Rio frozen and unresponsive on Arch Linux Wayland (nvidia)

[zackartz]

Hello!

I have been trying to get rio working on arch linux & hyprland with no luck. debug log here. If anyone knows what to do to fix this I would appreciate it!

[raphamorim]

Hey @zackartz thanks for the issue,
I suppose it renders but then get stuck right?

[zackartz]

@raphamorim yeah, it renders one frame and gets stuck, i have done a little digging and i believe this is an upstream winit issue as their examples don't seem to be working for me, and lapce also behaves similarly

[zackartz]

interestingly, neovide works fine, and uses the same winit version (0.29.4)

[raphamorim]

Hey @zackartz is this still happening in the 0.0.34?

[postsolar]

I have the same issue with Nix and Hyprland with an nvidia card, it renders the first prompt and doesn't respond to any inputs. Also closing the window doesn't work.

> rio --version
rioterm 0.0.34

[henriquekirchheck]

Same thing here in NixOS and Hyprland on Nvidia. I have some logs from the app when I ran it with RIO_LOG_LEVEL=debug

logs.txt

[raphamorim]

Thanks for all the debug data, looks the pty and sugarloaf didn't find any errors on initialization (at least based on what I saw from the logs). I also don't think is winit, otherwise the type of error would be another.

I have a question though, if you try to run from main with debug binary make dev and put a println! in the render function on screen, what it happens? is rio loop still running?

[zackartz]

i added a println! on the render event in the main event loop, and it is only called once

[zackartz]

adding a println! to the top of the event loop, logging each event, outputs this before freezing.
log.txt

[raphamorim]

Also closing the window doesn't work.

Hmm, it does look as it's some Winit problem tbh. Wondering if it was recent winit update that has caused this.

Looks this issue is similar to alacritty/alacritty#5174 and alacritty/alacritty#2453

[yofriadi]

I also experienced this with RiverWM, also it doesnt get resized when I try resize it.

[TitanNano]

I'm also experiencing this on Fedora 39. Both on master and on 0.0.x

[zackartz]

Did some digging, looks like its stuck waiting to grab a new texture. This is the call that is stuck, nothing in either part of the match statement gets to run. Maybe a VSYNC related issue?

[johnghall]

Also experiencing this on Sway + nvidia

[zackartz]

found it! This is a wgpu issue gfx-rs/wgpu/#4775, and while the issue is still open, if you change the Cargo.toml to pull wgpu from git the problem goes away.

[raphamorim]

wow, such a good finding! Is the fix in the main branch?

[postsolar]

I can confirm that the fix suggested by @zackartz works.
Input latency is very high though. I used the flake provided in this repo to build Rio.

[zackartz]

@raphamorim i did some more testing, it looks like it works as long as wgpu is called from git, i tested the branch trunk, as well as tags from 1.19.3 to 1.18.2 (anything older wouldn't build) and they all perform about the same. I can also confirm the high input latency.

along with the input latency there are also some rendering errors
(notice the differing indentations, and totally broken blur, config for reference). it is definitely a nvidia driver/wgpu vulkan problem though.

update: Using nvidia driver 550.67 (currently in beta) fixes everything! Use nvidia-beta if you are on Arch or config.boot.kernelPackages.nvidiaPackages.beta; if you are on NixOS!