Assuming the CPU has already loaded 'this', the value of 'd' is just
an indirect load away. The value of d->size, however, is two indirect
loads away, one of which is the load of 'd'.
So it makes more sense to check for d-pointer equality first, as that
can proceed in parallel with the fetch for d->size, which the CPU
may speculatively trigger.
In addition, at least GCC in release mode after this change doesn't
set up the stack frame if the d-pointer check succeeds.
Change-Id: I61f9b245070dd1742fca6ccb8d4936a0b1aa7c07
Reviewed-by: Giuseppe D'Angelo <giuseppe.dangelo@kdab.com>
Simplify the temporary buffer usage by using the built-in WinRT Buffer
class. This also allows one use of the local event loop to be removed.
Change-Id: Ice552910227ffbe31c6e8716ff7896af7c4532ef
Reviewed-by: Maurice Kalinowski <maurice.kalinowski@digia.com>
This stores the socket pointer in the descriptor, rather than an
abitrary handle, so that it is easier to access from e.g. SSL socket.
To further support SSL sockets, a special case for SSL sockets is
made so that the the socket reader installation can be delayed until
after the socket is encrypted (as this is the only supported mode of
operation with StreamSocket).
Change-Id: I693229189722dc43b221b167e8256f5497a50346
Reviewed-by: Maurice Kalinowski <maurice.kalinowski@digia.com>
Similarly to 1b19f660 (which added socket flushing before close), the
socket should have a chance to disconnect if the write buffer is not
empty. Instead of flushing, the pendingClose is added so that the
backend may disconnect the socket once these bytes have been written.
Change-Id: I2d85b6356c3e158bade3d5d86161d3e33649cad6
Reviewed-by: Richard J. Moore <rich@kde.org>
This moves the socket backend's host name matching functions up to
QSslSocketPrivate so that they can be shared between backends. This
works, as there is no OpenSSL-specific code here.
Change-Id: I73c2081fdc2e60a44c90e90800d1e1877391a626
Reviewed-by: Richard J. Moore <rich@kde.org>
They're identical to the compiler-generated ones, and probably inhibit
move operators here and there.
Change-Id: I918f2946f6b6aa49aa883420dc87df44013938a5
Reviewed-by: Olivier Goffart <ogoffart@woboq.com>
If a (non-external) link string ends in '.html', qdoc assumed it is
a direct link to a generated html page. However, it could also refer
to an example file with .html extension.
This commit fixes a corner case where links to an example file page
were broken for such files.
Task-number: QTBUG-40831
Change-Id: I31acc141970b6768f0a93964723be82611d37a3d
Reviewed-by: Jerome Pasion <jerome.pasion@digia.com>
Adapt the code that generates the navigation bar to the recent
changes in node hierarchy - Module pages, QML module pages and
groups are no longer DocNodes.
This change will ensure that if a page has a title, it will be
visible in the navigation bar regardless of the the page (node)
type.
Change-Id: I697a12d5904d88f91771764ab7ed607b79e4eab1
Reviewed-by: Martin Smith <martin.smith@digia.com>
In the offline template, the floating elements in the navigation bar
interfere with the layout of any elements displayed next to it. In
particular, subtitles used for example files and nested class
documentation have wrong horizontal position.
This change adds CSS for the subtitles to be displayed as block
elements, and disallows floating elements on the left side - TOC
is still allowed to float on the right side.
Task-number: QTBUG-40924
Change-Id: I89be3844985e6cf95263db575768034d9270d140
Reviewed-by: Jerome Pasion <jerome.pasion@digia.com>
When writing the Qt Help Project XML file, QDoc traverses the
documentation nodes recursively, adding html filenames for each node
to the XML.
The logic that QDoc uses for this process is not perfect, and needs
to be kept up to date whenever the internal structure of the node
tree changes. This often leads to problems where some pages are
generated but not added to the .qhp, resulting in missing pages in
the offline documentation.
This change fixes this problem by having the generator keep track
of the created filenames, and passing that to the help project
writer.
Task-number: QTBUG-40572
Change-Id: Ife60a30724183a2b6dcd2397ea79bfbdc2addd04
Reviewed-by: Martin Smith <martin.smith@digia.com>
Reviewed-by: Jerome Pasion <jerome.pasion@digia.com>
Since the \br was promoted from a macro to a QDoc command, its output
has been enclosed in extra paragraph end/start tags, adding to the
visible vertical space.
This change fixes the issue by not closing the paragraph when QDoc
encounters a \br command.
Also removes the now-obsolete \br and \hr macros, as they are both
proper commands. \BR and \HR substitute macros are kept.
Task-number: QTBUG-37361
Change-Id: Iabbefb6e79268419792ccba42386f6342ccd175d
Reviewed-by: Martin Smith <martin.smith@digia.com>
Having QAsn1Element in place, we can have a common foundation
for the ssl certificate class for upcoming ports like WinRT
and SecureTransport. The only thing that has to be added to
the existing class is the handle() functionality.
Change-Id: I560a8e412b26f350855c7bc456fcdb8e9b750939
Reviewed-by: Richard J. Moore <rich@kde.org>
Gets rid of awkward wrapping of Qt::InputMethodQueries as integer in a
NSObject.
Change-Id: Ia7e368fc12ec7957ca8ab602d8cec1e0a071af1d
Reviewed-by: Richard Moe Gustavsen <richard.gustavsen@digia.com>
On iOS 8, the value 0 is returned as a kCFNumberIntType.
The code still works - CFNumberGetValue converts it
to a 0.0 double.
Change-Id: Ic50900b22e4fa19ad1481e8e0e293559bbfd8cd2
Reviewed-by: Tor Arne Vestbø <tor.arne.vestbo@digia.com>
We do this for QWidget::setFocus(), but were missing a call on clearing
the focus widget. Since QWidgetWindow::focusObject() will fall back to
returning itself if there is no focus widget, we need to pass the
result of window->focusObject() to the change signal instead of 0.
Change-Id: I52a5519a19bb20e74b4a7c2a1abc9d47e2ea1315
Reviewed-by: Jan Arve Sæther <jan-arve.saether@digia.com>
This function actually is worse than just calling
QAccessibleButton::text which it already does. It would mess up the
shortcut handling in addition to that.
Change-Id: I56cb95a44624da4c5fccb43e6835f6012a083337
Reviewed-by: Jan Arve Sæther <jan-arve.saether@digia.com>
QList<WidgetMapper>::takeLast() is a needlessly expensive operation
(involves copying a QPersistentModelIndex and a QPointer).
Instead of looping over takeLast() until empty, reverse-iterate
over the list and call removeEventFilter(), then clear the whole
list.
We reverse-iterate to preserve existing behavior. Also, since the
original code popped the WidgetMapper instance before calling
removeEventFilter(), we move the whole list into a local copy
before iterating. There's little chance that a removeEventFilter()
call will cause reentrancy into QDataWidgetMapper, but better safe
than sorry.
This and the previous change together save 448 bytes of text size in
libQt5Widgets.
Change-Id: I1bfe907751659f31e618aa05bbb7b840f0aa61f4
Reviewed-by: Giuseppe D'Angelo <giuseppe.dangelo@kdab.com>
Reviewed-by: Friedemann Kleint <Friedemann.Kleint@digia.com>
Don't use index-based iteration, but use iterators.
Change-Id: I57c9582aed644fc58ced1a1af940dcd20d11d970
Reviewed-by: Friedemann Kleint <Friedemann.Kleint@digia.com>
This adds support for the environment variable
QT_DEVICE_PIXEL_RATIO for the Windows platform plugin.
Task-number: QTBUG-38993
Task-number: QTBUG-38858
Change-Id: I6831eb6d3a09a80be7bbef46395e91531b61cc50
Reviewed-by: Alessandro Portale <alessandro.portale@digia.com>
Reviewed-by: Paul Olav Tvete <paul.tvete@digia.com>
A device removed very fast after being inserted might disappear while
we are still seting it up. We must therefore check if we indeed still
get a matching device
Task-number: QTBUG-40820
Change-Id: I4372fb1932264e5799f37cea0d016795e28ebed6
Reviewed-by: Allan Sandfeld Jensen <allan.jensen@digia.com>
Reviewed-by: Shawn Rutledge <shawn.rutledge@digia.com>
Follow the good practice of checking for isCreated() before calling
VAO functions like bind(). Use also the vao binder where applicable.
Change-Id: Ib827f3bce838396bf2e08f9480fa63801d4d3a50
Reviewed-by: Jørgen Lind <jorgen.lind@digia.com>
It gets somewhat complicated due to the fact that a RasterGLSurface window
(i.e. any widget window since 5.3) may behave either like an OpenGLSurface
or a RasterSurface, and the expected behavior may change on each backingstore
sync.
This does not fit designs where the platform window implementation is separated
and there is different behavior for raster and GL windows.
Therefore QAndroidPlatformOpenGLWindow is now made capable of behaving like the
raster one, based on a flag communicated from the widget stack via QWindowPrivate
(since the plugin knows nothing about widgets).
This means that widget windows that do not have renderToTexture children (QOpenGLWidget,
QQuickWidget) will go through the raster path, while the ones that have will behave
like an OpenGL window with the actual rendering happening in
QPlatformBackingStore::composeAndFlush().
The surface type is RasterGLSurface in both cases nonetheless.
Task-number: QTBUG-37907
Change-Id: I6f9261fc0fd993afcda7f30d379c5410069033d3
Reviewed-by: Paul Olav Tvete <paul.tvete@digia.com>
Like in all other places. This way the resulting QMenu is correctly
parented so it will show up at the proper position even on platforms
which do not have a way to position top-level windows.
Task-number: QTBUG-29025
Change-Id: I2aa6fe73699379029c44a3f379366a2133753190
Reviewed-by: Robin Burchell <robin.burchell@viroteck.net>
Show debug output for rejected and currentColorChanged signals,
because accepting is not the only scenario to be tested.
Task-number: QTBUG-40855
Change-Id: If741ab19392e7d4314e0eff82a939d202ae86b48
Reviewed-by: Gabriel de Dietrich <gabriel.dedietrich@digia.com>
but only if there is a Cancel button. A color dialog without a cancel
button might be kept open and apply to various selections, so it
doesn't make sense to reject such a dialog, only to close it.
Task-number: QTBUG-40855
Change-Id: Ifffb4ae81307c72259ed388a4776ba09543603e7
Reviewed-by: Gabriel de Dietrich <gabriel.dedietrich@digia.com>
Setting it initially to 0 is very important, otherwise we will
do a glDeleteTextures with the undefined value.
The result sometimes goes unnoticed and sometimes causes bizarre
issues: For example in the 'textures' example one face of one cube
out of the six did go blank from time to time since the corresponding
texture was deleted by the backingstore.
Change-Id: Id19eb2164471b542b08a277a65edfcb5d0f8248d
Reviewed-by: Paul Olav Tvete <paul.tvete@digia.com>
If the window was not visible when the geometry was set, then the
DirectFB window would not be resized.
Change-Id: I7790c90ed0fb755aebee0e32c877ebd9e48417cd
Reviewed-by: Jørgen Lind <jorgen.lind@digia.com>
The local and global coordinates for mouse events were being translated
incorrectly from the native DirectFB events.
Change-Id: Id904a4335459b87c92f4b8b46d535c78fb7dad8c
Reviewed-by: Jørgen Lind <jorgen.lind@digia.com>
Previously when we created any QWidget based application a
QDesktopWidget would be created as a physical window like any other, but
this window would steal input from the application. We now create a
DirectFB window for the Qt::Desktop type of widget now, but it does not
receive input events and can not be painted to or displayed.
Change-Id: I6a090c5384b1f83383e40680dbede5d0edc41983
Reviewed-by: Jørgen Lind <jorgen.lind@digia.com>
Using the same multiple inheiritance that is used in the EGLFS platform
plugin.
Change-Id: I016f904bfc365bec6266c3f5d638ab15ecefe63b
Reviewed-by: Laszlo Agocs <laszlo.agocs@digia.com>
The DirectFB platform plugin was missing support for services and
inputContext.
Change-Id: I010fdcbed5e172b019b4dce79f3beea0f9c5025d
Reviewed-by: Laszlo Agocs <laszlo.agocs@digia.com>
The QPanGesture recognizer requires single-point touch events. The touch
implementation in Qt 4 would test Qt::WA_TouchPadAcceptSingleTouchEvents
and forward single touch events if set.
Making this work in Qt 5 is a little bit more involved since the platform
plugins don't know about widgets.
Change the Cocoa touch implementation to send single-point touch events
to QWidgetWindow windows only. Make QApplication forward single-point
touch events only if the target widget has the
Qt::WA_TouchPadAcceptSingleTouchEvents attribute set.
Task-number: QTBUG-35893
Change-Id: I68712a5e3efb4ece7a81ca42f49c412e525eeb3a
Reviewed-by: Jake Petroules <jake.petroules@petroules.com>
The iterators for QJsonArray and QJsonObject are currently lacking an
operator-> definition. Unfortunately it is not possible to do in clean
way without redefining either the iterators or QJsonValueRef class.
This patch instead adds two fake pointer classes that are only used
to handle the operator-> return value.
Task-number: QTBUG-29573
Change-Id: Ief785a6afbbedc9e89cf3b6f3958c2c755997a66
Reviewed-by: Lars Knoll <lars.knoll@digia.com>
Writing out the alpha is re-enabled too early. When blitting the
AlwaysStackOnTop widgets as the last step of the composition, they
need the exact same settings as the backingstore content, meaning
blending but without writing out alpha.
Move the glColorMask call to fix this. This will avoid issues with
semi-transparent AlwaysStackOnTop widgets when the top-level has
alpha enabled too.
Task-number: QTBUG-40910
Change-Id: Id6d0d684cfa78bf79b65a097efd92de575e73b2c
Reviewed-by: Jørgen Lind <jorgen.lind@digia.com>
Use showMaximized(). Just calling show() without setting a size can result
in a zero-sized, invisible window on some platforms.
Change-Id: Ifa48258060e3d651c2fac3a1409a26a2c3db6bdb
Reviewed-by: Robin Burchell <robin.burchell@viroteck.net>
There are three reasons to do so:
1. This could be more efficient, depending on the STL implementation.
2. By using QTypedArrayData iterators (T*) instead of QVector ones,
we actually invoke the non-templated range ctor of std::vector,
at least in the common case that std::vector<T>::const_iterator
is also const T*.
3. The change turns a former NRVO return into a RVO one, potentially
allowing more compilers to perform the copy elision.
Change-Id: I70b35aaeae70ba06a971a36b8b1b1da997e8094f
Reviewed-by: Olivier Goffart <ogoffart@woboq.com>
