Added a delay to the "closed" signal with Qt::QueuedConnection so that "triggered" signal could be handled before close.

Fixes: QTBUG-128158
Change-Id: Ia4185198fa523a9c87bf56b61a63e9461412d478
Reviewed-by: Mitch Curtis <[email protected]>

There actually is a code path that returns it as-is and that saves a
copy this way.

Coverity-Id: 469518
Coverity-Id: 469514
Change-Id: I50bcfe6c30648baba192233003d4a77fb8a4d048
Reviewed-by: Fabian Kosmale <[email protected]>
Reviewed-by: Sami Shalayel <[email protected]>

Use QPointF instead of QPoint for accurate mapping of coordinates.
Amends 317ae95

[ChangeLog][QQuickItem] The function mapToItem(item, real x, real y)
now returns a point with real values instead of rounding to an integer.

Fixes: QTBUG-119034
Pick-to: 6.8 6.5
Change-Id: I2c11ec03bdc925c55f48f309f22083b754023cf0
Reviewed-by: Shawn Rutledge <[email protected]>
Reviewed-by: Ulf Hermann <[email protected]>

Split the test up in multiple tests, and move the test files in separate
folders (as the data folder is already way too big).

Change-Id: Id6c22e4e0c60fb8115c15c1a19ad10b454fbd3fc
Reviewed-by: Semih Yavuz <[email protected]>
Reviewed-by: Olivier De Cannière <[email protected]>

This broke after 477c15d.
countChanged would have normally been emitted in the call to
applyModelChanges that we skipped after adding the q->size().isNull()
check.

We can check hasPendingChanges() to know if a count change
is pending (which is also the first thing applyModelChanges checks
before doing its work), and if so, emit countChanged so that
any bindings that are relying on it can function as expected.

Fixes: QTBUG-129165
Pick-to: 6.8 6.7 6.5
Change-Id: Ic045f1870b39d192f6880e23daab03fd73a16d58
Reviewed-by: Fabian Kosmale <[email protected]>
(cherry picked from commit 0a1d8fb)

Pick-to: 6.8
Change-Id: I40c103020432540c038932a4fe0516cac3b1ad3b
Reviewed-by: Volker Hilsheimer <[email protected]>

This is the first step in chaining up the values in a graph. We can only
derive the next node of the graph if we have a full previous node.
QQmlJSRegisterContent will ultimately just be a container for a node.
Prepare for this by requiring QQmlJSRegisterContent as input when
creating a new register content.

In some cases we can't actually nest contents, yet. Therefore we chicken
out by creating synthetic QQmlJSRegisterContents.

Task-number: QTBUG-124670
Change-Id: Idad793692ca78d71ba17107d396e508bc1392b67
Reviewed-by: Olivier De Cannière <[email protected]>
Reviewed-by: Fabian Kosmale <[email protected]>
Reviewed-by: Sami Shalayel <[email protected]>

Also skip the first paragraph if it starts with <p><b>.
The actual documentation usually starts in the second
paragraph if the first paragraph contains the <b>.
This will remove the warnings contained in any accompanying
(\note or \warnings) text box. Also, this post
processing only happens in HtmlExtractor::Simplified mode.

Pick-to: 6.8
Task-number: QTBUG-120016
Change-Id: I367f8aded68972d34f31ed380ed3317865c289f3
Reviewed-by: Sami Shalayel <[email protected]>

There are two ways to obtain the help through help plugin. The first is
querying the keyword which will probably give a hell of a lot resulting
results to be processed. Better way is to search the keyword by id which
will give the precise documentation for the given name. id's are kept in
the form of DocumentName::keyword form in the relavant .qhp file.

So, to get the correct documentation for an identifier, one should find
the defining scope for that identifier and try to get the relavant link
by searching DefiningScopeName::identifierName.

Documentation for most of the qml properties are found in the Qml named
namespace in the .qhp file. For example, x,y,width,height of an Item are
defined in QQuickItem scope but documentation doesn't exist for
QQuickItem::x, but for Item::x. To get the proper mapping, utilise
typeResolver->importedNames() method and introduce the m_cppTypesToQmlTypes
hash map.

Pick-to: 6.8
Task-number: QTBUG-127051
Task-number: QTBUG-120016
Change-Id: I27d4c60b11cd2aff4b4cb8a19ab9817519799e06
Reviewed-by: Sami Shalayel <[email protected]>

Pick-to: 6.8
Fixes: QTBUG-128797
Change-Id: I53ac0af6f98c6ad291762d4ea8d195529b84c51a
Reviewed-by: Sami Shalayel <[email protected]>
Reviewed-by: Ulf Hermann <[email protected]>

This is necessary after 0b7ee7a.

If QQuickPointHandler::wantsEventPoint() ever returns false,
QQuickPointerHandler::handlePointerEvent() calls setActive(false).
If AA_SynthesizeMouseForUnhandledTabletEvents is enabled, and
PointHandler is the only handler trying to handle tablet events, it does
not accept the events; so QGuiApplicationPrivate::processTabletEvent()
sends a synth-mouse press for each tablet press. If the position doesn't
match the last-known mouse position, QGuiAppPriv::processMouseEvent()
sends a synth-mouse move, too. PointHandler doesn't care about either of
these synth-mouse events, but it has to say that it "wants" them to
avoid getting deactivated. But in spite of that, we don't want
QQuickSinglePointHandler::handlePointerEventImpl() to call
QQuickHandlerPoint::reset() with those synth-mouse events, because
then the tablet-specific data from the previous QTabletEvent would
get overwritten. In general, pointer handlers handle original
device-specific events only, and we just have to avoid letting these
synth-mouse events interfere.

There were a couple of places that were assuming that if an event
is not a QTouchEvent, then it's safe to cast it to QSinglePointEvent;
that's usually true, but isSinglePointEvent() exists since qtbase
9a1a15b42fb526ad4f80944afb7761bfff1b5c9d so we might as well use it.

Previously, we'd set acceptedButtons: Qt.LeftButton | Qt.RightButton
to indicate that the PointHandler should respond to either one; but
now it's better to set Qt.NoButton to indicate that it doesn't care.
The synth-mouse move has no buttons set; if PointHandler only "wants"
events with specific buttons, that buttonless move will deactivate it
temporarily, and then the synth-mouse move would re-activate it, causing
extra activeChanged and pointChanged signals. But so far, its default
for acceptedButtons is still Qt.LeftButton. Perhaps we should change it.

The special meaning in QQPointerDeviceHandler::wantsPointerEvent() for
Qt.NoButton started already with HoverHandler in
1982d1b and then
fc636af for PinchHandler. Those
handlers fundamentally don't need to care about buttons, so they declare
it by setting acceptedButtons = Qt::NoButton in their ctors, and the
user should not set it differently; whereas with PointHandler, we do
expect the user to set acceptedButtons.

tst_PointHandler::tabletStylus now has code to test both with and
without AA_SynthesizeMouseForUnhandledTabletEvents. But in the "true"
case (the default), it won't pass until we also change
QGuiApplicationPrivate::processMouseEvent() to send synth-mouse-moves
from the same device that sends the mouse press, which itself is
synthesized from the tablet press.

[ChangeLog][QtQuick][PointHandler] If PointHandler is declared with
acceptedButtons: Qt.NoButton it now means that the PointHandler does not
care which buttons are pressed or not pressed: but it ignores
synth-mouse events and responds only to events that come from the
appropriate kind of device. If specific acceptedButtons are set,
synthetic mouse-moves can deactivate it temporarily.

Pick-to: 6.8
Fixes: QTBUG-111336
Change-Id: Ic48565f74d247803e338d2e368a5ef9d249296c2
Reviewed-by: Doris Verria <[email protected]>

Amazingly, all it takes is to import the QtQuick.tooling module into a
namespace.

As a drive-by, also fix the public/private #define'ery in
tst_toolsupport.cpp. That's quite wrong.

Fixes: QTBUG-129196
Change-Id: I29785f068b88820926356fe5facd871187a56ea1
Reviewed-by: Olivier De Cannière <[email protected]>

There is only one JS global object. We should not re-create the value
whenever we use it. This removes some syntheticType().

Task-number: QTBUG-124670
Change-Id: If0ea6dee9a46a7b9f9442834cd6f44a5713c6b0a
Reviewed-by: Fabian Kosmale <[email protected]>
Reviewed-by: Olivier De Cannière <[email protected]>

We can resolve the attached properties function without querying the
type registry. The attached properties function can be resolved with
pure C++, given it's surrounding type, name, and signature. We already
have functionality to do so in qqmlprivate.h

This is not only faster, but also avoids the problem of having to
register dependent types before being able to use their attachments.

Pick-to: 6.8 6.5
Fixes: QTBUG-128895
Change-Id: I5c1df81a43195eb4dd97a4abfaa79f3cfe39355e
Reviewed-by: Olivier De Cannière <[email protected]>

Amends commit 573b6a5

Change-Id: I4fe93712303adcb32db5760694ae96c1516569d8
Reviewed-by: Fabian Kosmale <[email protected]>
Reviewed-by: Olivier De Cannière <[email protected]>

We'll need it in more places.

Change-Id: I8dc23a4b8ee873c0b8e84b3aed9872d48959a021
Reviewed-by: Olivier De Cannière <[email protected]>
Reviewed-by: Fabian Kosmale <[email protected]>

We want to avoid synthesizing the type. Instead the scope should be
specified.

Task-number: QTBUG-124670
Change-Id: If9f6a6c4dad1ec4ff2ad84d1b83a0bc720a8f589
Reviewed-by: Sami Shalayel <[email protected]>
Reviewed-by: Olivier De Cannière <[email protected]>
Reviewed-by: Fabian Kosmale <[email protected]>

The example code snippet in the documentation https://doc.qt.io/qt-5/qml-qtquick-controls2-menu.html caused a warning:
Parameter 'mouse' is not declared. Injection of parameters
into signal handlers is deprecated.
Changed the code snippet in the documentation to use lambda functions.
Now users can use the code snippet without triggering any warnings.

Fixes: QTBUG-128520
Change-Id: Id787af63521a46c82302176c8da8fca4d49bb1b3
Reviewed-by: Axel Spoerl <[email protected]>

Add a check in the lexer and fail in the parser if the file to be parsed
is too big for qsizetype and quint32.

QQmlJS::SourceLocation uses quint32 to save the offset, and code using
QQmlJS::SourceLocation's offset usually do their computation on
qsizetype.

Therefore, limit the QML file size to
std::numerical_limits<quint32>::max() on 64 bits machine when qsizetype
has a representation for std::numerical_limits<quint32>::max(), and to
std::numerical_limits<qsizetype>::max() otherwise.

Thats 2Gb on 32 bits machine and 4Gb on 64 bits machine.

Currently, using qml tooling on files bigger than the limits mentioned
above already leads to UB without this patch.

Add a helper to construct SourceLocation from qsizetype indexes.

Task-number: QTBUG-127833
Change-Id: Ic255964e13ebae08488ed160e59d504638f1b9ad
Reviewed-by: Olivier De Cannière <[email protected]>

Change begin() and end() to return a qsizetype, as now we only process
QML files where quint32 can safely be casted to qsizetype.

This allows to change all users of begin() and end() to use qsizetype,
and to silence all MSVC compile warnings about comparison of ints with
different signedness.

Fixes: QTBUG-127833
Change-Id: I251435aa598386effe0259549dbe94d17b0d806b
Reviewed-by: Olivier De Cannière <[email protected]>

Pick-to: 6.8
Change-Id: I723dc8740a37f6aa92c73cca1b5ecc364578185d
Reviewed-by: Mitch Curtis <[email protected]>

There are existing notifiers (QQuickItem::ItemChange and
QQuickItemChangeListener) for geometry (position and size), and rotation.
Scale has a public signal but no corresponding notifiers. Transform
is also affected by QQuickTransform instances attached to the item,
and when any of those change there is currently no mechanism available
to public or private API's to detect such changes. This change adds the
notifiers so the set is complete - Geometry, Rotation, Scale, and
Transform which catches all three and transform changes.

Task-number: QTBUG-10644
Change-Id: Ib57dbe23c600a493b4e08ce8edad078e10e28d09
Reviewed-by: Shawn Rutledge <[email protected]>
Reviewed-by: Tor Arne Vestbø <[email protected]>

During processing, a significant amount of intermediate ("uncooked")
vertex data is allocated for every fill and stroke node. After the
final, smaller vertex data is generated, the intermediate data was
intended to be deallocated right away. Since the QList API now
requires 2 calls to actually deallocate, that did not happen, and the
data would live on as long as the node (path) itself.

Pick-to: 6.8
Change-Id: I48efa5d0bc2b226c5eaac37dbba485b653604bfd
Reviewed-by: Eskil Abrahamsen Blomfeldt <[email protected]>

Instead model the retrieval of an extension as explicit separate step.
With this in place, we don't have to synthesize the
QQmlJSRegisterContent for extensions anymore.

Task-number: QTBUG-124670
Change-Id: I20394ac8b8a30d7eb9efc91c4e99ef9255a9520b
Reviewed-by: Sami Shalayel <[email protected]>
Reviewed-by: Fabian Kosmale <[email protected]>
Reviewed-by: Olivier De Cannière <[email protected]>

This creates build issues on Windows

Amends ffd39a4

Pick-to: 6.8
Change-Id: Ia031f2ed5275683da617df90bcab62d9fe56d4d2
Reviewed-by: Semih Yavuz <[email protected]>

Make tst_qmllint_import_rel_script configure correctly on static builds,
instead on failing on tst_qmllint_import_rel_script_init if that target
was generated.

Change-Id: Iaebf3159b506d29faac69e90eaa09d0e4c317ec5
Fixes: QTBUG-129585
Reviewed-by: Olivier De Cannière <[email protected]>

This should help find raw string literals in generated code

Task-number: QTBUG-129797
Change-Id: I66e5d6302319678b494030e8cb3459eb06f0f134
Reviewed-by: Semih Yavuz <[email protected]>

Amends e7146cd

Pick-to: 6.8
Fixes: QTBUG-129797
Change-Id: I5dc48412cf29bd2de877dd681ea309a6e74d75c5
Reviewed-by: Ulf Hermann <[email protected]>

This should fix the compiler error on XCode 16, by moving the default
constuction of std::unique_ptr<BindingValue> into the cpp file, where
BindingValue is known.

Pick-to: 6.7 6.8
Fixes: QTBUG-129766
Change-Id: I58aae92c28553d0a0bff0c39a352d4f1444bd6aa
Reviewed-by: Olivier De Cannière <[email protected]>
Reviewed-by: Fabian Kosmale <[email protected]>

Previously, when setting a new file address and pressing the close
button, the focus would move to the quick widget due to
QLineEdit::editingFinished() being called. This triggered an attempt
to close the old file and open the new one, causing unintended state
transitions.

To resolve this, we now reload the file content and update the quick
widget's source without transitioning to CloseState and OpenState. This
is achieved by updating the file path and emitting stateChanged() from
OpenState to OpenState in the StateController, notifying other widgets
to reload their content.

Pick-to: 6.8
Change-Id: Ic623d320fe1736e11fd538a6735632b19c507394
Reviewed-by: Jan Arve Sæther <[email protected]>
Reviewed-by: Doris Verria <[email protected]>

Fixes: QTBUG-129182
Pick-to: 6.8
Change-Id: I6b7624eb514c42587870d6c3e1c1f43fcf691eae
Reviewed-by: Axel Spoerl <[email protected]>

Add code snippet for how to supply source string templates to
Qt::qsTrId using /*% <string> */. The /begincomment and
/endcomment expand to /* and */, respectively.

Fixes: QTBUG-127794
Pick-to: 6.8 6.7 6.5 6.2
Change-Id: I84b77179892eea6dba2c8816b658916d665e718e
Reviewed-by: Joerg Bornemann <[email protected]>

Flickable/ListView/TableView would sometimes pause (or lag) for almost
a second if you dragged it in one direction, but then switched to
drag it in the opposite direction (without lifting your finger). This was
especially evident on iOS. It could be reproduced in, for example, the
Gallery app (the controls-list in the drawer), or any other app with
a Flickable.

The reason this happened, was because a "reversed" drag would cause
dx and dy to converge back to zero (compared to the starting point
of the drag), which would result in overThreshold becoming false
again. And since the deltas ended up below the threshold, the content
item stopped moving.

Rather than overThreshold becoming false when dragging back to the
starting point, this patch will change it so that it stays forever
true once dragging has started. It will also remove the
related dx/dy != 0 checks because of the same reason; Once in dragging
mode, the content item should always move, even if it was dragged back
to the starting point (dx/dy == 0).

Fixes: QTBUG-129599
Pick-to: 6.8 6.7 6.5 6.2
Change-Id: I86d4e6eba74bbb52884bac4782531ac90818323d
Reviewed-by: Jan Arve Sæther <[email protected]>

On macOS, readOnly/non-editable text controls don't accept tab
focus. This behavior is also respected in Qt since
c1d2bcf .
Since this behavior is particular to macOS and not to other
platforms, in Qt we have tabFocusBehavior() in
qGuiApp->styleHints() which represents the system setting.
For all platforms other than macOS this always returns
Qt::TabFocusAllControls, which hints that all enabled controls
should receive tab focus no matter if editable or not. On
macOS if this true, all controls will accept tab focus, and if not,
it means that only editable text inputs and lists will.

The problem was that on macOS readOnly text edits were receivng
tab focus even if tabFocusBehaviour was not
Qt::TabFocusAllControls. This was because they have an
EditableText accessible role, which took precedence over their
editable status. To fix, if tabFocusBehaviour allows,
non-editable/readOnly properties should not receive tab focus no
matter the accessible role.
While at it, move the tabFocusBehaviour check inside
QQuickItem::canAcceptTabFocus.
Remove expect_fail in test covering case.

Note that this patch assumes that if tabFocusBehaviour ==
Qt::TabFocusAllControls readOnly text inputs will receive tab
focus on macOS too, even if this is never the case on a native
application: such inputs never get tab focus on Mac. We could
introduce a new enum value for tabFocusBehaviour that more
accurately depicts the readOnly text input focus behavior.

Pick-to: 6.5 6.8
Change-Id: I098f75eb118b04f73dae3deec0c798f18870a202
Reviewed-by: Jan Arve Sæther <[email protected]>
Reviewed-by: Tor Arne Vestbø <[email protected]>

Basic needs to depend on Templates, as it reuses the Overlay attached
object from there. At runtime, this already works, but tooling needs a
hint; otherwise it will only find the uncreatable Overlay type, not the
attached one.

Fixes: QTBUG-129799
Pick-to: 6.8
Change-Id: I536482cfebfde8a5b61adcc87e3fa2a0b33814ff
Reviewed-by: Olivier De Cannière <[email protected]>
Reviewed-by: Sami Shalayel <[email protected]>
Reviewed-by: Ulf Hermann <[email protected]>

It's a rather intricate mechanism, so it's best to have it documented
for later memory refresh.

Task-number: QTBUG-128875
Change-Id: I1d66cc1f0453abfd80330ec4318c2af25a3609e3
Reviewed-by: Shawn Rutledge <[email protected]>

With this in place, we don't have to synthesize conversion origins
anymore and get to trace values through conversions.

Task-number: QTBUG-124670
Change-Id: Ib3646d410526eca7b982f86adef9d5a387ff56ea
Reviewed-by: Olivier De Cannière <[email protected]>
Reviewed-by: Fabian Kosmale <[email protected]>

With this in place we don't have to synthesize the QQmlJSRegisterContent
for base types anymore.

Task-number: QTBUG-124670
Change-Id: Iafc1838c64e6191bf9b84163e4c1b14ea14dd2c9
Reviewed-by: Sami Shalayel <[email protected]>
Reviewed-by: Olivier De Cannière <[email protected]>
Reviewed-by: Fabian Kosmale <[email protected]>

This property was added to QFontMetricsF in Qt 5.8, but was never added
to the Qt Quick API.

[ChangeLog][QtQuick][Text] Added FontMetrics.capitalHeight property to
match the QFontMetricsF::capHeight() property.

Change-Id: I9c348b05f9f6a52b6fc4645ca7f001d7cad87c41
Reviewed-by: Mitch Curtis <[email protected]>

Remove the two CMake warnings from the CMake output when building
QtDeclarative. The warnings come from a missing optional dependency for
the wearable example.

Pick-to: 6.7 6.8
Change-Id: Id9bf2ccc4d6db2ea44618b2119c69c89e271a870
Reviewed-by: Matthias Rauter <[email protected]>

Add a new logging category, "qt.qml.gc.stateTransitions", that can be
enabled to receive some simple log lines about the GC being on the
verge of executing the current state and about the new state that it
transitioned to after the execution.

As the new logs print the current state, which is stored through the
`GCState` enum, slightly modify the code to allow for registering the
enumeration with `Q_ENUM`.

In particular, move `GCState` and `GCStateInfo`, the latter due to the
dependencies between the types, under `GCStateMachine`.
Make `GCStateMachine` a `Q_GADGET` and `GCState` a `Q_ENUM` and provide
some using statements to reduce the impact of the changes.

Finally, fix the unqualified accesses to the variants of `GCState` to
support the new structure.

Task-number: QTBUG-128357
Pick-to: 6.8
Change-Id: I9d469ddb745f70b9c4553379f6d96719b3a2bb09
Reviewed-by: Fabian Kosmale <[email protected]>

Add a new logging category, "qt.qml.gc.forcedRuns", and issue a log line
when a GC run is forced to completion or fails to do so by being in a
critical section.

Task-number: QTBUG-128357
Pick-to: 6.8
Change-Id: I83ccb60636307e9e8f2aa47fdffb50e39b3e8f17
Reviewed-by: Fabian Kosmale <[email protected]>

Add a logging category `qt.qml.gc.stepExecution` that is used to enable
the additional logs.

`GCStateMachine` was modified to carry some additional state related to
the registered timings over multiple run of a single step.

Call to the execution parts of a state were modified to collect some
measurements and log them if the relevant logging category is enabled.

Task-number: QTBUG-128357
Pick-to: 6.8
Change-Id: I6fc620c85534edc82546130f32bf2af49af6bcdd
Reviewed-by: Fabian Kosmale <[email protected]>

The header seems to be currently unused.

Change-Id: I98967766ecf2c7a5b48dca7fb91a2f66d72223a6
Reviewed-by: Fabian Kosmale <[email protected]>

Gauge was part of Qt Quick Extras, but its implementation used Loader
and was very heavy. This patch implements it using C++ to avoid that.

As Gauge is quite complex and has a lot of possiblities for how users
could style it, this patch offers an example that can be used as a
starting point, and customized using a familiar delegate-based API.

The goal is to provide a set of C++ and QML files that users can simply
copy into their projects.

Fixes: QTBUG-111770
Change-Id: I389058296d65063e9ab6f6e484c59eb526b058b0
Reviewed-by: Jan Arve Sæther <[email protected]>

If the internal class is not set early enough, we will run into a crash
if the object to hold the internal class ends up on the markstack, and
the markstack gets drained. This can happen when the markstack runs out
of space (reaches the softlimit), and does early draining.
We force this condition in the test via a test-only function to
manipulate the soft-limit.

Pick-to: 6.8
Change-Id: I13598797811d697981ef7018c5aa87251331ab17
Reviewed-by: Ulf Hermann <[email protected]>

Clarify a type.

Coverity-Id: 469550
Change-Id: I2f307ef3d0c6a855bcb914094a3a9cd7dcd3038d
Reviewed-by: Olivier De Cannière <[email protected]>

This is the last place where we had to create synthetic types, so delete
the syntheticType() method, too.

Task-number: QTBUG-124670
Change-Id: I88ee90c9adedd86374ea5e5e8106ddd68c4b285e
Reviewed-by: Fabian Kosmale <[email protected]>
Reviewed-by: Sami Shalayel <[email protected]>
Reviewed-by: Olivier De Cannière <[email protected]>

When passing a 64bit integer to string.arg() we better read it as a
double rather than a 32bit integer. This might still lose some
precision, but at least it doesn't clamp.

Amends commit 372db48 before which
64bit integers were not considered integral and therefore took the other
code path that converts to double.

Pick-to: 6.8
Change-Id: Ie4bf770e9cf195aafcf44d72d59318aeb2f5dd70
Reviewed-by: Olivier De Cannière <[email protected]>

Tooltips didn't match the headings. But the headings seem to be right.

Change-Id: Ib75818324e5a7ddde69419187e5fd5dcbbab06b9
Reviewed-by: Eskil Abrahamsen Blomfeldt <[email protected]>

`QQmlSettings` can save and load settings values based on the properties
defined on its usages in a QML project.

The logic that handles which properties are to be considered, currently
uses the MetaObject of a specific instance/type to filter out all
properties that are not defined by the instance/type itself.

It is possible to have a component derive from a `QQmlSettings`
component that defines certain properties.

With the current logic, all properties that are not directly defined on
the child component will not be considered when saving/loading settings
values, so that changes to them might be lost.

To fix the issue, ensure that instances of `QQmlSettings` consider all
properties that are not base properties of `QQmlSettings`, so that
inherited properties will be saved/loaded.

Add a test case to `tst_qqmlsettings` that show such a usage.

Fixes: QTBUG-129515
Change-Id: I3034de3cbe9a49969bae1ba90a567e4d0eda514a
Reviewed-by: Ulf Hermann <[email protected]>

Test is flaky and can't be investigated due to broken VM accessibility
on openNebula.

Task-number: QTBUG-129947
Task-number: QTQAINFRA-6626
Change-Id: I5d5866498e60baa1d012ed5562b1cbe4fb91ca97
Reviewed-by: Ulf Hermann <[email protected]>

Fix qqmljstypepropagator to pass more correct sourcelocations to QQmlSA
plugins: replace approximate getCurrentBindingSourceLocation()
calls with more precise getCurrentNonEmptySourceLocation() calls.

getCurrentNonEmptySourceLocation() is a newly introduced helper method
that filters out empty sourcelocations.
Empty sourcelocations happens for example for:
someValue: 42
where the parser generates a return with an empty sourcelocation:
someValue: return 42

This also adds the ability to check signal handlers in QQmlSA plugins,
as those were previously ignored in
QQmlJSTypePropagator::generate_ret_SAcheck().

Add a test to make sure the new sourcelocations do make sense, and add
sourcelocations to the expected warnings in the quick plugin.

This has the advantage that warnings become deduplicated because of
their more precise sourcelocations. Previously, two warnings on the
same binding would share the same sourcelocation, which would
technically generate two identical warnings that complain about two
different things.

Also fix the lintplugin in tst_qmllint to use the correct
sourcelocations.

This is needed for the .ui.qml warnings QQmlSA plugin.

Task-number: QTBUG-129308
Change-Id: Id52da7cf9ba586edcdae523017720fb695dd2726
Reviewed-by: Ulf Hermann <[email protected]>

Instead of having a cross product of "scope" and "object" on the one
hand and "property", "method" and "enum" on the other hand, use the
scopeType to resolve what kind of element a property, method or enum
belongs to.

Task-number: QTBUG-124670
Change-Id: Ia9db05e13979e55e0ed7912afc2003dfd018987a
Reviewed-by: Olivier De Cannière <[email protected]>

We can infer from the method itself whether it's a JavaScript call or a
C++ call. Therefore, what we need to carry around is the
QQmlJSMetaMethod. This way we can do with only one content variant for
all kinds of calls.

Task-number: QTBUG-124670
Change-Id: I1d8193018f0224924b8aef5181c398674bde857d
Reviewed-by: Olivier De Cannière <[email protected]>

…irection"

This reverts commit dc5b8f5.
It seems to have broken the ability for MapView's DragHandler to
start dragging if a parent Flickable is still moving.

Pick-to: 6.8
Fixes: QTBUG-129941
Reopens: QTBUG-129599
Change-Id: I225f360faabe8a99dc9ecd7e649500679899a0a6
Reviewed-by: Volker Hilsheimer <[email protected]>

This reduces the amount of "global" and "builtin" we have to use.

Task-number: QTBUG-124670
Change-Id: I7da71c7751f852c065fe9f4736d4d8dc78b98c50
Reviewed-by: Olivier De Cannière <[email protected]>

This commit adds links to QEvent::accept() and QEvent::ignore() to the
see also section of QQuickItem::keepMouseGrab() and
QQuickItem::keepTouchGrab()

Task-number: QTBUG-128933
Pick-to: 6.8
Change-Id: I322b3f885296fe4fd95734cfc51b25cbe6a37385
Reviewed-by: Ulf Hermann <[email protected]>

This reduces the amount of global and builtin we have to use.

Task-number: QTBUG-124670
Change-Id: I9c5600e71047a8e27db526557253694a301700d5
Reviewed-by: Olivier De Cannière <[email protected]>

Now that resolving the types can be more expensive, especially for
composite types, we should be careful not to do it too often. Most of
the cases where we have to resolve types are for lookups. Most of the
lookups only resolve types in their "init" step, which ideally is only
performed once in the application life time.

However, so far calls had to pass the types of arguments and return
values in the actual "lookup" step, which causes the resolution to
happen on every call.

This change moves those type resolutions to the init step. We can do
this because:

1. Regular typed method calls are calls to a specific overload with
   specific types. The method itself already has the list of types and
   we can just remember which one that is. Then we don't need to pass
   the types.

2. Calls to shadowable methods are all-QVariant typed. The only thing
   we need to know is the number of arguments. Then we can construct
   the list of types in the lookup itself.

We can remember which one of those we're dealing with by adding further
"AsVariant" lookup functions. For the case of non-property-cached
regular methods we also need a new "Fallback" lookup like we already
have it for properties.

Change-Id: I74a3729131d6a5ea0ad79e276965a5167cd609be
Reviewed-by: Olivier De Cannière <[email protected]>

The pathData structure would keep a copy of the QuadPath used to
generate the stroke nodes. That copy was never used for anything, so
remove it.

This commit also cleans up the functions to generate stroke nodes, by
decoupling them from the internal PathData structure. This was earlier
done for the corresponding addFillNodes() function.

Pick-to: 6.8
Change-Id: I414b421dfc0cdf5c38e52a298476433f7643b5db
Reviewed-by: Eskil Abrahamsen Blomfeldt <[email protected]>

Change-Id: If3d1e61636b257a721ed7b22d57db1e3d9d11565
Reviewed-by: Ulf Hermann <[email protected]>

Since QQmlJSCodeGenerator has so many methods that do different things,
it's good to be explicit. The "literal" method, for example, did not
actually generate code for a literal of any kind.

Change-Id: Ife6ac68bb54bc5c7511fa2c58b84d141ddbf74f2
Reviewed-by: Olivier De Cannière <[email protected]>

Fixes: QTBUG-129926
Task-number: QTBUG-129929
Pick-to: 6.8
Change-Id: I8f9a91b3a9d222e94883cafcd5c6267c717f10f6
Reviewed-by: Mats Honkamaa <[email protected]>

This prevented direct mode from ever successfully completing a direct
access to a scope object property. There is no reason to consider
context object lookups since we cannot do anything with context objects
here.

Pick-to: 6.8
Change-Id: Ifcbaa5934e383db2fc65953340443a5c29f04f8d
Reviewed-by: Olivier De Cannière <[email protected]>

When we want to read a value of a specific type, this is in effect a
conversion to this type and should be marked as such.

Task-number: QTBUG-124670
Change-Id: I5ecec6e38f1f211bf527312ab3b93df580055ef9
Reviewed-by: Olivier De Cannière <[email protected]>

…anup"

This reverts commit 3787a3e.

Reason for revert: Issue has been fixed.

Task-number: QTBUG-127422
Change-Id: I219e27e1665dd4aab30ce51088b9051397d4e164
Reviewed-by: Assam Boudjelthia <[email protected]>

For the time being this is feasible since the Qt Quick library links to
(Qt)OpenGL still due to classes such as QQuickFbo or QQuickOpenGLUtils.
In principle, having OpenGL-specific code returning to the common code
base is far from ideal, but since Qt Quick is not entirely free from
this anyway, we might just live with it for now.

Import the Qt 5 code in the painter node, all under ifdef
QT_CONFIG(opengl) and a runtime check for the QRhi backend.

[ChangeLog][QtQuick][QQuickPaintedItem] QQuickPaintedItem is made
to behave identically to Qt 5 when the application is rendering with
OpenGL and the renderTarget property is set to FramebufferObject or
InvertedYFramebufferObject.

Change-Id: If5297368836ca8373aae0365c5ae00a7b1ad50d9
Reviewed-by: Andy Nichols <[email protected]>

We do the same thing in the QJSEnginePrivate dtor, too. Doing it
twice per engine shutdown creates a lot of unnecessary churn.

Task-number: QTBUG-129388
Change-Id: I1c33e2efa33541673b97758b954dc4b1de1685a9
Reviewed-by: Olivier De Cannière <[email protected]>

In paths that might be executed during destruction of items and
shut-downs, check that the agent is not nullptr already.

Reported crash is not reproducible in a test case, so no auto-test
added.

Fixes: QTBUG-129699
Pick-to: 6.8 6.5
Change-Id: Ic913b125655aed57fdb3924e9a4b8754e8095c47
Reviewed-by: Doris Verria <[email protected]>
Reviewed-by: Shawn Rutledge <[email protected]>

Call QQuickDeliveryAgentPrivate's static function directly.

Change-Id: Ie2d3c30d4ffb552ceb78b2cc8e56f7271adddbc3
Reviewed-by: Doris Verria <[email protected]>
Reviewed-by: Shawn Rutledge <[email protected]>

Call the QQuickDeliveryAgentPrivate API directly, using the QQDA we get
from the relevant item's private where applicable, otherwise using the
delivery agent we get from the QQuickWindow.

Task-number: QTBUG-129699
Change-Id: Iebf44dc299cb9709d3ff4a00cf26b0d9c3d45ceb
Reviewed-by: Shawn Rutledge <[email protected]>

MSVC thinks it can deviate from the standard and fold separate functions
into the same code block even if their addresses are taken. This makes
the addresses equal and breaks the lookup mechanism. We set up a dilemma
for the compiler by explicitly comparing the function addresses and
failing when they are equal. Now the new function has "different" code
that will however be optimized away.

Pick-to: 6.8 6.5 6.2
Fixes: QTBUG-128843
Change-Id: I42d9f764235b877c5c9e054502ea01a16369d27b
Reviewed-by: Sami Shalayel <[email protected]>
Reviewed-by: Olivier De Cannière <[email protected]>

MSVC thinks it can deviate from the standard and fold separate functions
into the same code block even if their addresses are taken. This makes
the addresses equal and breaks the lookup mechanism. We set up a dilemma
for the compiler by explicitly comparing the function addresses and
failing when they are equal. Now the new function has "different" code
that will however be optimized away.

Pick-to: 6.8 6.5 6.2
Task-number: QTBUG-128843
Change-Id: Id9512397fcbb238d880ee9966d8708fc576590c7
Reviewed-by: Fabian Kosmale <[email protected]>

We have the TypeByName content variant already for the related concept
of naming a type in script code, for example to perform a type
assertion. We can re-use it for signatures.

Task-number: QTBUG-124670
Change-Id: If13518df2b2217769d02c4618f57203907b3f5f9
Reviewed-by: Olivier De Cannière <[email protected]>

We should no longer be encouraging readers to use the old, string-based
constructor for QtQuickView. Changed the code examples in the
QtQuickView API documentation to use the new style of constructing.

Task-number: QTBUG-130112
Pick-to: 6.8
Change-Id: I162350d237ac73427dcb09ba6e78aa9a8e350755
Reviewed-by: Assam Boudjelthia <[email protected]>

Probably I was trying to paste something by pressing Control-C, and the
keyboard only got the c key. Amends
78640ff

Change-Id: Ia830a02bacd8c7c6c8ba45aa9ba0bace8e528771
Reviewed-by: Shawn Rutledge <[email protected]>

There are two QGfxSourceProxy classes in the codebase, one in
qt5compat/src/imports/graphicaleffects5/private and the other in
qtdeclarative/src/effects.

In a static build, if a project links and initializes both qml
modules, it is undetermined which QGfxSourceProxy class' meta object
will be loaded.
This can lead to the multi effects object version being loaded when
initializing a component from the graphical effects module. You then
get a warning message saying "Missing QML.Element class info for
QGfxSourceProxy" and then a failure to load the QML module.

This happened for the sample qmake project attached to the bug report
this change is fixing.

Rename the multi effects QGfxSourceProxy class to QGfxSourceProxyME
(where ME stands for Multi Effects) to avoid the meta object conflict.

Pick-to: 6.8
Fixes: QTBUG-130080
Change-Id: I0001f36b19e36deeddd32897d23c8bfa52011e02
Reviewed-by: Fabian Kosmale <[email protected]>
Reviewed-by: Kaj Grönholm <[email protected]>

The shared_qml_module cmake build test project does not have a
'shared_qml_module' inner test target, it only has a 'test_models'
one, so passing 'shared_qml_module' after the -V was incorrect, and
even if the test existed, it should have been passed to the -R
argument.

Unfortunately ctest versions prior to 3.30 happily ignore incorrect
arguments. This means that the -V argument was completely ignored
because it was treated as a single argument due to the double quotes.

Remove the incorrect test name, leaving just -V to run all found
tests, like we do for other build tests.

As a drive-by use the CMAKE_CTEST_COMMAND variable instead of plain
'ctest', like the other build tests.

This is needed to ensure CI passes when CMake is updated to 3.30.

Pick-to: 6.8
Task-number: QTQAINFRA-6484
Change-Id: Ic7d1c267985cf94d5806f3972d628ab1bf9352cf
Reviewed-by: Alexey Edelev <[email protected]>

Assume that a button with DialogButtonBox.DestructiveRole is added to a
dialog. When that button is clicked, it only emitted the discarded()
signal. Now after emitting, we call QQuickDialog::close().

[ChangeLog][Controls][Important Behavior Changes] Dialog is now closed
when a button with the DestructiveRole is clicked. Currently the only
button with this role is Discard.

Pick-to: 6.8
Fixes: QTBUG-67168
Change-Id: I3fbee383e10709630ba48cc420924ca71153e70f
Reviewed-by: Mitch Curtis <[email protected]>

For some reason, people keep cargo-culting the ugly tests instead
of using QQuickTest::showView(), which has been possible since
22c2008 and takes care of all the
usual stuff: make sure the QML loaded a root item, center the window,
move the mouse outside etc. This tends to stabilize tests too.

Change-Id: I9dc170be9b9fe941bff380d334880dd6978e6865
Reviewed-by: Richard Moe Gustavsen <[email protected]>

When these fail, we'll get more information in CI logs.

Pick-to: 6.8 6.5
Task-number: QTBUG-118062
Task-number: QTBUG-123499
Change-Id: I90cdb466ca29bf955c45d3918be429e5d4d86d1b
Reviewed-by: Shawn Rutledge <[email protected]>

While this makes no difference in practice, it was technically undefined
behavior before. You cannot call the object getter through the interface
of the context getter.

Amends commit 20d413a.

Pick-to: 6.8 6.5
Change-Id: Iab15d4b05a18aa1997e1e0a45aa9df19adf283df
Reviewed-by: Fabian Kosmale <[email protected]>
Reviewed-by: Olivier De Cannière <[email protected]>
Reviewed-by: Sami Shalayel <[email protected]>

Codechecker complains about ambiguous 1-letter names, and especially 'l'
vs 'I'.

Change-Id: Ia32c2a8eb04956bb10e7e49e2085b724e8feb749
Reviewed-by: Olivier De Cannière <[email protected]>
Reviewed-by: Sami Shalayel <[email protected]>
Reviewed-by: Edward Welbourne <[email protected]>

Commit 9ef4c123c39c642357c9e8530d59f32f220a7824 in qtbase made the
implicit conversions to QChar through one of the integer types invalid.
quint16_le can cast to quint16 (at a cost) but isn't quint16. Doing an
explicit casting is required now.

Change-Id: I88fdff5cee3898a3aadd15acd467ac2f3e3da79b
Reviewed-by: Thiago Macieira <[email protected]>

We must not delete a QQmlTypeData on the main thread since it interacts
with data on the type loader thread. If the type loader thread is still
alive, we have to drop any QQmlDataBlobs there. Otherwise the type
loader thread may crash when accessing a dangling pointer.

Pick-to: 6.8 6.5 6.2
Task-number: QTBUG-128269
Change-Id: If4cfbd40caef8cc8ac5d96f41dd08631170191c5
Reviewed-by: Fabian Kosmale <[email protected]>

If we remove a value type binding from a proxy binding, the proxy
binding is still there. We must not clear the toplevel binding bit then.

Amends commit c9d7620

Pick-to: 6.8 6.5 6.2
Fixes: QTBUG-130143
Change-Id: Ie4f1ba4a22a1e15eaa218fb8e72f78964ae8cf75
Reviewed-by: Fabian Kosmale <[email protected]>

Downstream tooling may be able to locate the type in some other module.
You can successfully build incompletely linked static libraries as long
as the final dynamic binary contains all the dependencies, after all.

Amends commit ad88383 which
accidentally removed that feature.

Pick-to: 6.8
Fixes: QTBUG-128820
Change-Id: I7c59837ade98a3e3b4005b071957380be7f96628
Reviewed-by: Sami Shalayel <[email protected]>
Reviewed-by: Fabian Kosmale <[email protected]>

Now the text document can change in QQuickTextEdit, we cannot keep
lingering connections to the old document.

Pick-to: 6.8
Task-number: QTBUG-35688
Change-Id: I02ce1eaa561170a2f2bc5150368788df7b0f4769
Reviewed-by: Shawn Rutledge <[email protected]>

APFS and HFS+ are both case preserving, in which case we can trust
that the canonical file path reported above is correct.

But if any of these file systems are mounted via SMB or Virtiofs
(via Virtualization.framework) macOS will report that the canonical
file name of "Foo" is "Foo", even if the underlying file on the host
file system has a canonical name of "foo".

This caused test failures when running tests on a VM with the
build and source directories mounted from the host via Virtiofs
or SMB, as we would pick up date.qml as if it was Date.qml, which
ended up shadowing the built-in JS Date type.

As we can't trust the canonical name in this case, we go though
QDirIterator instead, as the directory listing for the mounted
filesystem _does_ report the correct canonical file name.

As a drive-by optimization we also treat QRC paths as being the
correct file name, just like we do on Windows, as the QRC file
system is case sensitive.

Change-Id: Ic16c1685268afb893887719a0f012d01757e1052
Reviewed-by: Anton Kudryavtsev <[email protected]>
Reviewed-by: Morten Johan Sørvig <[email protected]>

- QQuickTest::showView() calls QCursor::setPos() (when possible) to
  move the mouse outside the window, so that PointHandler won't
  deactivate because of stray mouse events
- call QWindowSystemInterface::handleTabletEvent() with an explicit
  timestamp to ensure valid velocity calculation during delivery of
  the second tablet event

Fixes: QTBUG-129943
Change-Id: Ib89e92195dd7ab41e6409c95c77b291ae98848ea
Reviewed-by: Shawn Rutledge <[email protected]>

The test function causes an accessibility popup on certain macOS
versions.

Skip it, if the cursor can't be moved.
Remove blacklisting for macOS, but add for QNX.

Fixes: QTBUG-129947
Pick-to: 6.8
Change-Id: I63a1277fd097794bdcebffecb71fdba563bb51c5
Reviewed-by: Shawn Rutledge <[email protected]>
Reviewed-by: Thiago Macieira <[email protected]>

uic shouldn't be necessary here and it can fail the cmake generation
phase so just disable it.

Amends baddb27

Change-Id: I140797d0d999e13e58c2c1bd3e4a385e101b98bf
Reviewed-by: Fabian Kosmale <[email protected]>
Reviewed-by: Alexey Edelev <[email protected]>

A follow-up of e4577eb.

This ensures that the user won't inadvertently base the customizations
on top of the native styles.

Pick-to: 6.5 6.8
Change-Id: Ibfeebfdb5ddad377947481deeb8de46976109daa
Reviewed-by: Richard Moe Gustavsen <[email protected]>

Fixes: QTBUG-129850
Pick-to: 6.5 6.8
Change-Id: I4a88b19672479a452facb0f1ca62e4fbb1d50203
Reviewed-by: Richard Moe Gustavsen <[email protected]>