This partially reverts 51ada7734a
because it's necessary to keep some scenarios with vnc and
remote X servers working. When an application is starting,
if we don't find the xrandr outputs but we know the dimensions
of the screen, we should still be able to put windows onto that
screen; but when we already had known xrandr outputs and then they
were removed, that's the case where we want to stop rendering
(and have no screen instances) until the screen(s) are reattached.
Task-number: QTBUG-31389
Task-number: QTBUG-40174
Task-number: QTBUG-42985
Change-Id: I13d0996ba6ece78c4ebcd2c3a59f1617c1c7f0fa
Reviewed-by: Uli Schlachter <psychon@znc.in>
Reviewed-by: Gatis Paeglis <gatis.paeglis@digia.com>
Reviewed-by: Laszlo Agocs <laszlo.agocs@theqtcompany.com>
OSX was missing text, url and image preview while dragging causing the
drag operation to show a small dashed box around the mouse cursor.
Task-number: QTBUG-33824
Change-Id: I8d0acd0d6a48b7cb29ad2faba8b9ecb9cdf2a5ab
Reviewed-by: Morten Johan Sørvig <morten.sorvig@theqtcompany.com>
Reviewed-by: Shawn Rutledge <shawn.rutledge@digia.com>
We used to stop listening for keyboard visibility when quipickerview
was deallocated. The problem with that approach is that we don't
have a guarantee for when dealloc gets called. So what can happen
is that we get a call from Qt to dismiss the menu. That causes us
to close the keyboard, which will trigger the notification, which
will call dismiss recursively. This will hit our assert checking
if we have a valid picker view.
This patch will unsubscribe to the notification explicitly upon a
call to dismiss, to avoid recursive dismiss calls.
Change-Id: If1efa3438037e00a02bc186fdcb6c0b3d3d595e4
Reviewed-by: Tor Arne Vestbø <tor.arne.vestbo@theqtcompany.com>
When using native windows, notification were received after the platform
window pointer was cleaned up, which caused crashes.
Change-Id: I438fc29d1761a32a4705c4c802afe46908505756
Reviewed-by: Shawn Rutledge <shawn.rutledge@digia.com>
We treat windows as exposed even if the application state is still
inactive (e.g. during startup), otherwise there's a visible moment
of black screen between the launch screen fading out and the app
rendering its first pixel.
Change-Id: I48368459a8a46fa1c5b2853ea88adfe1ac61b6f7
Reviewed-by: Richard Moe Gustavsen <richard.gustavsen@theqtcompany.com>
Reviewed-by: Tor Arne Vestbø <tor.arne.vestbo@theqtcompany.com>
When a QWindow is closed and destroyed it resets QGuiApp::focus_window
before calling setVisible(false) on the platform window, so we don't
have the needed information anymore to check if the window we are
hiding was the focus-window. To work around this we use the native
state instead, which should still be valid if the window was the
focus window.
Change-Id: I98057e6393411471f03668e3e5ce544f6b49c01d
Reviewed-by: Richard Moe Gustavsen <richard.gustavsen@theqtcompany.com>
Reviewed-by: Tor Arne Vestbø <tor.arne.vestbo@theqtcompany.com>
When the 'debugWindowManagement' option is enabled, e.g. by passing the
argument -platform ios:debugWindowManagement to the application, the
plugin will add a background color and outline to all QUIViews, as
well as draw a grid for the underlying QUIViewController. This makes
it easier to see where windows are placed when the window itself
doesn't draw anything, e.g. in unit tests.
Change-Id: If9a59822e0b320b154ad8e338f9fb5ec7cf191a2
Reviewed-by: Richard Moe Gustavsen <richard.gustavsen@theqtcompany.com>
The directory URL will now have the scheme "file". So we need to convert
it to a local file before we inspect the file name
Change-Id: Ib50f879501f560a4d0fec41dce6d7d9f78f06a3c
Reviewed-by: Tor Arne Vestbø <tor.arne.vestbo@theqtcompany.com>
We need to pass the asset url around as a file url in the
application, so that QUrl::fromLocalFile()/toLocalFile()
works.
Note that QUrl::fromLocalFile() will remove double slashes.
We therefore need to check for this, and restore missing
slashes, when loading files in the file engine.
Change-Id: I2de6b91d7a112354590cf2981f7b403eacf92a59
Reviewed-by: Tor Arne Vestbø <tor.arne.vestbo@theqtcompany.com>
For KMS we can always know the correct rate so report it from the backend's custom
screen implementation.
For the rest, query from the framebuffer. If the fb driver publishes the timings then
we can calculate the vertical refresh rate from them. If not, default to 60.
Task-number: QTBUG-44971
Change-Id: I854a34e7c0d652790cc2ac967715828ec76f5733
Reviewed-by: Andy Nichols <andy.nichols@theqtcompany.com>
This deduplicates qmake rules in platformsupport and platform plugins.
Change-Id: Ie9c7d933c4433b96bf502e9753a12faa238b4569
Reviewed-by: Lars Knoll <lars.knoll@digia.com>
Reviewed-by: Oswald Buddenhagen <oswald.buddenhagen@theqtcompany.com>
Further Split apart Windows CE/Windows code paths to provide 2
implementations of populateFontDatabase():
- The desktop version uses a callback which only
registers the font families using
QPlatformFontDatabase::registerFontFamily() to be
populated on demand later by populateFamily().
- The Windows CE version is unchanged.
Task-number: QTBUG-43774
Task-number: QTBUG-44647
Change-Id: Iad4ec849006a4daf86a5caea9afc89e83ccc7c63
Reviewed-by: Konstantin Ritt <ritt.ks@gmail.com>
This is needed by qtx11extras now that we can meet a nullptr QScreen.
Change-Id: I755523cbbbbe188f454bd518d2d86a981542458f
Reviewed-by: Richard J. Moore <rich@kde.org>
This was motivated by a real segfault:
#0 QScreen::handle (this=0x0)
at qtgui-5.5.9999/work/qtgui-5.5.9999/src/gui/kernel/qscreen.cpp:112
#1 0x00007faaf5d5e85e in QXcbNativeInterface::connectionForWindow (this=this@entry=0x7fab04d684f0, window=window@entry=0x0)
at qtgui-5.5.9999/src/plugins/platforms/xcb/qxcbnativeinterface.cpp:483
#2 0x00007faaf5d5fa53 in QXcbNativeInterface::nativeResourceForWindow (this=this@entry=0x7fab04d684f0, resourceString=..., window=window@entry=0x0)
at qtgui-5.5.9999/work/qtgui-5.5.9999/src/plugins/platforms/xcb/qxcbnativeinterface.cpp:304
#3 0x00007fab048b8b57 in QX11Info::connection ()
at /var/tmp/portage/dev-qt/qtx11extras-5.5.9999/work/qtx11extras-5.5.9999/src/x11extras/qx11info_x11.cpp:358
#4 0x00007fab02ce14c6 in NETEventFilter::nativeEventFilter (this=0x7fab04ea5e70, ev=0x7faaec003a60)
at kwindowsystem-5.7.0/work/kwindowsystem-5.7.0/src/kwindowsystem_x11.cpp:229
...at which point I tried to stop playing the
get-backtrace-patch-rebuild-repeat cycle and looked at the stuff which
looked fishy to my untrained eye. So this is speculative in nature, but
I think that each of these cases can be hit and dereference a nullptr.
Change-Id: I046debaa1b49fa55e876247fc62f3eb924496fe8
Reviewed-by: Shawn Rutledge <shawn.rutledge@digia.com>
Since we use CoreText API to populate the font DB, we
should also make sure we use consistent data when doing
queries.
This partially reverts and ameds b619c35d85.
Change-Id: I6a3470fbee719ae1ea3085c252a4870040b9af1a
Task-number: QTBUG-41487
Reviewed-by: Liang Qi <liang.qi@theqtcompany.com>
Reviewed-by: Tor Arne Vestbø <tor.arne.vestbo@theqtcompany.com>
When showing a QWindow, we transfer first responder status to its
QUIView. If another QWindow was active with a text responder at
that point, the text responder will loose first responder status
in favor of the new view, and the keyboard will hide.
Now, if the new window has a focus object with the same IM state
as the previous focus object (in the previous window), m_imeState
will not change, and QIOSIntegration::update() will assume that
nothing needs to be done to show the keyboard, even if it's
actually hidden.
This patch will change the logic, so that we:
- show the keyboard if its supposed to be visible, even if
m_imeState did not change.
- Only recreate the text responder if it needs a different
configuration than the one we already got (not only
from changes to Qt::ImEnabled)
Task-number: QTBUG-40695
Change-Id: I6f6788af4cbff5c7abe4f5a29e23a7cefea6b711
Reviewed-by: Tor Arne Vestbø <tor.arne.vestbo@theqtcompany.com>
QGuiApplication sets options passed to the plugin through -platform
as properties on the QPlatformNativeInterface. To handle those we need
to inherit QPlatformNativeInterface first (which is the QObject
subclass), and include the Q_OBJECT macro.
Change-Id: Ia496851c64cbb0036c26e7ed0683d0ecfa8319cc
Reviewed-by: Richard Moe Gustavsen <richard.gustavsen@theqtcompany.com>
We need to track the active popup globally, having
parent windows track child popups is not sufficient
since there may be one or more intermediate windows
in between the event receiving window and popup window.
Add API to QOCocoaIntegration for tracking the global
popup window and use it instead of the per-window
tracking. Make sure to only close popups on clicks
outside the popup only. Add code to QNSView::handleMouseEvent
that redirects mouse events to the active popup.
Add manual test.
Change-Id: Ia3e3fd42d8fddf5c69f0c6879b333ca544521f61
Task-number: QTBUG-43464
Task-number: QTBUG-31937
Reviewed-by: Gabriel de Dietrich <gabriel.dedietrich@theqtcompany.com>
This was easy to hit with KDE Plasma 5.2.1; KWindowSystem called
QX11Info::appRootWindow() which in turn tried to dereference a nullptr
returned from QXcbConnection::primaryScreen().
#0 QXcbConnection::rootWindow (this=<optimized out>)
at qtgui-5.5.9999/src/plugins/platforms/xcb/qxcbconnection.cpp:1303
#1 0x00007fc26da096d7 in QXcbNativeInterface::rootWindow (this=this@entry=0x7fc27d1734d0)
at qtgui-5.5.9999/work/qtgui-5.5.9999/src/plugins/platforms/xcb/qxcbnativeinterface.cpp:425
#2 0x00007fc26da0ab21 in QXcbNativeInterface::nativeResourceForIntegration (this=0x7fc27d1734d0, resourceString=...)
at qtgui-5.5.9999/work/qtgui-5.5.9999/src/plugins/platforms/xcb/qxcbnativeinterface.cpp:223
#3 0x00007fc27c563148 in QX11Info::appRootWindow (screen=screen@entry=-1)
at qtx11extras-5.5.9999/work/qtx11extras-5.5.9999/src/x11extras/qx11info_x11.cpp:158
#4 0x00007fc27a98c444 in NETEventFilter::nativeEventFilter (this=0x7fc27d425b60, ev=0x7fc264004ad0)
at kwindowsystem-5.7.0/work/kwindowsystem-5.7.0/src/kwindowsystem_x11.cpp:192
#5 0x00007fc2795d0a8a in QAbstractEventDispatcher::filterNativeEvent (this=<optimized out>,
eventType=..., message=message@entry=0x7fc264004ad0, result=result@entry=0x7ffc96ecf348)
at qtcore-5.5.9999/work/qtcore-5.5.9999/src/corelib/kernel/qabstracteventdispatcher.cpp:460
#6 0x00007fc26d9ea941 in QXcbConnection::handleXcbEvent (this=this@entry=0x7fc27d173580, event=event@entry=0x7fc264004ad0)
at qtgui-5.5.9999/src/plugins/platforms/xcb/qxcbconnection.cpp:971
Change-Id: I98a5d767cd7e143f00666f6fc78e9dc10893513d
Reviewed-by: Shawn Rutledge <shawn.rutledge@digia.com>
Using a byte array here simplifies the logic while being more efficient.
It also removes warnings about using deprecated QString methods.
Change-Id: Ia1384dfb92e953c56179647b92a41b59cd26d9ea
Reviewed-by: Louai Al-Khanji <louai.al-khanji@theqtcompany.com>
Reviewed-by: Laszlo Agocs <laszlo.agocs@theqtcompany.com>
Use a built-in JSON file in case the QT_OPENGL_BUGLIST environment
variable is not set. When QT_OPENGL_BUGLIST is set, the built-in list
is ignored.
To make the implementation simpler and more readable, some of the code
in QWindowsOpenGLTester is reshuffled a bit. It also caches the results
now, so it is safe and fast to call supportedRenderers() and friends
multiple times.
The blacklist currently contains the Intel card from QTBUG-43263 (Intel
GMA / HD3000 ?) and may also apply to QTBUG-42240.
[ChangeLog][QtGui] Qt now contains a built-in GPU driver blacklist for
Windows that disables the usage of desktop OpenGL with some older cards
that are known to be unstable with opengl32.dll.
Task-number: QTBUG-42240
Task-number: QTBUG-43263
Change-Id: I1ecd65b51fca77925317d52048e7ab01d9b8797c
Reviewed-by: Friedemann Kleint <Friedemann.Kleint@theqtcompany.com>
When climbing the menu hierarchy, it's sounder to
check whether the actual QPA menu is enabled. This
way we can trigger modifier-less shortcuts even in
submenus.
Task-number: QTBUG-38256
Task-number: QTBUG-42584
Change-Id: I13a27027306bce0f0732b05bf9469f3b77028f73
Reviewed-by: Liang Qi <liang.qi@theqtcompany.com>
Apparently failures can occur not just when doing eglMakeCurrent() but
also when creating window surfaces.
Change-Id: Ife1210293d5120fd41352164d9c89e83fb5ce468
Reviewed-by: Michael Brüning <michael.bruning@theqtcompany.com>
Reviewed-by: Friedemann Kleint <Friedemann.Kleint@theqtcompany.com>
In QXcbWindow::setParent(), the window may not have a screen,
and in that case we cannot get the root window in this way.
Task-number: QTBUG-44719
Change-Id: I719e5e2f8cad13b1460b4d9df6ffd6c4a48e0d37
Reviewed-by: Laszlo Agocs <laszlo.agocs@theqtcompany.com>
Reviewed-by: Friedemann Kleint <Friedemann.Kleint@theqtcompany.com>
It is possible for there to be a HCBT_CREATEWND hook which can set the
touch window flags already while the window is being created. Therefore
we want to defer to those settings instead as they should take precedence.
Change-Id: If8dcbd34db2b3bbbfb1bc36731665fb17fb87c24
Reviewed-by: Friedemann Kleint <Friedemann.Kleint@theqtcompany.com>
Querying X server for data can be very expensive, especially when there
are multiple processes querying it at the same time (which is exactly what
happens when screen configuration changes and all Qt applications receive
XRandR change notifications). This patch is aiming to reduce the number of
queries to X server as much as possible by making use of detailed information
available in the RRCrtcChangeNotify and RROutputChangeNotify events.
Firstly, the backend now does not rebuild all QXcbScreens on any change (which
involved the very expensive xcb_randr_get_screen_resources() call), but only
builds the full set of QXcbScreens once in initializeScreens(), and then just
incrementally updates it.
Secondly, it avoids querying X server for all screens geometry as much as
possible, and only does so when CRTC/Output change notification for a particular
screen is delivered.
As a result, handling of all XRandR events on screen change is reduced from tens
of seconds to less then a seconds and applications are better responsive after
that, because we don't block the event loop for long. The X server is also more
responsive after the screen change, since we are not overloading it with requests.
Change-Id: I9b8308341cada71dfc9590030909b1e68a335a1f
Reviewed-by: Shawn Rutledge <shawn.rutledge@digia.com>
We used to support calling QEventLoop::exec() from within a processEvent
recursion and still jump down to the root native runloop, but this does
not work as intended due to how QEventDispatcherCoreFoundation uses
flags in its ProcessEventsState for e.g. deferred wake up or timer
updates. The logic in QEventDispatcherCoreFoundation assumes that
the next recursion to processEvents will be handled by itself, so
that it can interpret the flags in ProcessEventsState. The iOS
event dispatcher subclass, QIOSEventDispatcher, does neither of
these things, and should only be used from a 'clean' state.
Change-Id: I44fa156feecc45772806002465c35bef0797ead2
Reviewed-by: Richard Moe Gustavsen <richard.gustavsen@theqtcompany.com>
We release the UIWindow that retains QIOSViewController in the
QIOSScreen destructor, but other parts of the OS may have retained
the view controller, so the dealloc may not happen until later. In
the meantime we may receive calls to shouldAutorotate, so we need to
guard this code for the situation that m_screen has been deleted.
Change-Id: Iefeb75f4fc698b5e80417ffd3a971b7de625bcd5
Reviewed-by: Richard Moe Gustavsen <richard.gustavsen@theqtcompany.com>
The trick to get design metrics on Windows is to request the
font with the design size, however we were passing the em square
size in as the cell height instead of the em square size (aka
character height in Windows docs). This would give us hinted metrics
and thus the design metrics would differ from other platforms.
[ChangeLog][Windows][Text] Fixed design metrics for text
Task-number: QTBUG-44501
Change-Id: I4cffc3b86359cfdaf2ece07e1259f6fa862132bc
Reviewed-by: Konstantin Ritt <ritt.ks@gmail.com>
Reviewed-by: Lars Knoll <lars.knoll@digia.com>
Fixes "warning: ‘QString::QString(const char*)’ is deprecated"
While we're here, make the locals const.
Change-Id: Iee70253a46f91937b93e06cc08cd361716cd669d
Reviewed-by: Laszlo Agocs <laszlo.agocs@theqtcompany.com>
Whenever a QWindow is associated with a QScreen, the screen is expected
to be a real working one, so that rendering continues to be possible.
This partially reverts 52f5e50f11
[ChangeLog][QPA][Xcb] If all QScreens (xcb outputs) are disconnected
while an application is running, QGuiApplication::primaryScreen() will
return null until a screen is connected again.
Task-number: QTBUG-40174
Task-number: QTBUG-42985
Change-Id: Id1b29dd70eaf3f2e7fd477516ce7e2bf24e095f6
Reviewed-by: Laszlo Agocs <laszlo.agocs@theqtcompany.com>
Reviewed-by: Jørgen Lind <jorgen.lind@theqtcompany.com>
When destroying the QWindowsIntegration, the global instance is set to
null in the destructor. This is followed by a lot of additional steps
when destroying the members. Some of that cleanup calls to
staticOpenGLContext() which was not handling the case of the integration
global instance being null.
Change-Id: Ib74faf491d4c81635934547968ec91e9dceec6e7
Reviewed-by: Friedemann Kleint <Friedemann.Kleint@theqtcompany.com>
Gives a stack-trace at the point the error was caused, making it easier
to debug what's going on.
Change-Id: I0d65bb5061e9a97c142d41f6c99a1a1803cbe82d
Reviewed-by: Richard Moe Gustavsen <richard.gustavsen@theqtcompany.com>
Reviewed-by: Tor Arne Vestbø <tor.arne.vestbo@theqtcompany.com>
Start using signalfd where we can. Drop the crash (SIGSEGV, SIGBUS) handling completely.
The crash handling that was in place previously was not async-safe. It also prevented getting
a core dump. So just remove it. There is no safe solution for a single application process since
restoring the keyboard, video modes, etc. all need unsafe calls in the signal handler almost for sure.
We can however improve the handling of non-crash scenarios greatly:
Introduce support for SIGINT, allowing nicely and cleanly restoring the video
mode with the KMS backend when pressing Ctrl+C while QT_QPA_ENABLE_TERMINAL_KEYBOARD
is set.
Same goes for keyboard suspend (SIGTSTP, Ctrl+Z). When QT_QPA_ENABLE_TERMINAL_KEYBOARD is set,
platform plugins now have the possibility to act upon Ctrl+Z. As an example eglfs' KMS backend
is enhanced to handle this by restoring the video mode before suspending the process, and
reinitializing when brought into foreground again (SIGCONT).
SIGTERM is also handled. This is extremely handy when starting an application locally on the embedded
device and then kill-ing it via a remote ssh session. Keyboard and video mode is now cleanly restored.
Finally, when disabling the keyboard, try setting also KDSKBMUTE.
Change-Id: I2b3608dc23c798e2b39f74cb27f12dcb0e958435
Reviewed-by: Louai Al-Khanji <louai.al-khanji@theqtcompany.com>
eglfs does not depend on the device makespecs anymore when it comes to these device
integration backends (hooks). Instead, backends are autodetected by configure.
The name of the preferred plugin is still set in the device makespecs. This
is optional. When not set and there is more than one plugin present in the system,
the environment variable QT_QPA_EGLFS_INTEGRATION will have to be set at runtime.
In the absence of that, the order is undefined.
Change-Id: Ie1ced2c9aa1beff2adb13b4fdea7c499cb5a6aab
Reviewed-by: Oswald Buddenhagen <oswald.buddenhagen@theqtcompany.com>
Reviewed-by: Andy Nichols <andy.nichols@theqtcompany.com>
Change the shared pointer to store a QPlatformCursor instead of
QWindowsCursor, removing the dependency of qwindowsscreen.h on
qwindowscursor.h.
Change-Id: I8b4bbc9fd4d5046c30ac3784f14229a9cc6d8dc6
Reviewed-by: Joerg Bornemann <joerg.bornemann@theqtcompany.com>
The Activity onResume function is always called before the application
gets initialized and we let the applicationState set to the default
value which is AppicationInactive.
Change-Id: Ifc3c7e3dfc51f2b821f8ca87f8b711f485b6a6f8
Reviewed-by: Peter Rustler <peter.rustler@basyskom.com>
Reviewed-by: Alex Blasche <alexander.blasche@theqtcompany.com>
If a QWindow has never been shown, it has no platformWindow (it's null).
So it's a valid condition and we need to be sure that the pointer isn't
null before dereferencing it.
Task-number: QTBUG-44766
Change-Id: Ia0aac2f09e9245339951ffff13c7eb024d6a0773
Reviewed-by: Shawn Rutledge <shawn.rutledge@digia.com>
Register font's english name as alias in the callback of
EnumFontFamiliesEx() at the time
QWindowsFontDatabase::populateFontDatabase() is being called. This will
help us to resolve english font family name to its corresponding localized
alias once windows font database has been populated. It will also fix an
assertion in Chinese Windows.
Task-number: QTBUG-44647
Change-Id: I265d93c16a1677a7f31ff56d60c24f6e90666419
Reviewed-by: Konstantin Ritt <ritt.ks@gmail.com>
Reviewed-by: Friedemann Kleint <Friedemann.Kleint@theqtcompany.com>
Reviewed-by: Eskil Abrahamsen Blomfeldt <eskil.abrahamsen-blomfeldt@theqtcompany.com>
It produces too much log output, so now it must be enabled by both
defining Q_XCB_DEBUG and enabling the qt.qpa.input category.
Change-Id: Id2eb7545a7cd07dadd70da31a1f17c2999feec0d
Reviewed-by: Alexander Volkov <a.volkov@rusbitech.ru>
Reviewed-by: Laszlo Agocs <laszlo.agocs@theqtcompany.com>
Shawn Rutledge
Filipe Azevedo
Richard Moe Gustavsen
Tor Arne Vestbø
Laszlo Agocs
Konstantin Ritt
Friedemann Kleint
Jan Kundrát
Andrew Knight
Morten Johan Sørvig
Gabriel de Dietrich
Andy Shaw
Daniel Vrátil
Eskil Abrahamsen Blomfeldt
Thiago Macieira
Tor Arne Vestbø
BogDan Vatra
Oswald Buddenhagen
Liang Jian