14:48 mntirc: is there a trick to use apitrace on gnome / how to actually capture opengl traces of i.e. gtk4-demo? i have some reproducible graphics corruption on freedreno/a740, but don't know how to record it. apitrace only generates 1 frame and the replay is empty, even with GSK_RENDERER=gl
14:54 mntirc: problem goes away with FD_MESA_DEBUG=flush, some kind of cache issue?
14:54 mntirc: (the corruption i mean)
14:59 soreau: if you are able to capture the corruption with screencapture software, it would probably suffice to file an issue and explain how to reproduce the problem
15:00 soreau: but it sounds like a mesa or graphics driver bug since it doesn't happen with other drivers
15:16 mntirc: ok. here's a screenshot of what it looks like https://mastodon.social/@mntmn/116664087369079537
15:19 soreau: mntirc: FWIW there is a similar corruption that happens on gles2 renderer with wlroots compositors on proprietary nvidia driver and the fix is to glFlush() when rendering is complete so the driver knows when to sync
15:19 soreau: but compositors don't really use swap_buffers like I expect gtk does for the egl case
15:20 soreau: so the driver should know what to do, which is why you should probably file a mesa issue
15:20 mntirc: yeah, it must be freedreno/tu specific as the same software stack works fine on panfrost/panvk for example
15:20 soreau: right
15:20 mntirc: also, blender has really strange rendering bugs on freedreno, but they go away with zink
15:20 soreau: it's a big clue that you found an env var setting that fixes it
15:21 soreau: so I'd put your env var fix and the screenshot in the report
15:21 mntirc: ok, will write it up tomorrow night, thanks
15:22 soreau: np
18:06 Company: mntirc: renderdoc has worked for me with GTK
18:07 Company: mntirc: and in your gtk-demo screenshot it looks like your driver is screwing up the gradient in the titlebar, too