Decided to try writing a Wayland compositor for fun.
-
Yesterday dove into one of the older animation jank bugs in niri: quickly resizing a window back and forth would cause adjacent windows to jump. This problem is hard to trigger unless you're specifically trying to (then it's easy), but it was causing "downstream" problems for more complex actions.
The fix was simple enough, but I imagined enough "interesting" cases around it and decided to write a whole bulk of tests. Check this out, all thanks to our layout and clock code!
Merged the "window positions and sizes in IPC" PR. Please test and report if anything's broken or missing, we have time to fix it until the next release.
https://yalter.github.io/niri/niri_ipc/struct.Window.html#structfield.layout
Here's a quick quickshell test I threw together that draws floating windows as rectangles.
Thanks yrkv for getting the PR mostly complete!
-
Merged the "window positions and sizes in IPC" PR. Please test and report if anything's broken or missing, we have time to fix it until the next release.
https://yalter.github.io/niri/niri_ipc/struct.Window.html#structfield.layout
Here's a quick quickshell test I threw together that draws floating windows as rectangles.
Thanks yrkv for getting the PR mostly complete!
With massive thanks to cutenice, we have a new wiki site!!! With good search!
https://yalter.github.io/niri/
Make sure not to miss the new Since: annotations
The site uses Material for mkdocs, and we retained full compatibility with GitHub Wiki, meaning all existing links keep working.
Also thanks everyone for several suggestions and test wikis in the GitHub discussion!
-
With massive thanks to cutenice, we have a new wiki site!!! With good search!
https://yalter.github.io/niri/
Make sure not to miss the new Since: annotations
The site uses Material for mkdocs, and we retained full compatibility with GitHub Wiki, meaning all existing links keep working.
Also thanks everyone for several suggestions and test wikis in the GitHub discussion!
-
Rearranged and expanded quite a bit the design principles page on the wiki: https://yalter.github.io/niri/Development%3A-Design-Principles.html
Give it a read if you're curious about my thinking process about various niri behaviors and such
Implemented basic AccessKit integration for niri based on @DataTriny's proof-of-concept. I have little experience with screen readers, and also had several questions, so I would appreciate more eyes and testing on the PR:
-
Implemented basic AccessKit integration for niri based on @DataTriny's proof-of-concept. I have little experience with screen readers, and also had several questions, so I would appreciate more eyes and testing on the PR:
niri v25.08 is now released with a logo, xwayland-satellite integration, basic screen reader support, NVIDIA screencast flickering fix, window positions and sizes in IPC, and many other improvements!
Read the notes here: https://github.com/YaLTeR/niri/releases/tag/v25.08
now i need at least a month off from any release notes preparation
-
niri v25.08 is now released with a logo, xwayland-satellite integration, basic screen reader support, NVIDIA screencast flickering fix, window positions and sizes in IPC, and many other improvements!
Read the notes here: https://github.com/YaLTeR/niri/releases/tag/v25.08
now i need at least a month off from any release notes preparation
-
Fullscreen refactor pt. 1: track pending vs. current fullscreen state for columns, avoid vertical jump when waiting for window to commit a new buffer.
-
Fullscreen refactor pt. 2: added un/fullscreen animations. This one involved quite a lot of moving parts actually, so looking for testing.
Fullscreen refactor pt. 3 turned out to be a refactor of a good chunk of core Smithay xdg-shell/layer-shell/session-lock logic, making it more correct. The fullscreen PR in niri now includes that Smithay refactor, and needs testing even more than before
https://github.com/YaLTeR/niri/pull/2333#issuecomment-3263990200
Specifically, the refactor makes Smithay correctly track the last acked configure for each commit, also enforces the "must ack before committing first buffer" protocol rule.
-
Fullscreen refactor pt. 3 turned out to be a refactor of a good chunk of core Smithay xdg-shell/layer-shell/session-lock logic, making it more correct. The fullscreen PR in niri now includes that Smithay refactor, and needs testing even more than before
https://github.com/YaLTeR/niri/pull/2333#issuecomment-3263990200
Specifically, the refactor makes Smithay correctly track the last acked configure for each commit, also enforces the "must ack before committing first buffer" protocol rule.
With the fullscreen refactors in place, I got started on the thing that I wanted to do all along: maximize.
Niri didn't support normal Wayland maximize because it's very similar (yet slightly different) to our full-width columns—and can't be bound to our full-width state either. However, after plenty of requests, and thinking about it, I reconsidered. Henceforth, the maximize buttons and double-clicking on the titlebars will do the expected thing.
Details in PR: https://github.com/YaLTeR/niri/pull/2376
-
With the fullscreen refactors in place, I got started on the thing that I wanted to do all along: maximize.
Niri didn't support normal Wayland maximize because it's very similar (yet slightly different) to our full-width columns—and can't be bound to our full-width state either. However, after plenty of requests, and thinking about it, I reconsidered. Henceforth, the maximize buttons and double-clicking on the titlebars will do the expected thing.
Details in PR: https://github.com/YaLTeR/niri/pull/2376
My Smithay refactors were merged: the big one mentioned above, plus a fix for popup stacking order (e.g. open popup, then open tooltip also from the toplevel itself), plus a fix for root popup grabs from Qt layer-shell surfaces. All of this along with fullscreen refactors and animations is now merged to niri main. (maximize is still wip)
-
My Smithay refactors were merged: the big one mentioned above, plus a fix for popup stacking order (e.g. open popup, then open tooltip also from the toplevel itself), plus a fix for root popup grabs from Qt layer-shell surfaces. All of this along with fullscreen refactors and animations is now merged to niri main. (maximize is still wip)
There's been a long-standing request to add config includes to niri. They're useful for config organization, but also for custom desktop shells to be able to change colors without having to edit the main user's config.
Today I finished the first step towards this: a many-days-long refactor that makes the main config part, layout, mergeable, i.e., able to be combined from multiple parts. And building on this, per-output/workspace overrides.
https://github.com/YaLTeR/niri/pull/2449
Testing wanted here!
-
There's been a long-standing request to add config includes to niri. They're useful for config organization, but also for custom desktop shells to be able to change colors without having to edit the main user's config.
Today I finished the first step towards this: a many-days-long refactor that makes the main config part, layout, mergeable, i.e., able to be combined from multiple parts. And building on this, per-output/workspace overrides.
https://github.com/YaLTeR/niri/pull/2449
Testing wanted here!
Our config system is declarative, rather than command-like, which means that we parse the config into a tree of data types instead of reading lines and applying them one-by-one. This design requires a lot of work to properly support includes, but in turn we get atomic and selective reloading (if output part of the config didn't change, we don't override your transient output adjustments), better error messages and no problems with things like "spawn-at-startup" that should only work once.
-
Our config system is declarative, rather than command-like, which means that we parse the config into a tree of data types instead of reading lines and applying them one-by-one. This design requires a lot of work to properly support includes, but in turn we get atomic and selective reloading (if output part of the config didn't change, we don't override your transient output adjustments), better error messages and no problems with things like "spawn-at-startup" that should only work once.
After several more days of work, I fully finished config includes. All config sections merge together, live-reloading watches all included files (even if they fail to parse), error messages work across files, documentation is written.
https://github.com/YaLTeR/niri/pull/2482
Once again, this needs testing! There must be NO breakage to existing configs, so if something breaks, I want to know about it to fix it.
-
After several more days of work, I fully finished config includes. All config sections merge together, live-reloading watches all included files (even if they fail to parse), error messages work across files, documentation is written.
https://github.com/YaLTeR/niri/pull/2482
Once again, this needs testing! There must be NO breakage to existing configs, so if something breaks, I want to know about it to fix it.
I merged config includes, along with per-output and per-workspace layout config overrides. Play around with them at your nearest niri-git package.
- https://yalter.github.io/niri/Configuration%3A-Include.html
- https://yalter.github.io/niri/Configuration%3A-Outputs.html#layout-config-overrides
- https://yalter.github.io/niri/Configuration%3A-Named-Workspaces.html#layout-config-overridesAlso merged ignore-drm-device which should let you passthrough a GPU to VMs: https://yalter.github.io/niri/Configuration%3A-Debug-Options.html#ignore-drm-device
-
I merged config includes, along with per-output and per-workspace layout config overrides. Play around with them at your nearest niri-git package.
- https://yalter.github.io/niri/Configuration%3A-Include.html
- https://yalter.github.io/niri/Configuration%3A-Outputs.html#layout-config-overrides
- https://yalter.github.io/niri/Configuration%3A-Named-Workspaces.html#layout-config-overridesAlso merged ignore-drm-device which should let you passthrough a GPU to VMs: https://yalter.github.io/niri/Configuration%3A-Debug-Options.html#ignore-drm-device
After a detour to config includes and, again, several days/weeks of work implementing all edge cases and expected behaviors, true window maximize is ready and merged to main. Tricky cases like: windows requesting fullscreen and maximize after opening; windows failing to match the full maximized size; transparent windows with the niri border behind them.
https://yalter.github.io/niri/Fullscreen-and-Maximize.html
Give it a try! Ngl I mostly switched to maximize just because I'm too lazy to reach the keyboard for Mod+F.
-
After a detour to config includes and, again, several days/weeks of work implementing all edge cases and expected behaviors, true window maximize is ready and merged to main. Tricky cases like: windows requesting fullscreen and maximize after opening; windows failing to match the full maximized size; transparent windows with the niri border behind them.
https://yalter.github.io/niri/Fullscreen-and-Maximize.html
Give it a try! Ngl I mostly switched to maximize just because I'm too lazy to reach the keyboard for Mod+F.
-
One cool thing I noticed about true maximize is that apps like GIMP or Inkscape or Blender, that really want all available space, maximize themselves at startup, so you don't have to window-rule them manually in your config
-
Small change on niri-git for people using the foot terminal with CSD, or other apps that constrain their sizes to a grid: niri will now match the default column width to a preset width when a window opens. So opening foot sized "proportion 0.5" and then pressing Mod+R will switch you to the next preset width, even if foot actually opened slightly smaller to match its terminal grid. Before the change, the first Mod+R would pick the same "proportion 0.5" in this case and "do nothing".
-
Small change on niri-git for people using the foot terminal with CSD, or other apps that constrain their sizes to a grid: niri will now match the default column width to a preset width when a window opens. So opening foot sized "proportion 0.5" and then pressing Mod+R will switch you to the next preset width, even if foot actually opened slightly smaller to match its terminal grid. Before the change, the first Mod+R would pick the same "proportion 0.5" in this case and "do nothing".
Also, experimenting with this interaction tweak on a branch: what if dragging tiled windows horizontally scrolled the view instead of dragging them "out"? This makes it possible to scroll the view mouse-only without going through the Overview (the zooming gets quite tiring when it's frequent), and makes it possible to scroll the view touch-only. To drag the window out of the layout, you can still drag it downward.