Just to make sure instanced drawing does not regress. Relevant
particularly with Metal.
Fixes: QTBUG-114885
Change-Id: Ib39066d32985bf25ca02d5aa54d9cf654772be9a
Reviewed-by: Andy Nichols <andy.nichols@qt.io>
Map QNetworkError::HostNotFoundError and QNetworkError::ConnectionRefusedError
to ProxyNotFoundError resp. ProxyConnectionRefusedError when it originated
from the communication with the proxy server.
Fixes: QTBUG-68821
Pick-to: 6.5 6.6
Change-Id: I21b91f2667ba0cd329d4ece1fe543472cdab2d22
Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org>
Reviewed-by: Mårten Nordheim <marten.nordheim@qt.io>
The test turns out to be flaky on QEMU.
Fixes: QTBUG-114760
Pick-to: 6.6 6.5
Change-Id: I264f79e9a056e82d4e6735e6ead0710ddabd8eba
Reviewed-by: Volker Hilsheimer <volker.hilsheimer@qt.io>
This minimizes any multi-threading / file-locking issues as the file is
closed once the contents are read.
This change assumes /proc/self/mountinfo is available on Linux systems,
and doesn't fallback to setmntent(). It's been around since at least
Linux Kernel 2.4.0.
This requires exporting qstrntoll() for the unittests (using
QT_AUTOTEST_EXPORT and wrapping the those unittests in "#ifdef
QT_BUILD_INTERNAL"), otherwise linking fails.
Fixes: QTBUG-77059
Change-Id: I0363258a9979ea6dadfe5e36c02534ffbd3386c5
Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org>
Reviewed-by: Axel Spoerl <axel.spoerl@qt.io>
Some shortcomings and unexpected problems are not unlikely.
The basic feature, with 2 views going to a texture array's
0 and 1 elements, seems to be working with macOS, also with
MSAA. Instanced drawing has not been verified. (relevant
because layered rendering works via instancing in Metal
and the QRhi backend has to adjust the instance count
in every draw call)
Fixes: QTBUG-114774
Change-Id: I3655e0d2c658b88c4cd6b52a32f94134324e4ac9
Reviewed-by: Andy Nichols <andy.nichols@qt.io>
We were basically timing the qSleep, which is pointless. We don't need
to verify that qSleep(X) spends at least X time. Because it also
doesn't. Somehow, QNX can execute 1000 ms sleeps in 996 ms.
Amends commit 30e5ff3ff2.
Pick-to: 6.6 6.5
Change-Id: I46b5dede27114be29724fffd176a66c1799075b7
Reviewed-by: Volker Hilsheimer <volker.hilsheimer@qt.io>
This should not be there because 'u' is passed to beginPass().
Change-Id: I95ba8ed400baa06948b4d4c6bbf7ca2d07a5480f
Reviewed-by: Laszlo Agocs <laszlo.agocs@qt.io>
Gerrit says it refuses to color lines that exceed 500 characters, so
let's split this line to get colors back.
Pick-to: 6.6 6.5
Change-Id: Iab6d37332e27ecdeaf1420da6fa242ca65f0aab5
Reviewed-by: Timur Pocheptsov <timur.pocheptsov@qt.io>
The test needs investigation. Skip it for now so that we can enable
CI for WASM but leave a note to investigate it.
Task-number: QTBUG-109954
Change-Id: I445996e969e7016a4d92a7e70da10b6d84a0fc71
Reviewed-by: Morten Johan Sørvig <morten.sorvig@qt.io>
Two QCOMPARE calls aren't worth the extra indirection.
Change-Id: If3b54d303c28582c7df29a3f1c4dc1a2c2a974f4
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
When a table view adds its first row, QHeaderView::initializeSections()
is called. It initializes the vertical header view with the number of
added sections. Subsequently QHeaderView::sectionsInserted() is called
with the same amount of newly added rows/sections.
That leads to the initial amount of sections being 2x the number of
rows added in the first go. In other words, the table view will display
at least one row more than the underlying table model has.
This patch adds an OR condition to the early return check at the
beginning of QHeaderView::sectionsInserted(). The method returns early
if the number of sections equals the number of respective sections
(rows in this case) in the model.
An autotest is added in tst_QTableView::rowsInVerticalHeader().
Fixes: QTBUG-114225
Pick-to: 6.6 6.5
Change-Id: I895444f025591981965562e54e2335391db52357
Reviewed-by: David Faure <david.faure@kdab.com>
Reviewed-by: Giuseppe D'Angelo <giuseppe.dangelo@kdab.com>
33c88f86b5 added some logic to QAPM in
order to have it automatically emit headerDataChanged when rows/columns
were added or removed in the model. This was done as a stopgap measure
to prevent QAPM from asking for illegal indices in order to implement
automatic remapping of the section headings (since there's no
mapSectionToSource).
The commit seems to have introduced a regression in QHeaderView, which
isn't prepared to receive headerDataChanged while a row/column count
change is in progress. When receiving headerDataChanged, QHeaderView
will try to read the row/column count and will store it internally.
When it will then receive the signals for insertion/removal of
rows/columns, it will interpret it as a modification of the previously
stored value -- even if the value it stored was already correct.
Fix this by avoiding to have two signals in flight at the same time;
emit headerDataChanged as a queued invocation.
Task-number: QTBUG-114225
Change-Id: I521465a852b8c7135f22f730ead41dca760ba428
Pick-to: 6.5 6.6
Reviewed-by: Axel Spoerl <axel.spoerl@qt.io>
Reviewed-by: David Faure <david.faure@kdab.com>
Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org>
INTEGRITY has a pre-P1115 implementation of std::erase/erase_if that
returns void instead of the number of erased elements, so make q20's
implementation more specialized, so the compiler will pick it over
INTEGRITY's (Marc's idea from the code review).
Change-Id: I88d025a3f90cdd65f5bb73beb8a39e32ccf12d9b
Reviewed-by: Marc Mutz <marc.mutz@qt.io>
Wrappers around P1301 [[nodiscard("reason")]].
[ChangeLog][QtCore][Q_NODISCARD_X/Q_NODISCARD_CTOR_X] Added as
wrappers around C++20 [[nodiscard("reason")]].
Task-number: QTBUG-114767
Change-Id: Ie566d9c9d500ef632c7e243af97081f83506a752
Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org>
Reviewed-by: Ivan Solovev <ivan.solovev@qt.io>
This relies on qsb being able to invoke dxc instead of fxc when the
request HLSL (shader model) version is 6.1. (6.1 is required for
SV_ViewID) This currently works only when conditioning offline with
qsb (or via CMake), because qsb can easily invoke dxc instead of
fxc. When shipping HLSL inside the .qsb package (so when -c is not
specified or running the external tool fails), this won't work since
the D3D12 backend still uses D3DCompile(), not IDxcCompiler. Support
for that will be investigated separately.
We also need to bump to ID3D12Device2 and ID3D12GraphicsCommandList1.
With Windows 10 version 1703 being quite old now, this should not be a
problem at run time.
There are however issues at build time, namely that MinGW and
MinGW/LLVM and similar seems to have ancient Windows SDK headers
according to the CI test runs. None of the MSVC configurations have
this in the CI, they have reasonable versions of d3d12.h and similar.
Therefore, one important aspect of this change is that the D3D12
backend of QRhi will only be available from now on when the SDK
headers are new enough (meaning ID3D12Device2 is declared, which is a
several years old type now). Otherwise, QRhi::create() will simply
fail when asking for D3D12 with a helpful warning message about the Qt
build being crippled.
Implementation-wise, there are surprises in store as well:
The way the PSO is created needs to be revamped to follow the
extensible approach that uses a pipeline state stream
description. Both the graphics and compute pipeline creation is
changed to use CreatePipelineState() and the associated
machinery. This is only really essential for graphics pipelines since
we know have to include data for view instancing (multiview). For
compute the result is the same as before.
Additionally, the view count must now be baked into the
QRhiGraphicsPipeline. This means that applications must call
setMultiViewCount() with the same value (typically 2) *both* on the
render target's color attachment and on the pipeline. Backends that do
not care about the pipeline's view count (GL, Vulkan) will of course
ignore it, but if it's not set correctly D3D12 will fail. The manual
test is updated accordingly.
Fixes: QTBUG-114772
Change-Id: I93db7313377e711c2faeb956815899b12132d23b
Reviewed-by: Andy Nichols <andy.nichols@qt.io>
Under some circumstances, MSVC seems to complain about SlotArgumentCount
being unused
qobject.h(210): warning C4189: 'SlotArgumentCount': local variable is
initialized but not referenced
note: see reference to function template instantiation
'QMetaObject::Connection QObject::connect<void(__cdecl QAction::* )(bool),
main::<lambda_1>>(const QAction *,Func1,
const QtPrivate::ContextTypeForFunctor<main::<lambda_1>,void>::ContextType *,
Func2 &&,Qt::ConnectionType)' being compiled
This is nonsense, as SlotArgumentCount is used in the next line, to
construct the list of signal arguments, but the workaround to declare
the variable as [[maybe_unused]] is trivial.
Add a connect statement to the test case that creates such a connection.
This does not produce any warning with or without the attribute (and if
it did, the build would fail for CI configuratinos setting -Werror).
Pick-to: 6.6
Fixes: QTBUG-114781
Change-Id: I4ee6f7d57c2836ef3dd9741d037d48181af2cdec
Reviewed-by: Marc Mutz <marc.mutz@qt.io>
And take the opportunity to clarify what the QtVforkSafe namespace is
doing. Amends commit e71c226d6f.
Pick-to: 6.6
Change-Id: I443cf0c8a76243eead33fffd1767f3fa390a7cdd
Reviewed-by: Volker Hilsheimer <volker.hilsheimer@qt.io>
Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org>
Reviewed-by: Oswald Buddenhagen <oswald.buddenhagen@gmx.de>
When setFuture() was handed an element of m_futures, it would hold the
reference to past the clear(), which invalidates said reference.
Fix by taking the future by value instead of by cref.
While append() is not affected, as QList::append() already guards
against aliasing, do the same change there, both for consistency as
well as to optimize the common case of passing rvalues. It also means
we can use the rvalue overload of QList::append(), skipping the alias
analysis in the lvalue QList::append().
[ChangeLog][QtConcurrent][QFutureSynchronizer] Fixed a crash in
setFuture() if the argument was already a member of
QFutureSynchronizer::futures().
Pick-to: 6.6 6.5 6.2
Change-Id: Ic0b212b9f265a746df9a6beb6272a5415d131442
Reviewed-by: Fabian Kosmale <fabian.kosmale@qt.io>
Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org>
QWidgetPrivate::focusObject() always returns nullptr. That has lead to
mismatches between QGuiApplication::focusObject() and
QApplication::focusWidget(), when a widget got focus by the window
system (e.g. mouse click).
This patch implements QWidgetPrivate::focusObject.
It returns the current widget, if it doesn't have a focus proxy.
If it has a focus proxy, it resolves the proxy chain and returns the
deepest focus proxy.
(Note: It does not return QWidget::focusWidget(), because the focus
widget might not yet have been set, when the method is called).
Fixes: QTBUG-92464
Fixes: QTBUG-108522
Pick-to: 6.6 6.5 6.2
Done-With: Liang Qi <liang.qi@qt.io>
Change-Id: Icf01e8ac4fc5f722fbf8e0ca5a562617ae9ae8f2
Reviewed-by: Volker Hilsheimer <volker.hilsheimer@qt.io>
Reviewed-by: Tor Arne Vestbø <tor.arne.vestbo@qt.io>
Reviewed-by: Liang Qi <liang.qi@qt.io>
Use the new argument to bundle the OpenSSL prebuilt libararies so that
ssl operations are actually tested at runtime.
Task-number: QTBUG-110025
Pick-to: 6.6 6.5
Change-Id: I0d73fed463a1724a9e0ee84ba603aa2ff1bc649b
Reviewed-by: Mårten Nordheim <marten.nordheim@qt.io>
This is in line with how QMetaType handles QMetaTypeInterface*. You can
retrieve a const pointer to it.
Pick-to: 6.6
Task-number: QTBUG-113690
Change-Id: Iaf3c10603dc6049a5553987c90006807867abc0d
Reviewed-by: Fabian Kosmale <fabian.kosmale@qt.io>
ACE-encode the domain names, replace dashes with underscores,
prepend underscore to domain name parts that start with a digit.
Add a regression test into tst_qdbusinterface.
Fixes: QTBUG-71674
Change-Id: I92e0c6889163c0eccc4c833f2058d759631f562c
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org>
Or old and new load hints in mergeLoadHints() instead of just storing
new ones. Andjust QLibraryPrivate::setLoadHints() to handle objects
with no file name differently and just set load hints directly.
Mention that load hints are merged once the file name is set
in the documentation for QLibrary::setLoadHints().
Add a regression test into tst_qfactoryloader.
Update and extend tst_QPluginLoader::loadHints() to take into account
load hints merging.
Fixes: QTBUG-114480
Change-Id: I3b9afaec7acde1f5ff992d913f8d7217392c7e00
Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org>
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
It's not a real leak in that the string data is being freed on program
exit (or, more recently, QPixmapCache::clear()), but it can lead to
lots of memory being bound for much longer than expected when users
put in new QString keys without attempting to retrive them again. It
can also lead to problems with QStringLiterals lingering around until
after their underlying data has been freed. A bug in the Fusion style,
generating new string keys for identical state, exposed this
misbehavior, and one way to fix the resulting issue for the user is to
make sure that QPixmapCache doesn't leak QString keys.
The Fusion style issue with generating non-repeating keys for use with
QPixmapCache should also be fixed, eventually, but this patch
relegates that to an optimization issue (the caching is effectively
non-existent), the resource exhaustion is gone now.
The issue exists because the QString keys are internally mapped to
QPixmapCache::Key's by way of a QHash<QString, Key> cacheKeys data
structure. When the QCache, indexed by Key, not QString, decides to
evict an entry, the Key is invalidated, but no-one was removing the
corresponding entry from cacheKeys. So make the existing releaseKey(),
used to invalidate copies of Keys referring to evicted pixmaps, do
that, now. So as not to have to scan the whole cacheKeys QHash for the
right Key, store the QString key, if any, inside the Key, so
releaseKey() can retrieve it and use it for O(1) erasure from
cacheKey.
This allows removing the previous work-around in clear()
(6ab0d25a09), greatly simplify
object(QString), and requires to rewrite all code that holds iterators
or references into cacheKeys over an insertion into or removal from
the QCache. Two (insert() and remove()) have already been done in
prequel commits, so only flushDetachedPixmaps() was left.
Fixes: QTBUG-112200
Pick-to: 6.6 6.5
Change-Id: Ic93b0ed388ae963267fe242b491c6c941d146b99
Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org>
Reviewed-by: Volker Hilsheimer <volker.hilsheimer@qt.io>
Amends 55fe46fd58. Before that change,
key release events were ignored as the QLineEdit::event reimplementation
continued to call QWidget::event, and as QLineEdit didn't override
keyReleaseEvent, the default implementation in QWidget got called to
ignore the event.
Restore that behavior by explicitly calling the QWidget implementation
after updating QLineEdit-specific states, and add a test case.
Fixes: QTBUG-114654
Pick-to: 6.5 6.6
Change-Id: Ic8aa35a1c915b446aece47aaf03ef5cf1884b936
Reviewed-by: Axel Spoerl <axel.spoerl@qt.io>
Previously, failure handlers did not propagate cancellation. This would
lead to crashes when a QPromise was cancelled without having generated
any result. Subsequent continuations would be invoked and try to access
the result (which was nonexistent) and then crash.
This patch propagates cancellation through failure handlers to prevent
subsequent continuations from being called in the first place.
Fixes: QTBUG-114606
Pick-to: 6.6
Pick-to: 6.5
Change-Id: I23b28a8e70a76e1ba6416be4440360c6dbaef2a3
Reviewed-by: Ivan Solovev <ivan.solovev@qt.io>
The export name is now ${TARGET_NAME}Entry. This can also be overridden
by using QT_WASM_EXPORT_NAME, both in CMake and qmake
Change-Id: I59c97ae6e22f0b2720716e9d7eff7b6b13d37ab5
Reviewed-by: Morten Johan Sørvig <morten.sorvig@qt.io>
QObjectPrivate::getPropertyAdaptorSlotObject called
connectionsForSignal.
Calling this function is only safe after it has been ensured beforehand that the vector has size > signalIndex. As getPropertyAdaptorSlotObject
is not supposed to modify the vector, it does not resize the vector and it could consequently end up with an out-of-bounds read.
To avoid that issue, we instead first check if the vector can
potentially contain an entry for the signal. If not, we simply return
nullptr, and avoid the call to connectionsForSignal.
The issue and its fix can be verified by running the modified
tst_qproperty test with ASAN enabled. The test is modified in the
following way:
- We first create a signal connection to a dummy slot. Otherwise,
connections.loadRelaxed() would return a nullptr, and the problematic
code would never be reached.
- We add enough signals to ensure that the fooChanged signal will
actually be out of reach (which means >= 8 signals, as the initial
capacity of the vector is 8)
Running the test without ASAN will most likely not result in a failure,
as then the out-of-bounds read will simply read garbage, and the most
likely result is that the cast below will fail.
Pick-to: 6.6 6.5
Change-Id: I18a3c4f52769c2b6491a685abb84f6fcfb44e4d8
Reviewed-by: Ivan Solovev <ivan.solovev@qt.io>
This property can take a either a WebAssembly.Module
or a promise to a module, and we don't have to specify
the exact type in the property name.
Pick-to: 6.6
Change-Id: Iebaf52178253afe8c93cf78bbe0853461bf48b67
Reviewed-by: Mikołaj Boc <Mikolaj.Boc@qt.io>
The test checks whether correct windows get correct keyboard input,
having been activated with mouse clicks.
Also fixes the event propagation error found. Key events would propagate
to parent windows, which is something that should not happen with
window managers.
Change-Id: I18bae8c14cce5ccae151c2a00c84ffdba3b4587f
Reviewed-by: Mikołaj Boc <Mikolaj.Boc@qt.io>
Reviewed-by: Piotr Wierciński <piotr.wiercinski@qt.io>
Fixes an issue with generated code where the name of an enclosing
namespace is identical to an enum class type, when Q_ENUM_NS is used.
Consider:
namespace a {
Q_NAMESPACE
namespace b {
enum class b { Key, Key2 };
Q_ENUM_NS(b);
}
}
moc generated code such as:
Q_CONSTINIT const QMetaObject a:🅱️:staticMetaObject = { {
...
qt_incomplete_metaTypeArray<qt_meta_stringdata_CLASSaSCOPEbENDCLASS_t,
// enum 'TestEnum'
QtPrivate::TypeAndForceComplete<b::b, std::true_type>,
// Q_OBJECT / Q_GADGET
QtPrivate::TypeAndForceComplete<void, std::true_type>
>,
nullptr
} };
which confused the compiler:
error: ‘b’ is not a member of ‘a:🅱️:b
83 | QtPrivate::TypeAndForceComplete<b::b, std::true_type>,
Fixes: QTBUG-112996
Pick-to: 6.6
Change-Id: I37aee83c32efe96cc9d6c2bd0bdb9ba80bb7b8a7
Reviewed-by: Fabian Kosmale <fabian.kosmale@qt.io>
We still need to consult fallbackThemeName() when computing the
parent list for an individual theme, as the Freedesktop Theme Icon
spec mandates that the "hicolor" theme comes last, but we no longer
need to do explicit fallback to fallbackThemeName() if a theme is
not found.
Pick-to: 6.5 6.6
Change-Id: I6c0b5a45d8258c5b6eaa761402944a735b1606ba
Reviewed-by: Axel Spoerl <axel.spoerl@qt.io>
The use of fallback icons should not depend on a theme being set.
[ChangeLog][QtGui][QIcon] QIcon::fallbackSearchPaths() will now be consulted
for fallback icons even if the current theme name is empty.
Pick-to: 6.5 6.6
Change-Id: Ia8d14062de7c53601fd9dac30f87a9e672aa2207
Reviewed-by: Axel Spoerl <axel.spoerl@qt.io>
So that building only tst_qprocess also builds the required test app.
Change-Id: I19a92f9dd2f4de08302d09cad0caf55c285a6dae
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
We would previously only use the fallback theme for themes that did
not exist, or for themes that did not declare any parent theme.
We now unconditionally use the fallback theme, even for themes that
declare their own parent themes, so that a QIcon::fromTheme("foo")
that doesn't exist in the current theme, nor any of its parents,
nor in "hicolor", will still be looked up in the fallback theme.
The reason this seemed to work in the existing tests was because
our test themes inherit system themes such as crystalsvg and gnome,
and we didn't provide a hicolor theme. Any of these themes missing
would lead us into the code path where we use the fallback theme
for a missing theme, masking that fact that we had not added the
fallback theme to the list of fallbacks for the theme that had
explicit parents declared.
The logic has been moved out of the theme parsing and into an
accessor in QIconTheme, so that we're not caching the fallback
theme lookup.
[ChangeLog][QtGui][QIcon] QIcon::fallbackThemeName() will now be
used as fallback even for themes that declare a parent theme.
Pick-to: 6.5 6.6
Change-Id: Ib0ce1dfe97030f23893460ed624073a719a3ebd1
Reviewed-by: Friedemann Kleint <Friedemann.Kleint@qt.io>
Reviewed-by: Axel Spoerl <axel.spoerl@qt.io>
Setting a fallback theme will affect the lookup strategy, so we need
to invalidate earlier lookups.
Pick-to: 6.5 6.6
Change-Id: I962245ddb3a20b7798d5ce831ed8a369b0ab76b4
Reviewed-by: Friedemann Kleint <Friedemann.Kleint@qt.io>
Reviewed-by: Axel Spoerl <axel.spoerl@qt.io>
