niri-spicy: A showcase of various developers' feature branches, rolled together for ease of use

[kode54]

I have rolled together a feature branch, of various developers' extensions to both Smithay and niri, which I am using, dogfooding as you say, mainly because I enjoy doing this, and enjoy using whatever my latest adventure in coding happens to be. Yes, I am using all of these.

Currently, it features:

• Vulkan renderer
• HDR
• Direct Scanout for fullscreen display
• Tearing
• FIFO control
• Explicit Sync
• Cursor Warp

In detail:

• Rebase on Smithay master (dividebysandwich)
• Initial HDR / Color Management protocol support (dividebysandwich niri smithay)
• Enough HDR / Color Management to satisfy Wine Wayland, among other things (kode54 - folded into the same branch as above)
• Downmixing HDR screens to SDR for SDR consumers, such as screen captures, screen shots with SDR format requested, and the transition animations of resizing windows (kakxem - also folded into the above branch)
• Tearing control, DRM SyncObj, FIFO (urayde - niri smithay - Still based on older Smithay, not master, so I had to restyle the changes)
• Color pipeline support for direct scanout of fullscreen SDR on HDR, scRGB on HDR, and HDR10 on HDR, which also allowed re-enabling hardware cursor and the default disabled overlay planes (kode54 - niri smithay)
• SurfaceBarrier additions and commit timing protocol (cmeissl, reworked from delayed blocker + syncobj + fifo + commit timing #1449 by kode54)
• Pointer Warp (zakissimo, PR Add wp_pointer_warp_v1 protocol support #4001 , with a minor fix by kode54)
• Vulkan renderer, gated behind a debug {} option, which is only checked on startup. Adds a runtime requirement of shaderc for GLSL shader compatibility, which ends up adding a build time requirement of having CMake installed. (kode54)

(edited the following)

https://github.com/losnoco/niri/tree/spicy-main
https://github.com/losnoco/smithay/tree/spicy-master

And I made an accompanying AUR package based on niri-git for Arch and derivatives:

https://aur.archlinux.org/packages/niri-spicy-git

I am also asking for any feedback on this, including, but not limited to:

Suggested order for surgically splicing up the Smithay and niri contributions in their current form for (re)submission to those upstreams is welcome, since some of their original versions were either not updated in over a year, or were never submitted to upstream for approval.

[Axlefublr]

all of those changes are heavily internals-focused, which is fine and all, but not very exciting :c skimming over the bullet list, everything sounds like “boongles a tong to blammo a grufus”. my niri fork features a few changes that are a bit more clear in how they are beneficial to the end user, so mixing something like that in might make your fork more exciting for potential users :>

[Sempyos]

I would recommend just putting a few bullet points such as:
Features:

• HDR
• Tearing
• Pointer Warp

And then provide a more in depth description below (as you do somewhat already) or link to the respective PRs and repos

[kode54]

Added brief list, kept verbose list, added some fork/repo links where they were missing.

@Axlefublr I consider some of the features in the above fork, but I have to wonder. Why remove the Screen Shot UI? Did you replace it?

[Axlefublr]

After you see the ui a few times, what value does it bring? I already know I can press p to toggle showing my cursor, and I definitely know I can press space to finish the screenshot. It's clutter that also makes making screenshots of the area at the bottom of the screen more annoying

[veneficium42]

After you see the ui a few times, what value does it bring? I already know I can press p to toggle showing my cursor, and I definitely know I can press space to finish the screenshot. It's clutter that also makes making screenshots of the area at the bottom of the screen more annoying

You should probably make this a toggle in the configuration then. You don't need it but some people who don't know the keybinds do.

[Axlefublr]

that's why it's a personal fork and not an upstream attempt :p I might've made it sound like my fork is meant to be used by others, but nope! note how there's no readme

[Brisingr05]

I'm getting a 404 error for the niri branch.

[kode54]

Dang, I am dumb, I typed the wrong repo in my links.

[kode54]

Updated several times. Too much?

[kode54]

Bah, I need to rewrite history. I am otherwise being dishonest, and I was only initially being dishonest because I thought the other contributors who were telling me their code would never be accepted upstream might also be hiding something. I can't speak for them, though. Will be rewriting history of my own commits to the entire tree down back to the first post-master/post-main commit.

It really seems dumb, since it's probably quite obvious from the timing of the commits. Hopefully I haven't produced something that's totally unmaintainable as a result of this.

[cspoelstra]

Please have the AUR package provide "wayland-compositor", it currently breaks the dependency chain if niri is the only installed wayland compositor.

[kode54]

Done. Thanks for pointing that out.

[hwamil]

thanks man, not sure what everything else does but really enjoying the hdr

[ColdLemonTea]

Hi @kode54,

Thanks for maintaining this awesome niri-spicy branch with so many cutting-edge features!

I would like to request adding support for the ext-image-copy-capture-v1 protocol to this branch.

Why this is needed

As a next-generation standard Wayland protocol, ext-image-copy-capture-v1 introduces much better continuous stream capturing, damage tracking, and native cursor capture support compared to the older wlr-screencopy.

Many modern tools (such as WayVNC, newer versions of OBS/grim, and various remote desktop/screenshot tools) are transitioning to or already require this protocol for optimal performance and per-window/toplevel capturing.

Benefit for niri-spicy

Since Niri upstream is still tracking this feature (#1558), having it integrated into niri-spicy beforehand would be amazing and perfectly aligns with this branch's goal of experimenting with advanced, performance-focused extensions (like Vulkan and Explicit Sync).

Could you please consider adding this protocol to your branch or downstream Smithay modifications when you have time?

Thank you so much for your effort!

[veneficium42]

this reads painfully written by an LLM :/

iirc the wlr screencopy protocol has similarities with the new copy-capture protocol and the infrastructure for this in niri, compared to other features in this fork, is largely there so i would rather see a PR to be merged into mainstream niri compared to something part of an already pretty featureful fork.

[Axlefublr]

thanks for saying it. I also wanted to express this, but couldn't find the right words for it

[ColdLemonTea]

this reads painfully written by an LLM :/

iirc the wlr screencopy protocol has similarities with the new copy-capture protocol and the infrastructure for this in niri, compared to other features in this fork, is largely there so i would rather see a PR to be merged into mainstream niri compared to something part of an already pretty featureful fork.

sry for using llm to write that comment. English is not my first language, and i want to make sure my point came across clearly. thx for your understanding.

[Axlefublr]

that's understandable, but note that in modern day, human words, even if poorly laid out, are a lot more valuable and appreciated than inhuman, well laid out words

[Skyrocket1653]

Hi, I was playing Risk: Global Domination with the latest proton-cachyos-slr and I had the game set to fullscreen and in my niri config I have block minimize to true for all windows but it seems that the game minimizes itself when it is not focused despite block minimize set to true. For now I switched the game to borderless window mode. I believe this was not happening before. Did I miss something?

[kode54]

You have to use Wine Wayland and set an environment variable to disable it from propagating focus loss to the app.

It will forcibly minimize itself otherwise, and if there is no minimize implementation, it will just restore itself to a floating window that hovers in place regardless of where you scroll your desktop.

This is a limitation of Windows and Wine, not Niri.

[Skyrocket1653]

Oh, what is the environment variable for stopping propagating focus loss to the app? I can't seem to find it.

I'm not sure whether this is relevant but I tried it with normal niri and using wine wayland and it does not minimize itself in normal niri.

[Skyrocket1653]

I've found the environment variable (WAYLANDDRV_FOCUS_LOSS=0). But I'm still curious why it does not minimizes itself under normal niri or whether this can be improved upon.

[Guara92]

Thanks a lot for this! I'm testing it (from AUR package) on my machine with an Intel Arc B580 and I noticed a couple of things that could worth pointing out for others:

• Currently with vulkan-renderer Firefox with gfx.webrender.compositor enabled crash at the first frame decoded

(niri) Wayland protocol error: zwp_linux_buffer_params_v1#85: error 4: Format DrmFourcc(NV12)/3231564e is not supported.

disabling the compositor force Firefox to do a format conversion and workaround the limitation, also I see a lot of noise around shader-cache: Failed program_binary and Failed to load a program object but I'm not sure it's directly related to this work

• On the HDR enabled monitor I notice small and not so random fluctuation of the brightness, it seems on 4/5 different "level" incrementally then after a while back on the lowest and step up again, it's not too annoying but definitively strange (default reference-luminance)

other than that the desktop runs very smoothly and solid!

[Guara92]

edit: the firefox crash happens only with https://github.com/CYFARE/HellFire, regular ff build (even developer) works fine

[IlChitarrista]

Hi, so far it's working well, though I've found a regression with Direct Scanout in Peak running Wayland Native (DX11):
Direct Scanout no longer works unless the window isn't focused (tested both on single and multi-monitor setup)

I can't reproduce the same issue in The Witness and Outer Wilds which are also DX11 games running under the same Proton version with Wayland enabled.

[IlChitarrista]

This is only related to using Immediate rendering, if FIFO or Mailbox are used it's not an issue. Technically not a regression then, if Direct Scanout isn't implemented for Immediate presentation.
Setting MESA_VK_WSI_PRESENT_MODE=mailbox fixes it even though immediate rendering is still reported by Mangohud... I wonder if it's misreporting.