07:58 <daniels> mntmn: I have no idea, and even more weird that it’s fixed in new versions :(

10:11 <agx__> marex: no chromium here, sorry

10:13 <agx__> marex: if you just need a deb https://salsa.debian.org/sdl-team/libsdl2/-/merge_requests/11 should work - i doubt we'll pull all that in for bullseye but i'll look into keeping that in shape for bookworm

11:25 <mntmn> daniels: what needs to go wrong in the wayland compositor to sample a surface as linear, but which is tiled? (assuming that this is what's happening -- it could also be the other way around?)

11:28 <mntmn> the same xwayland binary works correctly with another compositor (sway). is there something special about xwayland's wayland surface?

11:30 <lynxeye> mntmn: If the compositor doesn't properly specify the modifier when importing the client buffer, etnaviv will just assume that it is linear. Etnaviv doesn't have the magic side-channel to get the tiling information, unlike intel or amd.

11:34 <daniels> I'd look at $WAYLAND_DEBUG=server output from Mutter vs. sway to try to see which buffers are coming in and out

11:41 <mntmn> lynxeye, daniels: ok, thanks. i'm gonna try that

12:05 <marex> agx__: I'll keep tracking testing, so Im not really affected, but thanks :)

13:49 <mntmn> hmm, gnome-shell seems to ignore WAYLAND_DEBUG.

13:50 <mntmn> new puzzle piece: xwayland looks ok when XWAYLAND_NO_GLAMOR is set

14:23 <mntmn> ok so i added some debugging to xwayland. xwl_glamor_gbm_create_pixmap behaves differently on wlroots vs mutter. glamor_get_modifiers() returns 0 modifiers on mutter but 3 modifiers on wlroots.

14:23 <mntmn> on sway, it's format: 34325258 num_modifiers: 3 modifiers: 0x5589ab3a40

14:23 <mntmn> but on mutter its 0, NULL.

14:45 <mntmn> daniels: i have determined that under mutter, there are only 9 xwl_formats and none of them have modifiers, while under sway, there are 27 formats with 3 modifiers each. does this have to do with how the compositor sets up the screen or is it something related to wayland protocls (versions)?

14:46 <mntmn> oh https://gitlab.gnome.org/GNOME/mutter/-/issues/1229

14:46 <daniels> oooooooooooohhhh actually ... I don’t know why I didn’t realise this earlier

14:46 <mntmn> daniels: realised what? :D

14:46 <daniels> https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/75

14:47 <daniels> you need to set a gsettings key

14:47 <mntmn> wat

14:48 <daniels> Intel screwed it up

14:48 <mntmn> omg

14:49 <mntmn> this would explain why the same version works on my other dev system?!

14:49 <daniels> modifiers on Intel unlocks better (= more efficient) compression modes, but it seriously blows through display FIFO

14:49 <daniels> so Mutter disabled it by default

14:49 <lynxeye> daniels: org.gnome.mutter.pretty-please-dont-screw-etnaviv ?

14:49 <daniels> really that should be on Intel only

14:49 <mntmn> lol

14:49 <daniels> lynxeye: srsly

14:50 <daniels> mntmn: sorry, I don’t know why I didn’t think of this earlier :( well, I do actually, everything is on fire, but ...

14:50 <mntmn> *googles how to change gsettings things*

14:51 <mntmn> am i doing this right: gsettings set org.gnome.mutter experimental-features "['kms-modifiers']"

14:57 <mntmn> daniels: thank you! this solved it

14:57 <mntmn> man this was painful

14:57 <mntmn> i hope i won't forget to check for this modifiers business the next time everything gets trashed

14:58 <daniels> heh

15:00 <daniels> mntmn: btw, MOD_INVALID isn't -1, it's (1 << 56) - 1

15:00 <mntmn> daniels: lol ok. i just didn't want to type a million fs

15:00 <daniels> (yes it should be 0)

15:01 <mntmn> if you know the people who turned off the modifiers in mutter please say hi from me

15:10 <mntmn> the good news is that gnome3 on etnaviv works and is even performant

15:18 <pcercuei> nice!

15:19 <pcercuei> I want to test it on my gcw-zero at some point (GC680), but I need to get the HDMI to work reliably first

15:22 <daniels> \o/