Bug 324782

Summary:	Focus switches to opposite X Screen when closing window with two heads
Product:	[Plasma] kwin	Reporter:	Alex Leach <beamesleach>
Component:	multi-screen	Assignee:	KWin default assignee <kwin-bugs-null>
Status:	REPORTED ---
Severity:	normal	CC:	baron, beamesleach, thomas.luebking
Priority:	NOR
Version:	4.11.1
Target Milestone:	---
Platform:	Arch Linux
OS:	Linux
See Also:	https://bugs.kde.org/show_bug.cgi?id=333595 https://bugs.kde.org/show_bug.cgi?id=157219
Latest Commit:		Version Fixed In:
Attachments:	Debugging work on multihead task switching patch attempt base patch

Description Alex Leach 2013-09-11 11:10:00 UTC

Discussion began on BUG 256242 (See https://bugs.kde.org/show_bug.cgi?id=256242#c117), but wanted to stop bloating that old report.

Briefly, in a dual-head setup on Linux, where two X screens are running and KDE_MULTIHEAD=true, closing a window on one X DISPLAY will cause kwin to activate a window (or the desktop if no window is open), on the opposite X display.

Attaching detailed debug information here, generated in a tmux session, controlled over ssh from my laptop.

Reproducible: Always

Steps to Reproduce:
Hardware: Two graphics cards, each connected to its own monitor (well, one's a TV over HDMI, other a monitor over DVI).

Customise files in /etc/X11/x.org.conf.d/, to include two sections each of Monitor, Device and Screen. The active ServerLayout has two Screen's.

Steps taken to debug kwin
--------------

* Build and install debug version of kwin. (cmake -DCMAKE_BUILD_TYPE=Debug ...)
* Start a tmux session on DISPLAY :0.0
* In a tmux window, run `kwin --replace`
* Open new tmux window for gdb
* Attach to same tmux session from laptop over ssh.

* Setup some windows on DISPLAY :0.0

i. empty konsole:: ;WMCLASS: "konsole" : "konsole" ;Caption: "albl500 : bash â Konsole <2>â"
ii. dolphin:: ;WMCLASS: "dolphin" : "dolphin" ;Caption: "albl500 â Dolphin" '

* Keep these windows focused. Closing dolphin (which is active,) should then raise konsole

Bug traces have been generated under a couple of conditions. I will attach (to preserve monospaced characters) as a plain text file, after submitting.

Worth mentioning a couple of patches applied to the code, which were suggested in BugId#256242 to help with multihead setups. Specifically, in comments:-

- https://bugs.kde.org/show_bug.cgi?id=256242#c115
- https://bugs.kde.org/show_bug.cgi?id=256242#c119

Comment #115 patch is a bugfix;
Comment #119 patch prints message to console running `kwin --replace`.

Comment 1 Alex Leach 2013-09-11 11:13:13 UTC

Created attachment 82271 [details]
Debugging work on multihead task switching

Comment 2 Thomas Lübking 2013-09-11 20:43:58 UTC

events.cpp ~ line 290

            if (focus == None || focus == PointerRoot) {
                //kWarning( 1212 ) << "X focus set to None/PointerRoot, reseting focus" ;
                Client *c = mostRecentlyActivatedClient();
                if (c != NULL)
                    requestFocus(c, true);
                else if (activateNextClient(NULL))
                    ; // ok, activated
                else
                    focusToNull();
            }

I assume you're ending up here when/before this happens and preventing this block (* that is NOT USABLE FIX!) would also prevent the unwanted focus switch?

*
- if (focus == None || focus == PointerRoot) {
+ if (false /*focus == None || focus == PointerRoot*/) {

Comment 3 Thomas Lübking 2015-11-16 23:09:08 UTC

*** Bug 344671 has been marked as a duplicate of this bug. ***

Comment 4 Thomas Lübking 2015-11-16 23:10:50 UTC

Created attachment 95545 [details]
patch attempt

Patch, should apply to 5.4
I briefly tried it, but guess Martin would like to have it confirmed by somebody who actually uses zaphod multihead on a regular base

Comment 5 Thomas Lübking 2015-11-16 23:22:56 UTC

Created attachment 95546 [details]
base patch

patch for bug #348935 that's likely required to apply the actual patch cleanly; sorry.

Comment 6 Nicholas Redgrave 2015-11-17 12:57:19 UTC

I went back to clean sources and applied the patches.
It is, however, throwing errors like this from the "isOnCurrentHead" function:

QXcbConnection: XCB error: 9 (BadDrawable), sequence: 14659, resource id: 1, major code: 14 (GetGeometry), minor code: 0

which I suppose is sort of lucky because previous versions of the source just crashed there.

As I explained here:
https://bugs.kde.org/show_bug.cgi?id=344671#c9

the "Xcb::CurrentInput currentInput" call can have three types of return; XCB_WINDOW_NONE(0), XCB_INPUT_FOCUS_POINTER_ROOT(1) or the actual Window ID.

Your code only handles XCB_WINDOW_NONE and the actual Window ID.  If you attempt to execute the "Xcb::WindowGeometry geometry(currentInput.window())" line when "currentInput.window()" is "1" (i.e.  XCB_INPUT_FOCUS_POINTER_ROOT), the result is undefined at best and a crash at worst.

Ironically, if you had changed the window revert setting to "XCB_INPUT_FOCUS_NONE" as you threatened here:
https://bugs.kde.org/show_bug.cgi?id=344671#c18
then the code would probably have run without errors although had you ever changed it back to "XCB_INPUT_FOCUS_POINTER_ROOT", then your new patch would have broken too.

I look forward to testing your corrected patch.

Comment 7 Justin Zobel 2021-03-10 00:32:32 UTC

Thank you for the bug report.

As this report hasn't seen any changes in 5 years or more, we ask if you can please confirm that the issue still persists.

If this bug is no longer persisting or relevant please change the status to resolved.