After f08492c6fd was
merged this bug would manifest as an entry appearing twice
in the chain when a updating an existing entry (insert with
an existing key). This could sometimes result in crashes later
as the list filled up and the list was used in trim() to remove
various entries.
Fixes: QTBUG-89176
Pick-to: 6.0
Change-Id: Ide80160fb4317dc0aefe79eec5dce7ec6813e790
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Historic QDateTime behavior when being asked to create a
QDateTime in the DST gap was to interpret the given date
as if it was in the time before that gap, mapping it to a point
in time after the gap. This has changed with
a04411119e .
Since then, the given date is interpreted as if it was in the
time after the gap, thus being mapped to a point in time
before the gap.
This patch restores the historic behavior.
This was not caught by Coin because machines ran in timezone
"Atlantic/Reykjavik" which does not have DST since 1967.
This patch changes tests to always run in "Europe/Oslo".
Driveby: Test function "findSpring" did some operations in
local time, even though being asked to work in a specific
time zone. Fixed that.
Fixes: QTBUG-86960
Fixes: QTBUG-89208
Pick-to: 6.0 5.15
Change-Id: Iecce5898bf9711a10e7dfc0a25e4bbeaed1c8ade
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Technically, having a single constructor limits the use-cases for this
class. We should take into account that:
- an opened socket descriptor must be available at the moment of
construction;
- the constructor unconditionally enables the notifier (the possible
solution
notifier = new QSocketNotifier(...);
notifier->setEnabled(false);
is suboptimal due to heavy operations inside the event dispatcher);
- for these reasons, QSocketNotifier most often cannot be a member of
another class (we have an extra allocation and indirect access).
This patch addresses this shortcoming by making it possible to set the
socket descriptor at a later point:
[ChangeLog][QtCore][QSocketNotifier] Added setSocket() and an additional
constructor which requires no socket.
Change-Id: I2eb2edf33ddafe99e528aac7d3774ade40795e7a
Reviewed-by: Oswald Buddenhagen <oswald.buddenhagen@gmx.de>
Added overloads of .then()/.onFailed()/.onCanceled() which take a
pointer of a context object, and invoke the continuations in the
object's thread.
Task-number: QTBUG-86794
Change-Id: I0f3cbb0500695673fc4087af5d4b96b416e3e1ce
Reviewed-by: Andrei Golubev <andrei.golubev@qt.io>
Reviewed-by: Mårten Nordheim <marten.nordheim@qt.io>
Reviewed-by: Leena Miettinen <riitta-leena.miettinen@qt.io>
QTableViewPrivate::heightHintForIndex()/widthHintForIndex() did not
honor spans and therefore returned too big values.
Fixes: QTBUG-89116
Change-Id: I52948902b7eaaa27c092ed39da68950c3840e8e4
Pick-to: 5.15
Reviewed-by: Richard Moe Gustavsen <richard.gustavsen@qt.io>
/dev/zero and /dev/null are expected to always be present in any system
(even containers). Unlike /dev/null, you *can* read from /dev/zero so
test that QIODevice doesn't think it is random-access because of that.
/dev/tty is also always present but has an interesting semantic. Could
also try /dev/full, /dev/random and /dev/urandom.
Change-Id: Ia2aa807ffa8a4c798425fffd15d84b60573f2c26
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
Reviewed-by: Oswald Buddenhagen <oswald.buddenhagen@gmx.de>
This is the most basic way to represent custom transfer functions.
Change-Id: I529fb647ece82c03e85ef77b056d9daf13fe5a61
Reviewed-by: Morten Johan Sørvig <morten.sorvig@qt.io>
Reviewed-by: Eirik Aavitsland <eirik.aavitsland@qt.io>
Otherwise, it would report that lastIndexOf of an empty pattern
in an empty string doesn't exist. Next commit adds extensive autotests;
for now, disable a broken autotest (which already features a comment
about why it's broken).
Change-Id: I9a0e5c0142007f81f5cf93e356c8bd82f00066f7
Pick-to: 5.15 6.0
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
There's an assertion. Found by Google fuzz scan of CBOR data.
Pick-to: 6.0 5.15
Change-Id: I55083c2909f64a1f8868fffd164f1ff3af71605b
Reviewed-by: David Faure <david.faure@kdab.com>
There is an off by one in the implementation of count(): a match
must be attempted even at the very end of the string, because
a 0-length match can happen there. While at it, improve
the documentation on the counter-intuitive behavior of count(),
which doesn't merely count how many times a regexp matches
into a string using ordinary global matching.
[ChangeLog][QtCore][QString] Fixed a corner case when using
QString::count(QRegularExpression), causing an empty in the
last position not to be accounted for in the returned result.
Change-Id: I064497839a96979abfbac2d0a96546ce160bbc46
Pick-to: 5.15 6.0
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
test_QFINDTESTDATA is a test which is built and run in the source
tree. It tests usage of relative paths of the Ninja generator, which
only occurs when building in the source tree, so this test cannot
be done outside of the source tree.
Developer's expectation is, however, that an out-of-source build
does not change the source tree. Having "git status" showing
differences after running the tests is irritating. This patch removes
the in-source build files after executing the test.
Change-Id: Ia9fd368c9d54b97a415b63254b45e17bc95ecf45
Reviewed-by: Craig Scott <craig.scott@qt.io>
Reviewed-by: Alexandru Croitor <alexandru.croitor@qt.io>
When an eager binding triggers a setBinding call, we end up with a
special kind of binding loop:
setBinding() -> evaluate -> notifyObserver
^ |
| /
----------------------------
We now catch set condition, and set the binding status to BindingLoop
(with a distinct description).
Task-number: QTBUG-87153
Task-number: QTBUG-87733
Pick-to: 6.0
Change-Id: I9f9915797d82eab820fc279baceaf89d7e5a3f4a
Reviewed-by: Ulf Hermann <ulf.hermann@qt.io>
tst_QLineEdit::QTBUG13520_textNotVisible checks that text is visible
if a QLineEdit is set to Qt::AlignRight. To do that, it writes
some text into a line edit and checks afterwards that the first
character is in the left half of the window. This fails if the window
is larger than twice the length of the text used and thus might fail
in multiple situations where Qt is not in full control over the size
of the windows created, as is the case with tiling window managers.
This patch changes the test to not check for the first character in
the left half of the window, but instead check for the first character
be approximately at the expected position.
Pick-to: 6.0
Change-Id: I18f6de356ea20f4744f3a58cd2b1d76f6a9545a4
Reviewed-by: Ivan Solovev <ivan.solovev@qt.io>
Reviewed-by: Sona Kurazyan <sona.kurazyan@qt.io>
cborlargedatavalidation.cpp:93:60: warning: format ‘%llx’ expects argument of type ‘long long unsigned int’, but argument 2 has type ‘qsizetype’ {aka ‘int’} [-Wformat=]
93 | QTest::addRow("bytearray-too-big-for-qbytearray-%llx", v)
| ~~~^ ~
| | |
| | qsizetype {aka int}
| long long unsigned int
| %x
The cast to size_t is required to make the 64-bit not complain due to
the long vs long long difference.
Pick-to: 5.15 6.0
Change-Id: I00b01c01a66748508ea1fffd164a9add2a2650cf
Reviewed-by: Allan Sandfeld Jensen <allan.jensen@qt.io>
I don't know which of the previous commits caused this: as far as I can
tell, this test should never have passed.
Pick-to: 5.15 6.0
Change-Id: I55083c2909f64a1f8868fffd164e7e8c24eec1f8
Reviewed-by: Volker Hilsheimer <volker.hilsheimer@qt.io>
Because CBOR strings are encoded in UTF-8, it's possible that the string
that won't fit a QString in UTF-16 would still fit QByteArray in UTF-8
(e.g., anything US-ASCII and most Latin text).
The previous solution was an improvement because we used to read into a
QByteArray then convert the QByteArray to QString, thus using 3x the
amount of memory (1x in QByteArray, 2x in QString). The previous commit
skipped the middle allocation and made the regular readString() function
do the decoding either directly on source memory or by reading in small
chunks (16 kB).
Future improvement for Qt 6.1: add readStringChunk(char16_t *, qsizetype)
so we can do the validation / skipping at O(1) memory.
Pick-to: 5.15 6.0
Change-Id: I7b9b97ae9b32412abdc6fffd1645458c655cc566
Reviewed-by: Allan Sandfeld Jensen <allan.jensen@qt.io>
Looking at grafana it rarely fails in dev so unblacklisting it.
Though it is a little more flaky after switching to http 2 by default
because then we only have one channel and more requests end up queued in
the same channel, which will get errored out when the server
disconnects.
Task-number: QTBUG-88943
Change-Id: If5d6335864ce6bbc35f519b2c6d7068e4181afd2
Reviewed-by: Timur Pocheptsov <timur.pocheptsov@qt.io>
headFromHttp hasn't been flaky since 5.14 times according to grafana,
does not fail locally.
Same situation with ioHttpRedirect as above.
ioHttpRedirectMultipartPost has not failed on Windows since october
2019, assumed stable now.
backgroundRequestInterruption no longer exists.
ioPostToHttpFromSocket would fail in debug MSVC builds but was
fixed in 710886fbdd.
Task-number: QTBUG-88943
Change-Id: Ida640179ef15a3452291745e4e94a71a385f57ae
Reviewed-by: Timur Pocheptsov <timur.pocheptsov@qt.io>
The intention was always that you can define properties that do
not require a changed signal. But having to explicitly pass
a nullptr as signal parameter into the macro is ugly, so
use the cool QT_OVERLOADED_MACRO to make it optional.
Pick-to: 6.0
Change-Id: I0ce366d043850f983c968d73c544d89933c48df9
Reviewed-by: Fabian Kosmale <fabian.kosmale@qt.io>
Similar to the previous commit which applied to QCborStreamReader, don't
allocate too much data before checking that the stream actually has that
much.
Pick-to: 5.15 6.0
Fixes: QTBUG-88256
Change-Id: I7b9b97ae9b32412abdc6fffd16454b7568a063ba
Reviewed-by: Allan Sandfeld Jensen <allan.jensen@qt.io>
By calling the internal readStringChunk() function with a QByteArray
pointer, QCborStreamReader::readByteArray() can now avoid allocating the
resulting buffer until the internals have confirmed that there is
sufficient data in the incoming buffer. As a result, we first detect the
EOF condition before we conclude the payload would have been too big for
QByteArray (validation()) test. Meanwhile, the hugeDeviceValidation()
test ends up with a few conditions where it would have copied 1 GB of
data, so limit that too.
We make a choice of reporting OOM vs DataTooLarge only if QByteArray
fails to allocate in the first place (QByteArray::resize() ->
Q_CHECK_PTR -> qBadAlloc, QtCore is always built with exceptions on).
The QCborValue unit test needed a temporary work around until we apply
the same allocation fix (see next commit).
Pick-to: 5.15 6.0
Fixes: QTBUG-88253
Change-Id: I7b9b97ae9b32412abdc6fffd164523eeae49cdfe
Reviewed-by: Allan Sandfeld Jensen <allan.jensen@qt.io>
Previously a default constructor was required for the result type
of mappedReduced() and filteredReduced(), even if a default value
was provided.
This patch fixes the problem.
The issue was in the ResultReporter type, that was calling
QList::resize() to adjust the size of expected reported results.
A default-value parameter was added to the class, so that
a corresponding overload of QList::resize could be invoked.
Task-number: QTBUG-88452
Change-Id: I51113753e314d76aa74d201b5a7e327a6ca75f47
Reviewed-by: Sona Kurazyan <sona.kurazyan@qt.io>
QColors were not premultiplied before being set.
Pick-to: 6.0 5.15 5.12
Change-Id: Id3765b6932a72374ddfd788fae4bb628a4edf0b7
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
Just like it was done in the .pro file.
Pick-to: 6.0
Change-Id: I7def52127f4bab6f0ef490ac7eee2de2da479352
Reviewed-by: Alexandru Croitor <alexandru.croitor@qt.io>
If a QBindable is created from a computed property, it is not possible
to actually set a value or a binding. If we try to do it anyway, we'd
get a crash. Thus we now check whether the function pointer is null
before invoking it.
Pick-to: 6.0
Task-number: QTBUG-87153
Change-Id: I5bedb9080ccf79d9b8166b80d5733d095ed76f8d
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
The hash and equality operators used need to be consistent with
each other. Unfortunately, QPMI::operator==() is not suitable to do
this. So specialize qHashEquals() for QPMI.
Fixes: QTBUG-88966
Change-Id: If5f19a722ae9fc4e78e93537e7ea15726f148768
Reviewed-by: Jarek Kobus <jaroslaw.kobus@qt.io>
Reviewed-by: Volker Hilsheimer <volker.hilsheimer@qt.io>
(cherry picked from commit 83e95956ed58e88b11e2cc3cb61c5beacb7985db)
Reviewed-by: Qt Cherry-pick Bot <cherrypick_bot@qt-project.org>
Amend 53b7cb1bd7, match() is only used
to provoke data races.
Change-Id: Id20b11fedf7f20e74baab15bbb60c995c1a0c794
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
It was previously untested
Task-number: QTBUG-88183
Pick-to: 6.0
Change-Id: Icc59fc632957a75cac8c7f5e2a1aed88a1c9ff9d
Reviewed-by: Timur Pocheptsov <timur.pocheptsov@qt.io>
The test in general is fine, but it was making an assumption that the
first 5 readyRead emissions would never result in the whole message
being received. In certain scenarios with slowdown however it was still
possible that we would receive the whole message after just a few
readyReady emissions. While I didn't check it's most likely due to a
mechanic in the QNetworkReply machinery where we suppress some
emissions if we know there's more data just about to be available.
Task-number: QTBUG-88943
Change-Id: I0cf06edb34d4e86cc8a42c0f1cd7e8c35765f6ee
Reviewed-by: Timur Pocheptsov <timur.pocheptsov@qt.io>
It's not _wrong_ to time out when connecting to something unreachable
(it's just a different way of handling it) so we shouldn't fail when
this happens either.
In local testing (windows) it times out after 8 seconds, so bump
the timer to 10 seconds. On systems where it's faster there'll be
no difference as long as things don't go wrong.
Pick-to: 6.0
Fixes: QTBUG-89089
Change-Id: I8437cf8e4fbecedea2391ed87fdce1213085b964
Reviewed-by: Timur Pocheptsov <timur.pocheptsov@qt.io>
Add test for qRound that covers some edge cases for rounding. Note that
as of right now, this test fails and the docs have been updated to warn
that it should not be depended on for strict correctness.
Change-Id: I1a61bca47abd77855fe7c13ded44e913cc7e8722
Reviewed-by: Volker Hilsheimer <volker.hilsheimer@qt.io>
Which contains a lot of 'macos'/'osx' black-listed auto-tests. They
mostly fail with SecureTransport (on BigSur) because SecTrustEvaluate()
does not like our old certificate. Instead, since SecureTransport
is deprecated anyway and we are not planning to develop it in future,
skip the related auto-test depending on QT_CONFIG(securetransport).
Task-number: QTBUG-88943
Change-Id: I5f6cb7b2d0ea15c445603c1ff3e1700f123c28d1
Reviewed-by: Mårten Nordheim <marten.nordheim@qt.io>
It was blacklisted due to the issue there. The issue is resolved, but
the blacklist stayed.
Task-number: QTBUG-88943
Change-Id: I7d9a660a17c1463dd8b654752ed5787fe5f5af24
Reviewed-by: Timur Pocheptsov <timur.pocheptsov@qt.io>
If the first font in the families list happens to have been
disambiguated because of duplicates, two things went wrong:
1. hasFamily() would return false for the font family, because
it does not disambiguate when checking for the family name and
only checks if the families list contains the exact string.
2. Adding aliases to the full disambiguated string is not supported,
only the family name.
The first issue has been reported separately as QTBUG-89068.
The test failure is fixed by just avoiding the fonts that
are ambiguous in the test, as it really doesn't matter which
font we pick.
Fixes: QTBUG-89008
Pick-to: 5.15 6.0
Change-Id: I829778c2e7bb6090475c34dcf9cdce58862729d6
Reviewed-by: Liang Qi <liang.qi@qt.io>
Use a trick similar to the one we use for their ranged
constructors: support predicates that either take a
container's iterator, or that take a std::pair (for STL
compatibility).
[ChangeLog][QtCore][QMap] Added removeIf() and erase_if().
[ChangeLog][QtCore][QMultiMap] Added removeIf() and erase_if().
[ChangeLog][QtCore][QHash] Added removeIf() and erase_if().
[ChangeLog][QtCore][QMultiHash] Added removeIf() and erase_if().
Change-Id: Ie40aadf6217d7a4126a626c390d530812ebcf020
Reviewed-by: Andrei Golubev <andrei.golubev@qt.io>
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
The requests will remove themselves once they get deleted
but since the deletion is done through a _queued_ invokeMethod
to 'deleteLater' we will call
QHttpNetworkConnection::_q_startNextRequest first which may
end up starting a reconnect of the TCP socket which we had the error on.
In this specific instance it manifested as a race condition where we
either don't get a proxyAuthorizationRequired signal at all (it was
emitted while we didn't have any valid replies), or we get the signal
emitted too late and it gets emitted on whatever the next reply was.
Task-number: QTBUG-88417
Pick-to: 5.15 6.0
Change-Id: If3f8ececc5550f1868c90124559cb8e3029646d8
Reviewed-by: Timur Pocheptsov <timur.pocheptsov@qt.io>
We already have all we need in QHash to support this, so the addition
is simple enough.
Add test checking how many copies and/or moves are needed for a single
insert.
As a drive-by: remove some unneeded static_cast
Change-Id: Iaf768657644afa45f78f5c81ffcf89ba9607be96
Reviewed-by: Lars Knoll <lars.knoll@qt.io>
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
