The QWeakPointer conversion SMFs cannot actually be used for QObject
payloads, as, for unknown reasons (some comment about vtable this
author doesn't understand), conversion goes through QSharedPointer,
the creation of which throws the checkQObjectShared() warning and
yields a nullptr.
We need to continue to use the QWeakPointer(T*, bool) constructor the
QPointer(T*) ctor also uses.
It's high time we dissociated QPointer from QWeakPointer...
Amends 5f28d367d9.
Fixes: QTBUG-112464
Change-Id: I2f93843af3daf02323d77a4259eaa3745d8de3a8
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
The constructor's documentation already had the note saying that we
could apply this optimization, but it wasn't there and looks like it
never was. This only applies to the default constructor, which sets t1
to zero, which is the epoch. This could be a problem for an OS whose
monotonic clock starts with negative values, but I don't know such an OS
exists and this hasn't been a problem so far.
This commit doesn't change the behavior of the old-style, non-
std::chrono getters, because we always transform any negative remaining
time to 0. It is only possible to observe it by asking the remaining
time or the deadline using the std::chrono API.
Change-Id: I6f518d59e63249ddbf43fffd175a3fc2911e91fb
Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org>
Reviewed-by: Mårten Nordheim <marten.nordheim@qt.io>
We have quite a few Qt API that assumes this, so making this change
helps transitioning them to QDeadlineTimer.
[ChangeLog][Important Behavior Changes] QDeadlineTimer will now
interpret negative millisecond remaining times as "forever", instead of
only the value -1. This brings the API closer in line with other API
like QMutex. This change does not apply to the nanosecond counts in the
API, nor to the API based on std::chrono.
Change-Id: I6f518d59e63249ddbf43fffd175a3e5bead564ae
Reviewed-by: Mårten Nordheim <marten.nordheim@qt.io>
In case the time is of less than 1 ns, we want to make sure it remains
non-zero.
Change-Id: I3e3bfef633af4130a03afffd175d9c4903271f80
Reviewed-by: Mårten Nordheim <marten.nordheim@qt.io>
Reviewed-by: Ahmad Samir <a.samirh78@gmail.com>
Commit b498e1ae3a removed the last
distinction. And since there was no distinction, the code that was
previously under a conditional for CoarseTimer must work for precise
too.
Change-Id: I6f518d59e63249ddbf43fffd175a3eddbd41611a
Reviewed-by: Mårten Nordheim <marten.nordheim@qt.io>
qabstractanimation_p.h is pulled in from qmltc generated code, that
users may decide to build with `QT_NO_KEYWORDS`
Change-Id: I6ce1e82b3c68f6b7e1e349b40945a6f398d8baf5
Pick-to: 6.5
Reviewed-by: Fabian Kosmale <fabian.kosmale@qt.io>
As part of ed2b110b6a the implicit caching
of some of the system locale queries was removed, resulting in a
three orders of magnitude performance regression on converting
dates to strings via QLocale::system() on macOS and iOS.
We now cache the most critical of these, the zero digit, in the Darwin
backend, and clear the cache in response to the system locale changing,
restoring the performance two orders of a magnitude closer to that of Qt 5.
We're still one order of magnitude off, which seems to stem from the
auto release pool in QSystemLocale::query().
Fixes: QTBUG-104785
Pick-to: 6.5 6.2
Change-Id: I26f96147d430f08a721c55a1048d586a4af3a76c
Reviewed-by: Edward Welbourne <edward.welbourne@qt.io>
Add some tests inspired by the initial form of a bug report (before we
found out what the real issue was), that a small fraction with a large
exponent is correctly handled. This should work as long as the result
is representable, even if the fraction itself is too small to be
represented by the floating-point type.
Pick-to: 6.5
Task-number: QTBUG-113443
Change-Id: Ie004197961fc7b603e5024a6ebc5928261a0e2bb
Reviewed-by: Ievgenii Meshcheriakov <ievgenii.meshcheriakov@qt.io>
Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org>
It was blacklisted some years ago, the bug was closed as
cannot reproduce, and no one unblacklisted it.
Pick-to: 6.5 6.2
Change-Id: I51f1fe4d819e0f90bf18c19b67fa0dca198914d4
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org>
Having the size potentially be larger than the capacity can't
be considered anything other than a bug.
Pick-to: 6.5
Change-Id: Id059c0f2c7320f992d897011d7aa944c5cb86058
Reviewed-by: Laszlo Agocs <laszlo.agocs@qt.io>
Support for PRE_TARGETDEPS was added for iOS applications in
53ac8094b1, even if the Xcode
generator doesn't support PRE_TARGETDEPS, by taking advantage
of the glue Makefile we use to run xcodebuild.
And we add our own Qt libraries to PRE_TARGETDEPS in qt.prf,
as you would expect. But since Xcode supports both debug and
release, we always set debug_and_release for this glue Makefile.
The result is that when computing the Qt library PRE_TARGETDEPS,
we fail to apply a _debug suffix from qtPlatformTargetSuffix(),
since we've enabled debug_and_release.
In a debug only build, this means that 'make' of the glue Makefile
will fail to find the release versions of our Qt libraries.
To work around this we skip adding Qt to the target deps when
generating the xcodebuild Makefile, as we know these libraries
are added to the target in the Xcode project.
Pick-to: 6.5 6.2
Change-Id: Icafc103e34a6f83240fa8187181d885fb0172a86
Reviewed-by: Joerg Bornemann <joerg.bornemann@qt.io>
Use qUtf16Printable() to convert QString to somthing consumable by
%ls. Fixes the format/argument mismatch on non-Windows platforms:
tst_qfont.cpp: In member function ‘void tst_QFont::italicOblique()’:
tst_qfont.cpp:153:67: warning: format ‘%ls’ expects argument of type ‘wchar_t*’, but argument 2 has type ‘const ushort*’ {aka ‘const short unsigned int*’} [-Wformat=]
153 | QVERIFY2(f.italic(), qPrintable(QString::asprintf("Failed for font \"%ls\"", f.family().utf16())));
| ^~~~~~~~~~~~~~~~~~~~~~~~~ ~~~~~~~~~~~~~~~~~~
| |
| const ushort* {aka const short unsigned int*}
Amends 4bf82909f1.
Change-Id: I0c7e2dca91a093835d7dba8bff2e5ea78d3a926e
Reviewed-by: Mårten Nordheim <marten.nordheim@qt.io>
This moves the padding that exists on 64-bit architectures from between
the two fields to the end of the the structure. In turns, this allows
certain ABIs to reuse the tail padding space of the base class to store
some information. It can only be used if the functor has alignment of 4
or less, but the interesting case is when it's an empty but final
functor.
The pahole report goes from:
struct CallableObject : QSlotObjectBase {
/* class QSlotObjectBase <ancestor>; */ /* 0 16 */
struct FinalFunctor func; /* 16 0 */
/* XXX last struct has 1 byte of padding */
/* size: 24, cachelines: 1, members: 2 */
/* padding: 8 */
/* paddings: 1, sum paddings: 1 */
} __attribute__((__aligned__(8)));
to (pahole gets very confused by this trick):
struct CallableObject : QSlotObjectBase {
/* class QSlotObjectBase <ancestor>; */ /* 0 16 */
/* XXX last struct has 4 bytes of padding */
/* XXX 65532 bytes hole, try to pack */
struct FinalFunctor func; /* 12 0 */
/* size: 16, cachelines: 1, members: 2 */
/* padding: 4 */
/* paddings: 2, sum paddings: 5 */
/* BRAIN FART ALERT! 16 bytes != 0 (member bytes) + 0 (member bits) + 65532 (byte holes) + 0 (bit holes), diff = -524160 bits */
} __attribute__((__aligned__(8)));
Change-Id: I3e3bfef633af4130a03afffd175d3be98511bae5
Reviewed-by: Volker Hilsheimer <volker.hilsheimer@qt.io>
After the recent changes we only have a single implementation of
QSlotObjectBase, which can handle free functions, member functions,
functors, and lambdas. Rename it to callable, and explicitly hide
the static implementation function so that it doesn't become a symbol
of static libraries using Qt.
Also rename makeSlotObject to makeCallableObject, and polish coding
style and comments in the qobjectdefs_impl header a bit.
Change-Id: Id19107cedfe9c624f807cd8089beb80e9eb99f50
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
QSet's iterator doesn't have operator--, but the
Q_DECLARE_SEQUENTIAL_ITERATOR macro declared and implemented functions
hasPrevious/previous/peekPrevious/findPrevious for the QSetIterator
anyway, depending on that operator. The resulting code couldn't compile.
Use SFINAE to remove the various "previous" functions from Java-style
iterators if operator--() is not present.
This removes the hasPrevious() API as well, even though it could compile
(as we only check whether the iterator points at constBegin). But since
nothing useful can be done with that information, it's best to remove
that member function as well.
[ChangeLog][Potentially source-incompatible changes] QSetIterator no
longer has a hasPrevious() member function. The underlying iterator
doesn't implement operator--(), so couldn't be moved backwards anyway.
Fixes: QTBUG-113379
Change-Id: I47b0ba384d8fcd127123d8fa509cd89e10ea8c99
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
QDoc adds a 'Status' field to the table on top of type reference pages.
One common status for Qt modules/types is 'Technical preview'. Add
an icon that is displayed next to the status description, highlighting
the fact that the type is in tech. preview and therefore subject to
change.
Pick-to: 6.5
Task-number: QTBUG-113026
Change-Id: Ibe6ca2a562cc7810fe27e7dcf514c711cd022894
Reviewed-by: Nicholas Bennett <nicholas.bennett@qt.io>
Reviewed-by: Leena Miettinen <riitta-leena.miettinen@qt.io>
The reason this worked before is unclear. It could be suspected
that we have made a dpi awareness change or Microsoft changed
the behavior of the OpenThemeData function.
Regardless, we expect the result to match the primary display which
OpenThemeData doesn't do (anymore). Instead it returns a value based on
the hwnd screen (which btw didn't always match the widget) and the cache
system would then re-use that theme also for hwnds on other screens.
The most obvious solution is to use OpenThemeDataForDpi to make sure
we get a theme result matching the primary sceen. Then our correction
of the result by with multiplying
QWindowsStylePrivate::nativeMetricScaleFactor(widget)
works again.
This fix does not only fix QMenu sizes. It fixes the size for all
widgets that use this theme function, which could return near
random results before.
We load this library dynamically since MinGW 11.2.0 won't link with it.
[ChangeLog][QWidgets][QMenu] Fixed menu sizes on Windows systems
with more screens.
Fixes: QTBUG-112911
Pick-to: 6.5
Change-Id: I8fdfde2ef5b2aa407cbc74c85afe2c0b74026cff
Reviewed-by: Richard Moe Gustavsen <richard.gustavsen@qt.io>
Reviewed-by: Yuhang Zhao <yuhangzhao@deepin.org>
Reviewed-by: Santhosh Kumar <santhosh.kumar.selvaraj@qt.io>
If the event dispatcher is interrupted we propagate the interrupt to
lower event loop levels, in case they too need to be interrupted. And
we defer the actual interrupt of the NSApplication to the next time we
process Qt events, to avoid AppKit dropping queued events on the floor.
This logic relies on QCocoaEventDispatcher::processEvents() setting the
interrupt flag to false, which signals that we should not continue to
tear down any further event loops.
Unfortunately, native run loops such as running application modal
sessions, are not driven by QCocoaEventDispatcher::processEvents(),
so we never reset the interrupt, and end up ending the session
immediately.
To work around this we need to explicitly clear the interrupt flag
before starting native modal sessions. This also fixes the issue
seen in QTBUG-111524 with showing native alerts from nested event
loops.
Fixes: QTBUG-112697
Task-number: QTBUG-111524
Pick-to: 6.5 6.5.1
Change-Id: I6aaec97011fd18c4a513c1dde3173b1cc4d50112
Reviewed-by: Volker Hilsheimer <volker.hilsheimer@qt.io>
If someone has, for unknown reasons, re-parented the scroll bar of a
scroll area to outside the scroll area, we should not blindly add the
parent widget of the scroll bar as an a11y child of the scroll area.
We don't need to explicitly add the scroll bar itself as a child either,
as that will be handled by whoever is the new parent widget, as a normal
scroll bar would.
Fixes: QTBUG-93768
Pick-to: 6.5 6.2
Change-Id: Ib26f31674602e2221311e864ad31bbf141cad8f6
Reviewed-by: Richard Moe Gustavsen <richard.gustavsen@qt.io>
Similar to the font-features-settings in CSS, this is a low-level
API that allows you to pass the information to the shaper in order
to enable or disable specific font features by name.
[ChangeLog][QtGui][Text] Added an API to QFont which makes it
possible to enable and disable specific typographic features
in OpenType fonts.
Change-Id: Ib48c678f3b97a5a562b08ae34dc895800c8885c0
Reviewed-by: Lars Knoll <lars@knoll.priv.no>
It was new'd and unparented, so its destructor was never getting called.
A stack instance will do just fine.
Pick-to: 6.5
Task-number: QTBUG-111228
Change-Id: If362b0e1e5b50d711a8fe8e01a8830034edf5363
Reviewed-by: Ivan Solovev <ivan.solovev@qt.io>
Reviewed-by: Dennis Oberst <dennis.oberst@qt.io>
Incidentally using reversed video, too, as that's what came naturally
to the terminal I was using. File is smaller, but text in the new
image is considerably more readable and the image is bigger.
The examples guideline calls for an image giving some idea what the
example is about, with a 4:3 or 5:4 aspect ratio. An image shorter
than its width wasn't practical, and getting a useful amount of
information in a height less than 3/2 of the width would have required
widening into blank space purely to satisfy the aspect ratio. The
prior image's ratio was about 12:7, the new one's is roughly 2:3.
The old image showed the start of cbordump -a output, which is rather
less illuminating than the default output of formatted data; it's also
rather verbose, so was cut off part way through. It does, however,
illustrate how the annotated form might be useful for debug
purposes. The image also had a lot of blank space to the right of the
text of interest, wasting roughly half of its width; it could easily
have been cropped to 1:1 without loss of information. As example
data-set, it used ../savegame/save.dat, but without any hint to how to
generate that.
I decided to include a hint to how to generate a CBOR data file to
dump, then show the standard output and only the beginnings of the
annotated output, as the latter (even on its own) would make the
aspect ratio considerably taller. In the context of the standard
display, the preamble of the annotated one gives a reasonable feel for
what's going on, without belabouring the point.
Pick-to: 6.5
Task-number: QTBUG-111228
Change-Id: Ic54893c483607e95627a50db76a0d9918744c07e
Reviewed-by: Mårten Nordheim <marten.nordheim@qt.io>
Without enabling FEATURE_msvc_obj_debug_info ccache does not work
for me with msvc.
So, enable it along with -ccache to make things easier.
Change-Id: Ic6cf6ebddb4a5dce6a04987fba1a1f437b286e90
Reviewed-by: Alexey Edelev <alexey.edelev@qt.io>
Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org>
...optimizing the storage of empty function objects.
Change-Id: I6db7384e1ebb87249d5b93922a6c92f0767cc401
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org>
This is a generalization of QPrivateSlotObject's FunctionStorage.
Changes from the old API:
- uses a more neutral API (s/func/object/)
- preserves cv-qualifiers and value categories
- preserves constexpr'ness
- adds a disambiguation tag (e.g. for use in a compressed_pair).
The main feature is that it transparently uses the empty base-class
optimization to stow away empty classes without allocating separate
bytes for them.
To be used for the public QSlotObject and QGenericRunnable, at least,
so finally pulled the trigger and made it a separate component.
Change-Id: I9a13bbdd1dcf64aaf5fc7b2d40a60050d61800b6
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
While parsing the markdown document, each addition of text and
formats was triggering a relayout of the so far existing document,
leading to quadratic behavior and very bad performance for any
non-trivial markdown document.
Guard the changes while parsing with a begin/endEditBlock to
avoid these intermediate updates.
The performance impact can for example be observed with the
markdown editor in Qt Creator (11+).
Pick-to: 6.5
Change-Id: I5f89441ea41bc3c6281b616f0c5528b49b48e432
Reviewed-by: Shawn Rutledge <shawn.rutledge@qt.io>
The exception for MSVC was probably added due to similar logic in
qlogging.h, but in this case, qAbort never returns also with MSVC.
Change-Id: Ica2155c34d4b32b20af5b1fcb5af35bb3ff6f668
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Reviewed-by: Tor Arne Vestbø <tor.arne.vestbo@qt.io>
Basically, instead of re-creating QTextStreams all the time, create it
once, in main(), and then pass it to print() alongside the int
indentation.
Also fix a hard-coded indentation value that should have been relative
to the caller's indentation level.
Pick-to: 6.5 6.2
Task-number: QTBUG-108857
Change-Id: I811447295c9c3fdef23f61aff31ebe82941eb3b4
Reviewed-by: Edward Welbourne <edward.welbourne@qt.io>
Reviewed-by: Mårten Nordheim <marten.nordheim@qt.io>
Header field names are always considered to be case-insensitive.
Pick-to: 6.5 6.5.1 6.2 5.15
Fixes: QTBUG-113392
Change-Id: Ifb4def4bb7f2ac070416cdc76581a769f1e52b43
Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org>
Reviewed-by: Edward Welbourne <edward.welbourne@qt.io>
Reviewed-by: Volker Hilsheimer <volker.hilsheimer@qt.io>
Polymorphic class QMimeXMLProvider is copyable. Potential slicing.
[-Wclazy-copyable-polymorphic].
Change-Id: I1eff1389140ce9426db32a37df89935a7b3db3a5
Reviewed-by: Giuseppe D'Angelo <giuseppe.dangelo@kdab.com>
Drive-by change: use range-for with a QHash::asKeyValueRange().
Change-Id: I173d28d759bef8f23330338127eeff3018af4a55
Reviewed-by: David Faure <david.faure@kdab.com>
According to the Freedesktop spec[1], a mimetype that has glob-deleteall
overwrites other glob-pattern definitions for a mimetype if it is in a
higher precedence dir, the default order is (from high to low)
~/.local/share/mime, /usr/local/share/mime, /usr/share/mime. Or if the
XDG_DATA_DIRS env var is set, then it takes precedence. The
QMime*ProviderS in m_providers are constructed/stored in that same
order, high to low).
For QMimeXMLProvider, we can just clear the glob patterns associated
with those mimetypes from the lists/maps. For the QMimeBinaryProvider
however, we can't change the binary (mmap'ed) cache file, instead check
mimetype names against the exclusion list before modifying a
QMimeGlobMatchResult.
[1] https://specifications.freedesktop.org/shared-mime-info-spec/latest/ar01s02.html
This test uses XDG_DATA_DIRS so only viable when USE_XDG_DATA_DIRS is
defined.
Fixes: QTBUG-101755
Pick-to: 6.5
Change-Id: Icadbdf1027155296377c5a6ab3be8e41b6668325
Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org>
Reviewed-by: David Faure <david.faure@kdab.com>
[ChangeLog][QtCore][QDebug] Added pretty formatting of C++ <chrono>
durations.
[ChangeLog][QtTest] Added pretty formatting of C++ <chrono> durations
for QCOMPARE expressions.
Change-Id: I3b169860d8bd41e9be6bfffd1757cc087ba957fa
Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org>
Reviewed-by: Volker Hilsheimer <volker.hilsheimer@qt.io>
The frame size is resize border size + padded border size [1], so our
current calculation is not correct, and thus the title bar size is
also wrong.
[1] We also use this calculation in the Windows QPA, and it's also
widely used by many open-source repositories, including Microsoft's
own products.
Change-Id: I5a4c1f91a31a22cade6227131659a6de73bb6808
Reviewed-by: Wladimir Leuschner <wladimir.leuschner@qt.io>
Reviewed-by: Volker Hilsheimer <volker.hilsheimer@qt.io>
This amends commit 115f828ae4
"QTestEventLoop: stop when the test fails" from Qt 6.3 to not attempt to
stop the QTestEventLoop instance if the failure happened outside the
main thread. This is to prevent the QTEL from being created outside the
main thread and thus cause warnings about QObject in addition to and
logged before the real test failure.
[ChangeLog][QtTest][QTestEventLoop] The QTestEventLoop no longer
attempts to exit its event loop if the failure was detected outside the
main thread.
QWARN : tst_QMutex::tryLock_non_recursive() QObject: Cannot create children for a parent that is in a different thread.
(Parent is QCoreApplication(0x7fffffffe710), parent's thread is QThread(0x802e29000), current thread is QThread(0x7fffffffd5a8)
FAIL! : tst_QMutex::tryLock_non_recursive() 'timer.elapsed() >= waitTime - systemTimersResolution' returned FALSE. ()
Change-Id: Ieab617d69f3b4b54ab30fffd175b139322a1f02c
Reviewed-by: Edward Welbourne <edward.welbourne@qt.io>
Reviewed-by: Tor Arne Vestbø <tor.arne.vestbo@qt.io>
Since we don't have different environment variables for the plugin paths,
users have to set QT_PLUGIN_PATH to where plugins for both Qt 5 and 6
(and future versions) are located. This causes Qt to print warnings that
those couldn't be loaded because the major version mismatches. So don't
print them any more.
QT_DEBUG_PLUGINS and the category logging filter can still be used to
enable them.
Fixes: QTBUG-107459
Pick-to: 6.5
Change-Id: Idd5e1bb52be047d7b4fffffd175318ca1f8017bd
Reviewed-by: Lars Knoll <lars@knoll.priv.no>
This commit implements the manufacturer, model, and serialNumber
properties of QScreen for the Windows operating system. These were not
available previously because the Display Devices API allows piecemeal
access to the EDID fields, and the GDI API only returns the i-th device
name, e.g. "\\.\DISPLAY1".
Accessing the EDID of a given screen is possible by, given a
WindowsScreenData instance storing the device path, pivoting from
SetupDiOpenDeviceInterfaceW and then extracting the corresponding blob
from the Registry through a key handle retrieved with
SetupDiOpenDevRegKey. This blob can be parsed just like in Linux with
the QEdidParser class. The resulting metadata is applied to
the WindowsScreenData instance.
Additionally, this commit implements support for clone groups by making
getPathInfo return a list of the matching DISPLAYCONFIG_PATH_INFO
instances, and then concatenating the monitorFriendlyDeviceName
and the EDID manufacturer, model and serialNumber properties.
This commit makes the Windows and Direct2D QPA plugins dependent on
setupapi, and extends the QEdidParser class availability condition to
include these platforms.
Pick-to: 6.5
Change-Id: I56886b035a3d15e6f90aad5d797aeda21f99ff74
Reviewed-by: Oliver Wolff <oliver.wolff@qt.io>
How did this ever work? Q_GLOBAL_STATIC has been a structure for a long
time, but somehow this doesn't work with my latest changes.
qprinterinfo.cpp:73:49: error: operands to ‘?:’ have different types ‘QtGlobalStatic::GlobalImpl<{anonymous}::Q_QGS_shared_null>’
{aka ‘const QtGlobalStatic::Impl<QtGlobalStatic::Holder<{anonymous}::Q_QGS_shared_null, {anonymous}::Q_QGS_shared_null> >’} and ‘QPrinterInfoPrivate*’
Pick-to: 6.5
Change-Id: I6f518d59e63249ddbf43fffd175a264225f0480f
Reviewed-by: Marc Mutz <marc.mutz@qt.io>
MSVC's own headers trigger the C4530 warning just by being included.
C:\Program Files\Microsoft Visual Studio\2022\Professional\VC\Tools\MSVC\14.33.31629\include\chrono(2206):
error C2220: the following warning is treated as an error
warning C4530: C++ exception handler used, but unwind semantics are not enabled. Specify /EHsc
Change-Id: Ieab617d69f3b4b54ab30fffd175c02c23f677d75
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
