#
✕ nick:marcan
#asahi · 2022-12-08 12:15
marcanpiroko9: thanks :)
#asahi · 2022-12-08 12:13
marcanboth that and the scaling settings are stored independently for wayland and X
#asahi · 2022-12-08 12:13
marcankh: Check your DPI settings as the post mentioned. DPI forcing should be disabled.
#asahi · 2022-12-08 09:43
marcanyup
#asahi · 2022-12-08 09:42
marcanit's clear ~nobody cares any more
#asahi · 2022-12-08 09:42
marcanbut so much is broken about xorg, really...
#asahi · 2022-12-08 09:42
marcannot that it matters since neither work for us
#asahi · 2022-12-08 09:42
marcanit tries to use a newer API and fall back to an older one, but the logic for that is busted so it always uses the newer one
#asahi · 2022-12-08 09:42
marcanincidentally, I looked at the relevant Xorg code and it's also broken :)
#asahi · 2022-12-08 09:40
marcanI don't think we want to go there
#asahi · 2022-12-08 09:40
marcanI think in principle you could emulate the vblank stuff by queuing dummy page flips and waiting on those, but it sounds like a recipe for corner cases and pain and higher power usage
#asahi · 2022-12-08 09:39
marcanbut are the right speed in Wayland
#asahi · 2022-12-08 09:39
marcansame reason some KDE animations are way too fast on xorg (as they've always been on Asahi, since simpledrmfb also doesn't have vblank IRQs of course)
#asahi · 2022-12-08 09:38
marcanbecause Xorg apparently doesn't tie compositor page flips to the actual app swap interval, it instead relies on vblank IRQs to throttle clients and those don't work
#asahi · 2022-12-08 09:37
marcanso without TearFree you get tearing in glxgears at a bajillion FPS. with TearFree you get a smooth glxgears, still at a bajillion FPS behind the scenes
#asahi · 2022-12-08 09:36
marcanxorg gets errors on the relevant ioctls and so never waits
#asahi · 2022-12-08 09:36
marcanthere is no vblank emulation/IRQs without the actual vblank stuff
#asahi · 2022-12-08 09:36
marcan_jannau_: that function is misnamed, it's not a vblank event, what it really is is a swap complete event
#asahi · 2022-12-08 09:35
marcanand DCP is strictly built with this model in mind, hence it works well for macOS and Wayland, not so much Xorg
#asahi · 2022-12-08 09:35
marcanmacOS switched to this model ages ago (Wayland is basically doing what macOS does)
#asahi · 2022-12-08 09:34
marcanbut as we know Xorg is stuck 20-30 years in the past :)
#asahi · 2022-12-08 09:34
marcanso every display controller implementation switched to hardware double-buffering of the registers with automatic flips
#asahi · 2022-12-08 09:34
marcan(because you cannot guarantee latencies any more so you cannot guarantee the actual flip happens during vblank)
#asahi · 2022-12-08 09:33
marcansoftware buffer flipping has been broken for ~20 years
#asahi · 2022-12-08 09:33
marcanpresumably you just queue a flip and the hardware issues it at the earliest possible time, quantized to whatever it can do or whatever
#asahi · 2022-12-08 09:33
marcanthat's how it works on every modern display controller
#asahi · 2022-12-08 09:32
marcanthe hardware flips buffers
#asahi · 2022-12-08 09:32
marcanthey'd have to implement the vblank emulation in xorg proper for this to work somewhat better
#asahi · 2022-12-08 09:32
marcanTearFree doesn't change the way xorg implements client vsync (which is apparently based on explicit vblank queries which don't work)
#asahi · 2022-12-08 09:31
marcanlina removed the fake IRQ thing because it was broken, right?
#asahi · 2022-12-08 09:31
marcan_jannau_: there is no vblank IRQ support so there is no vblank ioctl support so clients just freewheel even when they think they are vsynced
#asahi · 2022-12-08 09:30
marcanyeah, I'm looking forward to the state of things mid-2023 or so, looks like ~everything is coming together nicely on the desktop side
#asahi · 2022-12-08 09:29
marcanbut fractional scaling creates piles of corner cases for interactions, and as far as I know that's why that protocol took so long to work out
#asahi · 2022-12-08 09:28
marcanQt has natively supported fractional scaling for ages
#asahi · 2022-12-08 09:28
marcankujeger: I think KDE has been doing its own thing for a while now
#asahi · 2022-12-08 09:27
marcan(because the disadvantage of output scaling is you burn more GPU power by pushing more pixels)
#asahi · 2022-12-08 09:27
marcantrue native fractional scaling is hard and Wayland will have a leg up on macOS there if it all gets implemented nicely and works
#asahi · 2022-12-08 09:27
marcanthough FWIW: as far as I know non-KDE desktops mostly implement fractional scaling via output scaling and... guess what, that's how macOS does it too.
#asahi · 2022-12-08 09:27
marcankujeger: it works well in kwin except for some glitching in weird cases. I hope the bikeshedding about how to do the rounding "perfectly" I hear happened trickles down into that and fixes it
#asahi · 2022-12-08 09:26
marcan*personally* I still have a partial Xorg blocker (inputleap/barrierd support) but that's me, and it looks like progress is being made there plus there's at least a Wayland version of the client anyway
#asahi · 2022-12-08 09:25
marcanthose concerns no longer apply; the Xwayland on KWin stuff got fixed and Wayland is a significantly smoother experience than Xorg with drivers
#asahi · 2022-12-08 09:25
marcanFWIW, when Asahi first shipped the alpha I chose Xorg because I think it was faster than Wayland without a GPU driver, and because KWin/Xwayland fractional scaling still wasn't working which made Firefox end up blurry, and I didn't have time to chase down these things since the actual "desktop experience" part of everything was quite rushed
#asahi · 2022-12-08 09:19
marcanthe only way it will get reset again is if you start playing really weird games with pacman package versions, it won't happen by accident
#asahi · 2022-12-08 09:19
marcanif you don't like it, choose xorg again on the login screen
#asahi · 2022-12-08 09:18
marcanthat is a one-time action on package upgrade
#asahi · 2022-12-08 09:18
marcanby "force switch" I mean I will edit their SDDM config to switch to wayland, because all existing users on xorg should be switched to wayland, because there is no way to determine intent ("xorg by default" vs "xorg on purpose") and the majority will be of the former category
#asahi · 2022-12-08 09:17
marcaneven though the blogpost told people to use wayland
#asahi · 2022-12-08 09:17
marcanI even shipped an xorg conf with the alpha GPU driver stuff to make xorg actually work (it doesn't work by default because its device selection algorithm is broken by default, see what kettenis said)
#asahi · 2022-12-08 09:16
marcanmps: dude I'm literally just talking of changing the default to wayland, I already said that, I'm not sure what you understood
#asahi · 2022-12-08 09:16
marcanand that is unlikely to ever be supported because it doesn't look like the hardware does at all

plus ›