Hi! Another month, another status update. Let’s dig in!

The highlight of this month is the launch of chat.sr.ht, a hosted IRC bouncer service based on soju and gamja. The service is in closed beta for now, feel free to ping me if you want to try it out!

It took some effort to setup the new server and iron out the initial set of issues (especially a deadlock I couldn’t reproduce locally). Special thanks to Hubert Hirtz who’s added PostgreSQL support to soju, which should scale a lot better than SQLite (SQLite is still great for small instances).

gamja has been extended to support a whole bunch of IRCv3 extensions (extended-join + account-notify + WHOX) and should now reliably display account names next to nicknames for authenticated users (note, soju itself is still missing WHOX at the moment). gamja now re-joins channels on reconnect, which is handy when running it without a bouncer.

wlroots has been pretty quiet. Thanks to nyorain and feedback from Vulkan experts, we’ve finally merged the Vulkan renderer, enabling a bunch of potential future optimizations and making it easier to write Vulkan-based Wayland compositors. I’ve slowly continued my work on dmabuf hints, now renamed to dmabuf feedback.

I’ve started a new project to help with the future migration of the Sway and wlroots projects off of GitHub. dalligi allows to plug builds.sr.ht to a GitLab instance. Since it appears as a runner from GitLab’s point of view, it integrates properly with things like merge request checks. On a different note, I’ve migrated swaywm.org to SourceHut Pages (it was previously hosted on GitHub pages).

I’ve worked on a number of kernel improvements. I’ve fixed corrupted fullscreen buffers on older AMD cards by adding a tiling check to amdgpu. After a lot of back-and-forth, it appears we found a consensus to re-enable the overlay plane on amdgpu, allowing libliftoff to better take advantage of the hardware. Some other patches are still in review. One patch eases seamless “tear-free” transitions between DRM clients (boot screen, display manager, and compositors). Another series adds more metadata to connector hotplug events sent by the kernel, and allows compositors to avoid force-probing all connectors each time (this can take quite some time).

That’s all for now, see you next month!