Log of /branches/KDE/3.5/kdelibs/kdeui/qxembed.cpp

And lo, here comes the lone hero, the one and only to wield the power to fix
the unsolvable 3.5.6 showstopper. No wonder he's so awfully grumpy about it.

Geometry handling of systray icons is ... somewhat borken. Where to start?
- QXEmbed used to by a mistake send configure notify events with an incorrect
window set, making them useless and even in extreme cases crashing apps
- Qt is apparently full of hacks for handling toplevel window geometry but code
for doing it properly appears to be missing, mistaking in this case local position
with global position. Surely somebody got confused by ICCCM again.
- Apps apparently so no problem with the fact that a toplevel widget reported local
coordinates as its position and decided to work around it themselves.
Now the first one got fixed by r609135 and there's no longer a bug to work around.
Count in broken boundary checks in apps and suddenly some widgets "appear" in various
broken locations. Count in apps rolling their own code instead of extending libraries
and finding a way of at least somehow fixing this becomes interesting.

- Keeping it as it is breaks apps and moreover makes it impossible to ever fix this.
- Reverting r609135 breaks apps.
- Not sending the configure notify at all may possibly break apps, I'm not really willing
to try.
This commit just sets and reset X window border, in the hope that it'll be a no-op
for everybody, but it will send a real configure notify that will confuse Qt and make
apps with the workaround "work". At the same time the configure notify sent from QXEmbed
is correct, giving a chance to fix this somewhen.

Try to work around the case when the WM first sets WM_STATE
to Withdrawn and only then reparents.
http://lists.kde.org/?l=kde-devel&m=116448102901184&w=2

ConfigureNotify should have event == window.
http://lists.kde.org/?l=kde-devel&m=116448102901184&w=2

SVN_SILENT: fix typo

branching of 3.5 and arts 1.5 

update FSF address

moved kdeaccessibility kdeaddons kdeadmin kdeartwork kdebindings kdeedu kdegames kdegraphics kdemultimedia kdenetwork kdesdk kdetoys kdeutils kdevelop kdewebdev kdepim kdebase kdelibs kde-i18n arts into /trunk/KDE/

The widget for XSetInputFocus() must be visible.
BUG: 99267

- Replaced
    (foo == true)
    (foo == false)
    (foo != 0)
    (foo == 0)
  with
    (foo)
    (!foo)
    (foo)
    (!foo)

Pass focus request properly further with multiple embed levels.

CVS SILENT: Comment changed for sendSyntheticConfigureNotifyEvent()

If the embedded client is reparented away, that means the end of the embedding
cycle as well -> emit embeddedWindowDestroyed() too ... which means the name
of the signal is somewhat wrong.

Detect failed embedding attempts.

As the client side detection of embedding termination is not done for 3.2,
revert back to mapping the embedded window when QXEmbed releases it.
Reviewed by Leon Bottou.

Fix problem with kppp's systray icon reappearing
(http://lists.kde.org/?l=kde-core-devel&m=106931968918881&w=2).

Fixed another anomaly with focus navigation with the tab key.

Focus navigation was still broken when using two levels of embedding.
Function QPublicWidget::focusNextPrevChild() was here to give access
to the otherwise protected focusNextPrevChild(). This is a virtual
function.  We do not want to call it as QWidget::focusNextPrevChild()
because that defeats virtual functions.  The only solution is to give
another name to QPublicWidget::focusNextPrevChild()...

Changed inexact comments.

Added XRemoveFromSaveSet in QXEmbed::~QXEmbed.
Added comment about possible corner case problems
with XRemoveFromSaveSet.

Change in comments only.

Following rev 1.51.
Protect calls to XSetInputFocus by first
testing whether qApp->activeWindow()==topLevelWidget().
This is not the same as isActiveWindow() because
function isActiveWindow() can return true when a
modal dialog child of this window is active.
We do not want to change the X11 focus in that case.

Reimplemented focus navigation with tab key.
See <http://bugs.kde.org/show_bug.cgi?id=66938>.

Second part of fixes for systray.
- QXEmbed needs to do XAddToSaveSet() when using XEMBED, otherwise
 e.g. tray icons will be lost after Kicker crash
- QXEmbed has to reparent windows always back to the root window if they're
 supposed to live longer than QXEmbed (even in autoDelete() case), otherwise
 they'll remain QXEmbed's inferior and will be destroyed by X11 together
 with QXEmbed's window

Be more paranoid before changing X focus.

Typos in comments.

MASSIVE INSERTION OF COMMENTS IN QXEMBED.CPP.
These changes only involve comments or indentation.
This commit makes no functional changes.

Minor changes to prepare next commit:
in QXEmbed::x11Event:
  Make sure intercepted buttonpress events are not
  passed to the widget.
in get_parent:
  Beautified code.

- Client did not set topdata()->parentWinId when receiving the initial
  XEMBED_EMBEDDED_NOTIFY message. Self-embedding clients were not affected by
  this because embedClientIntoWindow() also sets parentWinId. The bug only
  shows when using embed().
- A call to sendXEmbedMessage with XEMBED_FOCUS_IN dit not
  specify the detail parameter.
- Minor changes in debug messages.

Added function sendDelete().
Renamed private function send_xembed_message for consistency.
Made documentation in qxembed.h sightly more accurate.

Implemented changes discussed in thread
<http://lists.kde.org/?t=106597285500004&r=1&w=2>.

- Adding a test in QXEmbed::focusOutEvent to prevent stealing
  the X11 focus when receiving a QFocusOut event caused by a
  window activation change.
- Moving a part of embed() into the ReparentNotify section
  of QXEmbed::x11Event().
- Documenting and fixing the semantic of setAutoDelete().
- Getting rid of XAddToChangeSets because it usually does the
  wrong thing. Any embedding application that really need this
  can easily do it itself after calling QXEmbed::embed().

Fix #64553. When destroying QXEmbed, make sure the focus is not left
on the focus proxy, otherwise it will get destroyed, focus will be reverted
to the parent (=qxembed), that one will be destroyed too, and focus
will be set to None (RevertToParent semantics seem to be a bit
braindamaged).

CCMAIL: 64553-done@bugs.kde.org

Remove TRUE and FALSE from C++ files in favor of true and false.  This
bothers me when I run across it and well, we have an actual bool type in
C++, why not use it?  If the flames aren't too bad I'll apply this to
other modules (and I did read the diff by hand a couple of times and
compile and check for new warnings after this -- all seems to be ok).

reverted private comments

fixed some collisions...

speling fixes

Save a couple of roundtrips by checking the position only in case
when the move event could have changed it (ignore move events
not related to QXEmbed or its (indirect) parent.

speling fixes

*sigh*

Various namespace cleaness fixes.

Smallish improvement of focus signaling.

Simply amazing.  Wonderful fixes from Leon Bottou
Fixes various NS Plugin problems, including keyboard input.

Thank you!

CCMAIL: leon@bottou.org

By popular demand, move mouse grabbing from QXEmbed to KJavaAppletWidget

@@ -689,6 +689,7 @@ void QXEmbed::embed(WId w)
     XMapRaised(qt_xdisplay(), window);
+    sendSyntheticConfigureNotifyEvent();
     extraData()->xDndProxy = w;

So the embedded client gets pointer coordinates right even before the first
move.

Also send the embedded window a XFocus{In|Out}Event when getting/loosing the focus

Use XGrabButton/XUngrabButton to detect mouse clicks on the embedded application

Till's patch.  It fixes thing, so unless someone has a better patch...

Applied changes from javaembed (XConfigureNotify stuff). Makes javaembed obsolete :)

Fixed #38911: System tray entries disappear when Kicker restarts
  -        XFlush( qt_xdisplay() );
  +        XSync(qt_xdisplay(), FALSE);
Solution found by Matthias Ettrich, confirmed by me.

fix What's This mode (shows in embedded root modules in kcontrol)

CVS_SILENT: fixincludes

_WS_X11_ -> Q_WS_X11
_WS_QWS_ -> Q_WS_QWS

Allow to build with Qt/Embedded

compile with --enable-final

reverted nonsense patch from before xmas

small fix to avoid race conditions in embed(...)

* fixes keyboard problems in flash and hopefully java too

Don't show a window at 0,0.  Some people don't seem to like it.

unmap the swallowed window before deleting QXEmbed.

make kicker work again

update

sort of working whats's this? help in embedded windows. Not complete yet,
but ok for the next beta.

Progress on the NET_WM window manager front

small update

focus problem on initial activation fixed (test the exampleapplet)

improved embedder-side embedding (as used by kapplet)

minor fix

new iteration of QXEmbed

#if 0'ed lots of code to make it compile with Qt 2.1beta1.
FocusIn/FocusOut, key events and WheelEvents will stop working until
it's fixed correctly :-/

add embedded windows to the save set

panel applets framework

forgot something

nicer

several fixes to make --enable-final work. Most work is done by changing
the order of the files in _SOURCES

Added qxembed. Several apps use it now (kcontrol, koffice, ...)