If adding a QLineEdit to a QGraphicsScene as a QGraphicsProxyWidget, the popup
completion of that QLineEdit is drawn in the wrong location. When the completer
getting the rect of screen, it gets the rect of the QGraphicsScene where
QLineEdit is located rather than the rect of the screen, resulting in an error
in the following calculation.
Note that as long as the completer popup is a toplevel widget not parented
to the target widget, it will not be automatically embedded into the graphics
view via QGraphicsProxyWidget. So with multiple views for a scene, or in views
that use transformations, the geometry will still be off.
Pick-to: 6.3
Task-number: QTBUG-20531
Change-Id: If5d8a707ca35a9e4709117b077978145c6143e46
Reviewed-by: Volker Hilsheimer <volker.hilsheimer@qt.io>
tst_QFutureWatcher was deactivated for Android. This patch activates it.
Fixes: QTBUG-88136
Change-Id: Iead82e22d73eb15c9ecd2756eb33925910bbffc0
Reviewed-by: Assam Boudjelthia <assam.boudjelthia@qt.io>
Reviewed-by: Sona Kurazyan <sona.kurazyan@qt.io>
- current INTEGRITY development pack don't support denormals for float and double.
All values are rounded to 0.
Task-number: QTBUG-99123
Pick-to: 6.2 6.3
Change-Id: Iaaacdc4210c7ac2ec3ec337c61164a1ade0efb01
Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org>
Reviewed-by: Giuseppe D'Angelo <giuseppe.dangelo@kdab.com>
Reviewed-by: Edward Welbourne <edward.welbourne@qt.io>
... which implements or assumes something about the
broken^Wnon-STL-compliant insertion behavior.
Once this has integrated into all module dependencies, we can
re-implement these APIs using STL-compatible semantics.
Task-number: QTBUG-100092
Change-Id: I54f4f5ce7addd9543866d2c399f48aff50983b88
Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org>
Reviewed-by: Mårten Nordheim <marten.nordheim@qt.io>
Reviewed-by: Fabian Kosmale <fabian.kosmale@qt.io>
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 macOS standard library doesn't have std::contiguous_iterator yet, and
it doesn't seem like libc++ has it either.
Checking __cpp_lib_concepts for the C++20 official version appears to work.
Pick-to: 6.3 6.2
Change-Id: I8c31cd64de24c03b3a3f37cb393bb2f9b55a834d
Reviewed-by: Marc Mutz <marc.mutz@qt.io>
Remove Integrity and Android specific code that explicitly adds
test data to the resource files. qt_internal_add_test functions
implicitly adds test data to resources for Android and Integrity
platforms by default.
Change-Id: Ia1d58755b47442e1953462e38606f70fec262368
Reviewed-by: Assam Boudjelthia <assam.boudjelthia@qt.io>
Reviewed-by: Alexandru Croitor <alexandru.croitor@qt.io>
Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org>
Multiple tests use qt_internal_add_resource that copies the
functionality that is already implemented inside the
qt_internal_add_test function. Simplify these test by replacing
the qt_internal_add_resource call with the new BUILTIN_TESTDATA
option.
Change-Id: I18475b817d6f87264f0de53817d6c26c5ccab4e2
Reviewed-by: Alexandru Croitor <alexandru.croitor@qt.io>
Size a QMdiSubWindow is no real toplevel widget, QLayout::activate() did
not properly set the minimum size based on it's children. Fix this by
treating a QMdiSubWindow as a toplevel widget during the calculation.
Fixes: QTBUG-100494
Change-Id: Ia2e6c519c7214c36383facd244711bd932231d40
Reviewed-by: Volker Hilsheimer <volker.hilsheimer@qt.io>
If one calls setPlainText("") before setting a placeholder text,
the placeholder visibility is not updated, and the placeholder is not
visible. Fix it by updating placeholderVisible properly.
Fixes: QTBUG-96212
Pick-to: 5.15 6.2 6.3
Change-Id: I1bd3f0cb4c59973a847bcf3787e35d7c17b6d673
Reviewed-by: Volker Hilsheimer <volker.hilsheimer@qt.io>
In one case, cast an int to qsizetype rather than the other way round,
in order to compare them.
Change-Id: Id8dffe61f9565040f1a5ee24867956397f814a55
Reviewed-by: Marc Mutz <marc.mutz@qt.io>
Reviewed-by: Andy Shaw <andy.shaw@qt.io>
Some of the else if lines exceeded 100 characters, so should have been
split; and some lines already were split. So each whole chain should
have had braces on its bodies. Instead make it a switch, as this makes
it more evident what's going on in any case. Furthermore, as each
branch did the same thing with dbType-specific strings, change to just
setting QLatin1String variables, so that the rest of the code needn't
be duplicated in each branch; it can simply be done once after the
switch, using the string's .arg() to embed fragments. In the process
break up the SQL query strings more gracefully, purge spaces just
inside C++ parentheses.
Change-Id: Ie26166e098ad74720bb6d7c4d9fe47718c33a13c
Reviewed-by: Marc Mutz <marc.mutz@qt.io>
Telling us the condition tested should have been true is what
QVERIFY() does anyway, so don't go to the whole trouble of saying the
same thing - and wrapping it in in a QString() merely in order to then
qPrintable() it back out again, pointlessly converting a C-string to
unicode and back again.
At the same time, skip one other qPrintable(QString("...")) without
.arg() formatting; and change the check it's the message for to use
QL1S::arg() instead of QString::arg().
Change-Id: Ie71a79da8017916d301a38b69fc422e55a5a3649
Reviewed-by: Marc Mutz <marc.mutz@qt.io>
Expected warnings are clutter for anyone reding the output, so
suppress them when they're correct behavior.
Change-Id: Idf47ba4fab8069237067d9b20afef3041e0c2f8c
Reviewed-by: Marc Mutz <marc.mutz@qt.io>
Doing the tidy-up after the last check has the problem that it gets
skipped if any check fails, as the premature return skips the tidy-up.
In the process, added the missing tidy-up to prematureExec().
Repackage the code for one tidy-up as a named lambda so that the
duplicate can share it with the scope guard. Made some existing scope
guards const, while I was about it, so that the one that isn't -
because it gets dismiss()ed - stand out.
Change-Id: I96d6834d5d7675f15018169a7093b0211db6f8a9
Reviewed-by: Marc Mutz <marc.mutz@qt.io>
Raster pipeline is not supported on webOS OSE: trying to use it causes
an exit(1).
Fixes: QTBUG-100654
Pick-to: 6.3
Change-Id: I00325fc1330a2d0d4abfdee054343ecfac767309
Reviewed-by: Laszlo Agocs <laszlo.agocs@qt.io>
[ChangeLog][Potentially Source-Incompatible Changes] The qtextstream
header no longer includes <QString>, <QStringEncoder> and
<QStringDecoder>. Code which relied on the implicit inclusion of those
classes might now need to include the headers explicitly.
Task-number: QTBUG-97601
Change-Id: Ifb8c8452026195a772c0588dbbbc53fb51cac548
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
It only stores one layer of state, and pausing twice in a row will just
overwrite the previous state. This doesn't happen often but can happen,
especially on Windows if a certificate needs to be looked up in the system
certificate stores (socket gets paused) and then a recoverable error
occurs in QNAM (socket gets paused again).
Fixes: QTBUG-100362
Fixes: QTBUG-63196
Fixes: QTBUG-98476
Pick-to: 6.3 6.2 5.15
Change-Id: Ie524c48e11b6fa8010b78cc1bf3931efe2ce3351
Reviewed-by: Edward Welbourne <edward.welbourne@qt.io>
Reviewed-by: Timur Pocheptsov <timur.pocheptsov@qt.io>
No need for a virtual do-nothing destructor in a class that isn't
inherited from. Defining an empty init merely makes QTest go to the
trouble of calling it, which it could just as well skip.
Change-Id: Ifd44c473c05fdeaaa7923d2ccdd0a13c8921b6bd
Reviewed-by: Marc Mutz <marc.mutz@qt.io>
Reviewed-by: Andy Shaw <andy.shaw@qt.io>
The offset fields for tst_QTimeZone::specificTransition() had a
comment claiming they are in minutes; they are in fact in seconds, so
fix that. At the same time, Moscow hasn't had a time-zone change since
2017, so the end-date for one of the test intervals can be nudged a
little closer to the present without harm.
Change-Id: I66822cb758f7e00d6added801466a6745be3e31a
Reviewed-by: Marc Mutz <marc.mutz@qt.io>
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Reviewed-by: Mårten Nordheim <marten.nordheim@qt.io>
Unless "." (or the empty string) is in $PATH, we're not supposed to find
executables in the current directory. This is how the Unix shells behave
and we match their behavior. It's also the behavior Qt had prior to 5.9
(commit 28666d167a). On Windows, searching
the current directory is the norm, so we keep that behavior.
This commit does not add an explicit check for an empty return from
QStandardPaths::findExecutable(). Instead, we allow that empty string to
go all the way to execve(2), which will fail with ENOENT. We could catch
it early, before fork(2), but why add code for the error case?
See https://kde.org/info/security/advisory-20220131-1.txt
[ChangeLog][Important Behavior Changes] When passed a simple program
name with no slashes, QProcess on Unix systems will now only search the
current directory if "." is one of the entries in the PATH environment
variable. This bug fix restores the behavior QProcess had before Qt 5.9.
If launching an executable in the directory set by setWorkingDirectory()
or inherited from the parent is intended, pass a program name starting
with "./". For more information and best practices about finding an
executable, see QProcess' documentation.
Pick-to: 5.15 6.2 6.3
Change-Id: I54f205f6b7314351b078fffd16cf7013c97ee9fb
Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org>
Reviewed-by: Mårten Nordheim <marten.nordheim@qt.io>
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
In C++20, any given relational operator is also considered in its
reversed form, so e.g.
given op==(X, Y)
and X x, Y y, then y == x will compile, by using the reversed op(X, Y)
This, unfortunately, makes some existing asymmetric operator overload
sets ambiguous, and instead of applying tie-breaker rules, at least
Clang is warning about these.
For us, this means we need to make our overload set non-ambiguous. The
QJsonValue{,Ref} classes failed this, because they only provide the
following member-operators:
- QJsonValue::op==(const QJsonValue&) const
- QJsonValueRef::op==(const QJsonValue &) const
For member functions, there are no implicit conversions on the LHS. So
in C++17, we have a nice dichotomous overload set:
- LHS is QJsonValue -> use QJsonValue::op==(QJsonValue)
- LHS is QJsonValueRef -> use QJsonValueRef::op==(QJsonValue)
In both of these, it the RHS is a QJsonValueRef, it's implicitly
converted to QJsonValue for the call.
Enter C++20, and the reversed signatures are suddenly available, too,
which is a problem for QJsonValueRef <> QJsonValueRef, which could be
resolved, as in C++17, using
lhs.QJVR::op==(QJV(rhs))
or it could now be
rhs.QJVR::op==(QJV(lhs)); // reversed
Says Clang 10:
tst_qtjson.cpp:990:5: warning: ISO C++20 considers use of overloaded operator '==' (with operand types 'QJsonValueRef' and 'QJsonValueRef') to be ambiguous despite there being a unique best viable function [-Wambiguous-reversed-operator]
CHECK(r0, a0, r1);
^ ~~ ~~
qjsonvalue.h:189:17: note: ambiguity is between a regular call to this operator and a call with the argument order reversed
inline bool operator==(const QJsonValue &other) const { return toValue() == other; }
^
A similar argument makes op!= ambiguous.
Says Clang 10:
tst_qtjson.cpp:988:5: error: use of overloaded operator '!=' is ambiguous (with operand types 'QJsonValueRef' and 'QJsonValueRef')
CHECK(r0, r0, r1);
^ ~~ ~~
qjsonvalue.h:190:17: note: candidate function
inline bool operator!=(const QJsonValue &other) const { return toValue() != other; }
^
qjsonvalue.h:189:17: note: candidate function
inline bool operator==(const QJsonValue &other) const { return toValue() == other; }
^
qjsonvalue.h:189:17: note: candidate function (with reversed parameter order)
To fix, provide the missing operators as free inline functions (so Qt
6.2 and 5.15 don't get new symbols added) so there's always exactly
one best match.
This is a fix for 6.2 and 5.15. At the time of writing, 6.3 isn't
released, yet, so there, we could QT_REMOVED_SINCE the pre-existing
member operators in favor of hidden friends (as per QTBUG-87973).
Use C++17'isms to prevent an automatic merge to 5.15, which requires
contains(QT_CONFIG,c++2a):CONFIG += c++2a
added to tst_qtjson.pro.
[ChangeLog][QtCore][QJsonValue] Fixed relational operators to not
cause warnings/ambiguities when compiling in C++20.
Pick-to: 6.3 6.2 5.15
Change-Id: Ic70f3cad9987c87f7346d426c29cc2079d85ad13
Reviewed-by: Allan Sandfeld Jensen <allan.jensen@qt.io>
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
Amends d42cfeb84f, which would result in
infinite recursion when the Edit menu was populated and added to the
menubar after the menubar has been shown.
Add a macOS-only test case that reproduces the crash without the fix.
Fixes: QTBUG-100441
Pick-to: 6.3
Change-Id: I018a7aa7f01558a3b9732b4d6d96a911dc7fbd19
Reviewed-by: Tor Arne Vestbø <tor.arne.vestbo@qt.io>
The value of __cplusplus has nothing to do with whether the library
implements wg21.link/P1115 (libstdc++ even before C++20) or not
(libc++, even in C++20).
Use the idiomatic check (#if defined(foo) && foo >= x) instead,
fixing the Android build.
Pick-to: 6.3 6.2
Change-Id: I11bcefe455a1f13865c15d4beecbd3fe32115328
Reviewed-by: Marc Mutz <marc.mutz@qt.io>
moc, qmake, rcc, and uic tests execute corresponding tools inside
the test executables. This adds dependencies explicitly, to execute
tests using _check targets, without the need to manually build
dependencies.
Pick-to: 6.3 6.2
Fixes: QTBUG-98576
Change-Id: I8b989276e21199b867591f7b0f6af9204695ac13
Reviewed-by: Alexandru Croitor <alexandru.croitor@qt.io>
Reviewed-by: Marc Mutz <marc.mutz@qt.io>
CI tests for QNX are run in QEMU which does not have enough RAM
to run this test successfully.
Pick-to: 6.2 6.3
Change-Id: Idab3e8b6a1e0ae3eddf5aedb82e6784a74ae9a3a
Reviewed-by: Volker Hilsheimer <volker.hilsheimer@qt.io>
Reviewed-by: Mårten Nordheim <marten.nordheim@qt.io>
There is a TODO for this in Qt Quick from the 6.0 times. To decide
upfront if Metal can be expected to function, or if a fallback to
OpenGL needs to be triggered (especially important with macOS virtual
machines, where, unlike any real macOS system, Metal may not be
present at all), the scenegraph calls create() and then drops the
result. The idea to make this less wasteful was back then to add a
dedicated probing function which can, possibly, perform the checks in
a more lightweight manner than full initialization. Implement this
now, focusing on Metal.
Brought to attention by QTBUG-100441: printing warnings about not
having an MTLDevice is confusing in a Metal-less macOS VM, because it
is not an actual error, only part of the probing at scenegraph
initialization. We can now avoid printing confusing warnings there.
Change-Id: Ie52c36af9224bedc3f5e4c23edb486d961c9f216
Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org>
Reviewed-by: Andy Nichols <andy.nichols@qt.io>
The auto-repeat timer of QScrollBar kicks in after 50ms, so if the test
takes too long and the timer fires during the qWait call or during event
delivery, then we might get multiple valueChanged emissions.
Detect that scenario and allow the test to XFAIL if things take
a significant time (more than 40ms vs the expected 1ms).
Pick-to: 6.3 6.2
Change-Id: Ie90aadc62372397db695fd2b34fe1f5252ce8d37
Reviewed-by: Shawn Rutledge <shawn.rutledge@qt.io>
As with QHoverEvent, it's better to require globalPos rather than
"initialized to QCursor::pos(), which may not be appropriate" as the
docs have pointed out for many years now. This removes the remaining
calls to QCursor::pos() in event constructors.
Task-number: QTBUG-52430
Task-number: QTBUG-69433
Task-number: QTBUG-100324
Change-Id: I076dae56f37abaad7085cc95dddee453a80a45f3
Reviewed-by: Volker Hilsheimer <volker.hilsheimer@qt.io>
Use alternative method for finding searchPath since
QT_TESTCASE_SOURCEDIR did not point to a correct location on webOS
QEMU test environment.
Fixes: QTBUG-99954
Pick-to: 6.3
Change-Id: I05f95b4aa02027976036d0842ca564a602e01d0e
Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org>
Reviewed-by: Sona Kurazyan <sona.kurazyan@qt.io>
When QCborValue referred to an empty array or map, toArray() and toMap()
would respectively return the default value instead of the empty object,
as expected.
Pick-to: 6.2 6.3
Change-Id: I5e52dc5b093c43a3b678fffd16b60456d0037ad7
Reviewed-by: Edward Welbourne <edward.welbourne@qt.io>
Switched to use runtime path instead of compile time path for the
helper executable. Some tests cases were failing on webOS because of
wrong path.
Fixes: QTBUG-99846
Pick-to: 6.3
Change-Id: I566bc04bdb96ac6e7dd0a875eadb50685aef8282
Reviewed-by: Edward Welbourne <edward.welbourne@qt.io>
It leaves the system in a state that breaks at least one later test,
so put it last. (This was first seen when picking back to 6.2, but I
am now able to reproduce it on dev.)
Amends commit 78eac57f3d
Pick-to: 6.3
Change-Id: I918cf43cdfc27357329a175518d6f9755747bae5
Reviewed-by: Andy Shaw <andy.shaw@qt.io>
Reviewed-by: Volker Hilsheimer <volker.hilsheimer@qt.io>
The QSinglePointEvent ctor assumes that the given globalPos is correct,
so it was wrong that the QHoverEvent ctor passed along a local position
as global. It's better to require globalPos as an argument; and in fact
it seems that everywhere we construct a QHoverEvent, global position is
available, or possible to get by transformation (which is better than
resorting to QCursor::pos()).
Also, don't convert to QPoint: pointer events have qreal resolution and
there's no reason to truncate them.
Fixes: QTBUG-100324
Change-Id: I919455da36265988d3d149eb97563c9ed0d2c660
Pick-to: 6.3
Reviewed-by: Volker Hilsheimer <volker.hilsheimer@qt.io>
On X11 we already had to wait as window managers might scroll windows
in. On macOS we also waited - albeit a bit shorter - also because
screens scroll in slowly.
We didn't wait on Windows, and on Windows the test is quite flaky.
Attempting to make it run stably by waiting for long enough so that all
paint events are flushed to the screen.
As a drive-by, break some overly long lines.
Task-number: QTBUG-100412
Pick-to: 6.3
Change-Id: I49b1e447aced2fe1af3c5d30b514f8df3cc1813c
Reviewed-by: Tor Arne Vestbø <tor.arne.vestbo@qt.io>
A default-constructed QPluginLoader erroneously reports that the
load hints are empty. However, setting a filename would then
automatically set the PreventUnload hint, surprising the user.
Return the correct flags instead.
Amends 494376f980
Change-Id: I7a95964cb680afd3adf2f71ed73d2f93023238f2
Fixes: QTBUG-100416
Pick-to: 5.15 6.2 6.3
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
When running a test under increased verbosity levels, QTestLib would
print "failure location" for every QTest function (QVERIFY, QCOMPARE,
...), even if there's no failure at all.
Keep the code centralized, but split the formatting of failures and
non-failures (other messages).
Task-number: QTBUG-96844
Change-Id: I3c508653176b68579dc0eb0cffcc153a52da2e2c
Pick-to: 5.15 6.2 6.3
Reviewed-by: Edward Welbourne <edward.welbourne@qt.io>
Reviewed-by: Volker Hilsheimer <volker.hilsheimer@qt.io>
While investigating an assertion failure I noticed that the existing
tests didn't even exercise these methods for local time or zone time.
Of course, we can't robustly test these time-specs, due to vagueries
of offset details and zone availability, but we can at least verify
that they return date-times on the specified date. Add a test-case for
the start of 1900, on which the assertions were first seen; it is the
earliest moment representable with tm_year >= 0, after all.
One of these tests fails on 6.2 but the fix for that (as opposed to
the the assertion) requires 6.3's improvements to the handling of
time_t's fuller range - too risky a change to pick back to 6.2.
Pick-to: 6.3
Task-number: QTBUG-99747
Change-Id: I98f5d7850a701972b2d8ea2ce203a2b3e7071354
Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org>
Reviewed-by: Marc Mutz <marc.mutz@qt.io>
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
qt_internal_undefine_global_definition disables an internal global
definition that is defined by the qt_internal_add_global_definition
function for a specific target.
Remove the ability to set the custom "undefine" flag for the
definitions since it's hard to control it using the introduced
function.
Pick-to: 6.2 6.3
Task-number: QTBUG-100334
Change-Id: Ic1637d97aa51bbdd06c5b191c57a941aa208d4dc
Reviewed-by: Alexandru Croitor <alexandru.croitor@qt.io>
Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org>
Not MSVC, but pretends to be.
Pick-to: 6.3 6.2
Change-Id: I5c201dd917e79a22d6ef15715bf1d3a7010d123e
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
In 63eea5e5c69434871eaef9d9dc7184d7e54d7276, the
Qt::ItemNeverHasChildren flag was introduced. QFileSystemModel
was modified to use this flag for items which are not directories,
but only if the QFileSystemModel is not read-only and
the directory is writable.
This patch modifies QFileSystemModel to use the ItemNeverHasChildren
flags also if the model is read-only and if the item
is read-only.
Amends 63eea5e5c69434871eaef9d9dc7184d7e54d7276
Change-Id: Ie7f7d58ecf7baade93f9f03d120da84d3c005d42
Reviewed-by: Thorbjørn Lund Martsum <tmartsum@gmail.com>
Amends e3b2b12a91, which made the
style sheet calculate the font-dependent text size, but didn't
add the space needed for the close button or a tab icon.
QTabBar's layout code already adds the space needed for icons
and margins to the size; so instead of overwriting that size,
subtract the size needed by the normal font, and then add the
size needed for the font form the style rule, considering both
vertical and horizontal tab bars.
Fix the test case to style only one tab rather than the entire
tab bar, otherwise the font is applied to the entire tab bar,
and not just to a specific tab, which is the style sheet style
code in question.
Pick-to: 6.2 6.3
Task-number: QTBUG-6905
Change-Id: Ieed0ba146a32e81229419adecaf41f467cfd5959
Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org>
Reviewed-by: Richard Moe Gustavsen <richard.gustavsen@qt.io>
Task-number: QTBUG-100362
Change-Id: Ib09aec5ea03867163d476a2c92562bc949fc1b1a
Reviewed-by: André de la Rocha <andre.rocha@qt.io>
Reviewed-by: Mårten Nordheim <marten.nordheim@qt.io>
The code was trying to avoid a detach in the case no element needed to
be removed, by first running find_if() on const_iterators, and then,
after converting its result to (mutable) iterators, start the
remove_if() algorithm where find_if() left off.
But this applies the predicate to the element found by find_if() (if
any) _twice_: first just before we exit the first find_if() and then
just as we enter remove_if(), which will start by running find_if()
again, with the result of the initial find_if as 'first'.
Apart from being needlessly inefficient, this violates the
specification of Uniform Erasure, which defines sequential erase_if()
as being equivalent to remove_if() + container erase(), with the
former being specified to apply the predicate exactly once per
element.
Fix by writing the remove_if() part by hand.
Instead of doing the dance with the loop invariant documentation
twice, simply implement erase() via erase_if() (complicated a bit by
the weird passing of predicates by lvalue reference instead of by
value, as would be idiomatic). This exposes users to:
[ChangeLog][QtCore][Potentially Source-Incompatible Changes] A fix in
the implementation of the erase-like algorithms of sequential Qt
container may re-enable signed/unsigned comparison warnings previously
suppressed by having occurred in std library code. To fix, cast the
value to look for such that it has the same signedness as the
container's elements.
... but the issue would be the same had we inlined std::remove()
instead of passing a lambda to sequential_erase_if(), so it's nothing
we can, nor should, work around.
[ChangeLog][QtCore][Containers] Fixed a bug in the implementation of
most sequential Qt container's erase-like algorithms (member
removeAll()/removeIf() and free erase()/erase_if()) where the equality
operator or the predicate, respectively, was applied to the first
matching element twice. Each element is now tested exactly once.
Pick-to: 6.3 6.2
Change-Id: Ib6d24b01b40866c125406f1cd6042d4cd083ea0d
Reviewed-by: Giuseppe D'Angelo <giuseppe.dangelo@kdab.com>
