Follow-up to c146941, which stopped exporting KDecorationBridgeUnstable.
The way the typeinfo and object comparisons are performed in a dynamic_cast
are not entirely defined in any ABI spec (or the C++ standard), and the
situation gets trickier when a hidden symbol (KDecorationBridgeUnstable)
gets passed between a program such as kwin or kcmshell4 and a dlopen()ed
library like kwin3_laptop.so: kwin is unaware of the symbol in
kwin3_laptop.so and vice-versa, so stricter implementations of the C++
runtime can choose to make the dynamic_cast call fail and return 0.
This was a source of crashes on GCC for a long time [1][2] until its
developers chose to make the typeinfo comparison in libstdc++ >= 4.5.x less
strict and just perform a string comparison of the type names by default.
[1] http://article.gmane.org/gmane.comp.gcc.patches/189813
[2] http://gcc.gnu.org/ml/gcc/2002-05/msg01970.html
It does not make things more correct, though, and after c146941 KWin started
crashing when built in debug mode and run with libcxxrt, another C++ runtime
implementation, since it performed a stricted typeinfo comparison by
default.
Simply removing the dynamic_cast should be fine: it is in an assertion, so
the code is not essential, and the assertion itself is not entirely correct.
Additionally, the assertion has already been removed from the master branch
in commit 4702bbd.
REVIEW: 113296
with 4 instead of just altering the default value
It seems the value is actually written for some ppl.
Worse: former patch didn't actually work as expected
anymore
CCBUG: 321897
The variable is set not from config nor anywhere else when kwin is
started through ksmserver by startkde.
In addition the KGlobal implementation is twisted compared to the kcminit
config behavior (the config value defaults to true, not false - ie. if the
variable isn't set (by kcm init) it's reasonable to assume true either.
Therefore and in alignment with PW/2, the environment is read directly and
on absence resolved to "true".
To control the behavior, please export KDE_MULTIHEAD=true/false before starting KWin
(eg. in /usr/bin/startkde)
This reverts commit ab6d5c048a25bcb2f5bdb822ba3eda64019c61bc.
REVIEW: 112579
CCBUG: 256242
repaints caused by effects so far polluted the timing calculations
since they started the timer on the old vsync offset
This (together with undercut timing) lead to multiple frames in
the buffer queue, and ultimately to a blocking swap
For unsynced painting, it simply caused wrong timings - leading to
"well, kinda around 60Hz - could be 75 as just well".
REVIEW: 112368
CCBUG: 322060
that part is fixed in 4.11.2
Since windows can place the decoration outside the screen
this needs to be still a valid condition when checking
whether we've to keep in area
In addition: calculate tx, ty and perform one move call
(include rule check and XMoveWindow unless there's a geometry
blocker ...)
BUG: 324792
FIXED-IN: 4.11.2
REVIEW: 112805
compensated by a QuartOut shape, this keeps the
effect alive while the sheet or similar effects run
without much visual stretch
Covers issue until there's AnimationEffect::determine()
CCBUG: 321897
semi Fixed in 4.11.2
REVIEW: 111798
whether or not a window is maximizable does really not
matter (it's not if the size is locked, so the
maximized size is the only size) but only if the
window can take the fullscreen geometry.
otoh, the size restrictions (maximum size, fixed aspect)
should be honored by default but still ignorable by rule
BUG: 324733
FIXED-IN: 4.11.2
REVIEW: 112654
the restack code stacks under all members of an application
this is a problem if the group contains a keep below or desktop
etc. (while the other window is a normal one) what resulted in
restacking the client "invalidly" above the window of the other
layer but below all other memberso of its own group for no reason
REVIEW: 112627
CCBUG: 110543
F(S)UM mean "the focus is where the mouse is"
the mouse is not on the other virtual desktop
(and it was even granted regardless of the actual geometry/position)
The "unreasonable" focus policies expose an issue about
the present linked handling of "allow activation" and
"allow raising" (see https://git.reviewboard.kde.org/r/110919/ )
Activation would match "extreme" (if the window maps on the same
virtual desktop, half a mile away from the mouse, it won't
receive the focus) but not regarding raising (which is actually
an issue entirely different from FSP)
REVIEW: 112585
CCBUG: 80897
also work around broken fbo texture clearing on fglrx
so far supports FBO/glClear and resorts to glTexSubImage2D
if the fbo cannot be created or is (in case of fglrx)
known to break, resort to glTexImage2D loading of an
argb array of zeros
BUG: 323065
FIXED-IN: 4.11.2
REVIEW: 112526
broke on interim Wrapper() constructor change
The Constructor needs to explicitly pass
XCB_WINDOW_NONE to the inherited Constructor to
trigger a request
Thanks to Alex Leach for finding this
CCBUG: 256242
FIXED-IN: 4.11.2
REVIEW: 112595
1. when adding a full damange, that must not replace existing (larger) repaints
2. emit geometryChanged before invoking and to update shadowGeometry through addRepaintFull
BUG: 324560
FIXED-IN: 4.11.2
When a window is added while the effect is running
it shall either be highlighted or hidden, but not exit
the effect what happened, because the property notification
is invoked to test whether the window has a highlight
property (questionable since the effect is running, but
valid - could be different X11 client) and that routine
exits the effect if no property is found (assuming it was
withdrawn)
REVIEW: 112494
Because that's where it's placed. Notable esp. when there're panels on opposing sides.
Since snapping keeps max'd windows in position, it should preferably keep it in the proper position ;-)
REVIEW: 112322
According to massif just creating the ColorCorrection costs 2 MiB, so
it makes sense to not create it if it is not used.
REVIEW: 111604
DIGEST: Smaller memory footprint in KWin 4.11.2 if color correction is disabled
Client::releaseClient() deletes all Client objects
referenced by stacking_order, thus those pointers
dangle and everything trying to touch it died an ugly death.
REVIEW: 112020
BUG: 323383
This flag seems to be ignored by several OpenGL implementations,
so don't set it for now. Mesa may also stop accepting this flag
until forward-compatible contexts are fully supported.
