Also remove duplication by centralizing the main code for
erase(), and implement erase(pos) in terms of erase(first, last).
Change-Id: Ie0272ebac92fd7da48c31f9d68e69a2faa583bbc
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Receiving an rvalue still requires to check whether the parameter
is detached, otherwise we can't steal its backing std::map.
Change-Id: Ie88dbf39fd777112ad7bb20a46d5c2d65be8eb3d
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
... and QMultiMap as std::multimap.
Just use the implementation from the STL; we can't really claim that
our code is much better than STL's, or does things any differently
(de facto they're both red-black trees).
Decouple QMultiMap from QMap, by making it NOT inherit from
QMap any longer. This completes the deprecation started in 5.15:
QMap now does not store duplicated keys any more.
Something to establish is where to put the
QExplictlySharedDataPointer replcement that is in there as an
ad-hoc solution. There's a number of patches in-flight by Marc
that try to introduce the same (or very similar) functionality.
Miscellanea changes to the Q(Multi)Map code itself:
* consistently use size_type instead of int;
* pass iterators by value;
* drop QT_STRICT_ITERATORS;
* iterators implictly convert to const_iterators, and APIs
take const_iterators;
* iterators are just bidirectional and not random access;
* added noexcept where it makes sense;
* "inline" dropped (churn);
* qMapLessThanKey dropped (undocumented, 0 hits in Qt, 1 hit in KDE);
* operator== on Q(Multi)Map requires operator== on the key type
(we're checking for equality, not equivalence!).
Very few breakages occur in qtbase.
[ChangeLog][Potentially Source-Incompatible Changes] QMap does not
support multiple equivalent keys any more. Any related functionality
has been removed from QMap, following the deprecation that happened
in Qt 5.15. Use QMultiMap for this use case.
[ChangeLog][Potentially Source-Incompatible Changes] QMap and
QMultiMap iterators random-access API have been removed. Note that
the iterators have always been just bidirectional; moving
an iterator by N positions can still be achieved using std::next
or std::advance, at the same cost as before (O(N)).
[ChangeLog][Potentially Source-Incompatible Changes] QMultiMap does
not inherit from QMap any more. Amongst other things, this means
that iterators on a QMultiMap now belong to the QMultiMap class
(and not to the QMap class); new Java iterators have been added.
Change-Id: I5a0fe9b020f92c21b37065a1defff783b5d2b7a9
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
In preparation for the QMap/QMultiMap split. The previous code
had a workaround for storing multimaps, because the CF classes
actually don't support it: build a dictionary from one key
to a _list_ of values. Stop doing that.
In principle, if QMultiMap support does get added to QVariant
(which it probably should), then a similar workaround could be
readded for QMultiMap support.
[ChangeLog][Important Behavior Changes][QSettings] On Apple platforms,
when using the native format, QSettings is no longer able to handle
QVariantMap values which are actually multimaps. Since the native storage
does not actually support multimaps, QSettings used to flatten and
unflatten the maps. However, with QMap being changed to no longer
allow for equivalent keys, flattening when writing does not make
sense any more (there cannot be equivalent keys, because QMap in Qt 6
is a single-key map). Reading existing settings is supported by having
a key in the map mapping to a QVariantList of values.
Support for QMultiMap may be added back to QVariant and QSettings in a
future version of Qt.
Change-Id: Iaa9535100fe5ef55693f22a2068454a84180b4a6
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
The build breaks by disabling ASan in this function because it also
removes its ability to emit SSE2 code. That's clearly broken because all
x86_64 can use SSE2. So this adds #warnings so people are told how to
choose their solution.
Clang doesn't currently define __SANITIZE_ADDRESS__ but I added a
conditional just in case some future version does.
Fixes: QTBUG-84856
Pick-to: 5.15
Change-Id: I552d244076a447ab92d7fffd1617875fdd8dbe62
Reviewed-by: Mitch Curtis <mitch.curtis@qt.io>
This allows us to skip the surrogate pair decoding too, since it can't
match anyway.
Change-Id: Ied637aece2a7427b8a2dfffd16118183e5d76794
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
The implementation was broken. The "high" in PUNPCKHBW's "unpack high
data" means the high 64-bit of the 128-bit, not the high byte of a 16-
bit word. This never worked. It always passed for me because I don't
build non-SSE4.2 code (too old, no longer relevant).
So just use the working version of simdTestMask.
Pick-to: 5.15 5.12
Change-Id: I35a1b4d0a19a43149daefffd16284542f0de3fa3
Reviewed-by: Allan Sandfeld Jensen <allan.jensen@qt.io>
Even added an internal char8_t type if the compiler doesn't have one.
Change-Id: Ied637aece2a7427b8a2dfffd1611813c345d10ec
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
The problem was that QtCore exported those classes deriving from
containers, which caused applications to try and import those same
functions from QtCore, depending on whether they saw the #include or
not.
Now we don't need the hack anymore.
Change-Id: I0a103569c81b4711a649fffd14ec9282454a1fdc
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
The logic is a bit involved in qmake.
The Qt internal qt_common.prf adds CONFIG += strict_c++ which applies
to qt modules, qt plugins, qml plugins, qt helper libs, winmain and
qt_apps, qt_tools, but NOT tests (which is important because the tests
on Windows MinGW fail to build without the GNU extensions).
Then default_post.prf checks for the strict_c++ value and either uses
the strict or non-strict C++ standard flags. default_post.prf is
loaded for all qmake projects, not just the Qt internal ones.
Now CMake doesn't provide a transitive based option to disable C++
GNU extensions with a mechanism similar to target_compile_features.
It only provides the CXX_EXTENSIONS property and it's associated
CMAKE_CXX_EXTENSIONS variable. We can't set the variable at a
directory scope, because that is too coarse grained.
So we rely on setting the property via a function in every relevant
qt_add_<target> function.
Now the naming of the function is weird.
We name the function as qt_internal_<...>, because it's not meant to be
used by Qt users.
We prepend an underscore to the name because we need to place it in
Qt6CoreMacros, so that the function can be called by
qt_add_qml_module which IS a public function.
That's because in Qt5 load(qml_plugin) was private API, but in Qt 6 +
CMake we decided to make qt_add_qml_module() as public API.
Change-Id: Id014626b087d590e25cb46843f93d0c67fc36e44
Reviewed-by: Joerg Bornemann <joerg.bornemann@qt.io>
Eliminate some needless parentheses, tidy up some spacing and
indentation and split some long lines. Change first += after
declaration to initializer.
Change-Id: I05ff2a6337b7ed14e0a2dc9c03fc784c92b63515
Reviewed-by: Konstantin Ritt <ritt.ks@gmail.com>
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Avoids the conversion from UTF-8 for uses that are not dumping a string.
Mass conversion of Qt sources left for future opportunity.
Fixes: QTBUG-85811
Change-Id: I4ca4a35b687b46c39030fffd1626ae6c3294cacf
Reviewed-by: Edward Welbourne <edward.welbourne@qt.io>
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
The operators were declared as friend function templates (so, free
functions in the Qt namespace). Unfortunately, the template argument
of the operators was also defaulted -- causing MSVC trying to
instantiate those functions at all times, causing interesting
recursive template instantiation errors (C2968). It's extremely
likely that we're facing a MSVC bug, but work around it by not
defaulting the template argument. This in turn requires to move
the function definition outside QList's definition, otherwise
an extern template definition (like the ones we have for QList<QPoint>)
would cause a template redefinition error...
Change-Id: If03477ac1fa0a72aa252bb9e08e2a19c2b517b1b
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
After adding QByteArrayView overloads of QByteArray methods that used
to take char*, the way of determining the length of passed fixed-sized
array data has changed, due to the QByteArraiyView's optimization to deduce
the length at compile time for arrays. Document the behavior and add a
changelog item.
[ChangeLog][QtCore][QByteArray] When passing fixed size C arrays to
QByteArray methods, the length of the data on which the method will operate
is determined by array size, and not by scanning for the first '\0'
terminator, as it was in Qt 5.
Task-number: QTBUG-85815
Change-Id: I11cba17d428791e06c9bd4c7a727b7bd6b6aec3c
Reviewed-by: Paul Wicking <paul.wicking@qt.io>
Reviewed-by: Edward Welbourne <edward.welbourne@qt.io>
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
QByteArrayView's comparison operators are declared in the global
namespace, which can collide with the ones declared for other types that
are implicitly convertible to QByteArrayView (see the example attached to
the bugreport). Changing them to be hidden friends will make them visible
only when at least one of the operands is a QByteArrayView, so the
ambiguity will be avoided.
Fixes: QTBUG-85880
Change-Id: Ic3582269d9bda9a2105336ef0f044ea619af37ba
Reviewed-by: Marco Bubke <marco.bubke@qt.io>
Reviewed-by: Mårten Nordheim <marten.nordheim@qt.io>
We can end up in a situation where a (soon to be destroyed) observer is
owned by a binding which is about to be deleted. If in that situation
the binding is destroyed first, we end up with a dangling pointer
and ensuing memory corruption. Instead, we now first transfer the
ownership of the observer and only destroy the binding afterwards.
Fixes: QTBUG-85824
Change-Id: I721c0319281ada981ae7896bd2e02e9a0cc901b8
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
We need them to implement qHash for QQuickPixmapKey.
Change-Id: Ia67de25ec0068b9e1b889bc9c6ee077e24eb71a8
Reviewed-by: Volker Hilsheimer <volker.hilsheimer@qt.io>
Made QPromise::swap public, added free standing swap() for
QFutureInterface and QPromise. Updated QPromise special member
functions. Extended tests
Task-number: QTBUG-84977
Change-Id: I5daf6876df306d082441dbcdf5ae4dee3bfc0ead
Reviewed-by: Sona Kurazyan <sona.kurazyan@qt.io>
We already had append(const QList &) and now there's an overload
taking an rvalue reference.
Change-Id: Id2fbc6c57badebebeee7b80d15bb333270fa4e19
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Commit daaa557259 changed the enum to
bool, but the warning is still happening.
Pick-To: 5.15
Change-Id: Ibdc95e9af7bd456a94ecfffd1603f332a356c788
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
I don't know why std::shared_ptr allows this, but why not.
[ChangeLog][Important Behavior Changes] QSharedPointer objects will now
call custom deleters even when the pointer being tracked was null. This
behavior is the same as std::shared_ptr.
Fixes: QTBUG-85285
Pick-to: 5.15
Change-Id: I24006db8360041f598c5fffd161c260df0313b55
Reviewed-by: Giuseppe D'Angelo <giuseppe.dangelo@kdab.com>
‘t’
Specifies that the argument is a ‘ptrdiff_t’.
This modifier was introduced in ISO C99.
We use qsizetype, which is the same width as ptrdiff_t, so it makes no
difference in va_arg().
Change-Id: Iea47e0f8fc8b40378df7fffd16255730109413a5
Reviewed-by: Tor Arne Vestbø <tor.arne.vestbo@qt.io>
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
Deprecated in 5.15 in favor of CBOR.
Fixes: QTBUG-81239
Change-Id: I711d4bd7dd1247f58e77ac9fa53304cbe5028918
Reviewed-by: Edward Welbourne <edward.welbourne@qt.io>
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
keys() and values() can be slower for large containers.
I ran into this recently when profiling, and was surprised that keys()
had to build the list of keys (by appending in a loop).
Pick-to: 5.15 5.12
Change-Id: I73215f5a917790236704ad7ef78cefc4a049cd89
Reviewed-by: Mårten Nordheim <marten.nordheim@qt.io>
Reviewed-by: Edward Welbourne <edward.welbourne@qt.io>
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
Our hardware optimized conversions of float to qfloat16 rounds to even
where our table based conversion truncated to zero.
The rounding is not in this patch exactly round to even like the
hardware implementation but much closer.
Change-Id: I4c5e72c15fef9079d3660680b2727ff7ba4e768a
Reviewed-by: Edward Welbourne <edward.welbourne@qt.io>
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
This adds the option to build Qt with libb2, which is especially
recommended, if you care about the performance of the BLAKE2 hashing
algorithms. The bundled version is the pure reference C implementation
of BLAKE2, while libb2 has many additional hardware optimizations.
However, the API of both is the same, so no changes to the code were
necessary here.
Change-Id: I3563982f4e07be300291fe103c38b16a404b3ebb
Reviewed-by: Alexandru Croitor <alexandru.croitor@qt.io>
BLAKE2 does not specify requirements about specific hash sizes and
since QCryptographicHash does not support dynamic hash sizes, only the
most common hash sizes could be covered by this.
The supported hash sizes were chosen to match the ones supported by the
Linux kernel.
The new hashing algorithms for QCryptographicHash are:
* BLAKE2b (160 bit, 256 bit, 384 bit, 512 bit)
* BLAKE2s (128 bit, 160 bit, 224 bit, 256 bit)
[ChangeLog][QtCore][QCryptographicHash] Added BLAKE2b and BLAKE2s
hashing algorithms.
Fixes: QTBUG-78198
Change-Id: Id9e0180a974093982fdf1cdd6180988a2e5e9f4f
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
QFileInfoGatherer creates QFileInfo objects in the worker thread to
offload the work from the UI thread, but it never calls any methods on
the QFileInfo objects that would trigger a stat'ing of the files. For
large directories on remote file system, that easily results in the
UI thread being blocked for a very long time.
Add a private 'stat' method to QFileInfo which allows forcing it to
stat all attributes from the worker thread, and make
QFileInfoGatherer a friend so that it can call the function from the
worker thread. This way, QFileSystemModel can access the cached data
for each QFileInfo object, without having to touch the file system
from the UI thread.
Also reduce the amount of signal emissions for drive information,
batch all drives (which can safely be assumed to be at most a two
digit figure) into a single emission instead.
Change-Id: Ifdcae150406187db9984d0fec9add93597b5f85b
Fixes: QTBUG-41373
Pick-to: 5.15
Reviewed-by: Edward Welbourne <edward.welbourne@qt.io>
Reviewed-by: Mårten Nordheim <marten.nordheim@qt.io>
This new function is meant to be used where load(qt_app) is used.
It delegates functionality to qt_add_executable, while handling
some additional behavior via a finalization function (mostly handling
of macOS Info.plist files and icons, as well as Windows icons and
resource files)
It uses a new PlatformAppInternal interface target.
Task-number: QTBUG-85757
Change-Id: I1a2d5851b137fcd4a6323e0e06fb154f91619800
Reviewed-by: Cristian Adam <cristian.adam@qt.io>
We already have the information in the QMetaTypeInterface, and provide
functions to access sizeof. Adding alignof support seems natural, and
should make it easier to handle over-aligned types.
This should also be helpful in QVariant.
Change-Id: I166be76f4b7d2d2e524a3a1e513bd2f361e887c1
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
size() - 1 is converted to int as the result, but size() is a qsizetype
and could be bigger than INT_MAX. So rewrite to not depend on the cast.
This was introduced on b2f79cceb1.
I could have replaced size() - 1 with int(size() > 1) - 1, but that's
even more complex. To simplify, I split the function in two while
retaining the C++11 constexpr requirements.
Bonus: removes the use of the ambiguously-named "empty()" function that
looks like a verb.
Fixes: QTBUG-85665
Pick-to: 5.15
Change-Id: Ieffc3453b88c4517a1dbfffd162338fdb084a376
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
No need to change the output variable from int to qsizetype. That would
complicate the use of libdouble-conversion, which uses ints.
Change-Id: Iea47e0f8fc8b40378df7fffd1624bfdba1189d81
Reviewed-by: Edward Welbourne <edward.welbourne@qt.io>
The docs didn't make clear that there's a case where a rather poor
locale-aware compare may be in use - which turns out to be in use on
Android.
In the process, stop duplicating (in two slightly different variants)
the same paragraph in many places when we can use a \sa to refer to a
sub-section of the class doc.
Pick-to: 5.15
Task-number: QTBUG-63518
Change-Id: I5a41b0a269e477e74236131310992e70462734d7
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Reviewed-by: Topi Reiniö <topi.reinio@qt.io>
The Windows implementation of QFileSystemEngine tries hard to fill the
metadata for a file, even if GetFileAttributesEx returns with error.
This is good in many situations, but when the error code indicates that
the drive on which the file resides has been disconnected, then we
should fail quickly.
Task-number: QTBUG-6039
Pick-to: 5.15
Change-Id: I7574c5a2e524e913306d0b470b4f227416442c13
Reviewed-by: Mårten Nordheim <marten.nordheim@qt.io>
The previous implementation multiplexed callback-based event
notification into a single proxy event (cf. 85403d0af), which was
in turn object-waited for (this was the case since the beginning
of public qt history). It makes more sense to multiplex into a
posted message, because that also works with foreign event loops
that do not know anything about our event objects.
Task-number: QTBUG-64443
Change-Id: I97945ac8b5d7c8582701077134c0aef4f3b5a18f
Reviewed-by: Oswald Buddenhagen <oswald.buddenhagen@gmx.de>
Reviewed-by: Volker Hilsheimer <volker.hilsheimer@qt.io>
Infinite is only when the mantissa is 0, everything else is NaN.
std::isnormal returns false on zero.
Pick-to: 5.15
Change-Id: I897fc0dc3b8a9c557bb1922ea7ca8df501e91859
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
That is not a valid conversion. An empty string is a valid conversion; a
string containing a null should fail.
Change-Id: Iea47e0f8fc8b40378df7fffd1624c088f3bd1b14
Reviewed-by: Edward Welbourne <edward.welbourne@qt.io>
Giuseppe D'Angelo
Thiago Macieira
Tor Arne Vestbø
Alexandru Croitor
Edward Welbourne
Oliver Wolff
Sona Kurazyan
Fabian Kosmale
Shawn Rutledge
Andrei Golubev
Mårten Nordheim
Allan Sandfeld Jensen
Mitch Curtis
Linus Jahn
Volker Hilsheimer
Alex Trotsenko