| Commit message (Collapse) | Author | Age | Files | Lines |
... | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | | |
When QGLWidget was used as a viewport for QGraphicsView, it was
still treating the window surface as RGB16. Use the screen's actual
pixel format.
Also ensure that PvrEglWindowSurface::image() returns a non-null
QImage if the drawable hasn't been created yet.
Reviewed-by: trustme
Back port of 53b3a0572242d0a425e74848afba1293f195d29b
|
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | | |
Incorrect optimisation - for popup windows, Qt's focus is moved before
hide_sys is called, resulting in the popup window keeping its elevated
position in the CONE control stack.
This can result in keyboard focus being in an invisible widget in some
conditions - e.g. QTBUG-4733
Task-number: QTBUG-4733
Reviewed-by: axis
|
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | | |
Not exporting the whole class prevents the capabilities example from
linking, because the vtable is not exported.
However windows builds don't support export of templated class, and
export the vtable implicitly when there is an exported member function.
Task-number: QTBUG-4593
Reviewed-by: Thierry Bastian
|
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | | |
It turns out that scrolling appears to be slow when using
non-mac mice with the carbon build. This patch introduces a
an acceleration algorithm that closer resembles the one
used by Cocoa.
Rev-By: prasanth
|
| | |
| | |
| | |
| | |
| | |
| | | |
Added the missing defines
Reviewed-by: Denis
|
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | | |
The reason for scrolling bug was that CFbsBitmapDevice wasn't
informed if CFbsBitmap was resized. However, it seems that
scroll implementation in QRasterPixmapData is faster that
CFbsBitGc->CopyRect so this is also a patch which changes
QS60PixmapData's scroll function to call QRasterPixmapData::scroll().
Reviewed-by: Janne Koskinen
|
| | |
| | |
| | |
| | |
| | |
| | |
| | | |
This restores the behavior in 4.5.
Task: QT-2257
RevBy: Simon Hausmann
|
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | | |
OpenC incorrectly caches DST information between localtime_r
calls, i.e. if previous call to localtime_r has been called for DST
affected date, also the second call will be affected by DST even
the date is such that DST should not be applied.
The workaround is to call mktime with non-DST affected date before
calling localtime_r. mktime call resets the OpenC internal DST cache
to right value and localtime_r will return correct values.
This commit can be reverted once Open C bug 9525 has been reverted.
AutoTest: Fixes tst_QDateTime::totime_t
Reviewed-by: Aleksandar Sasha Babic
|
| |\ \ |
|
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | | |
This code was kind of dead anyway because
- On X11 and Windows, most of the widgets have the StrongFocus FocusPolicy
- On Mac, e->modifiers always contains KeypadModifier for arrows, so the
code is never reached on mac with real applications
Morever, QAbstractButton already have a more complex handling of arrow keys.
And the code breaks the QButtonGroup arrowKeyNavigation test on Mac (as when
the test system simulates events, the KeyPadModifier is not set)
Reviewed-by: mbm
Reviewed-by: Paul
|
| | | |
| | | |
| | | |
| | | | |
Reviewed-by: Jens Bache-Wiig
|
| | | |
| | | |
| | | |
| | | |
| | | | |
Add Q_ENUMS/Q_FLAGS accordingly.
Reviewed-by: Olivier Goffart <ogoffart@trolltech.com>
|
| |/ /
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | | |
This fixes issue that both DialogButtonBox and softkeys are visible
(e.g. in Ftp example) at the same time.
Reason is that the previous implementation did the visiblitity setting when buttons were created.
Unfortunately buttons can be created without parent and let layout manager to find parent them
for you - so dialog as parent (=hide dialogbuttonbox as softkeys are in use) check failed.
Now similar check is done before showing the widget - it is layoutted here, so it has parent as well.
As a bonus, the event handling is already looking for dialog parent, so we can avoid doing another
for-loop.
Task-number: QTBUG-4739
|
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | | |
The error was: "*** No rule to make target `\Qt\tests\auto\kernel\
qguiplatformplugin_p.h_, needed by ..." This happens since Symbian does
not use same include semantics.
Reviewed-by: Miikka Heikkinen
|
| | |
| | |
| | |
| | | |
Reviewed-by: TrustMe
|
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | | |
When QGLWidget was used as a viewport for QGraphicsView on PowerVR/MBX
systems, it was double-flushing every frame because the window surface
flush() implementation was still trying to do a raster blit after painting.
This change suppresses the raster blit, leaving it up to the GL
swapBuffers() call to get the painted contents onto the screen.
Reviewed-by: Sarah Smith
|
| | |
| | |
| | |
| | | |
Reviewed-by: TrustMe
|
| |\ \ |
|
| | | |
| | | |
| | | |
| | | |
| | | | |
It's faster than calling getObject(), since getObject() will do
type checking of the value.
|
| | | |
| | | |
| | | |
| | | | |
Make allocation faster.
|
| | | |
| | | |
| | | |
| | | |
| | | |
| | | | |
There's no CSIDL_LOCAL_APPDATA on Windows CE. Use CSIDL_APPDATA instead.
Reviewed-by: TrustMe
|
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | | |
Making the triangle fan of each sub path start at the sub path's
centroid will on average improve performance for complex paths, because
less pixels that are outside the path need to be touched. The centroid
is a more balanced choice than just picking the first element of the
sub path as triangle fan origin.
A performance improvement of 20 % was measured for star formed paths.
Reviewed-by: Trond
|
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | | |
When no stencil clip is set we reduce the number of compositing passes
for rendering winding fill paths from four to two. When stencil clip is
set, the number of compositing passes is reduced from five to four.
For clipping with a winding fill path, the number of compositing passes
are reduced from five to four when stencil clipping is already enabled.
A performance improvement of up to 85 % was measured in certain cases.
Reviewed-by: Trond
|
| | | |
| | | |
| | | |
| | | |
| | | |
| | | | |
This might be useful for some people, so we should document it.
Reviewed-by: Bjørn Erik Nilsen
|
| |/ /
| | |
| | |
| | |
| | | |
This reverts commit 342fcb287b09d016d482e25482ddd1b36e2983a3.
Didn't work on all compilers
|
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | | |
Add a backend for QFileIconProvider in the platform plugin
Also change the QIcon::fromTheme backend in the platform plugin:
On KDE, we unfortunately can't use KIcon as backend, as the current
API doesn't let us know easily (and quickly) wether we should use
the fallback or not (KDE always fallback to the question mark
"unknown" icon)
So we will use the QIconLoader even on KDE.
But we need to make sure that the theme name and the icon search paths
are correct. Ask that to the platform plugin
Reviewed-by: Jens Bache-Wiig
|
| | |
| | |
| | |
| | | |
Reviewed-by: Olivier
|
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | |
| | | |
We should not create a QAction instance at program startup,
that is before QApplication was initialized. One reason not
to do that is that internal fonts are prematurely initialized
without QApplication::qt_is_gui_used set to true, which leads
to wrong font Dpis in qt_defaultDpiX().
This issue was detected due to the failure in tst_QTextLayout,
cases: testDefaultTabs, testTabs, testMultilineTab,
testRightTab, testTabsInAlignedParag, testCenteredTab,
testDelimiterTab, testMultiTab and tabsForRtl.
Fix: create a Q_GLOBAL_STATIC_WITH_ARGS for that QAction
instance.
Reviewed-by: Liang QI
|
| | |
| | |
| | |
| | |
| | |
| | | |
We need to test through extension tool button and menu for it when there is no enough width on embedded devices.
Reviewed-by: TrustMe
|
| | |
| | |
| | |
| | | |
We changed the behaviour of that, so the name should reflect that.
|
|\ \ \
| |/ /
|/| | |
|
| | |
| | |
| | |
| | |
| | | |
Makes the code a bit easier to read and speeds up setItemsGeometries()
in the normal use-case.
|
| | |
| | |
| | |
| | |
| | | |
For instance, if sizeof(qreal) == 4 (for instance on ARM) the stack size
needed will be the same, so it will just add an extra indirection.
|
|\ \ \ |
|
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | | |
We should not use the registry key when looking up the storage
location as this is not really supported by Microsoft and not
allways accessible. Instead we now use SHGetSpecialFolderPath.
Note that we avoid SHGetKnownFolderPath as it is not available
on all CE platforms though it is the reccommended function by
Microsoft.
Task-number: QTBUG-3241
Reviewed-by: prasanth
|
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | | |
Missing header file, resulting in compiler error about returning
an incomplete type.
Reviewed-by: Trust Me
|
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | | |
Made it thread-safe and actually make sure that we don't initialize the
data several times.
Reviewed-by: axis
|
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | | |
The crash was caused by the image data not being locked before being
accessed.
Also avoided an unnecessary detach copy by making the image variable
a reference.
RevBy: Jani Hautakangas
Task: QTBUG-4705
AutoTest: QWidget passed
|
| | | |
| | | |
| | | |
| | | | |
RevBy: Miikka Heikkinen
|
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | | |
Source-incompatible change
This reverts commit 8714892977269591bb9b348c6eb549a7f2c45cbc.
Rev-by: Trustme
|
| |\ \ \
| | |/ / |
|
| | | |
| | | |
| | | |
| | | |
| | | |
| | | | |
There is only one change in the actual code here, and it simply removes
an unnecessary initialization of hasSize in the ctors of the composite
anchors.
|
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | | |
items
Now every item starts with its geometry on ((0,0), preferredSize()).
By doing this we guarantee that every item will have a known initial value, which
can be specially useful on float cases for instance.
Signed-off-by: Jesus Sanchez-Palencia <jesus.palencia@openbossa.org>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@openbossa.org>
|
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | | |
situation
We make use of the anchor visited on findPaths() to keep track of how many items are visited when
walking on the graph. We can use this information of how many items are reachable (non-float) in order to check
if we have floating items (not reachable from any graph vertex). If so, we have a situation not supported by now.
Signed-off-by: Jesus Sanchez-Palencia <jesus.palencia@openbossa.org>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@openbossa.org>
|
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | | |
Since the AnchorData cleanup commit (c974aca8) all anchor initialization
is being done by refreshSizeHints. However that method was not able
to properly initialize the internal layout anchors.
This commit refactors that method both to add the functionality and
improve readability.
Signed-off-by: Eduardo M. Fleury <eduardo.fleury@openbossa.org>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@openbossa.org>
|
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | | |
Updating "refreshSizeHints" and "updateChildrenSizes" methods for standard,
parallel and sequential anchors, in order to support the expanding SizePolicy
flag.
This adds the logic required to calculate equivalent expected/nominal expanding
size (AnchorData::expSize) as well as the logic to propagate the effective
size when in the expanding state (AnchorData::sizeAtExpected).
Signed-off-by: Caio Marcelo de Oliveira Filho <caio.oliveira@openbossa.org>
Reviewed-by: Eduardo M. Fleury <eduardo.fleury@openbossa.org>
|
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | | |
This assertion started failing after the addition of expanding
state. After some investigation we felt that the assertion itself
was too strong and would fail in some valid cases.
The new assertion is formally right as it tests whether the
simplex solver was able to satisfy the simplex constraints,
_exactly_ what it is expected to do.
Signed-off-by: Eduardo M. Fleury <eduardo.fleury@openbossa.org>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@openbossa.org>
|
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | | |
The previous implementation of solveExpanding would assume that anchors
were either completely expanding (would grow to their sizeAtMaximum) or
non-expanding (would try to remain at their sizeAtPreferred).
What happens however is that with simplification enabled, we may have
anchors that are "partially" expanding. An example is a sequential
anchor that groups together two leaf anchors where only one of them
is expanding.
In this case, the expected size of that group anchor would be the sum
of the max size of the expanding child plus the preferred size of the
non-expanding child.
To handle the above case we added the "expSize" variable. It holds
the expected value at Expanding for each anchor. Based on this, and
the already calculated values of sizeAtMaximum and sizeAtPreferred,
the solver calculates the actual sizeAtExpanding for them.
Signed-off-by: Eduardo M. Fleury <eduardo.fleury@openbossa.org>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@openbossa.org>
|
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | | |
Now we can use refreshSizeHints_helper() to initialize complex
anchors (parallel and sequential). This avoids duplication of
code in the function simplifySequentialChunk().
This commit also 'disable' ExpandFlag temporarily, stabilizing the
tests for the simplification. The next commit should re-enable it
implementing the necessary code for simplification as well.
Signed-off-by: Caio Marcelo de Oliveira Filho <caio.oliveira@openbossa.org>
Reviewed-by: Eduardo M. Fleury <eduardo.fleury@openbossa.org>
|
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | | |
Use just one constructor for AnchorData and tweak the struct when
necessary. Also use the function refreshSizeHints() for Normal anchors
to get the item information instead of passing to the constructor.
This has the advantage that we don't need to replicate the code for
checking and updating the size hint and size policies. Note that the
code before only considered the size policies after the first
refreshSizeHints() -- which we know will be called after the
simplification, so the assumption seems to be correct, but we don't
depend on that anymore..
Signed-off-by: Caio Marcelo de Oliveira Filho <caio.oliveira@openbossa.org>
Reviewed-by: Eduardo M. Fleury <eduardo.fleury@openbossa.org>
|