some assumptions were incorrect: our test server immediately sends
its SETTINGS frame, as a result we have to reply with client preface +
SETTINGS(ACK). So QVERIFY(!prefaceOK) was wrong from the beginning and
was only passing by pure luck.
Change-Id: Ie43f0d4ac41deb0e5339badaae6149a9b2f9d9b3
Reviewed-by: Timur Pocheptsov <timur.pocheptsov@qt.io>
Reviewed-by: Volker Hilsheimer <volker.hilsheimer@qt.io>
Also sanitize the initial WebAssembly hack. Both eglfs and wasm lack the concept
of true raster windows. A QWindow with RasterSurface is rendered with OpenGL
no matter what. The two platforms took two different approaches to work around
the rest of the machinery:
- wasm disabled the QOpenGLContext warning for non-OpenGL QWindows,
- eglfs forced the QWindow surfaceType to OpenGLSurface whenever it was
originally set to RasterSurface.
Now, the latter breaks since c4e9eabc30, leaving
all raster window applications failing on eglfs, because flush in the backingstore
is now checking the surface type and disallows OpenGLSurface windows. (just like
how QOpenGLContext disallows RasterSurface windows)
To solve all this correctly, introduce a new platform capability,
OpenGLOnRasterSurface, and remove the special handling in the platform plugins.
Change-Id: I7785dfb1c955577bbdccdc14ebaaac5babdec57c
Fixes: QTBUG-77100
Reviewed-by: Tor Arne Vestbø <tor.arne.vestbo@qt.io>
QWidget does not handle QWindow and QPlatformWindow being destroyed
behind its back, and the QWidget state for internalWinId and the
Qt::WA_WState_Created attribute can easily get out of sync with
reality.
To avoid QWidgetBackingStore mistakenly thinking that a widget does
not have a platform window it can operate on we use the QWindow and
QPlatformWindow handles directly, instead of relying on the winId.
This is a stop gap until we can teach QWidget to deal with dynamic
changes to its underlying window handles.
Change-Id: Ib09bea2ad62c42e9667a20ca6b5faf0f957288da
Fixes: QTBUG-74559
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
The array of QAtomicPointer<QMutex> can be initialized using relaxed
stores of nullptr, since nullptr is the whole data. But once we store
an actual QMutex pointer in the array, we need to publish the indirect
data thus created. We did this, with testAndSetRelease(); what was
missing was a corresponding acquire fence on load, without which there
is no happens-before relationship between the writes performed by the
QMutex ctor and the reads performed by a subsequent mutex.lock(), say,
on the same data.
Fix by adding acquire fences to all loads. That includes the dtor,
since mutexes may have been created in different threads, and never
been imported into this_thread before the dtor is running.
As a drive-by, return a new'ed QMutex that was successfully installed
directly to the caller, without again going through a load-acquire.
Fixes: QTBUG-59164
Change-Id: Ia25d205b1127c8c4de0979cef997d1a88123c5c3
Reviewed-by: David Faure <david.faure@kdab.com>
Reviewed-by: Giuseppe D'Angelo <giuseppe.dangelo@kdab.com>
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
(cherry picked from commit 65b8f59e04)
(cherry picked from commit da38f0d691d9d7eacfac5fbcbd47b887bd59bd39)
If this function is called by multiple threads, more than one could
reach the mutex locking and call TlsAlloc(), but only the last one would
save the data. The others would be leaked and, worse, be used by those
other threads.
[ChangeLog][QtCore][QObject] Fixed a resource leak caused by a race
condition if multiple QObjects were created at the same time, for the
first time in an application, from multiple threads (implies threads not
started with QThread).
Fixes: QTBUG-77238
Change-Id: Ife213d861bb14c1787e1fffd15b63a5818bcc807
Reviewed-by: Marc Mutz <marc.mutz@kdab.com>
Reviewed-by: Olivier Goffart (Woboq GmbH) <ogoffart@woboq.com>
Libtool cannot cope with absolute paths in the dependency_libs entry.
We split absolute paths into -L and -l here.
Change-Id: I30bf11e490d1993d2a4d88c114e07bbae12def6d
Fixes: QTBUG-76625
Reviewed-by: Kai Koehne <kai.koehne@qt.io>
Change-Id: Iaa6eb4d64f549a31aa5c53145e8b37facec4ea78
Reviewed-by: Andy Nichols <andy.nichols@qt.io>
Reviewed-by: Tor Arne Vestbø <tor.arne.vestbo@qt.io>
Added checking corner cases (more specific formulas) in
color_dodge_op()/color_dodge_op_rgb64() and color_burn_op()/color_burn_op_rgb64()
to produce correct results for any input.
Task-number: QTBUG-77231
Change-Id: I274f80b356bd4236a9176a84a95604c2eb01787a
Reviewed-by: Konstantin Tokarev <annulen@yandex.ru>
Reviewed-by: Allan Sandfeld Jensen <allan.jensen@qt.io>
One of the tests above was unsetting a variable that enforces
the use of a temporary keychain. We have to set it back, otherwise
the test is failing. What surprises me though - why I had this
problem only locally and not on CI? Apparently, SecureTransport
is not covered by our configurations ...
Change-Id: I0ff1e3e304632869391ed61213c245b949d8c778
Reviewed-by: Volker Hilsheimer <volker.hilsheimer@qt.io>
Allow AppKit to resolve screen for NSWindow lazily in the case where the
position is outside any known screen. And explicitly set the style mask
if detecting the corner case of positioning a window in the unavailable
space on a rotated screen.
In testing the effect of creating the window with a borderless style mask
and then updating the mask did not seem to have any visual consequences,
but we try to limit this mode just in case by only enabling it in the
corner cases we detect.
Change-Id: I4b7fcc6755a1ad5ff2683bec79d80a78226edae0
Reviewed-by: Timur Pocheptsov <timur.pocheptsov@qt.io>
The user may have assigned the application to start up on a specific
display, in which case the window's screen is nil after creation, and
the resulting screen will be delivered as a normal screen change once
the window is ordered on screen.
Fixes: QTBUG-77154
Change-Id: Idade6d833e31654db239243f2430166b5d86eca2
Reviewed-by: Volker Hilsheimer <volker.hilsheimer@qt.io>
std::not1 is deprecated in C++17, removed in C++20.
Use its replacement, std::not_fn.
Change-Id: I37d4929c81c2a5befeb44f954ae77b23960d2ff0
Reviewed-by: Ville Voutilainen <ville.voutilainen@qt.io>
For some reason, the overload resolution of the
High DPI scale() functions introduced by
b6ded193ee chose the
wrong overloads for QPointF and/or QPoint; it fell
back to the generic template intended for qreal,
QSize, etc, ignoring the origin. Remove the
template and spell out all overloads.
Fixes: QTBUG-77255
Change-Id: I5661f16f7326f65156f646f430f5a0c71d5302d2
Reviewed-by: Tor Arne Vestbø <tor.arne.vestbo@qt.io>
The m_windowState member is never updated regarding active state;
isActive is instead reimplemented to query the window manager, so use
that.
To extend the area where the user can move the window over the entire
titlebar of fixed-height windows, just test whether the mouse position
is within the titlebar.
Change-Id: I6b87aacd0bdab511cfd4959df1114af5c6013852
Fixes: QTBUG-77220
Reviewed-by: Friedemann Kleint <Friedemann.Kleint@qt.io>
Since Qt5.1 it is possible to deselect the current selected item in
SingleSelection mode when pressing the Ctrl key during the click but
this was not mentioned in the docs.
Change-Id: I86652308215bf218ea959f869334b6077e4634f9
Reviewed-by: Paul Wicking <paul.wicking@qt.io>
After commit 9c7ebd191b, qmake would
complain about missing cmake tests even for internal modules that have
no application side C++ linkage that needs testing.
Change-Id: I23b23c81dbe6be2b6da5672cbd7b8f8454ec2f66
Reviewed-by: Alexandru Croitor <alexandru.croitor@qt.io>
The newly introduced clear() method left the path in an undefined
state: d_ptr allocated, but no elements. The elements vector is
otherwise never empty, since ensureData() inserts a dummy initial
moveTo element.
Fix by making sure that clear() leaves the path in the same state as
ensureData() (i.e. "empty" but not "null"), except possibly more
capacity allocated in the elements vector.
Fixes: QTBUG-76534
Change-Id: I7ad8b312913f5eb6e22023f5d2fd873e54b1e23c
Reviewed-by: Allan Sandfeld Jensen <allan.jensen@qt.io>
While it is correct not to call the functor when the context object has
been destroyed, we still need ot clean up the slotObj. It's a low-
probability memory leak: the context object has to disappear while
waiting for a host resolution, and for repeated requests for the same
host the cache takes over anyway.
Task-number: QTBUG-76276
Change-Id: Id9daf391353b8252443f3186a7d504d70c553b24
Reviewed-by: Timur Pocheptsov <timur.pocheptsov@qt.io>
An inlined exported function does not make sense and will cause a
warning.
Fixes: QTBUG-77242
Change-Id: I016b93d6b39c4db82148fdc5a8a92bc9d5751885
Reviewed-by: Jörg Bornemann <joerg.bornemann@qt.io>
Without treating -force_load as an option with an argument, we end up
leaving stray -force_loads in the linker line, resulting in build
failures when the following option is a random library then treated
as a file path.
Task-number: QTBUG-66091
Change-Id: I352c50ab67e32ef6b2b5c6a4f90455b20034e207
Reviewed-by: Jörg Bornemann <joerg.bornemann@qt.io>
The linker doesn't support -framworkFoo, so neither should we. The correct
syntax is -framework Foo.
Change-Id: I3f39ffc067871ce058542bf0068274b35f7b51f6
Reviewed-by: Jörg Bornemann <joerg.bornemann@qt.io>
QApplicationPrivate::focus_widget became a dangling pointer
in the following scenario:
A widget first gets focus and later on gets a focus proxy.
QApplicationPrivate::focus_widget was still pointing to the initial widget.
Upon destruction, QWidget::hasFocus() [which follows to the focus proxy
and then compares with focus_widget] was therefore false for both
widgets. So QWidget::clearFocus() didn't call
QApplicationPrivate::setFocusWidget(0) for either of them. As a
result, focus_widget remained set, and became dangling.
In real life, this happened with a QWebEngineView, which the application
gave focus to upon creation. At that time it doesn't have a focus proxy
yet. That happens later, in QWebEngineViewPrivate::widgetChanged.
https://bugs.kde.org/show_bug.cgi?id=381793
Change-Id: Ifee610bb76a2d4d2797b98ece9bffe5fffe3c6a6
Reviewed-by: Volker Hilsheimer <volker.hilsheimer@qt.io>
The Surface Pen eraser button generates Meta+F18/19/20 keystrokes, but
when it is not touching the screen the Fn WM_KEYDOWN message is eaten
and only a Fn WM_KEYUP message with the previous state as "not pressed"
is generated, which would be ignored. With this patch we detect this
case and synthesize the expected key events to allow the button to be
used in applications.
Fixes: QTBUG-77153
Change-Id: I075f5cbb903cb36c9ec241ee1bb31d436b725e3a
Reviewed-by: Friedemann Kleint <Friedemann.Kleint@qt.io>
Reviewed-by: Oliver Wolff <oliver.wolff@qt.io>
The same logic is needed for QWinRTFileEngine. To be able to reuse the
code, it was moved out of the class.
Task-number: QTBUG-77095
Change-Id: If52b2fc8a0f3056d32fc693775565a1c3803b7d4
Reviewed-by: André de la Rocha <andre.rocha@qt.io>
Reviewed-by: Edward Welbourne <edward.welbourne@qt.io>
Fixes: QTBUG-77095
Change-Id: I45b38fab779518c49b22077c493d8640572d40d9
Reviewed-by: André de la Rocha <andre.rocha@qt.io>
Reviewed-by: Maurice Kalinowski <maurice.kalinowski@qt.io>
This change allows using custom flags for ar. For instance, it is
now possible to create a thin archive by setting QMAKE_LIB += -T.
This uses and extends commit d92c25b1b4
which served a similar purpose for the linker.
Change-Id: Ie1d6a0b957dc4809957726de00911c8d91647fab
Reviewed-by: Jörg Bornemann <joerg.bornemann@qt.io>
Passes locally, and there is no reason why it shoulnd't. Use IPv6
addresses for Google and Cloudflare DNS servers, and as with the IPv4
tests, rely on Python and (as a fallback) nslookup to produce the
reference.
Change-Id: I584f8ae9bc89c66a1f59d7b1e7493d0ed8033e8a
Fixes: QTBUG-22287
Reviewed-by: Timur Pocheptsov <timur.pocheptsov@qt.io>
Fixes: QTBUG-77132
Change-Id: Ic0410297a2215f1b7b656966cbe84b925706532f
Reviewed-by: André de la Rocha <andre.rocha@qt.io>
Reviewed-by: Maurice Kalinowski <maurice.kalinowski@qt.io>
Lookups performed via QHostInfoRunnable must not synchronously call
the user-code's receiver objects, as that would execute user-code in
the wrong thread. Instead, post a metacall event through the event
loop of the receiver object, or the thread that initiated the lookup.
This was done correctly for the trivial cases of empty host name or
cached results, so the code generally existed. By moving it from a
global function into a member function of QHostInfoResult, we can
simply access the required data to construct and post the event.
As we process that posted event, we need to check that the context
object (which is already guarded via QPointer) is still alive, if
we had one in the first place. If we had one, and it's deleted, then
abort.
[ChangeLog][QtNetwork][QHostInfo] Functors used in the lookupHost
overloads are now called correctly in the thread of the context object.
When used without context object, the thread that initiates the lookup
will run the functor, and is required to run an event loop.
Change-Id: I9b38d4f9a23cfc4d9e07bc72de2d2cefe5d0d033
Fixes: QTBUG-76276
Reviewed-by: Edward Welbourne <edward.welbourne@qt.io>
Reviewed-by: Mårten Nordheim <marten.nordheim@qt.io>
By default Qt tries to avoid potentially costly color matching by
not assigning an sRGB color space to our backingstore, even if
that's what we in practice fill it with.
We used to do this by assigning the display's color space, which
effectively opts out of color matching, similar to the old behavior
of the device RGB color space (which nowadays implies sRGB).
By picking up the color space from the NSWindow instead, we allow
the user to override the color space to trigger color matching,
for example by explicitly setting it to NSColorSpace.sRGBColorSpace.
NSWindow will fall back to the screen's color space if the window
doesn't have one set.
Task-number: QTBUG-47660
Change-Id: Iac8177e85e86fe9044a41eb2c93fbf26bb83c248
Reviewed-by: Allan Sandfeld Jensen <allan.jensen@qt.io>
Reviewed-by: Timur Pocheptsov <timur.pocheptsov@qt.io>
Reviewed-by: Simon Hausmann <simon.hausmann@qt.io>
As a follow-up to fixed resize handling: the trick Gabriel wanted
to use to enforce a specific look on a slider's bar, never actually
worked due to misplaced statement which essentially is cancelling
the 'magic' before the bar is drawn. Now it's fixed: bar is centered
(between the rows of tickmarks above and below) + it's had a nice
blue filling back!
Change-Id: I3021c2b86e4c25981eeee015e32baa24ccebc3bd
Reviewed-by: Volker Hilsheimer <volker.hilsheimer@qt.io>
