The testAndSet operation is expensive if the lock is contended:
attempting to CAS that lock will cause the cacheline containing the lock
to be brought to the current CPU's most local cache in exclusive mode,
which in turn causes the CPU that has the lock to stall when it attempts
to release it. That's not desirable if we were just trying an untimed
tryLock*.
In the case of timed, contended tryLocks or unconditional locks, we
still need to perform an atomic operation to indicate we're about to
wait. For that case, this patch reduces the minimum number of atomic
operations from 2 to 1, which is a gain even in the case where no other
thread has changed the lock status at all. In case they have, either by
more threads attempting to lock or by the one that has the lock
unlocking it, this avoids the cacheline bouncing around between the
multiple CPUs between those two atomic operations. For QMutex, that
second atomic is a fetchAndStore, not testAndSet.
The above explanation is valid for architectures with Compare-And-Swap
instructions, such as x86 and ARMv8.1. For architectures using Load
Linked/Store Conditional instructions, the explanation doesn't apply but
the benefits still should because we avoid the expense of the LL.
See similar change to pthread_mutex_lock in
https://sourceware.org/git/?p=glibc.git;a=commit;h=d672a98a1af106bd68deb15576710cd61363f7a6
Change-Id: I3d728c4197df49169066fffd1756dcc26b2cf5f3
Reviewed-by: Marc Mutz <marc.mutz@qt.io>
Amend 3c6e9dcc623c9d7281a81174bb3a696e030f30a7 by making sure that
we explicitly move move-only functors into the slot object in the
respective tests, and that failing to do so doesn't compile.
Also add test coverage for mutable lambdas, which work as they do
with connected functors: the connection stores a copy, and calls
don't modify the original functor.
Change-Id: I637e6f407133e2f8f72109b3fe5369a11d19da93
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org>
This reverts commit 9958edba41, which
incorrectly tested a move-only functor without actually moving the
functor.
Change-Id: I3707f9f8e5055102f7edfb3e1cb9750978356dd7
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Use reinterpret_cast instead of C-style cast.
Take by QStringView directly so that we can use QSV::utf16().
Change-Id: I6d30669b0ed4bdf80a63043c364af695fd1d68b5
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
- Add missing includes
- Use std::all_of, that also fixes a narrowing conversion warning
(qsizetype).
Change-Id: I0f7f4b91bda4c187b8f8094e3039079c43fbf478
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
The old one kept all the indices in the type name, which was completely
unnecessary. This caused the symbol name to explode for some very large
meta object, notably that of the Qt namespace itself, causing gdb to
produce this warning at every start:
warning: internal error: string "StringData<3, 12, 7, 7, 6, 6, 9, 5, 10,
4, 6, 5, 5, 8, 7, 8, 10, 9, 9, 12, 11, 12, 12, 8, 6, 5, 13, 12, 9,
[many lines suppressed]
33, 6, 6, 5, 6, 17, 12, 17, 13, 8, 7>" failed to be canonicalized
To ensure there's no binary-compatibility problem with a StringData of
two entries, the first parameter is now of type int.
Pick-to: 6.5
Change-Id: I6f518d59e63249ddbf43fffd1759fbb66adc1299
Reviewed-by: Volker Hilsheimer <volker.hilsheimer@qt.io>
Reviewed-by: Lars Knoll <lars@knoll.priv.no>
When compiling with clang-cl 16.0.2, it gives me some warning like:
warning: use of bitwise '|' with boolean operands [-Wbitwise-instead-of-logical]
Change-Id: I2a1e80fa94b617c0f8fc6561c326c8ae98cada45
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Reviewed-by: Giuseppe D'Angelo <giuseppe.dangelo@kdab.com>
[ChangeLog][QtCore] Introduced Q_NODISCARD_CTOR which resolves to
[[nodiscard]] attribute for constructors on compilers that support
it, and does nothing on other compilers.
Using [[nodiscard]] attribute on a constructor is a C++20 feature,
however in practice it is supported on most of the compilers that
we use in Qt 6. Clang generates a [-Wunused-value] warning, GCC
and MinGW generate a [-Wunused-result] warnings, and MSVC
generates a C4834 warning.
However, there are some exceptions.
The Integrity compiler provides the following warning:
"tst_qglobal.cpp", line 699: warning #3435-D:
the "nodiscard" attribute doesn't apply to constructors,
destructors, or routines with void return type
[[nodiscard]] explicit Test(int val) : m_val(val) {}
The QNX compiler (QCC 8.3.0) and GCC 9.3.1 on OpenSUSE generate
the [-Wattributes] warning:
tst_qglobal.cpp: In member function
'void tst_QGlobal::nodiscardConstructor()':
tst_qglobal.cpp:699:44: warning: 'nodiscard' attribute applied to
'tst_QGlobal::nodiscardConstructor()::Test::Test(int)' with void
return type [-Wattributes]
[[nodiscard]] explicit Test(int val) : m_val(val) {}
These warnings will lead to build failures when compiled with
-warnings-are-errors flag, so for these compilers the macro
does not do anything.
An attempt to use __attribute__((__warn_unused_result__)) was
also unsuccessful on these compilers, so this patch goes for
an easy solution, and simply checks
__has_cpp_attribute(nodiscard) >= 201907L
to decide if the attribute is supported or not.
This commit also introduces a syntax-only test, and also applies
the new macro to QMutexLocker, because not all platforms in the
CI build and run unit tests.
Fixes: QTBUG-104161
Change-Id: Ib4230661a5ad5e8af0d67b21b034486ebcd67562
Reviewed-by: Marc Mutz <marc.mutz@qt.io>
Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org>
According to the upstream bug report [1], this bug has been fixed
already. So remove the now useless workaround.
[1] https://bugs.llvm.org/show_bug.cgi?id=41226
Change-Id: I7e17750f9e47fb9a0f8315ca15f89b741b107a35
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
QWeakPointer can do the same, so there's no reason to not allow it for
QPointer.
[ChangeLog][QtCore][QPointer] QPointer<T> can now be (move- and
copy-)constructed from QPointer<X>.
Fixes: QTBUG-112464
Change-Id: I77cf5d39974bf2b3ec849b4afc33e286e864821e
Reviewed-by: Giuseppe D'Angelo <giuseppe.dangelo@kdab.com>
No duplicate paths in any XDG_* env var.
Also borrow some text from the original commit 9deacd7f20 message.
Change-Id: I89f1fe3093dd7fea2d1bb555e36f299a73685f07
Reviewed-by: Edward Welbourne <edward.welbourne@qt.io>
Reviewed-by: David Faure <david.faure@kdab.com>
Now the tst_qstring is compiled three times:
- with QT_NO_CAST_FROM_ASCII defined
- with QT_RESTRICTED_CAST_FROM_ASCII defined
- with neither of the above defined
so as to cover more code paths.
Pick-to: 6.5
Task-number: QTBUG-109228
Change-Id: I65eca0f6f6aea66fed6eeda1eb77a50a97210807
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
These warnings slipped in during a time period where documentation
testing in the CI was disabled.
src/network/kernel/qhostinfo.cpp:254: (qdoc) warning: clang couldn't
find function when parsing \fn template<typename Functor>
int QHostInfo::lookupHost(const QString &name, Functor functor)
src/widgets/widgets/qcheckbox.cpp:102: (qdoc) warning: clang couldn't
find function when parsing
\fn void QCheckBox::stateChanged(Qt::CheckState state)
src/corelib/kernel/qcoreapplication.cpp:2769: (qdoc) warning: clang
couldn't find function when parsing
\fn template<typename Functor> void
QCoreApplication::requestPermission(
const QPermission &permission, Functor functor)
src/corelib/serialization/qxmlstream.cpp:3806: (qdoc) warning: clang
couldn't find function when parsing
\fn bool QXmlStreamAttributes::hasAttribute(
const QString &qualifiedName) const
src/corelib/text/qtliterals.qdoc:11: (qdoc) warning: Multiple topic
commands found in comment: \namespace and \headerfile.
Pick-to: 6.5
Change-Id: I38c605f358dbca1ef3e2bfe20a6424f7a4d44b4a
Reviewed-by: Volker Hilsheimer <volker.hilsheimer@qt.io>
During QCoreApplication initialization, we create the main thread's event
dispatcher, which for a GUI app happens via QGuiApplicationPrivate's
createEventDispatcher() override. This in turn relies on the platform
integration to create the event dispatcher, but to do that it needs to
create the platform integration first. And this might result in calling
APIs that themselves initialize the main thread event dispatcher, such
as QEventLoop, which non-lazily creates an event dispatcher for the thread.
We already had a check to catch the platform integration setting the
QCoreApplictionPrivate::eventDispatcher member, but not anything for
checking the current thread data.
On macOS this resulted in a leak of QEventDispatcherUNIX because
QCocoaDrag contained a QEventLoop member. We now track the event
loop temproarily via a pointer instead, like we do in other places.
An alternative fix would be to defer the initialization of QCocoaDrag
until QCocoaIntegration::drag() is called, but that would run the
risk of something calling the function during platform initialization
and we'd be back to the same problem.
It's unclear why QEventLoop is not lazily ensuring the event dispatcher,
and this might be a wider fix for similar issues.
Pick-to: 6.5
Change-Id: I643010ddb09945936ce9b0b94de0df96f6fe218f
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org>
This avoids recalculating the native key file name every time we call
handle() to ensure the handle exists. The POSIX backend already has the
code like this (whence the comment was copied) and the Win32 backend is
similar.
Change-Id: Idd5e1bb52be047d7b4fffffd175076ec09334ae1
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Reviewed-by: Fabian Kosmale <fabian.kosmale@qt.io>
Commit 72c2cdbc57 said "Move the helper
from qsharedmemory.cpp" but didn't actually move it probably because
commit 72c2cdbc57 got cherry-picked to 6.5 but the original
introduction to qsharedmemory.cpp didn't.
So do it now.
Change-Id: Idd5e1bb52be047d7b4fffffd175067701d4ea58c
Reviewed-by: Fabian Kosmale <fabian.kosmale@qt.io>
The result is retained so long as we don't attempt to change our locale,
but failing to change that is the reason why we printed anything.
==20227==ERROR: AddressSanitizer: heap-use-after-free on address 0x000107312696 at pc 0x000103c48088 bp 0x00016ee180c0 sp 0x00016ee17880
READ of size 9 at 0x000107312696 thread T0
#0 0x103c48084 in wrap_strlen+0x164 (libclang_rt.asan_osx_dynamic.dylib:arm64e+0x18084) (BuildId: f0a7ac5c49bc3abc851181b6f92b308a32000000200000000100000000000b00)
#1 0x1023804bc in QString::vasprintf(char const*, char*) qstring.cpp:7112
#2 0x102243578 in qt_message(QtMsgType, QMessageLogContext const&, char const*, char*) qlogging.cpp:368
#3 0x10252630c in QMessageLogger::warning(char const*, ...) const qlogging.cpp:647
#4 0x10229f940 in QCoreApplicationPrivate::initLocale() qcoreapplication.cpp:664
#5 0x10229fba0 in QCoreApplicationPrivate::init() qcoreapplication.cpp:826
#6 0x1022a07c0 in QCoreApplication::QCoreApplication(int&, char**, int) qcoreapplication.cpp:799
#7 0x101454ef8 in main+0xeb0 (WSgen:arm64+0x100470ef8) (BuildId: ae9b4fec1fd73c1693047a6b9d9ce91432000000200000000100000000000b00)
Pick-to: 6.5
Task-number: QTBUG-111443
Change-Id: I6f518d59e63249ddbf43fffd1759d28738124797
Reviewed-by: Tor Arne Vestbø <tor.arne.vestbo@qt.io>
The description was copied from QVector, and doesn't exactly fit
QList. For QList, the function is \since 6.0 when QList became
QVector, not in Qt 5.7, as indicated.
Be more precise.
Amends 13293d3308, which changed the
docs for clear().
Pick-to: 6.5 6.2
Fixes: QTBUG-112334
Change-Id: I457a1f699ddcdcdad2f1daf88f577007c136ee8f
Reviewed-by: Mårten Nordheim <marten.nordheim@qt.io>
Reviewed-by: Fabian Kosmale <fabian.kosmale@qt.io>
The other signal connections are broken because we create a new
delegate. But 'q' is still the same, so we should not connect to
it a second time.
Just connect it in the constructor instead.
Fixes: QDS-9687
Pick-to: 6.5 6.2 5.15
Change-Id: Id47492c2e734087281a3d3488649471814fd981d
Reviewed-by: Timur Pocheptsov <timur.pocheptsov@qt.io>
Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org>
Reviewed-by: Fabian Kosmale <fabian.kosmale@qt.io>
The existing overload unconditionally copies the QVariant, causing it
to unconditional detach in most cases (when calling data() later).
The rvalue overload need not preserve the source object, so doesn't
need to copy.
Implement the lvalue overload in terms of the rvalue one, but keep the
sanity check in both to avoid copying the QVariant just to find that
there's nothing to do.
Following the copy, there were some references to the source object,
but they can all be replaced by references to the target object
instead, which hasn't been touched by then, yet.
[ChangeLog][QtCore][QMetaProperty] Added write() overload taking an
rvalue QVariant.
Fixes: QTBUG-112762
Change-Id: I0a8a91aa32143f071ebc8dae8f1f64b07fad9764
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
This follows up on commit b906796af6.
Fix an off-by-one error - I was testing the last days of December 1799
and June 1800.
Change-Id: I79ab622978d35f91e3e1b1b8d00d93b0d4b31c07
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Use static constexpr int values instead of abusing enum.
[ChangeLog][QtCore][QTimeZone] The MinUtcOffsetSecs and
MaxUtcOffsetSecs constants are now static constexpr members of
QTimeZone, rather than members of an anonymous enum. Their values are
now 16 hours either side of zero, to allow for some historical zones.
Change-Id: I1c3a0f85a2b83b5010f021ca0f5ca5baefbf32e4
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Reviewed-by: Venugopal Shivashankar <Venugopal.Shivashankar@qt.io>
Reviewed-by: Paul Wicking <paul.wicking@qt.io>
On macOS, swiping with a single finger on the track pad (which Qt
identifies as a QInputDevice::DeviceType::TouchPad) results in a
TouchBegin event. For widgets that accept touch events (perhaps
implicitly because they want pan gestures, like QGraphicsView),
this results in a TouchBegin event to be delivered.
QApplication::notify will then check the widget's focus policy, and with
ClickFocus set, will set focus on the widget.
This is not what we want for a TouchBegin on a touch pad, so skip the
setting of the focus for that device type.
Pick-to: 6.5
Fixes: QTBUG-112922
Change-Id: Ie828793a784cc0e2fa47954bf5b396d6a44bd5e8
Reviewed-by: Tor Arne Vestbø <tor.arne.vestbo@qt.io>
In Qt >= 6.1, < 6.5, a trivially constructible type would have the
NeedsDestruction flag set, but it's dtor pointer would have been null.
In Qt 6.5, the meaning of the NeedsDestruction flag was changed to be
more aligned with what the name suggests, and thus would only be set for
non-trivially destructible types. For QMetaType this was fine, but
QVariant has a check for acceptable metatypes which attempts to verify
whether a QMetaType is usable for QVariant. The check assumes the
semantics of Qt 6.5, and thus fails for metatypes created by older Qt
versions.
To fix this issue, we increment the QMetaType revision field, and only
check the metatype's destruction support if the revision is high enough.
In theory, that allows passing unsuitable metatypes from older Qt
versions to QVariant; however, such code would have been broken in prior
Qt releases already (which didn't attempt the check), and no code that
used to work in any released Qt version will break (as we simply skip a
check that was passing before).
Fixes: QTBUG-113227
Pick-to: 6.5
Change-Id: I12e02bd97d2c410ea1a36efb0ce2389f21d50a30
Reviewed-by: Ulf Hermann <ulf.hermann@qt.io>
Three is not enough, need four. This is because in core profile contexts
querying GL_MAX_VARYING_COMPONENTS may apparently be an error (this is
enforced in macOS with 4.1 core contexts for example), so we need to
use something else: GL_MAX_VERTEX_OUTPUT_COMPONENTS.
Change-Id: I005c30e5798b0bf8bb7666bd732aaf996d123db9
Pick-to: 6.5 6.2
Fixes: QTBUG-112985
Reviewed-by: Mårten Nordheim <marten.nordheim@qt.io>
This closely matches QPlatformOpenGLContext which in turn, where it can
be used by backends to guard low level resources.
Change-Id: Ia44cebced6cdf94497279c47d3c35c0e02e4cb0e
Reviewed-by: Laszlo Agocs <laszlo.agocs@qt.io>
Amends 207aae5560 to make it easy to
create human-friendly error messages. Since the functor-accepting member
functions are not removed from the API, the first compile error will be
that there is no suitable overload of the makeSlotObject helper, which.
With the assert helper, the first error message is easier to understand.
Change-Id: I4878ec35a44ddfa5dc9d9e358d81c3fd40389c0c
Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org>
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Move-only functors must never be passed by value, so fix the
QFunctorSlotObject constructor accordingly.
This then requires adjustments to the various QMetaMethod::invokeMethod
overloads, as those must also perfectly forwad the functor type.
Enable the previously failing test case for move-only functors.
Change-Id: I9c544fd3ddbc5e1da3ca193236291a9f83d86211
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Amend 207aae5560, as code checker
complained that we std::move'd a potential lvalue. This warning was
valid if the public API did not accept the functor parameter by value.
Fix this by consistently std::forward'ing the parameters through the
call stack, and add a compile-time test. Writing that test revealed that
the helper API didn't work with free functions, so fix that as well. It
also revealed that QFunctorSlotObject couldn't work with a const
functor, which is also fixed by this change.
We cannot support move-only functors with that change, as it requires
a change to QFunctorSlotObject that breaks the QMetaObject test.
Change-Id: Iafd747baf4cb0213ecedb391ed46b4595388182b
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
We can't remove the export in Qt 6 because of how MSVC ABI works. In all
other compilers, this change is a no-op anyway (inline functions have
hidden visibility).
Change-Id: I3d728c4197df49169066fffd1756dc377b30c504
Reviewed-by: Marc Mutz <marc.mutz@qt.io>
Only instantiate with object/methods that are invokable with void and
returns void.
Change-Id: Iab2e43bb8e061e3875a6cca8e06ebbfbfa9e6fe8
Reviewed-by: Marc Mutz <marc.mutz@qt.io>
For now these will be used in QtQuick Flickable.
Task-number: QTBUG-35608
Task-number: QTBUG-35609
Task-number: QTBUG-97055
Pick-to: 6.5
Change-Id: I944d7f0271d535822ceeef610f232f56c85e0938
Reviewed-by: Axel Spoerl <axel.spoerl@qt.io>
...which comes at the expense of performance. Therefore this is
currently an environment variable-based opt-in solution.
Change-Id: If884de041f619fef0929ac2cfbdd035092723d51
Reviewed-by: Christian Strømme <christian.stromme@qt.io>
...and improve the comments since it is not directly obvious why
things are done the way they are.
Pick-to: 6.5
Change-Id: I6aa5f09b1b2f7f3fc18f74a4af2977d24278aae6
Reviewed-by: Christian Strømme <christian.stromme@qt.io>
Mention something some parts of Qt Quick rely on: that the update
batch objects stay valid across frames.
Change-Id: I2a21c19ceaa0c73a95062cc9b5cc3f4d1020e8a5
Reviewed-by: Christian Strømme <christian.stromme@qt.io>
Reviewed-by: Andy Nichols <andy.nichols@qt.io>
Exercise the validity of update batches, i.e. that one can
safely commit it in later frames as well, as long as all
related buffers and textures stay valid.
Change-Id: Ia943e4b37141fe17253eeae32010e0f8d92c1583
Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org>
Reviewed-by: Christian Strømme <christian.stromme@qt.io>
To fix the broken sliders reported in QTBUG-98093 a workaround was
added by 4bee9cdc0a where we would
call initWithFrame on an already initialized NSSlider.
This breaks the contract of object initialization in Objective-C,
as the class is free to allocate and prepare resources for the
instance without freeing previously acquired resources first.
As noted by the Object Initialization chapter of the Concepts
in Objective-C Programming guide, "Once an object is initialized,
you should not initialize it again.":
https://tinyurl.com/objc-object-init
And as observed in QTBUG-112899, the additional initialization
resulted in a memory leak.
The other part of 4bee9cdc0a
was that we now called startTrackingAt twice when drawing.
Both from setupSlider, for all consumers, and from
drawComplexControl, and as it turns out, this is the key
thing that "fixes" the pressed knob drawing of NSSlider.
For some reason, NSSlider needs the duplicate startTrackingAt
call both to draw the knob as pressed, and to not let one
drawing pass affect another drawing pass. This would benefit
from further investigation, but for now the removed leak
is an improvement.
Fixes: QTBUG-112899
Task-number: QTBUG-98093
Pick-to: 6.5
Change-Id: Ia7e6ef963910f1858d2fdb10e0323fc5bb3b2eda
Reviewed-by: Timur Pocheptsov <timur.pocheptsov@qt.io>
This avoids constructing an object just to copy (later: move) it into a
QVariant.
ChangeLog will be in a follow-up change adding emplace support.
Task-number: QTBUG-112187
Change-Id: I444e580c7d8927d41b3d21d5a521e7c475119e4c
Reviewed-by: Marc Mutz <marc.mutz@qt.io>
The macros no longer return QGenericArgument and QGenericReturnArgument
objects. As of 6.5, they return QMetaMethodArgument and
QMetaMethodReturnArgument, respectively.
Fixes: QTBUG-113147
Pick-to: 6.5
Change-Id: I06e0cf8255e6d4fee43048850f1717b1740d4846
Reviewed-by: Ulf Hermann <ulf.hermann@qt.io>
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Show more detail about the checks made against each entry to show why
a shortcut might not get delivered. Specifically, it's useful to know
if the contextMatcher returned true. In the case of Qt Quick Controls,
it can return false if the Shortcut is blocked by a modal popup, or
a popup with a CloseOnEscape policy.
With this patch, combining qt.gui.shortcutmap with
qt.quick.controls.shortcutcontext.matcher makes it possible to see
which popup blocks a shortcut. With only the former enabled, it's
already quite useful:
qt.gui.shortcutmap: Possible shortcut key sequences: QList(QKeySequence("Ctrl+N"))
qt.gui.shortcutmap: - checking entry 0 QKeySequence("Ctrl+N")
qt.gui.shortcutmap: - matches returned 2 for QKeySequence("Ctrl+N") QKeySequence("Ctrl+N") - correctContext()? false
qt.gui.shortcutmap: - matches returned 0 for QKeySequence("Ctrl+N") QKeySequence("Ctrl+O") - correctContext()? false
qt.gui.shortcutmap: Found better match ( QList(QKeySequence("Ctrl+N")) ), clearing key sequence list
qt.gui.shortcutmap: Added ok key sequence QList(QKeySequence("Ctrl+N"))
qt.gui.shortcutmap: Returning shortcut match == 0
qt.gui.shortcutmap: QShortcutMap::nextState(QKeyEvent(ShortcutOverride, Key_N, ControlModifier, text="N")) = 0
Change-Id: I6e96c94a8b62823553837eda3ef2764ca21775c4
Reviewed-by: Friedemann Kleint <Friedemann.Kleint@qt.io>
Reviewed-by: Volker Hilsheimer <volker.hilsheimer@qt.io>
On systems with weak memory ordering, it was possible for the
storeRelaxed(Initialized) to be observed by another thread performing a
loadRelaxed() without observing the contents of the object itself. The
mutex *does* release the contents of the object to memory, but without
the corresponding mutex acquisition, we couldn't guarantee that the
object's contents would be observed. Now we can.
We don't need to fix the load inside the mutex because the mutex will
have acquired everything from either a previous call to pointer() or to
reset(). The store inside reset() need not be storeRelease() either
because the effect of observing the Uninitialized state will be to lock
the mutex.
None of this is used to protect the data as it is being mutated by the
user in multiple threads, or their access simultaneously with reset()
(which is why the load outside the mutex was removed).
Thanks to litb on Slack for noticing this and bringing to my attention.
Pick-to: 6.5
Change-Id: Idd5e1bb52be047d7b4fffffd1752df5b4d9b2e3f
Reviewed-by: Marc Mutz <marc.mutz@qt.io>
Functions in Qt that take a callback need to support callables with or
without context objects, and member functions of an object. The
implementation of those overloads follows a pattern that ultimately
results in a QSlotObjectBase implementation being created and
passed to an implementation helper that takes care of the logic.
Factor that common pattern into a new helper template in QtPrivate
that returns a suitable QSlotObjectBase after checking that the
functor is compatible with the specified argument types.
Use that new helper template in the implementation of
QCoreApplication::requestPermission and QHostInfo::lookupHost.
The only disadvantage of centralizing this logic is that we cannot print
a more detailed error message indicating which argument types the
caller expects. However, that information is visible from the detailed
compiler errors anyway.
Change-Id: I24cf0b2442217857b96ffc4d2d6c997c4fae34e0
Reviewed-by: Mårten Nordheim <marten.nordheim@qt.io>
Add the move ctor and move operator for QSqlField
Task-number: QTBUG-109938
Change-Id: Ib66eff76c3a920de9cfb3288f4219555005e7ae5
Reviewed-by: Marc Mutz <marc.mutz@qt.io>
Use QSharedData for the private class instead a home-brew version.
Change-Id: Id3625bb0eb8f81c9caa672e2453dab3d44b15ea9
Reviewed-by: Marc Mutz <marc.mutz@qt.io>
When I broke them out from functions in the QLocalTime namespace I
didn't notice that's where I was putting them; they don't belong
there.
Change-Id: If4c9d996b3e46b3b46a29a97d0bcc2cac72c91ab
Reviewed-by: Mårten Nordheim <marten.nordheim@qt.io>