By changing it to unique_ptr.
Pick-to: 6.2 6.3
Change-Id: I91abb69445b537d4c95983ae735341882352b29d
Reviewed-by: Marc Mutz <marc.mutz@qt.io>
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Even QtCore alone cannot be built without the properties feature since
Qt 5.5. While fixing this is easy, other modules like dbus,
networking are also using QObject::property() and friends liberally.
All in all I doubt that anybody will miss the feature (otherwise it
would have been fixed in the last decade).
Change-Id: Iaf3cc20bda54ee2ff3b809fac8fa82b94ecc88c0
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
Reviewed-by: Edward Welbourne <edward.welbourne@qt.io>
The new binding system made it impossible to read from non-owning threads,
but we have code that did so with external locking. This patch makes it
safe again, assuming all reads and writes are locked. This is left
intentionally undocumented.
Pick-to: 6.3 6.2
Task-number: QTBUG-99775
Change-Id: I845afa5d545ca0ac762ac369181b1497dac52195
Reviewed-by: Volker Hilsheimer <volker.hilsheimer@qt.io>
QWidget and QWindow use bits in QObjectPrivate to provide for a couple
of shortcuts -- one in qobject_cast, and another in the isWidgetType() /
isWindowType() functions in QObject. These can be optimized by simply
looking at the bits, without actually doing more expensive runtime
casts.
These bits were set on construction, but not unset on destruction. The
result was for instance that destroying a QWidget would report that the
object was still a QWidget when ~QObject was reached.
Fix this
1) by setting the bits only when QWidget / QWindow constructors start;
2) by resetting the bits once ~QWidget / ~QWindow are completed.
Technically speaking this is not 100% correct in the presence of data
members, but luckily those classes don't have any.
Amend an existing test for QWidget (whose comment said exactly the
opposite of what the test actually did) and add a test for QWindow.
Some other code was wrongly relying on isWidgetType() returning true
for destroyed QWidgets; amend it as needed.
[ChangeLog][QtCore][QObject] Using qobject_cast on partially constructed
or destroyed QWidget/QWindow instances now yields correct results.
Similarly, using the convenience isWidgetType() / isWindowType()
functions now correctly return false on such instances. Before,
qobject_cast (and the convenience functions) would erroneously report
that a given object was a QWidget (resp. QWindow) even during that
object's construction (before QObject's constructor had completed) or
destruction (after QWidget's (resp. QWindow's) destructors had been
completed). This was semantically wrong and inconsistent with other ways
of gathering runtime type information regarding such an object (e.g.
dynamic_cast, obj->metaObject()->className() and so on).
Pick-to: 6.3
Change-Id: Ic45a887951755a9d1a3b838590f1e9f2c4ae6e92
Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org>
Reviewed-by: Marc Mutz <marc.mutz@qt.io>
Reviewed-by: Tor Arne Vestbø <tor.arne.vestbo@qt.io>
Like Q_NAMESPACE_EXPORT for Q_NAMESPACE, this variant of Q_GADGET
allows passing an export macro. This is useful to avoid exporting the
whole class just to get the staticMetaObject hidden therein exported.
Before anyone asks: No, we don't need Q_OBJECT_EXPORT, because QObject
subclasses, being polymorphic, always need to have a class-level
export macro (to export their vtable), but while that technique also
works for value classes (the Q_GADGET audience), it is not desirable
for them, because it makes inline functions exported in Windows debug
builds, which is not what we want, because it needlessly restricts
what you can to with the inline functions (e.g. remove).
[ChangeLog][QtCore] Added the Q_GADGET_EXPORT macro, which is like
Q_GADGET, but allows passing an export macro (like Q_NAMESPACE_EXPORT
for Q_NAMESPACE).
Fixes: QTBUG-55458
Change-Id: I546297de1e8aa45d83381991bcd3fbca61e1eef0
Reviewed-by: Edward Welbourne <edward.welbourne@qt.io>
In some bug reports we got code which does deleteLater()
on QCoreApplication, however this is not going to work as
the user may expect.
In cases where an application uses Qt WebEngine, this leads
to weird looking crashes on exit as webenginecontext is not
destroyed.
Pick-to: 6.2 5.15
Change-Id: I4d284f30b0c7cad15ba6da3d65cdf813c36ee036
Reviewed-by: Michael Brüning <michael.bruning@qt.io>
Reviewed-by: Allan Sandfeld Jensen <allan.jensen@qt.io>
QT_VERSION is now at least QT_VERSION_CHECK(6, 3, 0), so remove all
checks against Qt 6.0.0 or earlier. They are superfluous. Tidied up in
some places in the process, particularly #include order.
Change-Id: I2636b2fd13be5b976f5b043ef2f8cddc038a72a4
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
The connect() documentation makes it clear that UniqueConnection does
not work with free functions / function objects and the like; only with
actual PMFs. Rather than silently *ignoring* the flag, be vocal about its
presence by warning, and make the connection fail (as the user has
passed an illegal argument).
[ChangeLog][QtCore][QObject] QObject::connect() now will refuse to
connect a signal to a free function / function object if UniqueConnection
is passed. Note that UniqueConnection has never worked for such
connections -- the flag was simply ignored, and they were established
multiple times. Now, the flag is not ignored and results in a connection
failure (as well as a runtime warning by Qt).
Change-Id: I6509667018c74f9bd24910cde0a1b16c5f84f064
Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org>
Reviewed-by: Fabian Kosmale <fabian.kosmale@qt.io>
Reviewed-by: Mårten Nordheim <marten.nordheim@qt.io>
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Outside tests, all in-tree callers of QObject::findChildren() pass no
name to match, and in my experience that is also true for the vast
majority of out-of-tree users.
Avoid the temporary QString creation in the caller and the repeated
QString::isNull() checks in the implementation by overloading
findChildren() without a name argument and checking for name.isNull()
only once, forking off into separate helper functions.
Adjust in-tree callers that used an explicit `QString()` argument in
order to pass options, which goes to show that `name` should never
have been the first argument of findChilden() in the first place, even
though I appreciate the symmetry with findChild() (the use-cases of
which, however, are radically different).
Change a `findChildren().size() == 0` call found while scanning for
findChildren() calls to `!findChild()` as a drive-by.
Modernize loops in the various qt_qFindChild{,ren}_helper() overloads
to match how the new code looks.
[ChangeLog][QtCore][QObject] Added findChildren() overload taking no
name (thus optimizing this common case).
Change-Id: Ifc56e5438023d079b40c67f11ae274a3e128ad5e
Reviewed-by: Fabian Kosmale <fabian.kosmale@qt.io>
Reviewed-by: Mårten Nordheim <marten.nordheim@qt.io>
When a signal/slot connection is broken, it gets added to the
sender's list of "orphaned connections", to clean up later.
This cleanup happens when the sender gets destroyed or as soon as
it emits any signal.
This may cause soft memory leaks in case receivers get destroyed,
and the sender is a long living object and doesn't emit signals
for a while (e.g. QThread).
For some reason, an explicit disconnection cleans up the list
(either by using the QMetaObject::Connection object, or in case
of string-based connect, using a string-based disconnect). This
raises lots of doubts about why having this list in the first
place.
Fix the soft-leak by cleaning up the orphaned connection list when
destroying a receiver.
Note: I still believe that we shouldn't have any "orphaned"
connection list, and rather cleanup on disconnect/deletion
(otherwise, emitting a signal may cause a CPU spike because it
triggers a cleanup). If we allow for any "impredictability" during
signal activation we're just admitting that signals/slots aren't
suitable for e.g. low-latency codepaths. That's why I'm not marking
the problem as fixed.
Original-patch-by: Giuseppe D'Angelo <giuseppe.dangelo@kdab.com>
Task-number: QTBUG-88248
Task-number: QTBUG-87774
Pick-to: 6.2 6.1 5.15
Change-Id: Id25f67a45dff49f740132a44d36e88740eb12070
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org>
This does not fix all data races that we have in the system yet.
One major issue is the virtual disconnectNotify(), that can be
called from any thread and thus is inherently problematic, as it
can collide with the object getting destroyed at the same time
in another thread.
Pick-to: 6.2 6.1 5.15
Task-number: QTBUG-88248
Change-Id: I9d841eb363b7e4f0de1657aeb8f5340d0fd55190
Reviewed-by: Andrei Golubev <andrei.golubev@qt.io>
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
This function is/will be used in a few places where we already have a
lock. Temporarily unlocking and relocking invites all kinds of troubles.
By adding a flag we can instead tell the function that we already hold
the lock.
Pick-to: 6.2 6.1 5.15
Change-Id: Ibca089de61133661d5cd75290f2a55c22c5d013c
Reviewed-by: Andrei Golubev <andrei.golubev@qt.io>
queued_activate adds a reference to the slot object. It also attempts to
deref it again, but that did not work correctly so far. We could end up
with
T1 | T2
queued_activate |
checks isSlotObject |
adds ref |
locker.unlock() |
| QObject::~QObject
| //In disconnect all senders loop
| sets isSlotObject to false
| derefs slotObj, but not deleted
checks isSlotObject |
(no deref because it's null) |
To solve this issue and others caused by early returns, we now use a
RAII helper, which always takes care of calling destroyIfLastRef if the
ref count has been incremented.
Change-Id: I9c011cdb8faa5f344d7e70f024fc13f407e39ccf
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
Once we're in ~QObject, only methods of QObject are still valid.
Notably, no setter of any derived class is still valid. Thus, to be safe
we must no longer react to binding changes of those properties. To
ensure that this happens for QObjectCompatProperty properties, we
explicitly clear the binding storage.
Fixes a particles3d example crash.
Change-Id: I10d2bfa5e96621ce039d751cffaf3ac41893623e
Reviewed-by: Laszlo Agocs <laszlo.agocs@qt.io>
Reviewed-by: Ulf Hermann <ulf.hermann@qt.io>
Extended QObjectPrivate::ExtraData to store a pointer
to its parent, and reimplemented qGetBindingStorage()
function for QObjectPrivate::ExtraData.
This allows to use Q_OBJECT_COMPAT_PROPERTY macro
for a property, stored in QObjectPrivate::ExtraData
and solves all the problems with calling a custom
setter.
Task-number: QTBUG-85520
Change-Id: I40e01c29430846359ef9160fa1ae97c702be9a18
Reviewed-by: Edward Welbourne <edward.welbourne@qt.io>
Reviewed-by: Sona Kurazyan <sona.kurazyan@qt.io>
The Core_qobject target was problematic for a few reasons.
In the very beginning, Core_qobject was an OBJECT library, explaining
its existence with the following comment "requires to spill internals
into users". The object library contained qobject.cpp and
qproperty.cpp and their manually generated moc files.
It was was a dependency of Core.
Afterwards it got converted into a static library instead, presumably
to circumvent some issues with metatype generation.
Unfortunately making it a static library was wrong, because it leads
to linker failures (undefined symbols) in a static Qt build when
trying to link a project that uses only QtCore.
Both Core and Core_qobject refer to each other's symbols, and
whichever library came first on the link line, many of its symbols
were discarded by the linker.
Transforming Core_qobject back into an object library could be an
option, but that leads to trouble with creating proper prl files
(our infrastructure doesn't quite support that for object libraries
yet).
Furthermore, there was some internal discussion and concern
about how to merge the two metatype json file lists (Core and
Core_qobject) into a single file.
Instead of using an object library, move qobject.cpp and qproperty.cpp
to be directly a part of Core again and remove the Core_qobject
target.
In order to do that, we have to circumvent some issues with
AUTOMOC.
First, we have to disable AUTOMOC for qobject.cpp and its headers.
qobject.cpp contains Q_GADGET, Q_NAMESPACE, etc keywords
in its doc comments. That confuses AUTOMOC, because it expects a
qobject.moc file which is not included by qobject.cpp.
See e2ff9e3b9957f844d6530cc92096bed3c16fa46a for some details
on the Q_GADGET + AUTOMOC issue.
We still need to run moc on on qobject.h and compile moc_qobject.cpp,
so we manually run moc on the header file. Because qobject.cpp
directly includes moc_qobject.cpp, we need to adjust the header
include path to the generated file.
Second, we remove the direct inclusion of moc_qnamespace.cpp
in qobject.cpp to prevent duplicate symbol compilation, because
the symbols were compiled both as part of qobject.cpp and
mocs_compilation.cpp.
Finally, because manual moc is used on qobject.h, the generated json
file needs to be explicitly passed to Core's metatype generation
step.
Amends e57a94cbd8f28ccab8b3368afc4365237c533a4f
Amends 3c4c02b3f1cec8ebc0134fd05b0e51f1f771937d
Pick-to: 6.1
Task-number: QTBUG-87580
Change-Id: I859a7d2f95508eda211a0a0b4ec132540b78bda4
Reviewed-by: Alexey Edelev <alexey.edelev@qt.io>
Reviewed-by: Joerg Bornemann <joerg.bornemann@qt.io>
The code in qtdeclarative did not do anything at all anymore.
Change-Id: Idd97145cb74aeb4f43dfce2f282a765e90945073
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org>
Original QML-specific connection mechanism ignores the receiver argument
and uses sender as receiver. This causes uncontrollable memory growth
in certain cases as connections on receiver persist even after receiver
is destroyed
New connect() with receiver parameter uses underlying API correctly,
disconnect is provided for the symmetry (not sure it's really needed)
Task-number: QTBUG-86368
Pick-to: 5.15 6.0
Change-Id: I4580d75b617cb2c4dfb971a4dfb8e943e325572b
Reviewed-by: Fabian Kosmale <fabian.kosmale@qt.io>
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
There was a check "if (cd && cd->signalVectorCount())" in the code
which was faulty because "signalVectorCount()" could return -1 to
indicate that no signalVector is available. This patch fixes
the problem.
Fixes: QTBUG-89846
Pick-to: 5.15
Change-Id: I12c512ff8d889ba2bd298789ab6081930f739089
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
The only callers are QObject::findChild(ren), and they always
pass valid pointers, so there's no point of checking them.
Change-Id: I789abc3a53db523acf06c1a8a340094a71c79e41
Reviewed-by: Volker Hilsheimer <volker.hilsheimer@qt.io>
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
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>
Introduce a category for debug/warning prints coming from connect()
and related functions. As a drive-by, rename and complete the
usage of logging categories for connectSlotsByName().
Change-Id: Ieb2d8641df856c8c60317757eda80dc5f9375130
Reviewed-by: Fabian Kosmale <fabian.kosmale@qt.io>
Mostly related to qstrlen().
Change-Id: I69e2052c83766e4fc466ed398d0d0eac011a77ec
Reviewed-by: Mårten Nordheim <marten.nordheim@qt.io>
Reviewed-by: Oliver Wolff <oliver.wolff@qt.io>
QBindingStorage is a class that can store a set of binding objects
for the properties of a QObject. This will get used to reduce the
memory overhead of the property system when adding bindable properties
to QObject based classes.
The binding storage has a pointer to the TLS entry containing the
currently evaluating binding. Like that we avoid repeated TLS
lookups and reduce the overhead of the property system to one
pointer lookup and one compare for the case that properties
aren't being used.
Each QObject now owns one binding storage object, that can be used to
store binding data for properties that members of the QObject.
Change-Id: I27427c03c2ba281f072e074be96147bdbcaac246
Reviewed-by: Fabian Kosmale <fabian.kosmale@qt.io>
If one needed to listen to a signal just once, one had to
store the QMetaObject::Connection object returned by connect()
and use it to disconnect the slot after the first signal
activation.
This has led to a proliferation of using wrappers (and enough
TMP); they usually look like this:
1) create a shared_ptr<QMO::Connection>, allocating its payload;
2) create a lambda, capturing the shared_ptr by value;
3) in the lambda, disconnect the connection (through the shared_ptr),
and call the actual slot;
4) connect the signal to the lambda, storing the returned
QMO::Connection into the shared_ptr.
This is expensive, error prone for newcomers, and tricky to
support as a general facility inside one's projects.
We can do better, just support single shot connections right
in QObject.
[ChangeLog][QtCore][QObject] Added the Qt::SingleShotConnection
flag. When a connection is established with this flag set,
the slot is going to be activated at most once; when the signal
is emitted, the connection gets automatically broken by Qt.
Change-Id: I5f5feeae7f76c9c3d6323d841efba81c8f98ce7e
Fixes: QTBUG-44219
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
Reviewed-by: Fabian Kosmale <fabian.kosmale@qt.io>
Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org>
Reviewed-by: Oswald Buddenhagen <oswald.buddenhagen@gmx.de>
One could guess it by assuming that disconnecting for a destroyed
receiver and disconnect() with given receiver use the same
implementation, but without closely knowing the implementation a
reader of the documentation can't know for sure.
Also add a test to prove that what the new documentation says is
really true.
Also remove an unnecessary negation in the preceding sentence.
Change-Id: I9d24442bb1a4646b89f969bad1a4d0e1eafa7534
Reviewed-by: Mitch Curtis <mitch.curtis@qt.io>
Reviewed-by: Fabian Kosmale <fabian.kosmale@qt.io>
When a signal/slot connection is activated, a lock on the receiver
object is taken (to be sure it doesn't get destroyed).
The path for blocking queued connections however took the lock on
the sender by accident, fix that.
Pick-to: 5.15 5.12
Change-Id: I8cd938a50eca2bf71e7bfb86768ee0c8431afdfa
Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org>
Reviewed-by: Fabian Kosmale <fabian.kosmale@qt.io>
As there is now a chance that a QMetaMethod already contains the
metatypes for its arguments, we can just query it directly (and use the
fallback to name lookup logic that already exists there).
This also allows us to avoid creating a QList of names, and only
requires us to do a name lookup in case the connection actually fails.
Change-Id: Idda30bc4b538a94476ae6c533776c22340f0030d
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
And don't use int based type mapping anymore.
Change-Id: I456e76d1933ef646a7bd39ce565886b89e938a44
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
And remove one of the type id to name mapping that still
existed in QMetaType. QMetaTypeInterface can provide that,
so there's no need to have a second copy of the data.
qMetaTypeTypeInternal() can still map all the names of all
builtin types to ids. That functionality is for now still
required by moc and can't be removed yet.
Change-Id: Ib4f8e9c71e1e7d99d52da9e44477c9a1f1805e57
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
- Remove obsolete functions and enumeration values
- Remove QObject * parameter from QMetaProperty accessors
- Fix renamed enumerations in QSsl
- Fix list items to be \li
- Fix function signatures and variable names
Change-Id: I37c7e6bf2c8ff92bc7b82620bae0a27796f866ab
Reviewed-by: Paul Wicking <paul.wicking@qt.io>
Centralize the offset handling in one place and avoid lots of magic
numbers in various places. Expose the number of ints per method in
QMetaObjectPrivate as a constant, so that code in other places can
access it via private API.
Change-Id: I59790287a17ea47e6160ec65d9c8d0aaee748947
Reviewed-by: Simon Hausmann <hausmann@gmail.com>
Reviewed-by: Ulf Hermann <ulf.hermann@qt.io>
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
This does the analog of 46f407126ef3e94d59254012cdc34d6a4ad2faf2 for the
methods we care about (signals, slots, Q_INVOKABLEs). In addition to the
actual QMetaType, we store an array with offsets so that we later can do
a mapping from methodIndex to metatype.
The newly added QMetaMethod::{return,parameter}MetaType methods can then
be used to retrieve the metatypes.
This does however require that all involved types are complete. This is
unfortunately not a feasible requirement. Thus, we only populate the
metatype array on a best effort basis. For any incomplete type, we store
QMetaType::Unknown. Then, when accessing the metatype, we fall back to
the old string based code base if it's Unknown.
Squashes "moc: support incomplete types" and "Fix compile failures
after QMetaMethod change"
Fixes: QTBUG-82932
Change-Id: I6b7a587cc364b7cad0c158d6de54e8a204289ad4
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
Just opened some files with spell checker turned on,
and corrected what was underlined.
Also, see QTBUG-84427.
Change-Id: I702b5bf2cfe1f23c928a6a57f117a2f571ad86ea
Pick-to: 5.15
Reviewed-by: Mitch Curtis <mitch.curtis@qt.io>
