1622487 - Fullscreening a video on a second monitor doesn't work, creates a crazy checkerboarding effect

Status: RESOLVED INCOMPLETE

(Core :: Widget: Gtk, defect, P1)

Description

[Jeff Walden [:Waldo]]

Attachment: nhl-glitch.jpg

STR:

1. Have a laptop, and connect a second monitor to it. (Mine is a Samsung TV, through my laptop HDMI port.)
2. Open a window on the second monitor.
3. Go to some YouTube video -- I used https://www.youtube.com/watch?v=SR56GJ05IMs but it shouldn't matter which. (I could reproduce it on the NHL streaming site too.)
4. Start the video.
5. Hit the fullscreen button in the player.

Instead of the video fullscreening legibly, the video appears horrifically mangled, as in the attached picture. (The picture's from an NHL stream, but you get similar effects with a YouTube video.)

mozregression indicates the regression is in this range, which seems pretty plausible for my Linux/GTK system:

https://hg.mozilla.org/integration/autoland/pushloghtml?fromchange=57f9a3f2c4069bef8166df880d0b4ad8c0e92c45&tochange=0a6d64975cfa90788adf7851053f48ef523babb7

which fingers bug 1617115.

[jwalden@find-waldo-now src]$ uname -a
Linux find-waldo-now 5.4.18-100.fc30.x86_64 #1 SMP Fri Feb 7 14:37:00 UTC 2020 x86_64 x86_64 x86_64 GNU/Linux

And here's the gtk+ and gtk2 I have installed:

Name         : gtk+
Epoch        : 1
Version      : 1.2.10
Release      : 90.fc30
Architecture : x86_64
Size         : 2.7 M
Source       : gtk+-1.2.10-90.fc30.src.rpm

Name         : gtk2
Version      : 2.24.32
Release      : 6.fc30
Architecture : x86_64
Size         : 14 M
Source       : gtk2-2.24.32-6.fc30.src.rpm

Based on my very-quick skim of the bug that regressed this, this failure mode seems much worse than that in bug 1617115. That change probably only happened to stumble into a much larger problem -- but I'd say this really has to be fixed at least at the same time this is fixed, if it were me.

Comment 1

[Martin Stránský [:stransky] (ni? me)]

It's interesting how such innocent change as different window stacking can behaves on various systems :) I really don't wonder why we're missing HW acceleration on Linux when even such change leads to such ugly regressions.

Anyway, we need to revert bug 1617115 and just live with it I guess.

Comment 2

[Darkspirit]

Looks similar to bug 1494520 and bug 1569745 which might be the Gnome-equivalent of bug 1502519. (OpenGL/WebRender-only)

Comment 3

[Martin Stránský [:stransky] (ni? me)]

To be sure, can you try to run Firefox as

MOZ_GTK_TITLEBAR_DECORATION=system ./firefox

and confirm you don't see the bug? Because that's the only change of bug 1617115.

Thanks.

Comment 4

[Martin Stránský [:stransky] (ni? me)]

Guys, can anyone else confirm the regression from bug 1617115 here?
Thanks.

Comment 5

[Martin Stránský [:stransky] (ni? me)]

Jeff, please reopen if you can provide the info.
Thanks.