This avoids dynamic construction of the private class. According to
the benchmark we go from 4,550 to 3,900 instruction reads per iteration.
(without change 32629676 the baseline is 5,600)
Change-Id: I5df925e30dbd49bdde87173e481820574ce5abe1
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
According to Clang 3.3, this is an extension:
qpalette.h:178:9: error: anonymous types declared in an anonymous union are an extension [-Werror,-Wnested-anon-types]
qsizepolicy.h:148:9: error: anonymous types declared in an anonymous union are an extension [-Werror,-Wnested-anon-types]
If you try to simply give the struct a name, it complains with:
qpalette.h:178:16: error: types cannot be declared in an anonymous union
Change-Id: I61c69b8e42a1f4c4a15a0733f2d7efa0b3e44864
Reviewed-by: Robin Burchell <robin+qt@viroteck.net>
Ever since 76e0223619, qPopulationCount
was extracted from QBitArray and moved elsewhere. That unfortunately
meant that the 24-bit loads are completely useless, since
qPopulationCount always operates on 32 bits.
Instead, do a full loop on 32-bit and then do 16-bit and finally 8-bit
(for which there are qPopulationCount overloads).
Change-Id: If945609f075095257d12877c39434909ac190e54
Reviewed-by: Olivier Goffart <ogoffart@woboq.com>
Reviewed-by: Marc Mutz <marc.mutz@kdab.com>
QWS is gone.
Change-Id: Ic04c4dcfff51522dd538bf2cb198dc2e02fe7572
Reviewed-by: Gunnar Sletta <gunnar.sletta@digia.com>
Reviewed-by: Paul Olav Tvete <paul.tvete@digia.com>
Reviewed-by: Peter Kümmel <syntheticpp@gmx.net>
The QtSensors module defines the coordinate system relative to
QPlatformScreen::nativeOrientation(). However, that is not part
of the public Qt API.
Task-number: QTBUG-32144
Change-Id: I6941c5b8589219064749f18880775b803f23fbbd
Reviewed-by: Gunnar Sletta <gunnar.sletta@digia.com>
It handles window movement events, not mouse move events.
Change-Id: Ibc8fd0272756fcf87dfac210addbf2f87cfb39d5
Reviewed-by: aavit <eirik.aavitsland@digia.com>
Make sure we don't modify the lhs. Instead copy it and advance the
copy.
Change-Id: I3440e8e175bfc299f8f83b816faca54fa3f79e43
Reviewed-by: Olivier Goffart <ogoffart@woboq.com>
Prevent qdoc from reporting duplicate pages for external pages with
the same URL and title.
Task-number: QTBUG-33462
Change-Id: Idabdb241aaa4fe105f7b3ea78229ff1ae8776ecf
Reviewed-by: Jerome Pasion <jerome.pasion@digia.com>
Our previous event loop integration had two unfortunate flaws:
1. We would call qt_user_main() from a timer, after returning from
didFinishLaunchingWithOptions. This had the effect of showing the
iOS application window long before the Qt application UI had been
set up, resulting in a 1-2 second flash of black/pink between the
launch image disappearing and the actual application showing.
2. We spun a nested event loop, where our implementation of the
different event loop modes did not perfectly match the Apple
implementation. This resulted in scrolling being busted in
some cases such as when showing the virtual keyboard for
Emoji characters.
These two issues have now been solved by calling the user's main()
from didFinishLaunchingWithOptions. Normally this would not work, as
the user's main would call QApplication::exec() at the end of their
main(), which would block and we would never return back from the
didFinishLaunchingWithOptions callback, resulting in no UI on screen.
We work around this by longjmp'ing out of QApplication::exec(), back
into didFinishLaunchingWithOptions, so that it can return. Again,
this would normally not work, as the call stack where QApplication
and friends would live would get smashed as the application
continued executing. We work around this by allocating a block
of stack space at the start of main(), which we then redirect the
stack pointer to before calling the user's main. This results in
the whole stack of the user's main() and below being preserved, even
if we longjmp out of the call stack (which then restores the
stack pointer).
This approach should work fine together with garbage-collection as
well, since the mark-and-sweep phase will walk the stack from the
stack pointer to the stack base, including sections of the stack
that were part of qt_user_main() and live in the reserved area.
One case where GC will fail though is if it happens as part of the
qt_user_main() call, where the GC will not mark anything in the
'real' callstack below UIApplicationMain(), but this is not
expected to happen.
The size of the reserved stack can be controlled through the
Info.plist key 'QtRunLoopIntegrationStackSize', as well as the
'QtRunLoopIntegrationDisableSeparateStack' key to disable the
separate stack approach completely. This will fall back to the
old approach. The amount of stack space used by the user's
main can be determined by enabling a special debugging mode,
using the 'QtRunLoopIntegrationDebugStackUsage' key.
Change-Id: I2af7a6cfe1a006a80fd220ed83d8a66d4c45b523
Reviewed-by: Simon Hausmann <simon.hausmann@digia.com>
If both n1 and n2 are "Private", the comment says they're equivalent,
but the function is actually stating that n1 < n2 AND n2 < n1.
Change-Id: I8e30b32c1e5240551ab1808baf6fc476841dfde4
Reviewed-by: Martin Smith <martin.smith@digia.com>
Reviewed-by: Topi Reiniö <topi.reinio@digia.com>
Make sure enough pixels is sourced by multiplying
the layout rect size with the pixmaps devicePixelRatio.
Task-number: QTBUG-31477
Change-Id: I742603c09a65a520587da3eef9479ef7f0c87022
Reviewed-by: Gabriel de Dietrich <gabriel.dedietrich@digia.com>
A module plugin in qml belongs to a URI/namespace. This
uri is resolved run-time by QtDeclarative by knowing the
path of the qmldir that references the plugin.
For static plugins this becomes a problem, since we lost
the information regarding which plugin belongs to which
qmldir, since a static plugin has no file path.
To avoid pushing the responsibility of clarifying this
onto the application developer, it is better to embed this
information into the meta data of the plugins themselves.
Since this information can be resolved by the
build system, a new option to moc has been added:
-M<key=value>
that will let you add meta tags to the meta data from
the command line to each class that has an IID specified.
For the URI case, we can then e.g do:
-Muri=QtQuick.Controls -Muri=QtQuick.Controls.Private
Change-Id: I81a156660148fc94db6f3cac0473e9e1c8458c58
Reviewed-by: Olivier Goffart <ogoffart@woboq.com>
We plan to add support for native message/alert dialogs on Android
and iOS because it's otherwise impossible to have a true popup window.
Then we might as well have native message/alert dialogs on other
platforms too. It will become an alternative implementation behind
QMessageBox and perhaps QErrorMessage.
Task-number: QTBUG-30883
Task-number: QTBUG-29462
Change-Id: I73dcfc6438e696189b6d37091874c7ad69b4ec68
Reviewed-by: Eskil Abrahamsen Blomfeldt <eskil.abrahamsen-blomfeldt@digia.com>
Reviewed-by: Richard Moe Gustavsen <richard.gustavsen@digia.com>
Added QXcbSessionManager to the Xcb plugin.
QXcbSessionManager inherits from QPlatformSessionManager, it's a port of
QSessionManager as it is in Qt 4.8.
Minor changes also in QPlatformSessionManager and QGuiApplication to
hook it up.
Task-number: QTBUG-28228
Task-number: QTBUG-30011
Task-number: QTBUG-33033
Change-Id: I50b33d05a1e32c5278dea339f693713acc870a70
Reviewed-by: Friedemann Kleint <Friedemann.Kleint@digia.com>
Reviewed-by: David Faure <david.faure@kdab.com>
These structures do not need to be packed. With some compilers, Q_PACKED
was defined to be empty, which means that the code was working fine
without packing. Or there were some lingering problems on those
platforms (MSVC) and we're now exposing them in all platforms.
Actually, it shouldn't be a problem anywhere. QCharAttributes, quint24
and QFontEngineQPA::Glyph have only char/uchar members, so they're
packed already (they have alignof == 1). The only platform where that
wasn't true was ARM OABI, which we don't support anymore.
QFontEngineQPA::Header seems to always come from a QByteArray, so it
doesn't neeed to be packed either. However, just to be sure, I'm
inserting a check for alignment.
And QFontEngineQPA::Block isn't used anywhere.
Change-Id: Iacfa25edf336ef2a03aeb6e40ae90937a21661ae
Reviewed-by: Konstantin Ritt <ritt.ks@gmail.com>
Reviewed-by: Gunnar Sletta <gunnar.sletta@digia.com>
Structs defined with Q_PACKED *need* to be packed in all
platforms. Having an empty define means errors might go unnoticed for
a long time.
Change-Id: I9c388358fed27a8e018465f05375e4bbbee3484c
Reviewed-by: Konstantin Ritt <ritt.ks@gmail.com>
Reviewed-by: Olivier Goffart <ogoffart@woboq.com>
Leftover. It's not used anywhere.
Change-Id: I023a390d84c9ed5dacd2699145f183e8a2317153
Reviewed-by: Konstantin Ritt <ritt.ks@gmail.com>
Reviewed-by: Olivier Goffart <ogoffart@woboq.com>
Sorting is O(NlogN) complexity, while nth_element is linear.
Change-Id: Ic6596affe183494e87abe7bdaa7c9985f5b7cd58
Reviewed-by: Olivier Goffart <ogoffart@woboq.com>
In case this code is ever instantiated with a type that actually benefits from moving.
'value' is save to move since it was passed to the function by value.
'*mid' is save to move since it originates in a container that will be
discarded on the next line anyway.
Change-Id: I13587be10974ba6cf95d56e0f8912ff88167b60b
Reviewed-by: Olivier Goffart <ogoffart@woboq.com>
Call QPlatformCursor::pointerEvent() even when the mouse event is
stopped due to being blocked by a modal window. This is necessary for
relatively dumb platforms, like eglfs, because the mouse cursor is
expected to be functional across the entire screen regardless of
having a modal dialog shown.
Platforms where the cursor is handled by the system (xcb, windows,
etc.) are not affected as they don't implement pointerEvent() anyway.
Change-Id: I4996ca0db23eb7f6ca3e2432d8d7abb1a5113dca
Reviewed-by: Gunnar Sletta <gunnar.sletta@digia.com>
Mesa does not like eglSwapInterval calls without a current context.
Change-Id: I7ec2d4311586cf74da0461bc951a0e5d9399c35b
Reviewed-by: Gunnar Sletta <gunnar.sletta@digia.com>
This is a simple optimisation allowed to us by the fact that all platforms
we run on use two's complement for the signed integers. The trick works as
long as one of the two signed integers is known beforehand to be
non-negative:
- by definition, for any signed integer i, i <= INT_MAX
- by definition, for any unsigned integer u, u >= 0
- given a signed integer x >= 0, 0U <= uint(x) <= uint(INT_MAX)
- therefore, given another signed integer y of whatever
value, uint(x) < uint(y) ←→ x < y && y >= 0
The trick is an optimisation because the compiler doesn't know that one of
the two sides is always non-negative. Otherwise, it would do the
same optimisation.
Change-Id: If256ec0df4e06335805af8010bb67ce5fd3e065a
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
When trying to create more than one window, stop with a helpful error
message since this is not yet supported. Also, return a fake WId for
desktop windows.
Change-Id: I9859b62b1d4f6b6142982d2e5a90afc1fc3c6a28
Reviewed-by: Gunnar Sletta <gunnar.sletta@digia.com>
Q_GLOBAL_STATIC accessor is documented to return dangling pointer if
called after destruction.
Change-Id: Ieafd5619b20ad256d9d5ad007d939f1430ef681f
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Reviewed-by: Olivier Goffart <ogoffart@woboq.com>
It is not really fatal but fixing it gets rid of a Mesa warning.
Change-Id: I3045b2691e7457541d6524c3e3ff8a1882ca460b
Reviewed-by: Gunnar Sletta <gunnar.sletta@digia.com>
Probably added by a1898f4466.
qmetatype.h:70:29: error: inline function 'qMetaTypeId<QtMetaTypePrivate::QSequentialIterableImpl>' is not defined [-Werror,-Wundefined-inline]
qmetatype.h:1363:30: note: used here
const int toId = qMetaTypeId<QtMetaTypePrivate::QSequentialIterableImpl>();
qmetatype.h:70:29: error: inline function 'qMetaTypeId<QtMetaTypePrivate::QAssociativeIterableImpl>' is not defined [-Werror,-Wundefined-inline]
qmetatype.h:1386:30: note: used here
const int toId = qMetaTypeId<QtMetaTypePrivate::QAssociativeIterableImpl>();
Change-Id: I9afe1a3d50ab23eb701797e28b259966dc95a147
Reviewed-by: Stephen Kelly <stephen.kelly@kdab.com>
So far we've known that we want QCollator as public API. It hasn't
been possible yet due to the strong dependency that QCollator used
to have on ICU.
This patch adds collation support for the platforms where ICU is not
the best option by using native collation API. Namely Windows and
Mac OS X.
Additionally a fallback POSIX back-end is added, so that we can make
sure it will work on any posix-compliant platform.
Change-Id: Ia1734acbf5f596698a81f2af927cc15636e4c908
Reviewed-by: Lars Knoll <lars.knoll@digia.com>
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Propose the API to be changed so that we can implement QCollator in the
different platforms where Qt is available.
Change-Id: I7b3e933d7e3d1aa26c1b78d21ef75b71c692827f
Reviewed-by: Lars Knoll <lars.knoll@digia.com>
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
The class missed the feature freeze for Qt 5.0, but has been ready
for quite a while. So make it public in time for Qt 5.2.
Change-Id: I9ac3f579ff5e371925fad40684762fff7ee4abd8
Reviewed-by: Lars Knoll <lars.knoll@digia.com>
Move scriptRequiresOpenType() body right into QFontEngine::supportsScript(),
thus centralizing use of this performance cheat.
Change-Id: I5f494b086f8f900b631c491f41e9cb800002c0f6
Reviewed-by: Lars Knoll <lars.knoll@digia.com>
Get rid of duplicated includes, useless defines, and private enums
that just duplicate a public ones; avoid variable names collisions.
Change-Id: I540e812b6cd510eb1ed441d97e9af377611d804a
Reviewed-by: Lars Knoll <lars.knoll@digia.com>
Implement the foreignWindowCreated and foreignWindow closed signals, emitted when foreign windows are created and closed, respectively.
Change-Id: I72dd5380e6061f191eb8362fda5dd8fb8e9ed06b
Reviewed-by: Thomas McGuire <thomas.mcguire@kdab.com>
Reviewed-by: Kevin Krammer <kevin.krammer@kdab.com>
Reviewed-by: Rafael Roquetto <rafael.roquetto@kdab.com>
Implemented nativeResourceForScreen api to return QObject* which can be used to connect foreignWindowCreated, foreignWindowClosed signal in QQnxScreen.
Usecase is to connect signal in custom QML component as below
QObject * obs = interface->nativeResourceForScreen("QObject*", screen);
connect(obs, SIGNAL(foreignWindowCreated(void*)), d, SLOT(newForeignWindowCreated(void*)));
Change-Id: I512c3b6d188a2e90ef7b8e89c413ca420a29dd9b
Reviewed-by: Kevin Krammer <kevin.krammer@kdab.com>
Reviewed-by: Fabian Bumberger <fbumberger@rim.com>
Reviewed-by: Rafael Roquetto <rafael.roquetto@kdab.com>
As we're using CFRunLoopIsWaiting() to check for the possible presence
of system events hasPendingEvents() will never return false if called
on the main thread. We assume clients will not use this function to
determine whether or not to call processEvents(), but instead use the
return value from processEvents.
Change-Id: Ifd63892c6d35bb7da204072616bfe3ee69ca1d85
Reviewed-by: Richard Moe Gustavsen <richard.gustavsen@digia.com>
UIKit changes the run-loop mode during scrolling to UITrackingMode, which
presumably prioritizes touch events and other sources related to a
smooth scrolling experience. It signals this change by interrupting the
current run-loop pass, and the outer loop is responsible for re-entering
the run-loop in the new mode. Failing to enter the run-loop with this
new mode results in UIScrollViews losing their kinetic feel when
flicking. This can be observed by e.g. bringing up the Emoji keyboard
and scrolling it horizontally.
We keep track of the current run-loop mode by listening for push and pop
notifications on the UIApplication object. The current mode is then used
in our Q_FOREVER-loop when re-entering CFRunLoopRunInMode.
For now we don't add our posted event source or timer source to the
new modes, under the assumption that the system prefers to limit the
number of sources that will fire during scrolling. If this turns out
to give a bad user-experience for Qt applications we should consider
changing it.
Change-Id: I3a612b3cfc77c74b658963057732dc4d61684df8
Reviewed-by: Tor Arne Vestbø <tor.arne.vestbo@digia.com>
The median, for arrays of odd size, is found in [size/2], not [size/2+1].
For arrays of even size, the code doesn't work anyway, so nothing to do there.
Change-Id: Id23ff3fe9538c2d8ef8f88e23127cb92af08b444
Reviewed-by: Olivier Goffart <ogoffart@woboq.com>
Sorting is O(NlogN) complexity, while nth_element is linear.
Also remove the errornous +1 when calculating the median position.
Change-Id: Ib39085b59a6c5d15a3a940b1ce3377080340bc09
Reviewed-by: Olivier Goffart <ogoffart@woboq.com>
Instead of having separate code-paths for QEventLoop::EventLoopExec
and the non-blocking processEvent() we now have a single Q_FOREVER
loop where the logic can be shared. We make multiple loop-passes,
each time calling CFRunLoopRunInMode with potentially different
arguments, depending on the result of the previous run.
For the EventLoopExec case we'll continue making loop-passes until
the event-loop has been interrupted. For the non-EventLoopExec case,
we respect interruption, but will continue making loop-passes only
until we've processed all events in the queue, optionally waiting
for the initial event if WaitForMoreEvents is set. Limitations in
the CoreFoundation APIs unfortunately force us to keep some state
on whether or not we've processed events and timers for a given
processEvents() pass (with corresponding deferred scheduling of
timers and event source signaling).
The way we handle timers has also been rewritten to no longer defer
the timer activation to a special timer source. The constraint of
CoreFoundation timers is that they can not recurse (re-fire) in a
callback, but that only applies per timer, so using multiple CF
timers allows us to recurse. We still only use a single CF timer
for all the Qt timers though, and only spawn a new CF timer if
the user calls processEvents() inside a timer callback.
This commit removes the logic related to dealing with UITrackingMode,
as that logic was slighly problematic, but the feature will be
added back in a follow-up commit, in line with the new approach.
The result of this commit is that we're passing all event-loop,
event-dispatcher, and timer auto-tests, both for the QtCore dispatcher
and the GUI event-dispatcher.
Change-Id: I3c56fbc7857a25110064681257abb47075b5bd2d
Reviewed-by: Richard Moe Gustavsen <richard.gustavsen@digia.com>
Apple only provides simulators for 5.0+, and we now rely on 5.0+ APIs.
Change-Id: I9ec047767b5f5e1b33aeef186ec6aff2b9c70a05
Reviewed-by: Richard Moe Gustavsen <richard.gustavsen@digia.com>
It's a supported platform from 5.2, and we want build-scripts/CI/etc to
adapt to the change as soon as possible.
Change-Id: I8c78351191f59a6ecab33acc0829d2535379c787
Reviewed-by: Richard Moe Gustavsen <richard.gustavsen@digia.com>
Reviewed-by: Simo Fält <simo.falt@digia.com>
