And inline the fast checks inside the methods in QBindingStorage.
This allows QObjectBindableProperty and friends to inline all the
fast checks and almost completely eliminates the overhead for property
accesses when no bindings are being used.
Read and write times of QObject based properties when no bindings
are being used:
Read Write
Old style property: 3.8ns 7.3ns
QObjectBindableProperty (no notification): 4.5ns 4.3ns
QObjectBindableProperty (with signal): 4.5ns 7.6ns
QObjectBindableProperty (inline accessors): 3.2ns 3.4ns
Numbers without this patch:
Old style property: 3.8ns 7.9ns
QObjectBindableProperty (no notification): 7.2ns 7.7ns
QObjectBindableProperty (with signal): 7.2ns 16.0ns
QObjectBindableProperty (inline accessors): 6.3ns 6.7ns
Change-Id: Ifd1fa3a489c3be8b1468c0b88af547aac397f412
Reviewed-by: Fabian Kosmale <fabian.kosmale@qt.io>
(cherry picked from commit 98c82fb445acf45cc4c4bc86a5adda43358127bf)
Reviewed-by: Ulf Hermann <ulf.hermann@qt.io>
- QList iterators are now nested classes inside QList.
- Drop reference to Qt OpenGL Widgets landing page, there is no such
page.
- Fix typos and linking issues.
Fixes: QTBUG-86295
Pick-to: 6.0
Change-Id: I964843deb81aa55ff8ddb9a1c2b004cb72e68de9
Reviewed-by: Topi Reiniö <topi.reinio@qt.io>
Reviewed-by: Paul Wicking <paul.wicking@qt.io>
QMetaObject::indexfOfMethod returns the method corresponding to a
specific signature. In QML, we however only want any of the methods with
a given name (and do overload resolution at a later point).
For this usecase this patch introduces the internal
QMetaObject::firstMethod function.
Change-Id: Ie3820354edffb273c4cbe1399201a955ebe79344
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
std::optional<int> is the wrong datatype to use for compare.
First and foremost, it can't be used in the idiomatic form of
auto r = a.compare(b);
if (r < 0) ~~~ // a is less than b
if (r > 0) ~~~ // a is greater than b
which we *already* feature in Qt (QString, QByteArray).
Also, std::optional<int> (explicitly) converts to bool, which is
a trap, because the result of the comparison can be accidentally
tested as a bool:
if (a.compare(b)) ~~~ // oops! does NOT mean a<b
Not to mention extending this to algorithms:
auto lessThan = [](QVariant a, QVariant b) { return a.compare(b); }; // oops!
std::ranges::sort(vectorOfVariants, lessThan);
which thankfully doesn't compile as is -- std::optional has
an *explicit* operator bool, and the Compare concept requires an
implicit conversion. However, the error the user is going to face
will be "cannot convert to bool because the operator is explicit",
which is deceiving because the fix is NOT supposed to be:
auto lessThan = [](QVariant a, QVariant b) { return (bool)a.compare(b); }; // big oops!
Instead: backport to Qt the required subset of C++20's <compare>
API, and use that. This commits just adds the necessary parts
for compare() (i.e. partial ordering), the rest of <compare>
(classes, functions, conversions) can be added to 6.1.
Change-Id: I2b5522da47854da39f79993e1207fad033786f00
Reviewed-by: Volker Hilsheimer <volker.hilsheimer@qt.io>
(cherry picked from commit 3e59c97c3453926fc66479d9ceca03901df55f90)
Reviewed-by: Qt Cherry-pick Bot <cherrypick_bot@qt-project.org>
Q_MOVABLE_TYPE was conceived before C++ had move semantics. Now, with
move semantics, its name is misleading. Q_RELOCATABLE_TYPE was
introduced as a synonym to Q_MOVABLE_TYPE. Usage of Q_MOVABLE_TYPE
is discouraged now. This patch replaces all usages of Q_MOVABLE_TYPE
by Q_RELOCATABLE_TYPE in QtBase. As the two are synonymous, this
patch should have no impact on users.
Pick-to: 6.0
Change-Id: Ie653984363198c1aeb1f70f8e0fa189aae38eb5c
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
Reviewed-by: Andrei Golubev <andrei.golubev@qt.io>
Except for types marked as unresolved, we're doing it anyway - the only
difference is that now we skip looking up the metatype by typeid.
[ChangeLog][QMetaProperty][Important Behavior Change]
QMetaProperty::typeName returns now always the same name as name() of the
corresponding metatype. This can cause a change for enum properties
which were not fully-qualified.
Change-Id: I1f57743948b7262ac06095d3bbc838d620f6e481
Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org>
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
Reviewed-by: Ulf Hermann <ulf.hermann@qt.io>
Properties of non-dynamic classes always have the metatype stored, so we
can make use of it. Moreover, when the builder is converted into a
metaobject, we need to resolve the metatype anyway.
As a driveby, add a dedicated metatype test to tst_qmetaobjectbuilder.
Change-Id: I7eea0cd8fc2da5d92c7fc803f05cd81e3a9d4cf4
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
It was redundant with the timer info dictionary, and was even used
where using the dictionary would have been more efficient.
Change-Id: Ia656bf9b56c61e23df9f8743d8f8efbf65d37574
Reviewed-by: Oswald Buddenhagen <oswald.buddenhagen@gmx.de>
Reviewed-by: Alex Trotsenko <alex1973tr@gmail.com>
C++ allows overrides to return a covariant type, so if calling code has
already cast the callee down to the right type, don't require it to cast
the returned clone down as well.
Pick-to: 6.0 6.0.0
Change-Id: I802f00a3c78e03047046986d0ed0a479b47573b8
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
This information is required in the QML engine to handle list properties
(instances of QQmlListproperty<T> and list<T> types from QML).
Pick-to: 6.0.0 6.0
Change-Id: I1e30572f1c91f58b290cb9b4b07433af99a1db6f
Reviewed-by: Ulf Hermann <ulf.hermann@qt.io>
As of 19f9b0d5f5 we define a protected copy
constructor and assignment operator. Static analyzers warn about missing
move variants, so delete those in addition.
Pick-to: 6.0 6.0.0
Change-Id: I28c1a2afd4560ce35a258bde507dafd93245b75b
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Just shows how bad an idea to call this function it is.
Pick-to: 5.15
Fixes: QTBUG-88653
Change-Id: If51855da004b4f3fbf43fffd1648fc294065160d
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
This was so far problematic as it gave various link errors. The solution
to that seems to be to make the default constructor of QPairVariantInterfaceImpl
constexpr to get around one set of problems.
The other problem to solve where undefined references to metaobjects. The
reason for that is apparently that QMetaTypeInterface contains a direct
pointer to the meta object, something the linker doesn't like. Adding a
level of indirection by using a function that returns the pointer seems
to solve that problem.
Fixes: QTBUG-88468
Change-Id: I5612ae807ea3b7e49bc40349d8d1fca1be9bd7ee
Reviewed-by: Fabian Kosmale <fabian.kosmale@qt.io>
Polymorphic classes should not be copied.
However, we do rely on event copying in our propagation logic. So, make the
members protected, don't delete them, using a dedicated macro.
This way, QMutable*Event classes can be used to make copies.
Remove some last usage of copying of QInputMethod(Query)Events.
Change-Id: Ia0a8ae4ca9de97dcd7788ca3c6ed930b6460c43a
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
We have use cases for cloning of events, e.g. in the state machine and
for event propagation.
Provide the means to do so through a virtual method.
Adapt QFutureCallOutEvent::clone, which is now an override. No code
seems to be using that method.
Change-Id: I6864d6597f6de800343c4dc458a7994e84dc6fb4
Reviewed-by: Shawn Rutledge <shawn.rutledge@qt.io>
It's not been used ever since it was added in the Qt 2 days, other than
a hack in QGraphicsView that could be solved more elegantly.
With this change, QEvent is only 16 bytes large, which is nicely aligned
with C++'s default operator new alignment, and we still have
plenty of bits (plus an unused bool) left for extensions.
Change-Id: If9376d90b92983db7fc12575f3893064a1797adf
Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org>
Reviewed-by: Shawn Rutledge <shawn.rutledge@qt.io>
Make sure all bits reported by sizeof for the most important public
event classes are used optimally.
QEvent is 24bytes large due to default alignment in C++, so we might
just as well use bool instead of bitfields for the most important
data members. This generates less (and thus smaller and faster) code,
and we still have plenty of bits available for future needs.
Default the copy constructor and assignment operator, the assert and
tracing seem to be relics from the Qt 3/4 days.
Note: QEvent's d-pointer is currently unused, with the exception of a
hack in QGraphicsView. Removing that would save another 8 bytes
through the entire event hierarchy.
For the new classes in the QInputEvent hierarchy, apply the same
principle. Allocate bits in QInputEvent and QSinglePointEvent to fill
the 8-byte aligned space. Using some of those bits for QMouseEvent
and QWheelEvent makes sure we don't increase the size for those in
spite of additionally reserved bits.
As a result of this, several QInputEvent and subclasses become 8 bytes
smaller on clang and gcc (with the exception of QNativeGestureEvent)
while at the same we have more space for future extensions.
The sizeof's for the various classes on different compilers produce
these before and after result:
clang +/- gcc +/- msvc +/-
QEvent 24 0 24 0 24 0
QInputEvent 56 -8 56 -8 48 0
QPointerEvent 80 -8 80 -8 72 0
QSinglePointEvent 96 -8 96 -8 88 0
QMouseEvent 96 -8 96 -8 88 0
QTabletEvent 112 -8 112 -8 96 -16
QKeyEvent 104 -8 104 -8 96 0
QNativeGestureEvent 120 0 120 0 120 0
QWheelEvent 112 -8 112 -8 112 -8
So, with this change we save 8 bytes on gcc and clang for many
event types, esp on Linux systems.
As a drive-by: replace ulong with quint64, make QTabletEvent data
floating point, and rename some variables for clarity.
Change-Id: I4cf81c8283262cbf59ee3fb7064a59837332ced7
Reviewed-by: Shawn Rutledge <shawn.rutledge@qt.io>
Not used anywhere in Qt 6, and if whatever needs this ever comes back,
then we won't implement it like this.
Change-Id: I99655ba5825644cef3686bbe50ca688b68f5c2a4
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
QGraphicsSceneEvent carries a widget pointer, and we can add a new
event
type. Assert that QGraphicsScene doesn't get Leave events from
elsewhere.
This was the only "usage" of QEvent's d-pointer in Qt, so with this gone
we can reduce the size of QEvent by 8 bytes.
Change-Id: I0513daf742de9084d0469d55b7a7f41b667e8081
Reviewed-by: Shawn Rutledge <shawn.rutledge@qt.io>
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
Instead of multiplexing all notifications into a single Qt event for
the event dispatcher, we can send 'WinEventAct' event directly for each
notifier which activated. This trick improves the performance (esp.
on a large number of events) and allows us to remove notifiers handling
from the event dispatcher completely.
As an alternative to sending Qt events, use of Windows' APC queue in
conjunction with waking up the Qt event loop from within the Windows
thread pool has been considered. However, that would lead to signal
emission asynchronous to the Qt event loop's operation, which is not
acceptable.
Thanks to Oswald Buddenhagen for the proposed idea.
[ChangeLog][QtCore][QAbstractEventDispatcher] The
{un}registerEventNotifier() member functions have been removed.
QWinEventNotifier is no longer needed to be registered in the
event dispatcher.
Change-Id: I140892fb909eaae0eabf2e07ebabcab78c43841c
Reviewed-by: Oswald Buddenhagen <oswald.buddenhagen@gmx.de>
We used quint32 for 32-bit types and ushort for 16-bit ones,
but using explicit bit sizes looks more consistent.
Change-Id: I3106dd6ecb2367fef6f8012c28266e1b4b1abf4b
Reviewed-by: Volker Hilsheimer <volker.hilsheimer@qt.io>
The documentation of processEvents states that it processes all
pending events. However, this is not true for the glib backend
and probably was not true for at least a decade.
QEventDispatcherGlib::processEvents(ProcessEventsFlags)
calls "g_main_context_iteration" which is documented
to only process events of highest priority:
https://developer.gnome.org/glib/2.66/glib-The-Main-Event-Loop.html#g-main-context-iteration
Fixes: QTBUG-87307
Task-number: QTBUG-84291
Pick-to: 5.15
Change-Id: I82c6f9dba865bb07f981ad648a128316749682ca
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
The metaType is only used in the constexpr if branch where the callable
is not std::is_invocable_v. Therefore, we need to mark it as unused in
order to avoid compile errors in code that exercises the other branch.
Change-Id: I46e855b0f4b0a088f15ff41d4929fe010531b97e
Reviewed-by: Fabian Kosmale <fabian.kosmale@qt.io>
This makes high-level event dispatching easier: for example we often
need to cast an event to access getters like button() and buttons().
We can so far assume that any QPointerEvent that is not a QTouchEvent
is a QSinglePointEvent; but more explicit type-checking looks safer.
Implemented in a similar way as c7f7279969.
Change-Id: I980d759e2a7538b6b30fd3bdc3be0c351ec6c246
Reviewed-by: Volker Hilsheimer <volker.hilsheimer@qt.io>
Remove forward declarations, we have those already from qcontainerfwd.h.
Change-Id: I1b8a91ece912e6cdf747f4e2a750f0e85bcb5b2a
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Reviewed-by: Volker Hilsheimer <volker.hilsheimer@qt.io>
Reviewed-by: Ulf Hermann <ulf.hermann@qt.io>
Add missing declaration of QVariantList and friends. Replace
class with typename for template parameters.
Remove some left-over forward declarations in other headers.
Change-Id: I31d443019d48b619e02834395dafa40182cac7b9
Reviewed-by: Volker Hilsheimer <volker.hilsheimer@qt.io>
Reviewed-by: Mårten Nordheim <marten.nordheim@qt.io>
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
This ensurse that we do not do dobule notifications in setValue.
Moerover we avoid needless notifications in markDirtyAndNotifyObservers
when the value did not change. Lastly, if the value did actually change,
we pass that information along to notify, so that we do not evaluate the
eager property twice.
Fixes a test-case which errorneously relied on the old behavior, and
adds a new test which verifies that the fix works.
Change-Id: I8ec6fa2fe8611565dfc603ceab3ba5f92999b26c
Reviewed-by: Ulf Hermann <ulf.hermann@qt.io>
std::function as a type is rather unfortunate for us, as its SSO buffer
makes it rather large, and we can ensure that the function is never
empty.
Considering that we do need to allocate memory for
QPropertyBindingPrivate anyway, we can get rid of the SSO buffer and
instead coalesce the allocations (similar to how std::make_shared works).
The memory looks then like
[--QPropertyBindingPrivate--][Functor]
and QPropertyBindingPrivate can get a pointer to the functor via
reinterpret_cast<std::byte>(this)+sizeof(QPropertyBindingPrivate).
To actually do anything with the functor, we do however need a "vtable"
which describes how we can call, destroy and move the functor. This is
done by creating a constexpr struct of function pointers, and storing a
pointer to it in QPropertyBindingPrivate.
As a consequence of those changes, we cannot use QESDP anymore, as we
now have to carefully deallocate the buffer we used for both the
QPropertyBindingPrivate and the functor. We introduce a custom
refcounting pointer for that. While we're at it, we make the refcount
non-atomic, as bindings do not work across threads to begin with.
Moreover, we can now make the class non-virtual, as that was only needed
to hack around limitations of QESDP in the context of exported symbols.
Change-Id: Idc5507e4c120e28df5bd5aea717fe69f15e540dc
Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org>
Reviewed-by: Ulf Hermann <ulf.hermann@qt.io>
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
We'll need QContainerTraits as a class for changing properties
of our containers, so free up that name. This is not a problem,
as the namespace is new in Qt 6 and has only been used internally
so far.
Change-Id: I6d6b9d9c32b92b77e66323f1fc29b3ddd8baa98f
Reviewed-by: Mårten Nordheim <marten.nordheim@qt.io>
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
Reviewed-by: Ulf Hermann <ulf.hermann@qt.io>
No action taken at Qt 6, suggesting it shall never happen.
Four removed, one converted to Qt 7, others converted to unversioned TODOs.
Filed Jira tasks, and referenced in comments, for those retained.
There remain two "once bootstrap builds are obsolete" comments and
one other on which pending action may yet happen.
Fixes: QTBUG-85700
Change-Id: Ib140a6a21c63370e51e4734cc591f67573a29d9a
Reviewed-by: Volker Hilsheimer <volker.hilsheimer@qt.io>
We can't assume that the number of arguments we get from GetCommandLine
will match what we have from argc, as the former is not wildcard
expanded on MingGW.
mingw64 will ask __getmainargs to expand wildcards depending on
the mingw-specific variable _dowildcard. By default this is set
to 0 (false), but some mingw64-packagers (msys2, mingw-builds)
pass --enable-wildcard during build, changing this default. As
a result, the arguments we get through argc/__argc have already
been expanded. MinGW does not reflect this through GetCommandLine
though, like MSVC does, which triggered the assert.
Amends dff18b8e80.
Pick-to: 5.15
Fixes: QTBUG-67515
Task-number: QTBUG-84002
Change-Id: Ib87961c6901d2894b48ba0f5b9a3d6da2a595f24
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
Reviewed-by: Simon Hausmann <hausmann@gmail.com>
Reviewed-by: Volker Hilsheimer <volker.hilsheimer@qt.io>
Reviewed-by: Oliver Wolff <oliver.wolff@qt.io>
clang warned about result being uninitialized if iface is nullptr. Return
a properly initialized result in that case as well.
This might break RVO, but the alternative is to always initialize result,
even if we are going to call the getter anyway, which I assume would be
more expensive.
Change-Id: I5d6d243b3094b79bf021725d017be5c72b1089bb
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
This reverts commit 1918c689d7.
The template gets always instantiated in QObjectPrivate::connect, even
if the connection types is not Qt::(Blocking)QueuedConnection. For
non-queued connections we however support using incomplete types in
connect. The only way to fix this would be to make the connection type a
template parameter of QObjectPrivate::connect (or at lesat pass some
compile time constant indicating "blocking"-ness) along, so that we can
use if constexpr instead of if. As all involved classes are private, we
can postpone investigating this solution to 6.1
Change-Id: Ieffaf015f8e60ca6ac6f85eb9e2756e480060b4f
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
Reduce ADL noise.
Clean up documentation and parameter naming a bit, while at it.
Change-Id: Ie9eb2a63b8e87a9ffc019b8fff5bd1c6bafaaf43
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
QMetaSequence refers to those members, and QMetaType operates on void*s,
and documents such low-level APIs as well, so can't think of a reason
why we should not document the QVariant getters as well.
Change-Id: Id18de839bb0e4f5559b28bfca80eaabab1eb6479
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
Declare hidden friends like qdoc expects them, and other signature fixes
Document function parameters
Remove documentation for removed APIs.
Change-Id: I44c1caeed0d40be04612129d074acc30b75f5259
Reviewed-by: Allan Sandfeld Jensen <allan.jensen@qt.io>
It's been obsolete for a long time already. Make sure
the compiler now warns about it and remove all remaining
uses in qtbase.
Change-Id: I0ff80311184dba52d2ba5f4e2fabe0d47fdc59d7
Reviewed-by: Fabian Kosmale <fabian.kosmale@qt.io>
Lars Knoll
Topi Reinio
Fabian Kosmale
Giuseppe D'Angelo
Andreas Buhr
Maks Naumov
Volker Hilsheimer
Thiago Macieira
Zhang Sheng
Shawn Rutledge
Alex Trotsenko
Mitch Curtis
Ulf Hermann
David Skoland
Edward Welbourne
Tor Arne Vestbø
Allan Sandfeld Jensen