with the Debian update to mesa 18.1 I started having a complete blank screen on xfwm4 (4.13) when compositing is enabled. I've reported a bug on Debian BTS (https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=901789) but I'm unsure if the bug lies in mesa or xfwm4 so I'm reporting a bug here.
The xfwm4 4.13.0-1 Debian package is built with epoxy support. Disabling compositing or building without epoxy fixes the issue. I haven't tried yet to build with epoxy and run with XFWM4_USE_PRESENT as indicated in #13519 but I'll try and report back.
I have an Intel embedded video card and use the modesetting DDX and i915 kernel module.
If you need more information, don't hesitate to ask.
Yes, XFWM4_USE_PRESENT=1 works with compositing enabled even when built with libepoxy.
Without epoxy, you don't have GL support, so forcing Xpresent or building without epoxy is basically the same, you're not using GL. And if the bug is related to mesa, then no wonder not using GL avoids the issue.
(In reply to Olivier Fourdan from comment #2)
> Without epoxy, you don't have GL support, so forcing Xpresent or building
> without epoxy is basically the same, you're not using GL. And if the bug is
> related to mesa, then no wonder not using GL avoids the issue.
Yeah I agree. But any idea where to look at (whether in mesa or in xfwm4)? Can you reproduce the bug yourself?
Created attachment 7811
(In reply to Yves-Alexis Perez from comment #3)
> Yeah I agree. But any idea where to look at (whether in mesa or in xfwm4)?
> Can you reproduce the bug yourself?
Well, it's between the two, probably xfwm4 is not doing something correctly but hat was working until now, and that changed recently...
I guess you can use the attached source code as a reproducer, right?
Humm, looks like this example from khronos.org doesn't work either:
(In reply to Olivier Fourdan from comment #4)
> I guess you can use the attached source code as a reproducer, right?
Not entirely sure how it should behave, but indeed the popped window is just plain white, console output is:
Using root as parent
Using texture type GL_TEXTURE_RECTANGLE_ARB
1/26: xvisual id 0x13e != 0x21, skipped
2/26: xvisual id 0x146 != 0x21, skipped
3/26: xvisual id 0x161 != 0x21, skipped
4/26: xvisual id 0x169 != 0x21, skipped
5/26: xvisual id 0x153 != 0x21, skipped
6/26: xvisual id 0x176 != 0x21, skipped
7/26: xvisual id 0x154 != 0x21, skipped
8/26: xvisual id 0x177 != 0x21, skipped
9/26: xvisual id 0x155 != 0x21, skipped
10/26: xvisual id 0x178 != 0x21, skipped
Using texture target GLX_TEXTURE_RECTANGLE_EXT
Using texture format GLX_TEXTURE_FORMAT_RGBA_EXT
I think this is not Mesa related but the Xserver, on a laptop where the example was working fine with Xorg 1.9.6, I upgraded to Xorg 1.20 and the issue appeared, with the same exact same Mesa version which was working before (i.e. upgrading just the Xserver from 1.19 to 1.20)
So... Which Xserver version do you use?
(In reply to Olivier Fourdan from comment #7)
> I think this is not Mesa related but the Xserver, on a laptop where the
> example was working fine with Xorg 1.9.6, I upgraded to Xorg 1.20 and the
> issue appeared, with the same exact same Mesa version which was working
> before (i.e. upgrading just the Xserver from 1.19 to 1.20)
> So... Which Xserver version do you use?
I'm on Xserver 1.20 indeed. Upgrade from 1.19.6 to 1.20 was on May 26th and I think everything was ok at that point. Mesa update (from 18.0.5-1 to 18.1.1-1) was on June 13th. Maybe it's a combination of both then. Debian testing has xserver 1.19.6 so I'll try to downgrade it and report back
When running with Xserver 1.19.6 the tftp-test binary shows a red background instead of a white one.
I encountered a similar problem: I installed mesa 18.1.3 and xorg 1.20 from backports and got a white screen after rebooting
That's weird, I can get xfwm4 to work with GL just fine using xserver, mesa and libglvnd current, all from their respective git master.
$ MESA_DEBUG=1 MESA_INFO=1 ./tfp-test
on your system?
command not found
(In reply to vladimir from comment #13)
> command not found
Well, ahumm, my comment was assuming you had built the reproducer (attachment 7811 ) I posted previously in comment 4...
(In reply to Olivier Fourdan from comment #11)
> That's weird, I can get xfwm4 to work with GL just fine using xserver, mesa
> and libglvnd current, all from their respective git master.
Actually, I take that back, 18.0.x works., 18.1.x and git master fail.
I filed bug https://bugs.freedesktop.org/show_bug.cgi?id=107117 to get some insight here, as that might be a regression in mesa maybe.
Which DDX are you using for your tests?
I just figured that it works with the intel DDX so I wonder if that could be related to the modesettings DDX with glamor instead...
I've tested with composit enabled and xserver-xorg-video-intel Vers. 2:2.99.917+git20171229-1+b1 and it works fine.
Xfce 4.13 - complete compiled from git.xfce.org
xfwm4 4.13 - compiled with libepoxy
xserver-xorg-core Vers. 1.20.0-3
mesa Vers. 18.1.4-1
Intel Core i3-7100U Kabylake NUC
Intel HD Graphics 620
With composit and driver modesetting, I get the same error as described by yves in Debian Bug #901789 (white screen).
Thanks, your comment matches my own observation I posted in https://bugs.freedesktop.org/show_bug.cgi?id=107117#c19
Out of curiosity, does it work with "allow_rgb10_configs=false" set ? (you can use the simple reproducer from https://bugs.freedesktop.org/attachment.cgi?id=140540)
Created attachment 7835
Can you try with this updated test program see f that one works?
Olivier Fourdan referenced this bugreport in commit dc1610147ddfc77df40cf24c88c8083f24e531ef
compositor: Skip ARGB10 GLX configs
I've just pushed a couple of changes, can you retry with xfwm4 from current git master?
I have compiled from git master, but unfortunately the error (white screen) is still there.
Can you please try comment 19 and comment 20 as well?
I am not sure what to do with it. I am not a programmer, only a user.
Is this a patch or should the attachment be copied somewhere?
It's a test program. it can be compiled and run, it mimics what xfwm4 does so you can try without running xfwm4.
I wonder if setting the envvar "allow_rgb10_configs=false" globally (i.e. even before starting the Xserver) would help.
Created attachment 7836
This is the result of my test with tfp-test. The error is the same as before, you can see the white screen
test result of tfp-test
For the record, "allow_rgb10_configs=false" makes no difference here, not even before starting xorg with modesettings so this is unrelated to RGB10.
FWIW I've posted a patch for Mesa to address this issue, you may want to try with that patch:
Closing as "not our bug", the fix has been posted to mesa-devel and will be worked on from there.
*** Bug 15765 has been marked as a duplicate of this bug. ***