| Commit message (Collapse) | Author | Age | Files | Lines |
|\
| |
| |
| |
| |
| |
| |
| | |
git@scm.dev.nokia.troll.no:qt/kinetic into graphicseffects
Conflicts:
src/gui/graphicsview/qgraphicseffect.cpp
src/gui/graphicsview/qgraphicseffect_p.h
|
| |
| |
| |
| | |
This brings performance back up to where it was pre-merge.
|
| |\
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | | |
Conflicts:
src/opengl/gl2paintengineex/qglengineshadermanager.cpp
src/opengl/gl2paintengineex/qglengineshadermanager_p.h
src/opengl/gl2paintengineex/qpaintengineex_opengl2.cpp
src/opengl/gl2paintengineex/qpaintengineex_opengl2_p.h
Merge custom shader / GL blur pixmap filter implementation from graphics
team repo with implementation from kinetic graphics-team repo.
|
| | | |
|
| | | |
|
| | | |
|
| | |
| | |
| | |
| | |
| | | |
The FBO pool will be useful in other places as well, plus it makes it
easier to deal with graphics memory management issues.
|
| | |
| | |
| | |
| | |
| | | |
Ensures that the programs are evicted from the cache when the shaders
are destroyed.
|
| | |
| | |
| | |
| | |
| | | |
No raster engine based fallback so far... Also, performance is lacking
since the QGLShaderProgram isn't cached.
|
| | | |
|
| | |
| | |
| | |
| | | |
Works just like drawImage / drawPixmap but uses the given texture id.
|
| | |
| | |
| | |
| | |
| | | |
This will make it easier to implement pixmap filters, YUV->RGB
conversions, etc in other parts of Qt.
|
| | | |
|
| |\ \
| | | |
| | | |
| | | | |
git@scm.dev.nokia.troll.no:qt/kinetic into kinetic-graphicseffect
|
| | | | |
|
| | | | |
|
| | | | |
|
| | | | |
|
| |\ \ \
| | | | |
| | | | |
| | | | |
| | | | | |
Conflicts:
src/gui/graphicsview/qgraphicsitem_p.h
|
| | | | | |
|
| | | | | |
|
| | | | |
| | | | |
| | | | |
| | | | | |
Reviewed-By: Trustme
|
| | |\ \ \ |
|
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | | |
Change 34fde4a4 removes one bit from the flags bitfield, which was
added in change 7bc98d7b. This happened during resolving of a
merge conflict and caused some input method related autotests in
tst_QGraphicsView to fail.
Reviewed-by: mbm
|
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | | |
Following QWidget's behavior, you can not assign any item in the
same scene as a focus proxy for another item. Also supports nested
focus proxies. You can only assign items in the same scene as
focus proxies. Autotests are included.
Reviewed-By: mbm
|
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | | |
This fixes the build on older Solaris machines which don't have
the GLX_EXT_texture_from_pixmap defines in glxext.h.
Reviewed-By: Trustme
|
| | |/ / /
| | | | |
| | | | |
| | | | | |
Reviewed-by: akennedy
|
| | | | |
| | | | |
| | | | |
| | | | | |
Reviewed-by: Sarah Smith
|
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | | |
Something went wrong with the integrate from 4.5 to master.
Reviewed-by: TrustMe
|
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | | |
The dummy implementaion of QReadWriteLock wasn't source compatible with
the real implementation and this lead to compilation errors in
qabstractfileengine.cpp which now has a global static QReadWriteLock
that takes a Recursive argument.
Reviewed-by: Noam Rosenthal <noam.rosenthal@nokia.com>
|
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | | |
There's a possibility for deadlocking with user code in
QAbstractFileEngine. Changing the QMutex there to a QReadWriteLock
should reduce the possibilities for this happening.
Also reduced the scope of the lock in QAbstractFileEngine.
Reviewed-by: Thiago Macieira
|
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | | |
The texture_from_pixmap patch removed a bindTexture overload from
QGLContextPrivate which is actually needed by all architectures. It was
just it's use in the mac compat methods which broke the build and
highlighted the issue.
Reviewed-By: Trustme
|
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | | |
~QGLTexture wouldn't make the texture's context current if the current
context was zero, meaning the texture would leak. This also means
deleteBoundPixmap doesn't need to make the context currnet anymore (as
it's only called from ~QGLTexture).
Reviewed-By: Kim
|
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | | |
the same method was also added to QSslSocket.
previously, it was only possible to ignore all SSL errors; now, it is
also possible to only ignore specific SSL errors, given by a QList of
QSslErrors.
Moreover, it is possible to call this newly added method right after
connecting, not just when we get the SSL error.
Reviewed-by: Thiago
Task-number: 257322
|
| | |\ \ \ |
|
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | | |
I had this #ifdef __OPTIMIZE__ there so that the compiler would know
not to generate unnecessary calls and a long jump table for the switch
of the marshalling code.
Turns out that in release mode, the checks I added to make sure we
detect invalid object paths and signatures were never hit (we always
treated them as pure strings).
So use the signature- and object path-checking code in both release
and debug mode.
Task-number: reported via email (tst_qdbusmarshall failing)
Reviewed-by: Peter Hartmann
|
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | | |
In some cases we might get an invalid timestamp that is far away in
the future, so remembering it will break all consequent X calls that
require a timestamp because it just contains junk (for example
clipboard will stop working). This happens with XIM+SCIM pair -
whenever we start input method and type something to the widget, we
get a XKeyPress event with a commited string, however the 'serial' and
'time' members of the XEvent structure are not initialized (according
to valgrind) and contain junk.
This reverts commit 2ed015b8a0ffad63f0f59b0e2255057f416895fb.
Reviewed-By: Brad
|
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | | |
I'm not all to happy with this fix, but its the best that one can
acheive given the current design. The problem is that QPdfBaseEngine
sets a number of states as part of updateState(), but only when we are
playing back through the alpha engine. These states are used in some
draw functions, also when we are recording in the alpha engine. This
leads to the states and their checks being out of sync. So to follow
the existing pattern in the code we need to not touch d-> vars prior
to a check to usesAlphaEngine.
Reviewed-By: Eskil
|
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | | |
The patch tries to use texture_from_pixmap extentions on glX to properly
bind an X Pixmap to a texture in QGLContextPrivate::bindTexture(QPixmap,).
Because GL & X have different coordinate systems, the pixmap will be
inverted about the y-axis. The extension does however allow a
GLX_Y_INVERTED_EXT attribute to be set which will bind the pixmap the
correct way up. If the underlying driver doesn't support this,
texture_from_pixmap can't be used for QGLContext::bindTexture, because
that function expects the resulting texture to be the right way up.
However, it can still be used internally by the paint engine for
drawPixmap operations. For these cases, if the pixmap is inverted, the
paint engine can simply invert the texture coords to compensate. This is
why this patch also moves QGLTexture into qgl_p.h.
QGLContextPrivate::bindTexture(QPixmap,) now returns a QGLTexture which
the paint engine can inspect to see if it needs to invert the texture
coords.
Finally, it seems on some (probably all) drivers, deleting an X pixmap
which has been bound to a texture before calling glFinish/swapBuffers
renders garbage. Presumably this is because X deletes the pixmap behind
the driver's back before it's had a chance to use it. To fix this, we
reference all QPixmaps which have been bound to stop them being deleted
and only deref them after we swap the buffer, when they can be safely
deleted.
Reviewed-By: Kim
|
| | | | | | |
|
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | | |
This removes the need for a "root state" in the machine; or rather,
the machine _is_ the root state.
User code can now pass in a QStateMachine directly to the QState
constructor, instead of machine->rootState().
This also means we could get rid of the "proxying" from the machine
to the root state for things like properties (initialState et al),
finished() signal and auto-reparenting of states (the ChildAdded
event hack).
A fun little side-effect of this change is that it's now possible
to embed state machines within state machines. We can't think of
a good use case yet where you would rather embed a stand-alone
state machine (with its own event processing etc.) rather than
having just a regular nested state, but it's neat and it works.
Reviewed-by: Eskil Abrahamsen Blomfeldt
|
| | |\ \ \ \
| | | |/ / / |
|
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | | |
The signal could be connected a huge number of times
This is already fixed in master with Qt:UniqueConnection
Task-number: 258381
|
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | | |
I had missed this one in my rounds of updates. Bad me.
Reviewed-by: Thorbjorn
|
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | | |
Reviewed-by: Simon Hausmann
Reviewed-by: Frans Englich
|
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | | |
If the resource is valid, children should not be empty. If it happens to
be, hasNext() should then return false. Setting the index to 0 ensures
this and also that we don't keep trying the same thing over and over.
Reviewed-by: Olivier Goffart
|
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | | |
This was making the resource iterator return empty entries after listing
resources. This showed up after QDirIterator stopped filtering empty
entries.
Reviewed-by: Olivier Goffart
|
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | | |
Empty filenames should only show up from bugs in file engine iterators's
hasNext() function. We shouldn't try to hide those here.
Reviewed-by: Olivier Goffart
|
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | | |
Name filters and resulting regular expressions are stable, no need to
regenerate the latter on each iteration.
Reviewed-by: Olivier Goffart
|
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | | |
'*' is functionally the same as having no name filters. Equating the
equivalence in the constructor avoids repeated checks in the advance
"loop".
Reviewed-by: Olivier Goffart
|