wayland/core: move to new impl

[vaxerski]

It's the final countdown!

This MR moves the last parts of wayland over to Hyprland. After this is done, I'll focus on some cleanups for 1-2 weeks, then get onto the last bits of wlroots to be rewritten (backend)

TODO:

• wl_shm
• wl_compositor
• wl_buffer
• wl_subcompositor
• wp_viewporter
• mesa drm
• linux_dmabuf
• Testing
• Nix
• Meson

Bugs:

• shm seems to update oddly? (only tested with qt)
• shm issues with release timings
• foot won't show up
• rendering bugs: chromium
• ff/gtk won't open
• fuzzel crashes
• ff dt issues
• linux_dmabuf buffers won't pass through egl
• add more error checking for linux_dma
• dmabuf vram memory issues while resizing e.g. kitty
• viewporter issues in chromium

[rtgiskard]

Maybe add cpp modules?

[vaxerski]

no

[The-Briel-Deal]

Hey Vaxry, feel free to do whatever you like but is there a reason for the frequent force pushes on these large PR's? I love force pushes as much as the next guy but it makes it very hard to bisect for a bug when there is no commit history and its a 5000 line pr?

I've been trying to fix #6159 for a few days now and some sort of commit history on this wlroots rewrite pr (which was where gamescope broke) would help so much. Reflogs last in git for 90 days by default, is there any chance you could zip/tar your .git/logs folder so that I have something to bisect through to figure out what caused the issue?

Thanks for your hard work Vaxry! This issue has just been really bothering me for the last week and I really wanna get it fixed.

[vaxerski]

the reason is I rebase instead of squash when merging for these and thus want to keep commits to a few

is there any chance you could zip/tar your .git/logs folder so that I have something to bisect through to figure out what caused the issue?

I don't have access to my desktop rn. You can always manually click on a few commits and g reset --hard. They are not gone

[vaxerski]

If anyone wants to help, you can already test this MR, most stuff should work (outside of the unticked bugs above)

with firefox, it would be nice if someone could find what differs in the WAYLAND_DEBUG output between hyprland main (or sway) and this rewrite. I can't seem to get kdiff to work on my laptop and doing it by eye is painful

if any other bugs arose from this MR, let me know, I'll try to repro and fix.

[vaxerski]

cheers from the mountains fam :)

[Honkazel]

If anyone wants to help, you can already test this MR

It doesn't even build

[vaxerski]

hyprwayland-scanner-git needed

[Honkazel]

hyprwayland-scanner-git needed

FAILED: CMakeFiles/Hyprland.dir/src/render/Renderer.cpp.o 
/usr/bin/c++ -DHAS_EXECINFO -DHyprland_EXPORTS -DUSES_SYSTEMD -DWLR_USE_UNSTABLE -I/home/goose/srcmk/Hyprland/. -I/home/goose/srcmk/Hyprland/src -I/home/goose/srcmk/Hyprland/subprojects/wlroots-hyprland/include -I/home/goose/srcmk/Hyprland/subprojects/wlroots-hyprland/build/include -I/home/goose/srcmk/Hyprland/subprojects/udis86 -I/home/goose/srcmk/Hyprland/protocols -I/home/goose/srcmk/Hyprland/subprojects/udis86/libudis86 -isystem /usr/include/uuid -isystem /usr/include/pango-1.0 -isystem /usr/include/cairo -isystem /usr/include/glib-2.0 -isystem /usr/lib64/glib-2.0/include -isystem /usr/include/harfbuzz -isystem /usr/include/freetype2 -isystem /usr/include/pixman-1 -isystem /usr/include/libdrm -isystem /usr/include/libxml2 -isystem /usr/include/libpng16 -isystem /usr/include/libmount -isystem /usr/include/blkid -isystem /usr/include/sysprof-6 -isystem /usr/include/fribidi -O3 -DNDEBUG -std=gnu++23 -O3 -Wall -Wextra -Wno-unused-parameter -Wno-unused-value -Wno-missing-field-initializers -Wno-narrowing -Wno-pointer-arith -fmacro-prefix-map=/home/goose/srcmk/Hyprland/= -DWITH_GZFILEOP -pthread -Winvalid-pch -include /home/goose/srcmk/Hyprland/build/CMakeFiles/Hyprland.dir/cmake_pch.hxx -MD -MT CMakeFiles/Hyprland.dir/src/render/Renderer.cpp.o -MF CMakeFiles/Hyprland.dir/src/render/Renderer.cpp.o.d -o CMakeFiles/Hyprland.dir/src/render/Renderer.cpp.o -c /home/goose/srcmk/Hyprland/src/render/Renderer.cpp
/home/goose/srcmk/Hyprland/src/render/Renderer.cpp:3:10: fatal error: linux-dmabuf-unstable-v1-protocol.h: No such file or directory
    3 | #include "linux-dmabuf-unstable-v1-protocol.h"
      |          ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
compilation terminated.
[202/252] Building CXX object CMakeFiles/Hyprland.dir/src/render/OpenGL.cpp.o
ninja: build stopped: subcommand failed.
make[1]: *** [Makefile:15: release] Error 1
make[1]: Leaving directory '/home/goose/srcmk/Hyprland'
make: *** [Makefile:34: all] Error 2

[vaxerski]

ah uh oops heh you can remove that include. I'll nuke it when I am on my laptop

[VirtCode]

Aside from firefox not showing anything, wofi crashes for me after typing two letters. It also shows this warning right before the crash:

(wofi:41120): Gdk-WARNING **: 18:30:33.506: (../gtk/gdk/wayland/gdkwindow-wayland.c:918):buffer_release_callback: runtime check failed: (impl->staging_cairo_surface != cairo_surface)
wofi: ../cairo/src/cairo-surface.c:938: cairo_surface_reference: Assertion `CAIRO_REFERENCE_COUNT_HAS_REFERENCE (&surface->ref_count)' failed.
Aborted (core dumped)

Here's the coredump info (not sure whether that helps though): wofi-coredump.txt

Also, it currently doesn't build for me with meson, probably because of the missing quote there:

[vaxerski]

lmfao oops

[vaxerski]

I'll merge this, if it still happens open up a ticket with an ASan trace.

[vaxerski]

Thanks everyone!

[vaxerski]

heh on drm and release I also get the viewporter bug. Not on nested. I love programming

[vaxerski]

uhhh... I think there is something wrong with handling the buffers now that I am running it for a bit longer... :(

I've launched minecraft and I am getting nvidia experience'd with black flashing

[VirtCode]

hmm, minecraft works normally for me (on intel), in both fullscreen and tiled

[vaxerski]

20240608_124531.mp4

[VirtCode]

okay nvm, ran the wrong build,
can also reproduce it on intel

maybe it has something to do with glfw? both mc and kitty use that IIRC

[vaxerski]

it could - but it still happens only here

me rn

reeeeee.mp4

[vaxerski]

launched cs2, (SDL on X11) and no flickers are to be seen

[vaxerski]

what else uses glfw so I can do some testing?

[VirtCode]

i am also struggling to find good examples. basically everything made with lwjgl would be, but i don't have any other such games besides minecraft

I tried to run the official glfw examples (in their repo), and don't see any issue with them, but i also don't know whether they use all the features a normal program/game would

[vaxerski]

likely an issue with the interaction with glfw with the EGL mode as a backend, so kitty and minecraft fall into the category as they both use opengl

[VirtCode]

I think the issues with minecraft might be caused due to it running through xwayland. If i run the same version on native wayland (by using system glfw with -Dorg.lwjgl.glfw.libname=/usr/lib/libglfw.so), it runs fine for me without the stutters and black frames.

I tested with as vanilla a game as possible, so I'm using the official launcher (and bundled jre), with vanilla 1.20.4. My system glfw is that from the arch repos. Testing still has been a bit flakey with it sometimes refusing to start after adding the flag, but rebooting fixed most of that (a full reboot is probably not necessary, but I don't know where the issue is there)

However, it working through xwayland should still be supported because i think older versions require other versions of glfw which may not have a wayland patch.

[vaxerski]

it still is a regression from this MR, so I would guess I did some no-no with implicit sync somewhere.

[VirtCode]

yes, but that probably means that the issue isn't glfw but something else in the case of mc if I am not mistaken

[vaxerski]

chromium stretching popups fixed on main

[vaxerski]

nhekobug.mp4

@VirtCode you seein this shit too on intel? that's nheko

[VirtCode]

doesn't seem to happen for me neither on wayland nor xwayland

[vaxerski]

I have a patch ready. I'll just do some more testing

[vaxerski]

FIXED! 9994b73

[Aqa-Ib]

9994b73 also fixed the default mpv --vo=gpu.