This is for two reasons:
1. Fixes crash on tear-down when XGetXCBConnection returns junk
2. In Qt5 getting the connection is not cheap enough for our uses
REVIEW: 108826
Uses query_pict_formats to retrieve all formats and iterates over them to
find the one matching the requested depth. Results are now cached in an
QHash with the depth as key instead of an array using the depth as index.
In case of native graphics system nothing is changed for the moment - the
X11Pixmap-QPixmap bridge is continued to be used.
But in case of graphics system raster (or Qt5) this relationship is no
longer used. Instead the QPixmap is converted to a QImage and the image
bits are put into the created X11Pixmap for this XRenderPicture.
Note: Qt5 uses shm to transfer image data to drawables. This seems
unsuited in this case as it's only a one time transformation.
For Qt5 the native pixmap block needs to be removed and the ctor might be
changed to taking an QImage as argument to make more clear that there is
no mapping from QPixmap to X11Pixmap.
Instead of having a pointer to a QPixmap the offscreen target holds an
xcb_render_picture_t. To make this possible in SceneWindow the tempPixmap
is changed from a QPixmap* to a XRenderPicture*. QPixmap was only used
for convenience.
ScreenShot Effect as only user of the offscreen target is adjusted but
as it needs a QImage, still uses a QPixmap wrapper.
This follows how it is done for OpenGL where the renderRoundBox() got
dropped some time ago.
New implementation implements the box with round corners using xrender
directly instead of using a QPainter on a QPixmap.
For the time being the current design of Options is more or less kept to
not have to adjust KWin code all over the place. Also for some parts the
generated class from KConfigXT cannot be used due to inter-settings
dependencies defined in the setters.
Options now holds a pointer to a Settings object which is generated from
KConfigXT and uses it to read the default values and the individual
settings. This means the static default value methods are dropped and the
variables are initialized with a normal default value (all int 0, all
boolean false and so on) in the initializer list. Afterwards the values
are set to the correct default value through KConfigXT.
So far for the first step only Windows category is using KConfigXT.
REVIEW: 108572
Apparently it's not allowed to have two groups with the same entry names.
Parameterized groups are still a "TODO" and a group must have a name.
Looks like we won't be able to use KConfigXT for TabBox.
Tried to compile it and at least decobenchmark does not compile due to
missing include to QObject in header file using QObject. I don't know
how long that is a requirement, but it shows that nobody has tried to
use it for quite some time.
REVIEW: 108792
You don't need to cherry-pick if you find a conflict in a .desktop file
made by Scripty.
Just use git checkout --ours <path to .desktop file> and then git add
<path to .desktop file>.
Thomas, I picked the versions from master for these conflicts, please
review that I did not mess up anything.
CCMAIL: thomas.luebking@gmail.com
Conflicts:
kwin/layers.cpp
kwin/screenedge.cpp
kwin/workspace.h
plasma/generic/applets/batterymonitor/metadata.desktop
plasma/generic/applets/lock_logout/metadata.desktop
X counts the middle button as 2 and right as 3, Qt as 4 and 2 resp. the flags
... we shall at least match our GUI ;-)
BUG: 314756
FIXED-IN: 4.10.1
REVIEW: 108883
happens when the polling misses the press when doing more
important things
also shortcut buttons == oldButtons
BUG: 314762
FIXED-IN: 4.10.1
REVIEW: 108883
but drop screenedges below the supportWindow instead
that's why it exists, that's deterministic, that's faster
includes adaption to new screenedge and xcb invocation (compared to 4.10)
BUG: 314625
FIXED-IN: 4.10.1
REVIEW: 108867
X counts the middle button as 2 and right as 3, Qt as 4 and 2 resp. the flags
... we shall at least match our GUI ;-)
BUG: 314756
FIXED-IN: 4.10.1
REVIEW: 108883
happens when the polling misses the press when doing more
important things
also shortcut buttons == oldButtons
BUG: 314762
FIXED-IN: 4.10.1
REVIEW: 108883
and ensure to repaint labels for very tiny windows
Also move to LayerRepaints
BUG: 278137
FIXED-IN: 4.10.1
REVIEW: 108676
(cherry picked from commit 1365f04435f0bbbc74301988b778f7fdf1ad4fd8)
- don't cast Window's to pointers for no apparent reason
bug introduced with ac0f8bfb24403168199027a77bba0107bc6d42e1
- no stupid java style iterators, we've stl everywhere
- postfix in a for loop is a bug. period. ;-)
BUG: 313145
REVIEW: 108864
FIXE-IN: 4.10.1
The preview code still contained all the old logic for embedding the
preview into another widget. Most of the code had been dead for quite
some time.
The only preview rendering code now is the one to render into a pixmap.
This is actually changed to render the widget directly at the right
position obsoleting the previous widget positioning code.
The no preview available label is dropped as it did not get rendered into
the pixmap at all.
Last but not least we fake that compositing is on. This has always been
the case for Aurorae themes and just looks better. Especially on Oxygen I
had always had some corruption when rendering this preview with
compositing disabled.
REVIEW: 108777
This feature has been superseded by the close window button and as it
is a rather destructive action it makes sense to remove it.
Sorry to everyone who used it.
CCBUG: 314393
REVIEW: 108851
EffectsHandlerImpl starts to monitor DBus for the screen being locked and
provides this information to the Effect system by allowing them to ask
whether the screen is currently locked and by emitting a signal when the
screen gets locked/unlocked.
This information is needed to ensure that no private data is shown on the
screen. The following effects are adjusted:
* taskbar thumbnails
* thumbnail aside
* mouse mark
* screen shot
BUG: 255712
FIXED-IN: 4.11
REVIEW: 108670
Corners are still ours (it's a valid use case to still be able to switch
window through e.g. Present Windows even when running a fullscreen app).
How is it done? An Edge can be blocked and does no longer trigger if it
is blocked. For WindowBasedEdges the edge windows get unmapped in the
blocking case and mapped again when the blocking condition is no longer
valid.
The blocking is so far connected to:
* changes of active window
* changes of fullscreen windows
Whenever one of the events occurs it is checked whether there is:
1. an active client
2. it is fullscreen
3. on the same screen as the edge
If this is the case the edge will be blocked, otherwise unblocked.
BUG: 271607
FIXED-IN: 4.11
