Skia for content rendering has been a success and it’s already well integrated and optimized. We thought we could try to use Skia for the composition too and replace TextureMapper with Skia. TextureMapper was introduced in 2010 for the Qt port and later adopted by other ports. It uses the OpenGL ES API and maintains a collection of shader programs to paint different content. Nowadays TextureMapper is mostly the same code and shader programs, and it’s unmaintained and missing features. However, the performance was good and it has served us really well all these years. So, this time the goal was not to get better results in benchmarks, but to modernize the implementation, reduce the amount of code to maintain ourselves (like all shader programs) and make it easier to implement the missing features and fix existing bugs. This post is a summary of all the work we have done this cycle to implement the new Skia compositor.

SkiaCompositingLayer

The first step was adding an SkiaCompositingLayer class to replace TextureMapperLayer and adapt all the code to use one or the other depending on an environment variable. The initial implementation was based on the TextureMapper one for the things that are common like iterating the layer tree, computing transformations, etc. The way layers produced their contents didn’t change, so we were receiving textures for tiled content, video buffers, WebGL, accelerated 2D canvas, etc. SkiaCompositingLayer created a Ganesh Skia surface to draw those textures using SkCanvas::drawImageRect(). This initial implementation was enough to run the default MotionMark test suite, since it doesn’t use other composition features. Even though performance was not the goal, we had to make sure we didn’t regress. This initial implementation was neutral in MotionMark. We needed tests to implement those features and measure performance at the same time, so we decided to add a new set of tests to MotionMark, just extending the existing tests to require composition, which makes sure that filters, masks, path clipping, transformations, etc. were done by the compositor.

Filters

We first tried implementing filters using an intermediate surface like TextureMapper does. It worked, but the MotionMark score in the filters test was much worse. We realized that with Skia we could implement most of the filters without using an intermediate surface. All filter types except blur and drop shadow can be simplified to an SkColorFilter with SkImageFilter::asAColorFilter() which can be implemented without an intermediate surface, just by setting the color filter in the SkPaint we pass to SkCanvas::drawImageRect(). This not only fixed the performance regression, but also gave better results than TextureMapper, which always needs an intermediate surface.

Masks

There are two different kinds of masks: image mask, where the source mask is an image already, and clip path, where the mask is represented by a path to be clipped. In TextureMapper both are implemented the same way using intermediate surfaces. The mask is painted into a surface and then the masked layer creates an intermediate surface where its contents are first painted and then the mask contents on top using DstIn blend mode. Skia has APIs that allowed us to implement both cases in a much simpler and more efficient way. In the case of image masks, where we already have an image, we paint the mask contents once and keep it cached, and then the masked layer creates an SkShader for the image mask that is passed to SkCanvas::clipShader() without having to paint into an intermediate surface. Clip path masks are even easier, because we can just take the path we get and build an SkPath we can pass to SkCanvas::clipPath(), without having to paint the mask as an image at all or use any other intermediate surface. Once again, masks were not only easier to implement but they ended up being more performant too.

MotionMark composition suite, WPE with GPU rendering on a Raspberry Pi 4, comparing TextureMapper (312400@main) with the Skia compositor (313600@main). TextureMapper never implemented blend modes, so its high score on bouncing blend circles is the score for not doing the work.

3D contexts

The implementation of 3D layer contexts is fairly independent of TextureMapper and OpenGL, so we could just take it almost as it was, using SkPath to build the clips and a few other adaptations. We could also fix existing bugs like the z-ordering that has always been broken in TextureMapper.

The same page rendered by TextureMapper (left) and by the Skia compositor (right), WPE on the same build. The red box intersects the rotated green plane. TextureMapper draws the box flat against the plane, so the intersection is lost; the Skia compositor splits it, drawing the part in front of the plane and hiding the part behind it. Blend modes

TextureMapper never supported blend modes and they were easy to implement with Skia just using the SkPaint property for it. This made several layout tests start passing.

Batched painting

After implementing all the features we were at a point in which we had the same or better performance in all tests except for three MotionMark compositing tests that were giving much worse results. Those tests use small layers and give a high result which means we end up adding a lot of layers to the scene before we start skipping frames. The root cause was the large number of layers filling the command queue of Ganesh. Skia Ganesh queues the GL drawing operations instead of sending them to the GPU right away. When the surface is flushed for whatever reason, the queued GL drawing operations are then processed and sent to the GPU. This allows Skia to apply nice optimizations like merging several tasks and reducing the amount of draw operations we end up sending to the GPU. In those tests where a lot of layers are created and painted to the compositor Skia surface the internal command queue ends up being huge too. Processing and analyzing such a long queue to optimize what we send to the GPU required more CPU work than what we save by optimizing the GL draw operations. Skia provides an API that allows us to do the batching ourselves. Since the compositor already has information to decide what operations could be merged together, we could reduce the internal queue size in many cases. We can merge SkCanvas::drawImageRect() operations as long as they share the same color filter, blend modes and sampling options. In the best case scenario we could reduce the whole internal queue to just one operation. This time the change improved the results of those tests getting them to about 93% of the TextureMapper score, but still a bit behind.

Promise images

The Skia Ganesh backend requires that an SkImage backed by a texture is created for the current thread GrDirectContext, even if it’s borrowing an existing texture. In WebKit all textures are created with a sharing GL context so that they can be accessed and destroyed from different threads with the same sharing GL context. So, for a layer whose content is an image we had to create a texture in the compositing thread to upload the pixels if the image was not accelerated, or for accelerated images get the texture identifier of the image, and then create another SkImage from the compositing thread borrowing the texture for the current GrDirectContext. The Skia Ganesh backend provides an API to create promise images, which can be created from any thread but targeting a specific thread, providing a fulfill callback that will be called on the target thread when the SkImage is first used to retrieve the wrapped texture. This way we can create the SkImage from the main thread for the compositing thread without using OpenGL at creation time. For non-accelerated images we realized we don’t need to manually create the texture and upload the pixels in the compositor, we can just pass the unaccelerated SkImage to the compositor SkCanvas and Skia will handle it internally much more efficiently than we did. And this change improved those compositing tests much further than we expected. The reason turned out to be the batching from the previous section: Skia merges the entries of an image set by comparing texture proxy pointers, and until now we were wrapping the texture in a new SkImage on every frame for every layer, so hundreds of layers drawing the very same image produced hundreds of different proxies that Skia could not merge. Passing the same SkImage every time collapses all of them into a single draw operation, which is the best case we described above. With batched painting and promise images together we could beat TextureMapper significantly.

MotionMark composition suite, leaves subtests, WPE with GPU rendering. Score per revision; higher is better. The same two steps appear with CPU rendering. Deferred Display Lists (DDL)

When we switched to Skia for painting, we kept the threaded rendering model, just using a separate smaller queue for GPU rendering workers. The GPU workers created their own GrDirectContext to paint the layer tiles. The resulting textures were re-wrapped in the compositing thread for the compositor GrDirectContext using fences for the proper synchronization. We knew this was not the recommended way to use Skia Ganesh from multiple threads, but with TextureMapper we had no other option. However, with the Skia compositor we can do it the recommended way by using a single GrDirectContext in the compositing thread and use Deferred Display Lists (DDL) and promise images to paint the tiles. With DDL, GPU workers no longer use GL at all and they don’t need a GrDirectContext, they paint tiles into a display list that records the GL drawing operations, but without touching GL. For image drawing operations recorded into the DDL, promise images are used too. Since this is now all CPU work we can remove the smaller GPU worker queue and use a single queue with more workers. The compositor replays the DDL into an SkSurface that is then passed to the compositor SkCanvas.

This change fixed rendering glitches on Android and was performance neutral for the whole composition suite and for most of the MotionMark tests, but in MotionMark 1.3 at 15fps it cost 29% in Suits and 14% in Leaves, while improving Images by 9%. Correctness and the other benefits of DDL made us accept those regressions.

MotionMark 1.3 at 15fps, suits subtest, WPE with GPU rendering on a Raspberry Pi 4. Shaded regions are where deferred display lists were enabled by default. CPU rendering moves less than 1% at all three switches, since it has no GPU worker threads for DDL to change. Damage

TextureMapper already supported using damage information to optimize the painting while compositing, but it has always been disabled at run time because there were issues we never managed to fix. With the Skia compositor we decided to start from scratch and properly handle the damage information while compositing to render only the parts of the frame that actually changed. I’m not going to go into detail here because Nikolas Zimmermann has written an amazing blog post about it with all the details.

Current situation

The Skia compositor is finished and enabled by default in 2.54. Even though it was not the main goal, it performs better than TextureMapper in most of the benchmarks we run: the composition suite we added is 45% faster, and MotionMark 1.3.1 is 35% faster. The exception is MotionMark 1.3 at 15fps with GPU rendering, which comes out flat, because the Suits and Leaves tests are still about 26% and 10% behind due to the deferred display lists trade-off described above.

We are already working on fixing existing issues in composition that we never fixed in TextureMapper. In the main branch TextureMapper is now disabled by default at build time, and support will be removed soon for the GTK and WPE ports. In 2.54 it’s still a run-time decision so if you find any issue with 2.54, you can check if it’s a Skia compositor regression by trying TextureMapper with WEBKIT_USE_SKIA_FOR_COMPOSITION=0 environment variable.

Overall (geometric mean) score, WPE on a Raspberry Pi 4: 320000@main and later against the TextureMapper baseline at 312400-313296@main. Bars start at the baseline. Part of the gain in the MotionMark suites is Skia rendering work rather than the compositor. WPE on a Raspberry Pi 4, change from the TextureMapper baseline (312400-313296@main) to 320000@main and later. Suits and leaves are the deferred display lists trade-off, not the compositor switch, which was neutral in this suite. Future plans

We are already working on further improvements like using promise images for all external textures we have to pass to the compositor. We will explore the possibility of using Vulkan with the Ganesh backend instead of GL and eventually try the new Graphite backend. And of course we will continue fixing any existing issues related to the compositor.

Bombarded by the deception and lies of the AI industry I chose to sample boy Amodei for the ironic outrage about Chinese companies stealing their dataset. Thus the tune title.

Usually I barely manage to finish up my weekly beats track on a Sunday night. This week I've somehow had some extra time to sink into polishing an actual full track on the Dirtywave M8. Built around the bassline where I've mimicked the approach used on the Analog 4 of fading in a modulated filter and volume pulse over time using slight different tools (the M8 has 4 LFOs and ability to modulate a modulator).

[Originally written in Czech]

At the very beginning of the last year, Jiří Eischmann wrote a post on his blog about the status of the Czech translation at GNOME, tl;dr: the translation was slowly dying. I would really love to say that it is resolved, but that would be oversimplified. What changed?

During this year we decided to restructure our team and to restore the translation’s former scope and quality. Since spring, I’ve taken on the coordinator role in our Czech translation team—this release is under my lead. Fortunately I have Daniel Rusek beside me, who makes sure nothing goes unnoticed and who proposes further direction of our team, and I am very lucky to work with him.

I am happy to announce that our Czech translation is slowly forming to a pretty nice form, maybe soon as it was before. The first action I have done as coordinator was updating our manual for translators—I made sure it was easy to comprehend without a need for bigger changes from the previous one. I thought it would attract new contributors, which happened in summer right before the release was available to translate.

The core is almost fully translated to Czech, only sysprof is not. There is now also a new translation of foundry. Does it mean GNOME 51 is fully Czech? Unfortunately no. While using GNOME you can still find untranslated strings from the modules that GNOME depends on, like NetworkManager, which is used for VPN connections—but we are still responsible for translating some of them. We also translated a few apps from GNOME Circle, some websites, and some of the modules from freedesktop.org.

Even though we had small updates of user documentation, it’s largely stagnating. But…thanks to Petr Kovář’s awesome work help.gnome.org is now translatable and even translated to Czech language.

(You can find whole overview of translated modules on Damned Lies.)

During this cycle we got new members to our team, half of whom have already translated at least one module. I am very optimistic, and I believe these are not one-off translations but the beginning of long-term collaboration. Daniel Rusek remains reviewing, and I am joining him with doing so too.

That doesn’t mean that the translation is somehow resolved. You have to care about translations as if it were your garden, just having seeds does not imply a harvest, you need to take care of your plants first. We are still just a small group of people who gave up their leisure time and a few hours of sleep for the others. It’s not easy, and we possibly cannot translate for the eternity, that’s why we take your help seriously, we are more thankful for it than maybe you imagine.

Thank y’all.

Don’t let us down

We are grateful for every help with translating. If you want to make GNOME closer to Czech users, we are willing to teach you and navigate through translation. All needed information is listed on our page, or you can directly reach me via Matrix.

It’s been about 4 months since my last GNOME Foundation update. Time flies. I’m sorry that it’s been so long. I will try to do more regular posts again in the future, but perhaps not at the same tempo as before. While I would love to post every other week, it’s hard to sustain.

With that said, let’s jump in. Given the time since my last post, I’m going to focus on the bigger and more recent news items that have happened at the GNOME Foundation.

New board, new officers

The Foundation’s board elections happen every year, and this year’s election completed in July. The election resulted in a number of changes to the board:

• Sri Ramkrishna, Jonathan Blandford and Adrian Vovk joined the board as new/returning directors
• Deepa Venkatraman, our treasurer, secured a new two year term
• Robert McQueen, Federico Mena Quintero and myself all ceased to be directors (Rob failed to be re-elected, Federico didn’t run, I withdrew part-way through the process)

The election was a difficult one for me personally, and left me reconsidering my involvement in the Foundation. This was not because I lacked motivation or commitment, but because the situation around the election had become untenable for me personally. However, I’ve spent a good deal of time since I withdrew my candidacy thinking about my role at the Foundation, and I’ve concluded that I care about this organisation and the progress we’ve made, and I want to see that work through. Conversations I’ve recently had with members of the community have also given me confidence that we can move forward together. In short: I’m happy to be sticking around.

The new board held its annual meeting in August, which is when officers and committees are appointed for the next 12 months. The Board decided to put me into position as Interim Executive Director, with Sri Ramkrishna taking my place as President. This is a good move from my perspective: it recognises that I’ve been doing a lot of the day to day management work (which I will continue to do), and gives the Board more ability to hold me accountable. Sri stepping into the role of President means that he will be my backup.

Other officer changes include Jonathan coming in as Second Vice-President, Cassidy moving from Vice-Secretary to Secretary, and Adrian stepping up as Vice-Secretary. Our other officers remain in post, with Maria as chair, Deepa as Treasurer, and Arun as Vice-President.

Huge thanks to everyone who volunteered for these positions!

In terms of committees, the Executive Committee had a minor reshuffle, with Jonathan, Adrian, and Sri joining, and Julian and Rob departing. The new members of the exec are already taking on work, which is great, and I’m hopeful for the newly reconstructed committee. The Finance Committee had some slight membership changes, with Rob leaving and Sri joining.

Finance and Operations Director

Last April we opened the search for a new paid team member, to join us as our Finance and Operations Director. There are a number of goals for this new position: to enhance the finance and accounting expertise that we have internally, to lead the development of our internal systems and budgets, to ensure the sustainability of finance and compliance tasks, to manage our fiscally sponsored projects, and more generally take ownership of the business side of the organisation.

We had a huge number of applicants apply for the position, and had some extremely high quality candidates to choose from. After going through several rounds of interviews we selected Dawn Matlak for the role, who we are extremely excited about joining us. Those of you who have read my previous posts might remember Dawn’s name: she initially started working with us as a consultant last year, in order to help us prepare for our first formal audit, which happened in March this year. As part of this work she helped us to transform many of our internal systems and processes. We’re thrilled that she is joining the Foundation on an ongoing basis, and are confident that our internal operations will continue to improve under her stewardship.

Dawn is already doing a small number of hours for us each week, which she will continue to do until she properly starts in the role in November.

Many thanks to Arun and Deepa who helped enormously with the hiring process.

FY27 Budget

The Foundation’s financial year runs from 1 October to 30 September, and each financial year requires a new budget, both for planning and as the basis of reporting and spending authorisation. We have all therefore been working hard on the new budget that will come into effect on 1 October. The new budget has been in the works for a while, and has been a major focus for the board over the past few months. Thankfully we got the initial budget approval done last week at the board’s regular September meeting. We’ll follow-up with a more detailed post about the budget as soon as we’re able, so the community can have some insight into how we’re managing our finances.

Events

With GUADEC 2026 wrapped up, Kristi has turned her attention to the next event in our schedule: GNOME.Asia 2026. This is being held in Terengganu, Malaysia, from 31 October to 2 November. There’s a great venue lined up, and Kristi is busy working on the details with a fantastic local team.

Aside from GNOME.Asia, the other recent focus has been GUADEC 2027. We have a couple of options for locations right now, and are in the process of confirming details before we commit to one of them for next year. We’ll share updates as soon as we have more details confirmed.

Fundraising

The end of the calendar year is an important time for non-profit fundraising, and we are currently busy planning our campaign for the end of 2026. I’ll be posting more about this soon, in particular in relation to the budget, but for now I will say that this campaign is going to be critical for our ability to grow and support the GNOME project.

Other

As ever, many other things have been happening at the Foundation, and there’s too much to go into detail about here. Work on GNOME’s infrastructure and Flathub continues, our back office operation continues with finances and other routine paperwork, and the board continues to discuss our long-term plans.

That’s it for now. Many thanks for reading, and feel free to leave questions in the comments.

Back in April I wrote an informal history of the BuildStream project: Status update: 23rd April 2026.

Things escalated and somehow I ended doing a podcast interview with Rich Bowen of the Apache Software Foundation recently, on the Apache PlusOne podcast:

Apache BuildStream — with Sam Thursfield – YouTube

Fame at last!

I didn’t get much time to prepare for this so excuse any clunky explanations or inaccuracies. My main aim was to place BuildStream and Freedesktop in context for an audience who don’t live and breathe operating system integration tools. I’m interested in your thoughts on how successful that was. Comments are enabled on the YouTube video so you can also fact-check us there as needed.

Not a lot of things have landed this cycle, but there's still a bit to list, so let's do that.

Android support

As part of his effort to port GTK to Android, Florian also ported libadwaita demo. The builds are available from CI and the GTK 4 Android page.

He also implemented a settings backend, meaning that libadwaita apps now support system dark mode and accent color on Android (not high contrast or document/monospace fonts though).

Ministream

AdwAboutDialog can be populated from an AppStream metainfo file, via libappstream. While useful, it also causes problems on other platforms, such as Windows (libappstream can't be built using msvc) or Android, due to its dependencies.

Since we only use a small part of appstream (for example, we don't need composing or anything related to networking), he reimplemented the subset libadwaita uses as ministream. It only depends on GLib and it should build fine with msvc, so it should make building libadwaita outside of Linux easier.

CSS class bindings

A fairly common pattern is having property that toggles a style class - e.g. for use with breakpoints. Currently implementing it is a bit annoying, so Jamie Murphy added API for automating it - adw_bind_property_to_css_class().

It's modeled after g_object_bind_property() and works much the same way, incl. allowing bidirectional bindings.

A variant with mapping functions is also available, allowing to bind properties of arbitrary types and not just booleans.

Sidebar additions

AdwSidebar and AdwViewSwitcherSidebar have received a number of additions.

Sidebar prefix and suffix

First, both sidebar widgets now support having prefix and suffix widgets. This can be used for things like adding an account switcher, a prominent title, or a help button at the bottom. While it's not used a lot in GNOME apps at the moment (the only app I'm aware of is a development version of Crosswords), it's a common pattern on other platforms, so it's good to have API for this.

Section suffix

Next, sections can have suffixes in their headers, similar to AdwPreferencesGroup. This can be used to put a spinner or a button in the sidebar sections, similar to what Polari has.

Item prefix

Finally, sidebar items can have prefix widgets. They can be used instead of the icon or together with it, in that case it will be displayed before it. This can be used to display avatars, checkboxes and so on.

Icon changes

Last cycle I announced the new icon work. Unfortunately, it's still not ready, but a few smaller things have landed. First, larger icon sizes now use smaller weight, so new icons in AdwStatusPage and in images with the icon-size (but not pixel-size!) property set to LARGEwill look thinner.

Second, AdwSpinner now also follows icon weight and will look consistent with icons. Apps that use spinners at large sizes outside of AdwStatusPage may have to adjust the weight manually using the -gtk-icon-weight CSS property.

Other changes

• AdwShortcutLabel now uses proper labels for keys like ⌘ or ⌥ on macOS, as well as more natural ordering for modifiers everywhere.

• GtkDropDown can now be used with the .flat style class, and automatically becomes flat in toolbars (which can be undone with the .raised style class, same as for other buttons)

• AdwAboutDialog now has the :other-apps-title property, allowing to override the title of the "Other Apps" section.

Overall, not a lot has happened. Even this blog post is late, for the first time.

Part of the reason is various health issues, both physical and mental, another part is the state of the world at large and software industry in particular. It's hard to focus at the best of times, let alone when everything is falling apart.

I've been working on a personal project as a means of escapism, but it does mean libadwaita is getting less attention.

Thanks to the GNOME Foundation for their support and thanks to all the contributors who made this release possible.

Guadec happened. It was a extremely well organized event, with plenty of good talk, people that I was longing to see again, and new faces to put a name on.

My experience was however very much soured by interactions with other people within the community, against myself and other members of the community. To the point that I had to take the time to relieve the distress it has caused me, hence the time it took me to write this down.

The mobile shell initiative

The development for a “mobile” GNOME Shell started somewhere around 2022, and at least the supporting parts of it passed as a project sponsored by the first and only round of GNOME projects sponsored by Germany’s Sovereign Tech Fund, as “Increase Range and Quality of Hardware Support”, this was acknowledged in the final STF report.

Making GNOME Shell behave natively on a new form factor is a massive undertaking, and the planning was not sufficient. The first, most glaring mistake was to drive this project from the start with minimal involvement from the existing project maintainers. No consultation happened at any point in planning, not just to ensure the goals were in scope, but also to ensure the project is stewarded towards a point that everyone can walk away with a sense of completion.

The second biggest mistake was in planning for upstreaming, instead the work piled up on a branch in a personal repository.

To be fair, the supporting bits got merged over time, and there’s got to be some breathing room for new initiatives. But sooner or later there’s the harsh reality that the work has to be divided into tractable pieces and pushed in a structured manner through the review process, in order to end up with the work merged upstream.

The Mutter low level pieces were merged over the course of 1 year after the STF project, divided in 3 (1 2 3) merge requests, and some of the corresponding GNOME Shell changes to make use of this (no longer) new infrastructure were merged as well.

But meanwhile the mobile-shell branch kept piling on, north of 300 patches, with substantial changes to code (diff is +120437 -7156, by my accounting) and UI. The original author did not make attempts to upstream the changes, and the few efforts from other participants to upstream bits of it or as a whole did not last long, unfortunately. The work is nowadays sitting on its separate repository, using a separate issue tracker.

The mobile BoF at Guadec

This jagged interaction between the people acting as maintainers and the people driving the mobile-shell initiative lead to difficulties in making this work upstreamed in a timely fashion, much to everyone’s disappointment. In this situation, we arrive to this year’s Guadec. There is of course an interest in upstreaming the changes, from Mutter+Shell core developers and maintainers included, so we attend this BoF.

What happened there could be best described as a maintainer shaming session, specifically towards Jonas Adahl, Florian Müllner and myself, for “discriminating/demotivating newcomers”, “wanting to steal the spotlight”, “stalling things on purpose”, … essentially choosing slander as a way to put the blame on us for not having merged the work as-is. Even though this attack was driven by a few closely related to the initiative, it happened in front of 20+ people.

This was not ok

Even though I understand the frustration behind, I find the tactics used on us inexcusable.

Look, the community guidelines at conduct.gnome.org are a bar to meet for everyone, towards everyone. And the guiding principle of them all is pretty simple, we all row the boat roughly in the same direction. Once the basic principles of respect are lost between us, the boat does sink. The main asset that keeps Free Software moving forward is not the code, but the people.

Jonas, Florian and myself are lucky to be paid by our employer to work upstream on GNOME, but I can say for myself (and perhaps the three of us) that I work at Red Hat because I work on GNOME, rather than the other way around. We go well beyond our duties, in ways our employer does not care in the slightest if we do, and in ways it consumes our free time as well. We have looked to nurture the community, mentoring in GSoC and Outreachy more times than it’s worth counting. The accusations of discrimination fall entirely flat.

We so far had no trouble in collaborating with the main developer behind the mobile shell work either, there’s 226 merge requests merged from him in GNOME Shell and 151 in Mutter attesting that.

This strain on relationships is not baggage free. I don’t see myself working with the individuals that drove this attack pretty much anymore with any productive outcomes, and I will avoid that to the extent of my capabilities. There is an unbelievably long road to undo the damage they’ve done.

How to improve from here

The mobile fork is currently sitting in a separate repository, based on a now old release, and contains a number of back-and-forths, FIXMEs, WIPs, and code that has been either already done (albeit differently) or entirely refurbished upstream.

A rebase that is mindful to all these and brings the branch to a plausible up-to-date state is likely to take days to weeks. At this point, it could make more sense to identify the possible topics to split into multiple (many) merge requests, and cherry-pick the patches individually.

After these merge requests are done, they should go through review, and the style/architectural differences between the original author and the maintainers’ mindset be settled. Changes could be merged incrementally, advancing towards the common end goal.

It looks like I just described the software review process in a nutshell, and I very much did! But I feel it is important to point out that nothing of this has happened yet with these patches in a proper or substantial way.

In a shred of constructivism during the Mobile BoF, Markus Göllnitz offered himself to do this on behalf of the original author. I am looking forward for these steps to happen, and will collaborate with Markus on it.

I deep down hope that this blog post also serves as a cautionary tale about how wanting to rock the boat instead of rowing together may stall initiatives, and eventually poison relationships.

Andreas Gabriel Berbescu has reported several root privilege escalation vulnerabilities in various NetworkManager VPN plugins. If the VPN plugin is installed, then an unprivileged user can escalate to root by loading a malicious VPN configuration file:

• (CVE-2026-91837) network-manager-iodine: Option confusion reaches iodine’s pre-drop root shell
• (CVE-2026-91838) network-manager-sstp profile data reaches root pppd pty shell
• (CVE-2026-91839) NetworkManager-fortisslvpn credential newline injection permits local root code execution
• (CVE-2026-91840) NetworkManager-vpnc: Top-level VPN username newline injection reaches a root password helper
• (CVE-2026-91841) NetworkManager-vpnc: Incomplete fix of CVE-2018-10900

While most obviously bad for multi-user systems, root privilege escalation is also a serious defense in depth problem for single user systems. You are vulnerable if you have the VPN plugin installed; it does not matter whether you actually use it or not.

These are not vulnerabilities in NetworkManager itself. The VPN plugins are each separate projects, with their own separate maintainers, hosted by GNOME rather than by freedesktop.org. The status of each project is a little different:

• The NetworkManager-vpnc and Network-Manager-fortisslvpn git repos have both been archived. Contributions are no longer accepted, and you should uninstall them immediately. NetworkManager-vpnc users should migrate to NetworkManager-libreswan, and NetworkManager-fortisslvpn users should migrate to NetworkManager-openconnect.
• network-manager-sstp is currently unmaintained, but it is not obsolete. If you are interested in SSTP, this project needs a new maintainer.
• network-manager-iodine is maintained, and the maintainer has created a merge request to resolve this issue.

For more information on NetworkManager VPN plugins, see Josephine’s VPN plugin overview and announcement.

In July, I shared an update about my new role as AI Alignment Community Architect at Red Hat, focused on Fedora. This post clarifies what that role entails, why "AI alignment" is more than a technical term, and how I plan to support the Fedora community in leveraging LLM-gen-AI.

I organized this blog post into three sections:

1. Reclaiming AI Alignment: The meaning behind the term.

2. Model builder engagement: Why do we need bidirectional feedback loops?

3. My work in Fedora: Upcoming priorities this quarter and the path forward.

Note

I use the term “LLM-gen-AI” throughout this article aligned to the Software Freedom Conservancy’s recommendations. This is in support of addressing this technology in community-first terms.

Reclaiming AI Alignment: The meaning behind a term

As I defined my new role, I carefully considered the title. The Fedora community sentiment toward LLM-gen-AI is deeply divided: some rally against it, while others push for rapid adoption without wider community consensus. I needed a title that signaled a neutral, balanced approach. My mandate at Red Hat is to support upstream projects in adopting AI, focused primarily on Fedora. Working in the “AI” space at Red Hat is fascinating because I am exposed to diverse ways that customers use and deploy innovative open source technology. Additionally, Red Hat provides real value by supporting customers in their “hybrid AI” journeys. Many of my Red Hat colleagues consistently push for more Free Software and open source answers for customers and enterprises building infrastructure to support AI inference, local models, and more. Red Hat has a responsibility to innovate when new technology opportunities emerge that its customers are acting upon. With this in mind, I am more convinced that LLM-gen-AI is something that open source maintainers and contributors can leverage for real workloads. There are opportunities to solve real problems and routine maintenance tasks for complex projects. LLM-gen-AI used right can support maintainers in automating boring, cyclical work so they can focus more on the exciting work of innovation and focused engineering efforts within their projects. Or even going outside and spending time offline.

However, I distinguish "AI alignment" from "AI adoption." "AI adoption" communicates a pre-defined, non-negotiable stance where the goal is simply to increase usage. "AI alignment," by contrast, communicates that LLM-gen-AI use exists on a spectrum.

My intent in Fedora is not to insist, but to negotiate and compromise. I want to align how our community uses these tools with our existing values, norms, and culture.

CHAOSS AI Alignment Working Group & model builders

My approach to "AI alignment" is influenced by the CHAOSS Project. I co-chair the CHAOSS AI Alignment Working Group with Emma Irwin and Coraline Ada Ehmke. Recently, Emma, Adrian Edwards, and I presented at FOSSY 2026 on this topic in greater detail. This experience frames my definition of "alignment" as I move forward in my new role.

Traditionally, "AI alignment" is a term used by model builders to describe processes where communities have little influence. As LLM-gen-AI grows in widespread use, the power gap between those model builders and the communities they impact will widen. This creates an unsustainable dynamic in the safety and well-being of our communities with these new tools.

We need more than just "AI adoption". We need bidirectional feedback loops. Free Software communities deserve a seat at the table. This is not just for iterating on model development, but for defining how we integrate LLM-gen-AI tools sustainably and responsibly.

To achieve this, we (i.e., open source community citizens) need a stronger value proposition for engagement with model builders. Whether commercial or altruistic, we must persuade model builders that a community-driven approach is a critical advantage. If we refuse to engage, we risk Free Software values and culture being shut out of conversations entirely. So, I believe it is better to be an advisor than a bystander. Advising is its own form of open source contribution. Therefore, I lend my support toward the wider notion that "AI alignment" fosters two-way conversations that ensure model builders actually listen to the communities their work impacts.

My work in Fedora: What I hope to work on next

September 2026 is my first full month in this role. While there is still much to define, a few priorities have emerged. Here is where I will be focusing my initial energy:

• Migrating "This Week in Fedora": Aurélien Bompard (@abompard) created a useful tool for AI-curated, human-reviewed weekly summaries. Currently, it lives on a personal fedorapeople.org space. I am beginning to work with Aurélien to migrate this to a weekly WordPress article on the Fedora Community Blog, since we first began talking about this in July. I am already submitting pull requests to support this transition.

• Launching LLM-gen-AI Agent Skills: Together with the AI/ML SIG, we are building the Fedora Agent Skills Library. These define best practices for using LLM-gen-AI agents to automate routine maintenance. My immediate task is community architecture: setting up the repository for contributions and improving documentation so these skills are accessible and scalable.

• AI/ML SIG Documentation: As the Fedora Docs Team works to identify "team captains" for specific topics, I am volunteering to lead AI/ML SIG documentation. This involves importing or deprecating content in the Quick Docs site and migrating extensive Wiki documentation to the Fedora Docs site, creating a central, discoverable home for all things LLM-gen-AI in Fedora.

• An update to the Fedora AI-Assisted Contributions Policy?: Honestly, I am not sure about this one yet. But it seems apparent to me that eleven months after the Fedora Council first introduced the Fedora AI-Assisted Contributions Policy, it is time for an update. Nearly a full year of lived experience has happened within the framework of this lightweight policy. Furthermore, a coalition of Fedora contributors agree that an update is needed, but there is not a single, shared view of what those updates should be. It will take more community input and feedback to shape the next iteration to that policy. I anticipate facilitating inclusive future community conversations about what those changes should be.

I am both excited and nervous to work with fellow Fedorans on these initiatives. I know there are strongly-held opinions on both sides of the LLM-gen-AI debate. (An understatement!) I accepted this role because I believe participation is more constructive than standing on the sidelines. My goal is to navigate this work in alignment with Fedora’s Four Foundations: Freedom, Friends, Features, First.

Until next time!

Since March 2026, I invested a lot of time into migrating my blog to a new publishing system and improving the website user interface. However, now that my site is fully functioning on a technical level, it is refreshing to begin writing content again here. There is more to come from me in this space. Expect new content about my work in Fedora and the LLM-gen-AI space, and other open source and personal items too.