Allow opting-out of V-Synced cursor updates.

[ratmarrow]

(Before it is mentioned, yes, I have tried forcing hardware cursors, forcing mouse acceleration to be disabled, the works. I combed through as much of the documentation as I could and found no configuration of options that creates the desired behavior.)

Initially, when I was much less knowledgeable on these sorts of things, I had made an issue (#8439) lamenting the lack of a global tearing toggle. After having raised the issue, and looking more into it, I realized it wouldn't have solved the problem I was experiencing.

I had learned, both from rudimentary experiments done by Mort in his blog post Hard numbers in the Wayland vs X11 input latency discussion and the following post by Asahi Lina about it, that Wayland compositors perform V-Synced cursor updates, unlike X11.

This creates a trade-off of cursor latency for virtually perfect frames.

For the record, I don't think this is a bad thing in isolation, people are allowed to like this behavior and I fully support that! I wouldn't want to see this removed, I just personally think that allowing users (insert "there are DOZENS of us!" line) who are willing to sacrifice perfect frames the option to avoid the behavior if possible.

Personally, this one option alone would be the catalyst for me to jump ship from X11, no matter how many hoops I'd have to jump through to have this on my machine.

[ratmarrow]

Oops! Forgot to mention the setup I am on!

Distro: Arch Linux x86_64
Kernel: 6.15.3-arch1-1
CPU: AMD Ryzen 7 5700x (16) @ 4.67 GHz
GPU: NVIDIA GeForce RTX 4060 [Discrete]
Monitor: 1920x1080 @ 60.000hz
Memory: 16GB (2x8GB)
Mouse: Endgame Gear OP18k (1600 DPI, 2000hz Polling)

[vaxerski]

afaik AMS doesn't allow for this, you either tear everything or tear nothing. When tearing is active, cursor is also torn

[ratmarrow]

All I have to go off of is Lina’s mention of implementing the legacy ioctls to get the behavior, and her insistence on it not being too difficult, but unlikely to be seen in compositors. Beyond that, it’s mostly crickets as far as I was able to find.

I wish I could be of more help, but truthfully I am a little out of my depth.

[ratmarrow]

Would you possibly be able to point me in the right direction so I could attempt to cobble together some hack for personal use?

I don't know if it would go anywhere, but I figure giving it a Hail Mary would ease my soul a little bit.

[vaxerski]

you'd need support for actual proper legacy ioctls in aquamarine (the current state of legacy is very broken there) and get hyprland to be able to render using that, would probably require a lot of code that you're unlikely to get right

[ratmarrow]

Well that's unfortunate :(

Thanks for responding anyhow, hopefully in the future someone more equipped for the task will step up.

✌️ best wishes! :)

[vaxerski]

my hope is people at dri-devel/drm will realize this is a thing we want and actually add an option to do this in ams

[ratmarrow]

Huge necro-post, but I took it upon myself to reach out (and effectively plea) to dri-devel about this issue.

I received a response from Michel Dänzer, who told me that, at least at the moment,
it’s up to the compositor to implement a solution.

He put work into using the legacy functionality X11 uses as a workaround in Mutter until a modern solution can be found. It didn’t end up going through, and this is what he had to say to me about it:

I was asked by kernel KMS developers not to do this though, because it might tickle kernel bugs (in which case Xorg should tickle at least some of them as well though...), so I abandoned it.

I wonder if something similar to this could maybe be considered as an experimental workaround in Hyprland?