Screen Flicker With Virtualbox 5.x

I opened a bug report at Oracle for Virtualbox;

https://www.virtualbox.org/ticket/14701

There seems to be some sort of conflict going on, either from compton or virtualbox...

Hmm

This was happening on the last version of Slackware 14.1, and I've recently upgraded to 14.2 as a clean install.

Here are some system specs;

Slackware 14.2 x86_64 (multilib)
Openbox 3.6.1
Nvidia GTX 660m GPU
Nvidia drivers 352.63
xorg-server-1.18.3
latest version of compton from git

From VirtualBox 5.0.3 onwards I noticed when I start the app the entire desktop flickers for a brief second as it starts.

I made a video of this you can download to watch;

https://ln.sync.com/dl/e841038d0#9rbbx4r8-9bgvy5hg-s7m92shh-zzd4q97e

Disabling compton or adding a rule to remove Virtualbox from shadows, I no longer see the issue.

This is my compton config;

http://dpaste.com/21DAWBG

[richardgv]

I am not able to reproduce the slow start issue with VirtualBox 5.1.2 on Gentoo ~amd64. Are you still experiencing it?

As the the flickering, it is caused by the shadow (and possibly fading) of a short-lived maximized frameless window in VirtualBox created in UIDesktopWidgetWatchdog::sltRecalculateHostScreenAvailableGeometry() (VirtualBox-5.1.2/src/VBox/Frontends/VirtualBox/src/globals/UIDesktopWidgetWatchdog.cpp). The window is seemingly used to detect the available space on the desktop. It does not have a unique characteristic to be matched against, but you could try the following pattern (tested on Openbox 3.6.1):

--shadow-exclude 'class_g = "VirtualBox" && name = "VirtualBox" && (_NET_WM_STATE@[0]:a = "_NET_WM_STATE_MAXIMIZED_VERT" && _NET_WM_STATE@[1]:a = "_NET_WM_STATE_MAXIMIZED_HORZ" || _NET_WM_STATE@[0]:a = "_NET_WM_STATE_MAXIMIZED_HORZ" && _NET_WM_STATE@[1]:a = "_NET_WM_STATE_MAXIMIZED_VERT") && _NET_WM_WINDOW_TYPE@[0]:a = "_KDE_NET_WM_WINDOW_TYPE_OVERRIDE"'

It is probably abnormal that the window contents do not appear while the shadow does. I am still investigating.

@richardgv,

Hello, nice to see you back in action...

I'm using Openbox, so why would I have the last line with KDE in it?

I'm not using Virtualbox at the moment, so it may be some long time, before I get back to using and testing this...

But thank you for looking into this....

[richardgv]

Well, further analysis showed the window contents do not appear because the window VirtualBox created has a 1x1 bounding region. The shadow of a window is not affected by the X Shape bounding region, so this window has a big full-screen shadow but only 1x1 contents displayed.

It seems there is not much we could do currently except disabling shadow for the window. Here is a revised version of the option to do it, that takes the bounding shape into account:

--shadow-exclude 'class_g = "VirtualBox" && name = "VirtualBox" && bounding_shaped && (_NET_WM_STATE@[0]:a = "_NET_WM_STATE_MAXIMIZED_VERT" && _NET_WM_STATE@[1]:a = "_NET_WM_STATE_MAXIMIZED_HORZ" || _NET_WM_STATE@[0]:a = "_NET_WM_STATE_MAXIMIZED_HORZ" && _NET_WM_STATE@[1]:a = "_NET_WM_STATE_MAXIMIZED_VERT") && _NET_WM_WINDOW_TYPE@[0]:a = "_KDE_NET_WM_WINDOW_TYPE_OVERRIDE"'

I'm using Openbox, so why would I have the last line with KDE in it?

The property _NET_WM_WINDOW_TYPE has nothing to do with the WM: It is set by the client (VirtualBox) instead of WM.

The window VirtualBox created has Qt::FramelessWindowHint attribute, and for such a window Qt 5 XCB plugin:

• Adds _KDE_NET_WM_WINDOW_TYPE_OVERRIDE to the _NET_WM_WINDOW_TYPE property of the window in QXcbWindow::setWmWindowType().
• Sets _MOTIF_WM_HINTS to hide the decorations in QXcbWindow::setMotifWindowFlags().

As _MOTIF_WM_HINTS is actually a set of bits, and we lack the bit masking functionality required to properly match against _MOTIF_WM_HINTS, I chose matching against _NET_WM_WINDOW_TYPE, even though matching against _MOTIF_WM_HINTS is actually more correct.

In simple to understand terms, if possible, why did Oracle make this change to virtualbox 5.x, that it didn't have before, to what benefit?

[richardgv]

In simple to understand terms, if possible, why did Oracle make this change to virtualbox 5.x, that it didn't have before, to what benefit?

Well, you should consult the developers of VirtualBox if you wish know it. The code base of VirtualBox and Qt is huge and it takes too much effort to figure out the exact effects for me.

Here is some known information:

• The changes were initially introduced in this changeset, but the description did not explain the reason very well: https://www.virtualbox.org/changeset/57480/vbox
• With the original code path, Qt QDesktopWidget::availableGeometry() seemingly will be used to query the available geometry on the screen. Internally probably QXcbScreen::updateGeometry() generally determines the geometry returned, as the intersection of the rectangle in the _NET_WORKAREA property of the root window, and the size of the screen. I am not sure how the value may be different from the one produced by the new code path.
• There are too many references to the available geometry on the screen in VirtualBox, and it requires more comprehension of the internal workings to understand their exact effects, unfortunately.

So simply said, going to take some time to figure out hehe :)

THANKS richardgv