This was a regression from Qt 4.
Before this patch, we supported filtering events only at QWindowSystemInterface
level, but to properly support filtering in QAbstractEventDispatcher::filterNativeEvent,
we have to filter the events earlier. Now it is possible to enable/disable this
feature for platforms that support native event filtering.
The mapping of which events are user input events were taken from
QWindowSystemInterfacePrivate::EventType.
Task-number: QTBUG-69687
Change-Id: I9a5fb9f999451c47abcdc83fdcc129b5eeb55447
Reviewed-by: Paul Wicking <paul.wicking@qt.io>
Reviewed-by: Allan Sandfeld Jensen <allan.jensen@qt.io>
This patch reworks 0b1ce5db9e.
Just by looking at the source code it was unclear why is this
signal-and-slot connection necessary. It doesn't do anything on
normal exit - at the time we call dtor of this class,
QCoreApplication::instance() already is nullptr, which means that
no further event processing happens and we never get this signal.
Without digging into git history it may appear that the goal was
to process the remaining events on application exit, which would
be a questionable code by itself.
Change-Id: I202d11584901e3b52254d8e3c347aece17844b72
Reviewed-by: Allan Sandfeld Jensen <allan.jensen@qt.io>
The pdf engine uses a resource file, but Q_INIT_RESOURCE() was lacking.
Task-number: QTBUG-71070
Change-Id: I685961b3f2eea0ffe6b5313c72d504a8ad9a98e5
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
Also, call it only if the state really changes. If we stay disconnected
the whole time, there is no point in trying to create the session over
and over.
Change-Id: Ic3a92dd0575bed1a23ae36a944cc51b9741fb64a
Fixes: QTBUG-49760
Reviewed-by: Jesus Fernandez <Jesus.Fernandez@qt.io>
Reviewed-by: Mårten Nordheim <marten.nordheim@qt.io>
Reviewed-by: Timur Pocheptsov <timur.pocheptsov@qt.io>
[ChangeLog][Third-Party Code] Sqlite was updated to version 3.25.2
Change-Id: I2d74ee924745a5e1edd6fe511777965313a4b77a
Reviewed-by: Oliver Wolff <oliver.wolff@qt.io>
Added eglext_angle.h and gl2ext_angle.h to the list of header files
copied to include/QtANGLE during install. These two header files were
introduced by the recent ANGLE update but were not added to the install.
This was causing build failures when including the ANGLE headers (e.g.
in qtmultimedia).
Task-number: QTBUG-71158
Change-Id: If2f1a9ecfcdf509cccf2b3671adf575cc39892d4
Reviewed-by: Andre de la Rocha <andre.rocha@qt.io>
Reviewed-by: Oliver Wolff <oliver.wolff@qt.io>
- The only place where window flag Qt::WindowDoesNotAcceptFocus changes is inside
QXcbWindow::setWindowFlags and from there we call updateDoesNotAcceptFocus(). The
current code was redundantly calling xcb_wm_hints_set_input in various places.
- Similar as above: call xcb_wm_hints_set_normal/iconic only where it can change.
This hint depends on window state, so update it only from setWindowState().
Removed unnecessary call to setTransparentForMouseEvents(), which is already called
few lines above from setWindowFlags().
Change-Id: I8da919b135a4dfda0c8c1dad51d85d3e706153ab
Reviewed-by: Allan Sandfeld Jensen <allan.jensen@qt.io>
QAbstractEventDispatcher::wakeUp is a thread-safe method, using
a queued connection to invoke it is wasteful. This type of connection
involves allocating temporary QMetaCallEvent on a heap and locking of
destination thread's post-event queue. In most use cases this is ok,
and really convenient when target method is not thread-safe. But in
this case the existing solution was suboptimal, especially because
the events we are reading can be high frequency.
The solution that we use here is lock-free. There can be only one
time when it might need to wait for the lock, which is upon exiting
the application. If we have entered the critical section in
QXcbEventReader::run(), then the registered post routine (qAddPostRoutine)
will block the QCoreApplication's dtor (this is where dispatcher is
set to 0) until we exit the critical section. We also know when not
to enter the critical section, in case dtor is already running.
With this approach we might need to compete for the lock at most
once, instead of whole application lifetime, which was the case
with the existing code.
Change-Id: If6737329c972347b0050d67658e28dbaa6f552e8
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
It is necessary for QTBUG-69687. The original code processes the xcb
event queue only when new events have arrived, but if we want to do an
event filtering that buffers some events and processes them later based
on set/unset flags (e.g. QEventLoop::ExcludeUserInputEvents), we need
to call processXcbEvents() on every event loop iteration, not only when
new events have arrived from X server.
The required functionality is implemented by having custom event dispatchers,
instead of using the generic ones from QtGenericUnixDispatcher::
createUnixEventDispatcher() / eventdispatcher_support-private. This also
enables for further customizations, as might be necessary by QTBUG-70095.
Task-number: QTBUG-69687
Change-Id: I1f8b2400d26cccf17279d57bb4b678e40c615f33
Reviewed-by: Gatis Paeglis <gatis.paeglis@qt.io>
We need to override this snippet for the documentation
we generate for Qt for Python, and it is easier to have
it on a separate file.
Task-number: PYSIDE-801
Task-number: PYSIDE-691
Change-Id: Ideb5b6af25024279f167137d3b65660bb9c96a7e
Reviewed-by: Topi Reiniö <topi.reinio@qt.io>
In some cases, a tool tip may be shown outside screen geometry, i.e. if:
- QToolTip::showText is invoked manually with a position outside.
- In tst_QToolTip::setPalette if there is no screen at (0, 0). This might
happen in a multi-monitor setups where one screen is taller than the other.
- On Wayland windows are (by design) not allowed to know their position on
the screen. This means that global positions can't be trusted.
This started crashing when QDesktopWidget::screenGeometry(pos) was replaced
with QGuiApplication::screenAt(pos)->geometry() because screenAt will return
null if no screen is found, while screenGeometry defaulted to the primary
screen.
This reverts to the old behavior of falling back to the primary screen.
This won't solve the issue completely for the Wayland case, but at least we
will stop crashing.
Change-Id: I42dd07cc21c2f9f0ea0d69f0c25bd46d8a2615a0
Reviewed-by: Filipe Azevedo <filipe.azevedo@kdab.com>
Reviewed-by: Christian Ehrlicher <ch.ehrlicher@gmx.de>
Reviewed-by: Richard Moe Gustavsen <richard.gustavsen@qt.io>
QImage's operator>>(QDataStream&) did not set an error mode on the
stream on read failures. That would break QDataStream transactions.
Since the current QImage serialization cannot differentiate between
truncated and corrupted data, we set the ReadPastEnd error as expected
by the transaction system.
Also specify the expected file format on decoding QImage from stream,
to avoid all the format handlers' canRead() being invoked. This is
necessary since some of them may call ungetChar(), which fails when
the stream is in a transaction.
Also add testing of this feature to the QDataStram transaction
autotest. That required a slight rewrite of the fake sequential
QIODevice subclass. The previous implementation had incorrect
behavior of peek(), which is required by QImage decoders.
Task-number: QTBUG-70875
Change-Id: If3f1ca7186ad1e6ca0e6e8ea81d2b2fbece6ea01
Reviewed-by: Alex Trotsenko <alex1973tr@gmail.com>
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
The dependencies.json file allows to tweak the list of imports the
module is depending on, so that types implicitly imported are not
listed twice.
Task-number: QTBUG-70264
Change-Id: I7a3800e5ea713a8aaae0cddbf4e1607f92c41497
Reviewed-by: Oswald Buddenhagen <oswald.buddenhagen@qt.io>
Reviewed-by: Shawn Rutledge <shawn.rutledge@qt.io>
The correct name is c++1z. Anyhow, this is easy enough to get wrong,
so make sure CONFIG += c++17 works as well.
Task-number: QTBUG-67527
Change-Id: Iea26b18824b38b1b5170f85987cf5c750b8e10ab
Reviewed-by: Oswald Buddenhagen <oswald.buddenhagen@qt.io>
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Reviewed-by: André Hartmann <aha_1980@gmx.de>
This fixes qopenglwindow shader in the triangle
Task-number: QTBUG-67338
Change-Id: I8552183bf9ca45e9b56760b340d014ddd34c21f4
Reviewed-by: Laszlo Agocs <laszlo.agocs@qt.io>
Our fallback logic for inexact matches was not very good at accepting
better suggestions.
Change-Id: I40fb78bf583171105725156148e4a2245fb81354
Reviewed-by: Gatis Paeglis <gatis.paeglis@qt.io>
Wrong option was used to set zlib compression level for png
setCompression with negative value uses default compression
setCompression with value between 0-100 converts to zlib compression level 0-9
setCompression with positive value overrides Quality option
Change-Id: Ic4b048a1e30d6940019c2a00a6c24d0c11e3f821
Reviewed-by: Alexandra Cherdantseva <neluhus.vagus@gmail.com>
Reviewed-by: Eirik Aavitsland <eirik.aavitsland@qt.io>
Handle both of them inside the QXcbConnection::compressEvent().
Change-Id: Ibe7184ba5c5b636013145e887c817dca701345ad
Reviewed-by: Shawn Rutledge <shawn.rutledge@qt.io>
For details how this works refer to the documentation in the patch.
The follow-up patches will switch to calling processXcbEvents() on every
event loop iteration. With the existing code that would mean frequent
locking of shared data (event queue). Acquiring a lock is fast, but
lock contention isn't. To avoid potential problems, reimplement xcb event
processing to be lock-free. Besides theoretical performance benefits,
this definitally improves code readability in qxcbconnection.cpp. Thanks
to Mikhail Svetkin for questioning the design of the existing code.
Done-with: Mikhail Svetkin <mikhail.svetkin@qt.io>
Change-Id: I935f2b6ca802580f5c80205aef7b2f9afc172d26
Reviewed-by: Mikhail Svetkin <mikhail.svetkin@qt.io>
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
This reverts commit 67cc8fea10.
I forgot about this patch and now it makes rebasing the local changes
too time-consuming. Besides, 67cc8fea10 broke a build for -no-xcb-xlib.
I will restore this patch, with adaptations to the new QXcb*Connection
hierarchy.
Task-number: QTBUG-68859
Change-Id: I938f32b5da22ce18f95d761f9b34e77fff923e24
Reviewed-by: Gatis Paeglis <gatis.paeglis@qt.io>
This is a better solution for fbb485d4f6.
The existing solution was working fine, but it was exposing logic that is
internal to QWindowSystemInterface and platform plugin interaction. Some
platform plugins do event filtering at native event level - those that
support QAbstractEventDispatcher::filterNativeEvent(). Other plugins rely
on QWindowSystemInterface to do the filtering. Dispatchers should not care
about this.
The new logic rely on the fact that QWindowSystemInterfacePrivate::handleWindowSystemEvent
calls QAbstractEventDispatcher::wakeUp(). The same way postEventSourcePrepare()
rely on QCoreApplication::postEvent() to call QAbstractEventDispatcher::wakeUp().
Event sources run in the order they are attached, postEventSourcePrepare runs
before userEventSourcePrepare(). We rely on that order to pass wakeUpCalled
value.
Change-Id: I0327f4f2398fd863fb2421b8033bb1df8d65f5af
Reviewed-by: Allan Sandfeld Jensen <allan.jensen@qt.io>
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
While this requires from us calling a deprecated method, a (non-deprecated)
method we were using gives a wrong color which is too bright/saturated.
Task-number: QTBUG-70676
Change-Id: Icebeb53e351caa646c533595ca1a886e5eb6b5b8
Reviewed-by: Tor Arne Vestbø <tor.arne.vestbo@qt.io>
that resulted in 'Cmd' reported as combo of Qt::Meta/Qt::Control and
Qt::KeypadModifier.
Task-number: QTBUG-71006
Change-Id: I3dddc56f4d404a1ceefb21d57ac120b6273456ec
Reviewed-by: Tor Arne Vestbø <tor.arne.vestbo@qt.io>
When there were additional spaces between the function definition and
the first parameter, the parser failed to parse it when it contained
another function (e.g. 'qlineargradient(... rgb() ...)').
The reason for this was that ::until() needs the function at index-1 so
it can correctly count the opening parenthesis.
Fixes: QTBUG-61795
Change-Id: I992f556e7f8cd45550f83bc90aa8de2b4e905574
Reviewed-by: Richard Moe Gustavsen <richard.gustavsen@qt.io>
it must end up in front of QMAKE_LIBS{,_PRIVATE}, but not of
LIBS{,_PRIVATE} (which are preceded by QMAKE_LIBDIR).
Task-number: QTBUG-61982
Started-by: Liang Qi <liang.qi@qt.io>
Change-Id: Id3de01ee0e9b66af02f79949aeb5a0eabd55363f
Reviewed-by: Liang Qi <liang.qi@qt.io>
Reviewed-by: Simon Hausmann <simon.hausmann@qt.io>
In NDK r18, libc++.so was renamed to libc++.so.XX where XX is the Android
API level.
[ChangeLog][Android] Fixed build issue with NDK r18+.
Task-number: QTBUG-70631
Change-Id: Id0d2955648197e3054e3c69263b5a90d57964f6c
Reviewed-by: Oswald Buddenhagen <oswald.buddenhagen@qt.io>
Reviewed-by: Liang Qi <liang.qi@qt.io>
the early merging of LIBS* into QMAKE_LIBS* meant that we could not
interleave them properly. defer the merging until the points of use.
Task-number: QTBUG-70779
Started-by: BogDan Vatra <bogdan@kdab.com>
Change-Id: I890f98016c3721396a1f0f6f149a9e2b37d56d8e
Reviewed-by: Liang Qi <liang.qi@qt.io>
it's unused now, and just complicates matters. its interaction with
LIBS_PRIVATE & co. has always been a bit shaky. google produces no
public hits outside qt itself, so let's assume it really remained
internal.
Change-Id: I6606bbabd44f1b76d84e97219e155e38d6f1b3a6
Reviewed-by: Liang Qi <liang.qi@qt.io>
Reviewed-by: BogDan Vatra <bogdan@kdab.com>
Reviewed-by: Simon Hausmann <simon.hausmann@qt.io>
this was introduced in 2002 supposedly for qnx4, but doesn't appear to
have actually been used ever. remove it, as it's in the way now.
Change-Id: I54dcabb61e1d3609a1e7a9fa4ff4b25509cfdb7a
Reviewed-by: Liang Qi <liang.qi@qt.io>
Reviewed-by: BogDan Vatra <bogdan@kdab.com>
Reviewed-by: Simon Hausmann <simon.hausmann@qt.io>
Those system calls are present in glibc 2.28. Instead of using
syscall(3) to place the system calls directly, let's use only the glibc
functions. That also means we no longer accept ENOSYS from either
function, if they were detected in glibc.
Change-Id: I44e7d800c68141bdaae0fffd1555b4b8fe63786b
Reviewed-by: Oswald Buddenhagen <oswald.buddenhagen@qt.io>
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
Reviewed-by: Jüri Valdmann <juri.valdmann@qt.io>
QWinRTBackingStore::flush() was not considering the possibility
that the supplied region may fall partially outside the bounds
of the paint device. This resulted in possible accesses to invalid
memory addresses, causing a crash. This bug was exposed by an update
in ANGLE that was causing a crash when running tst_QTableView::bigMode
with a small screen size. With this fix the function will use the
intersection of the supplied region with the paint device bounds.
Change-Id: I2f0f0f7f5510688bfa1459320a0c146df6be65d1
Reviewed-by: Miguel Costa <miguel.costa@qt.io>
lld for coff/mingw doesn't support linker scripts, which qmake used
for passing larger numbers of input file names to the linker.
Instead of using a fullblown linker script for this, just use a plain
response file, which both lld and binutils ld support.
Change-Id: I3aace7902fa6ca861a0a9fe67feaa236e7ea417b
Reviewed-by: Oswald Buddenhagen <oswald.buddenhagen@qt.io>
It is not always easy to spot the compiler version in the build or
configure log, so report it explicitly to make our lives easier when
trying to figure out why a specific build might have failed.
Change-Id: I1c84199aad4a98a30b0b4c4fbf2554008dc3ba2d
Reviewed-by: Oswald Buddenhagen <oswald.buddenhagen@qt.io>
This reverts commit 341bfcd1ea.
As it turns out there might be use cases where we want to have proper
windowing system event integration with glib dispatcher via g_source_attach().
For example with gtk_dialog_run, where GTK blocks in a recursive main loop.
We want to continue dispatcing our windowing system events during this nested
event loop. Not having a proper glib integration can result in rendering issues,
e.g. when resizing parent window via mouse while GTK-based dialog is shown. Can
be seen on examples/widgets/richtext/textedit/ -> Format (from menu) -> "Color..."
The issue from 341bfcd1ea actually should be fixed inside XCB platform plugin,
by improving integration with event dispatcher. That is handled in follow-up patches.
Change-Id: Icabc6d841a554aefbdd460765a3165d22e65f651
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Reviewed-by: Allan Sandfeld Jensen <allan.jensen@qt.io>
xcb_poll_for_queued_event() was introduced in libxcb 1.8.
The minimal required libxcb version was bumped up to 1.9 in
1f5d791708. Before this version
bump we needed the runtime check to support older versions
of libxcb.
Updated connections in the event reader to use the new signal
and slot syntax. Removed threadedEventHandling() method because
now it is always 'true'.
Change-Id: I0bce61fd478a871d35e676239ee5280c4f40be8a
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
The remaining diff to clean 1.6.35 is archived in the qtpatches.diff file.
[ChangeLog][Third-Party Code] libpng was updated to version 1.6.35
Change-Id: I925b3d95f6da20e12fc3274b4713e3fea05094f7
Reviewed-by: Kai Koehne <kai.koehne@qt.io>
Reviewed-by: Liang Qi <liang.qi@qt.io>
