My KWIN fork where I work on features like per-output virtual desktops
Find a file
Luboš Luňák 03fedc56dd No effects when not compositing.
svn path=/branches/work/kwin_composite/; revision=560594
2006-07-10 18:34:57 +00:00
clients merge bleeding edge. bleeding edge is now again dead 2006-07-04 17:09:53 +00:00
data merge bleeding edge. bleeding edge is now again dead 2006-07-04 17:09:53 +00:00
kcmkwin remove line that wasn't suppost to be there. 2006-07-06 18:08:14 +00:00
killer kde4_header and kde4_footer() are gone now 2006-05-12 18:12:37 +00:00
lib These warnings were pretty annoying. 2006-07-04 20:23:29 +00:00
pics merge bleeding edge. bleeding edge is now again dead 2006-07-04 17:09:53 +00:00
tools fixing some EBN issues 2006-05-20 15:56:37 +00:00
wm-spec
activation.cpp Remove kompmgr-related code except for the generic parts like 2006-07-04 23:30:05 +00:00
atoms.cpp Some convenience functions for QX11Info stuff. 2006-07-04 19:36:19 +00:00
atoms.h svn merge svn+ssh://coolo@svn.kde.org/home/kde/trunk/KDE/kdebase@438057 2005-07-28 14:59:42 +00:00
bridge.cpp deprecated-- 2006-04-12 15:13:31 +00:00
bridge.h Adapt to new kde4 api 2005-11-08 10:20:07 +00:00
client.cpp Add support for translation and add two "plugins" that 2006-07-06 13:17:44 +00:00
client.h Add support for translation and add two "plugins" that 2006-07-06 13:17:44 +00:00
CMakeLists.txt svn merge -r558690:558691 https://lunakl@svn.kde.org:/home/kde/trunk/KDE/kdebase/workspace/kwin 2006-07-05 20:55:16 +00:00
COMPLIANCE Fix usage of PBaseSize and PMinSize in size hints. Makes geometry tip 2005-05-18 13:59:59 +00:00
composite.cpp No effects when not compositing. 2006-07-10 18:34:57 +00:00
cr16-app-kwin.png Losslessly optimize file size of *.png files in /trunk/KDE/kdebase with optipng -i0 -o5 and advdef -z -4. 2006-03-13 17:07:19 +00:00
cr32-app-kwin.png Losslessly optimize file size of *.png files in /trunk/KDE/kdebase with optipng -i0 -o5 and advdef -z -4. 2006-03-13 17:07:19 +00:00
cr48-app-kwin.png Losslessly optimize file size of *.png files in /trunk/KDE/kdebase with optipng -i0 -o5 and advdef -z -4. 2006-03-13 17:07:19 +00:00
effects.cpp No effects when not compositing. 2006-07-10 18:34:57 +00:00
effects.h No effects when not compositing. 2006-07-10 18:34:57 +00:00
events.cpp Add support for translation and add two "plugins" that 2006-07-06 13:17:44 +00:00
eventsrc SVN_SILENT made messages (.desktop file) 2006-02-19 08:04:22 +00:00
geometry.cpp Add support for translation and add two "plugins" that 2006-07-06 13:17:44 +00:00
geometrytip.cpp Some convenience functions for QX11Info stuff. 2006-07-04 19:36:19 +00:00
geometrytip.h #include <q...h> -> #include <Q...> 2006-05-11 08:04:23 +00:00
group.cpp Some convenience functions for QX11Info stuff. 2006-07-04 19:36:19 +00:00
group.h Fix #72074 - when trying to close a window, update user timestamp 2004-02-27 13:35:09 +00:00
HACKING Deprecated-- 2006-02-06 14:47:18 +00:00
killwindow.cpp Some convenience functions for QX11Info stuff. 2006-07-04 19:36:19 +00:00
killwindow.h Merging kwin core from kwin_iii back to HEAD. 2003-09-16 19:28:03 +00:00
kwin.kcfg #67406 - config option kwinrc:Windows:ShowDesktopIsMinimizeAll. 2006-05-30 11:47:16 +00:00
kwinadaptor.cpp Remove kompmgr-related code except for the generic parts like 2006-07-04 23:30:05 +00:00
kwinadaptor.h Remove kompmgr-related code except for the generic parts like 2006-07-04 23:30:05 +00:00
kwinbindings.cpp Merging in from bleeding edge kdebase branch 2006-04-10 11:57:13 +00:00
layers.cpp Some convenience functions for QX11Info stuff. 2006-07-04 19:36:19 +00:00
LICENSE I'm glad you didn't actually write the "General Public Li_n_cense" :P 2003-11-09 13:40:49 +00:00
main.cpp No effects when not compositing. 2006-07-10 18:34:57 +00:00
main.h Added composite (translucency/shadows) support 2005-01-15 17:07:48 +00:00
Makefile.am remove the messages targets already in separate scripts 2006-05-01 16:48:29 +00:00
manage.cpp Add support for translation and add two "plugins" that 2006-07-06 13:17:44 +00:00
Messages.sh Convert the "messages" targets of kdebase into Messages.sh scripts 2006-02-22 01:47:34 +00:00
NEWCOLORSCHEME.README
notifications.cpp TRUE->true, FALSE->false 2006-02-19 01:33:48 +00:00
notifications.h #include <q...h> -> #include <Q...> 2006-05-13 16:33:18 +00:00
options.cpp Use kwinrc:[Translucency]:UseTranslucency for controlling compositing. 2006-07-05 21:46:01 +00:00
options.h Remove kompmgr-related code except for the generic parts like 2006-07-04 23:30:05 +00:00
org.kde.KWin.xml Remove kompmgr-related code except for the generic parts like 2006-07-04 23:30:05 +00:00
placement.cpp #include <q...h> -> #include <Q...> 2006-05-13 16:33:18 +00:00
placement.h #include <q...h> -> #include <Q...> 2006-05-11 08:04:23 +00:00
plugins.cpp #include <q...h> -> #include <Q...> 2006-05-13 16:33:18 +00:00
plugins.h Added KDecorationFactory::checkRequirements(), which allows the decoration 2003-11-24 14:22:04 +00:00
popupinfo.cpp Some convenience functions for QX11Info stuff. 2006-07-04 19:36:19 +00:00
popupinfo.h #include <q...h> -> #include <Q...> 2006-05-11 08:04:23 +00:00
README svn merge svn+ssh://coolo@svn.kde.org/home/kde/trunk/KDE/kdebase@438057 2005-07-28 14:59:42 +00:00
rules.cpp Remove kompmgr-related code except for the generic parts like 2006-07-04 23:30:05 +00:00
rules.h #include <q...h> -> #include <Q...> 2006-05-13 16:33:18 +00:00
scene.cpp Ok, I give up, whoever wants more complicated transformations than just 2006-07-06 18:22:01 +00:00
scene.h Ok, I give up, whoever wants more complicated transformations than just 2006-07-06 18:22:01 +00:00
scene_basic.cpp Ok, I give up, whoever wants more complicated transformations than just 2006-07-06 18:22:01 +00:00
scene_basic.h Ok, I give up, whoever wants more complicated transformations than just 2006-07-06 18:22:01 +00:00
scene_xrender.cpp Oops. 2006-07-06 19:14:38 +00:00
scene_xrender.h Ok, I give up, whoever wants more complicated transformations than just 2006-07-06 18:22:01 +00:00
sm.cpp Not needed later after all. 2006-07-10 17:07:29 +00:00
sm.h Save also stacking order when doing session save. 2006-04-20 14:57:10 +00:00
tabbox.cpp Some convenience functions for QX11Info stuff. 2006-07-04 19:36:19 +00:00
tabbox.h #include <q...h> -> #include <Q...> 2006-05-11 08:04:23 +00:00
toplevel.cpp Make the opacity slider and wheel operations work again. 2006-07-05 22:26:34 +00:00
toplevel.h Add support for translation and add two "plugins" that 2006-07-06 13:17:44 +00:00
unmanaged.cpp Don't track InputOnly windows. 2006-07-10 18:20:25 +00:00
unmanaged.h Add support for translation and add two "plugins" that 2006-07-06 13:17:44 +00:00
useractions.cpp Make the opacity slider and wheel operations work again. 2006-07-05 22:26:34 +00:00
utils.cpp Add support for translation and add two "plugins" that 2006-07-06 13:17:44 +00:00
utils.h Add support for translation and add two "plugins" that 2006-07-06 13:17:44 +00:00
workspace.cpp Don't track InputOnly windows. 2006-07-10 18:20:25 +00:00
workspace.h Add support for translation and add two "plugins" that 2006-07-06 13:17:44 +00:00

 This README is meant as an explanation of various window manager related
mechanisms that application developers need to be aware of. As some of these
concepts may be difficult to understand for people not having the required
background knowledge (since sometimes it's difficult even for people who
do have the knowledge), the mechanisms are first briefly explained, and
then an example of fixing the various problems is given.

 For comments, questions, suggestions and whatever use the kwin@kde.org
mailing list.


Table of contents:
==================

- Window relations
    - how to make the window manager know which windows belong together
- Focus stealing prevention
    - how to solve cases where focus stealing prevention doesn't work
      properly automatically



Window relations:
=================

(For now, this explanation of window relations is mainly meant for
focus stealing prevention. To be extended later.)

 All windows created by an application should be organized in a tree
with the root being the application's main window. Note that this is about
toplevel windows, not widgets inside the windows. For example, if you
have KWrite running, with a torn-off toolbar (i.e. a standalone toolbar),
a file save dialog open, and the file save dialog showing a dialog
for creating a directory, the window hiearchy should look like this:


             KWrite mainwindow
              /             \
             /               \
      file save dialog      torn-off toolbar
            \
             \
          create directory dialog

 Each subwindow (i.e. all except for the KWrite mainwindow) points to its
main window (which in turn may have another main window, as in the case
of the file save dialog). When the window manager knows these relations,
it can better arrange the windows (keeping subwindows above their
main windows, preventing activation of a main window of a modal dialog,
and similar). Failing to provide this information to the window manager
may have various results, for example having dialogs positioned below
the main window,

The window property used by subwindows to point to their mainwindows is
called WM_TRANSIENT_FOR. It can be seen by running
'xprop | grep WM_TRANSIENT_FOR' and clicking on a window. If the property
is not present, the window does not (claim to) have any mainwindow.
If the property is present, it's value is the window id of its main window;
window id of any window can be found out by running 'xwininfo'. A window
having WM_TRANSIENT_FOR poiting to another window is said to be transient
for that window.

 In some cases, the WM_TRANSIENT_FOR property may not point to any other
existing window, having value of 0, or pointing to the screen number
('xwininfo -root'). These special values mean that the window is transient
for all other windows in its window group. This should be used only
in rare cases, everytime a specific main window is known, WM_TRANSIENT_FOR
should be pointing to it instead of using one of these special values.
(The explanation why is beyond the scope of this document - just accept it
as a fact.)

 With Qt, the WM_TRANSIENT_FOR property is set by Qt automatically, based
on the toplevel widget's parent. If the toplevel widget is of a normal
type (i.e. not a dialog, toolbar, etc.), Qt doesn't set WM_TRANSIENT_FOR
on it. For special widgets, such as dialogs, WM_TRANSIENT_FOR is set
to point to the widget's parent, if it has a specific parent, otherwise
WM_TRANSIENT_FOR points to the root window.

 As already said above, WM_TRANSIENT_FOR poiting to the root window should
be usually avoided, so everytime the widget's main widget is known, the widget
should get it passed as a parent in its constructor.
(TODO KDialog etc. classes should not have a default argument for the parent
argument, and comments like 'just pass 0 as the parent' should go.)



Focus stealing prevention:
==========================

 Since KDE3.2 KWin has a feature called focus stealing prevention. As the name
suggests, it prevents unexpected changes of focus. With older versions of KWin,
if any application opened a new dialog, it became active, and
if the application's main window was on another virtual desktop, also
the virtual desktop was changed. This was annoying, and also sometimes led
to dialogs mistakenly being closed because they received keyboard input that
was meant for the previously active window.

 The basic principle of focus stealing prevention is that the window with most
recent user activity wins. Any window of an application will become active
when being shown only if this application was the most recently used one.
KWin itself, and some of the related kdecore classes should take care
of the common cases, so usually there's no need for any special handling
in applications. Qt/KDE applications, that is. Applications using other
toolkits should in most cases work fine too. If they don't support
the window property _NET_WM_USER_TIME, the window manager may fail to detect
the user timestamp properly, resulting either in other windows becoming active
while the user works with this application, or this application may sometimes
steal focus (this second case should be very rare though).

 There are also cases where KDE applications needs special handling. The two
most common cases are when windows relations are not setup properly to make
KWin realize that they belong to the same application, and when the user
activity is not represented by manipulating with the application windows
themselves.

 Also note that focus stealing prevention implemented in the window manager
can only help with focus stealing between different applications.
If an application itself suddenly pops up a dialog, KWin cannot do anything about
it, and its the application's job to handle this case.


Window relations:
-----------------

 The common case here is when a dialog is shown for an application, but this
dialog is not provided by the application itself, but by some other process.
For example, dialogs with warnings about accepted cookies are provided
by KCookieJar, instead of being shown by Konqueror. In the normal case,
from KWin's point of view the cookie dialog would be an attempt of another
application to show a dialog, and KWin wouldn't allow activation of this
window.

 The solution is to tell the window manager about the relation between
the Konqueror main window and the cookie dialog, by making the dialog
point to the mainwindow. Note that this is not special to focus stealing
prevention, subwindows such as dialogs, toolbars and similar should always
point to their mainwindow. See the section on window relations for full
description.

 The WM_TRANSIENT_FOR property that's set on dialogs to point to their
mainwindow should in the cookie dialog case point to the Konqueror window
for which it has been shown. This is solved in kcookiejar by including
the window id in the DCOP call. When the cookie dialog is shown, its
WM_TRANSIENT_FOR property is manually set using the XSetTransientForHint()
call (see kdelibs/kioslave/http/kcookiejar/kcookiewin.cpp). The arguments
to XSetTransientForHint() call are the X display (i.e. QX11Info::display()),
the window id on which the WM_TRANSIENT_FOR property is to be set
(i.e. use QWidget::winId()), and the window id of the mainwindow.


  Simple short HOWTO:
  
 To put it simply: Let's say you have a daemon application that has
DCOP call "showDialog( QString text )", and when this is called, it shows
a dialog with the given text. This won't work properly with focus stealing
prevention. The DCOP call should be changed to
"showDialog( QString text, long id )". The caller should pass something like
myMainWindow->winId() as the second argument. In the daemon, before
the dialog is shown, a call to XSetTransientHint() should be added:

 XSetTransientForHint( QX11Info::display(), dialog->winId(), id_of_mainwindow );
 
 That's it.

Non-standard user activity:
---------------------------

 The most common case in KDE will be DCOP calls. For example, KDesktop's DCOP
call "KDesktopIface popupExecuteCommand". Executing this DCOP call e.g.
from Konsole as 'dcop kdesktop KDesktopIface popupExecuteCommand" will lead
to showing the minicli, but the last user activity timestamp gained from events
sent by X server will be older than user activity timestamp of Konsole, and
would normally result in minicli not being active. Therefore, before showing
the minicli, kdesktop needs to call KApplication::updateUserTimestamp().

 However, this shouldn't be done with all DCOP calls. If a DCOP call is not
a result of direct user action, calling KApplication::updateUserTimestamp()
would lead to focus stealing. For example, let's assume for a moment
that KMail would use this DCOP call in case it detects the modem is not
connected, allowing to you to start KPPP or whatever tool you use. If KMail
would be configured to check mail every 10 minutes, this would lead to minicli
possibly suddenly showing up at every check. Basically, doing the above change
to kdesktop's minicli means that the popupExecuteCommand() DCOP call is only
for user scripting. (TODO write about focus transferring?)

 Simply said, KApplication::updateUserTimestamp() should be called only
as a result of user action. Unfortunately, I'm not aware of any universal
way how to handle this, so every case will have to be considered separately.