This is similar to and extends the prevention of thread cancellation
introduced by commit ba05af82d3. This
prevents the situation in which a signal gets delivered (usually because
of a crash) and the parent process' handler is run, doing things it
shouldn't between vfork() and execve().
Most C libraries (all that I've investigated) unblock SIGABRT on
abort(), so this doesn't affect them. Likewise, on most OSes, crashes
ignore the signal block and terminate the application -- Darwin appears
to be an exception, but vfork() is not enabled there. Both situations
are tested by terminateInChildProcessModifier().
Task-number: QTBUG-113822
Change-Id: Ib5ce7a497e034ebabb2cfffd17628ca33969b7af
Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org>
Reviewed-by: Volker Hilsheimer <volker.hilsheimer@qt.io>
Simplify the code a little bit.
Change-Id: I7fdb14e676c5b0aa8752b6ee17f0644056ddfcea
Reviewed-by: Timothée Keller <timothee.keller@qt.io>
Reviewed-by: Axel Spoerl <axel.spoerl@qt.io>
Reviewed-by: Santhosh Kumar <santhosh.kumar.selvaraj@qt.io>
Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org>
The 3-arg connect is error-prone and makes the lifetime of the
connection unclear.
Change-Id: I99aa3575a7f901ac52f451f9ef51aa903640d097
Reviewed-by: Volker Hilsheimer <volker.hilsheimer@qt.io>
The 3-arg connect is error-prone and makes the lifetime of the
connection unclear.
Change-Id: I4e518dd4a9733cc0a42b02639cce9c6136ad5afc
Reviewed-by: Volker Hilsheimer <volker.hilsheimer@qt.io>
Replace const char * based connect statements by modern API and remove
'_q_' slot prefixes.
This requires explicit disconnection in the destructor -> add.
Replace bool traps for layouting and removing buttons by enum classes.
Replace if/elseif event handler by switch.
Replace iterator typedef with auto.
Encapsulate logic to ensure defaulting to first accept button in a
member function for better readability.
Remove dead code.
Move private header declaration from cpp to a new _p.h.
Task-number: QTBUG-114377
Pick-to: 6.6 6.5
Change-Id: I8a2bc355e3816c3c826c10cd96194c89bf0ae510
Reviewed-by: Volker Hilsheimer <volker.hilsheimer@qt.io>
There are two temporaries, reply.arguments() returns a temporary QList
and list.at(0) returns a temporary reference to the first element. The
local reference variable would only extend the lifetime of the temporary
object it's bound to, list.at(0), but not the temporary list itself.
Even though this a false positive in this case because QList is
implicilty shared, the compiler can't tell the difference and the fix is
simple.
tests/auto/dbus/qdbusabstractadaptor/tst_qdbusabstractadaptor.cpp:1845:21:
warning: possibly dangling reference to a temporary
[-Wdangling-reference]
1845 | const QVariant &retval = reply.arguments().at(0);
| ^~~~~~
tests/auto/dbus/qdbusabstractadaptor/tst_qdbusabstractadaptor.cpp:1845:50:
note: the temporary was destroyed at the end of the full expression
‘QDBusMessage::arguments() const().QList<QVariant>::at(0)’
1845 | const QVariant &retval = reply.arguments().at(0);
| ~~~~~~~~~~~~~~~~~~~~^~~
Pick-to: 6.6 6.5 5.15
Change-Id: I03d54b56769cbd0f9f1165e4679ec4947267181a
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
So no-one 'fixes' the test by pinning synchronizer.futures() into a
named variable or collapsing the two lines into one. Both would break
the premiss of the test.
Amends e8dcbaaaf6.
Pick-to: 6.6 6.5 6.2
Change-Id: I7a98c382a7db0bf3ff369dcaf61af0942796d6a3
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
Reviewed-by: Mårten Nordheim <marten.nordheim@qt.io>
This fixes a crash while parsing an XML file with garbage data, the file
starts with '<' then garbage data:
- The loop in the parse() keeps iterating until it hits "case 262:",
which calls fastScanName()
- fastScanName() iterates over the text buffer scanning for the
attribute name (e.g. "xml:lang"), until it finds ':'
- Consider a Value val, fastScanName() is called on it, it would set
val.prefix to a number > val.len, then it would hit the 4096 condition
and return (returned 0, now it returns the equivalent of
std::null_opt), which means that val.len doesn't get modified, making
it smaller than val.prefix
- The code would try constructing an XmlStringRef with negative length,
which would hit an assert in one of QStringView's constructors
Add an assert to the XmlStringRef constructor.
Add unittest based on the file from the bug report.
Later on I will replace FastScanNameResult with std::optional<qsizetype>
(std::optional is C++17, which isn't required by Qt 5.15, and we want to
backport this fix).
Credit to OSS-Fuzz.
Fixes: QTBUG-109781
Fixes: QTBUG-114829
Pick-to: 6.6 6.5 6.2 5.15
Change-Id: I455a5eeb47870c2ac9ffd0cbcdcd99c1ae2dd374
Reviewed-by: Allan Sandfeld Jensen <allan.jensen@qt.io>
For easier debugging, e.g. to print out value.len and value.prefix.
Pick-to: 6.6 6.5 6.5.2 6.2 5.15
Change-Id: Ib0eed38772f899502962f578775d34ea2744fdde
Reviewed-by: Marc Mutz <marc.mutz@qt.io>
As requested in code review. Big improvement, code-wise.
tst_Sleep::wait() was failing on the CI, so be more accurate by using
QElapsedTimer::durationElapsed(), which returns nanoseconds.
Pick-to: 6.6
Change-Id: I5bed6d6bd768adfdecab2475e6cbe245c20aabd7
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
The syncqt tool was originally designed to run at build time, as a
part of dependency chain of Qt modules. This works well unless we need
the code model of the Qt project in IDE, since Qt source code actively
uses header aliases, and creating them at build time breaks the code
model until the initial build is done. So we made syncqt the configure
time tool to not break the developer experience.
It's more likely that developers build Qt using command line or don't
need the code model before the first build. So running the tool at
configure time should be optional.
QT_SYNC_HEADERS_AT_CONFIGURE_TIME switches the "mode" of the syncqt
tool from build time tool to the configure time tool. Without the
option enabled build procedure runs all the syncing targets at build
time only. The exception are the developer builds, if the
'-developer-build' option is enabled, QT_SYNC_HEADERS_AT_CONFIGURE_TIME
is set to TRUE by default. This gives better development experience for
the developers that don't use the code model in their IDE or don't
require it before the first build is finished. Also this build time
mode is preferred for the CI or similar build procedures where code
model is not required at all.
By default, the option initialized from the
QT_INTERNAL_CONFIGURE_FROM_IDE CMake variable.
TODO: The option is forced to TRUE for the static Ninja Multi-Config
builds. See QTBUG-113974 for details.
[ChangeLog][Build System] When building Qt from sources, syncqt and Qt
header files are now created at build time, not configure time. This
should speed up the configuration step. You can set the CMake variable
QT_CONFIGURE_TIME_SYNC_HEADERS to ON to use the previous behavior,
though. The old behavior is also preserved if cmake/configure is run
from inside an IDE - Qt Creator, Visual Studio Code, and CLion are
currently detected.
Task-number: QTBUG-111163
Task-number: QTBUG-109792
Pick-to: 6.6
Change-Id: Ib61bda9546e58492be874a8591c37e100313d02c
Reviewed-by: Alexandru Croitor <alexandru.croitor@qt.io>
Vcpkg detection is enabled by default, but we did not have a flag to
disable it, and it was not showing up in config.summary either. By
adding a -vcpkg flag, we get to use `-no-vcpkg` when necessary, as well
as adding an entry to config summary indicating whether vcpkg is in use
or not. Besides `-no-vcpkg`, one can pass `-DQT_USE_VCPKG=OFF` to cmake
command in order to disable the automatic vcpkg detection/integration.
[ChangeLog][configure] vcpkg detection, and integration can be disabled
by passing the -no-vcpkg flag to the configure command, or by passing
`-DQT_USE_VCPKG=OFF` to the cmake command.
Pick-to: 6.6
Change-Id: Ide8da70a7b473ec23995104d162356e75e6d1240
Reviewed-by: Alexandru Croitor <alexandru.croitor@qt.io>
This is dysfunctional since Qt 6.0.0.
Change-Id: I22043868c4114c0ea6e18f3be7e056296b663987
Reviewed-by: Alexey Edelev <alexey.edelev@qt.io>
Reviewed-by: Amir Masoud Abdol <amir.abdol@qt.io>
We were storing the DPR as int; change to qreal.
Task-number: QTBUG-114175
Pick-to: 6.2 6.5 6.6
Change-Id: I7dc7df82f584cddbbb3f690f1df74e7a30369ab2
Reviewed-by: Tor Arne Vestbø <tor.arne.vestbo@qt.io>
When moving a window with keyboard shortcuts, popup menus currently do
not follow along. Allow this to happen by accounting for a window's
transient children, and moving them after the owning window has finished
its move.
Fixes: QTBUG-106483
Pick-to: 6.6 6.5
Change-Id: Id51a7c0163e4fdd2d139565f2bf500a3fc997488
Reviewed-by: Oliver Wolff <oliver.wolff@qt.io>
Just to make sure instanced drawing does not regress. Relevant
particularly with Metal.
Fixes: QTBUG-114885
Change-Id: Ib39066d32985bf25ca02d5aa54d9cf654772be9a
Reviewed-by: Andy Nichols <andy.nichols@qt.io>
The most common limit for the maximum number of allowed documentation
warnings is zero. Use a global value for 'warninglimit', adopted by
all Qt module documentation projects that include the configuration
from qtbase/doc/global.
This allows for a temporary increase of the limit across all modules as
needed - for example, when updating the QDoc binary that the CI
provisions to a version that introduces new types of documentation
warnings.
Increase this base limit temporarily to 10 to help re-enable
documentation testing in CI as it's currently disabled.
Task-number: QTBUG-113326
Change-Id: I8b66951ca9324bcfaec3b5a7ec2cff544c62feb0
Reviewed-by: Paul Wicking <paul.wicking@qt.io>
Map QNetworkError::HostNotFoundError and QNetworkError::ConnectionRefusedError
to ProxyNotFoundError resp. ProxyConnectionRefusedError when it originated
from the communication with the proxy server.
Fixes: QTBUG-68821
Pick-to: 6.5 6.6
Change-Id: I21b91f2667ba0cd329d4ece1fe543472cdab2d22
Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org>
Reviewed-by: Mårten Nordheim <marten.nordheim@qt.io>
The test turns out to be flaky on QEMU.
Fixes: QTBUG-114760
Pick-to: 6.6 6.5
Change-Id: I264f79e9a056e82d4e6735e6ead0710ddabd8eba
Reviewed-by: Volker Hilsheimer <volker.hilsheimer@qt.io>
Basically an adjusted version of findD3dCompiler() that looks for two, not one, DLLs. The logic is the same, with added support for ARM64, though this has not been verified in practice.
Fixes: QTBUG-114789
Change-Id: I1fec51fc98a5146e2770e13cf2f3b160ac4282d6
Reviewed-by: Jonas Karlsson <jonas.karlsson@qt.io>
We have been using as default DH parameters the 1024-bit MODP group.
This is now considered insecure, and applications should use the
2048-bit at a minimum [1]. This commit therefore replaces the parameters
with the 2048-bit MODP group from [2].
To double check the data, use openssl asn1parse to verify that the prime
matches. For instance:
1) put the encoded string in a `encoded.txt` file (c&p from the source,
removing the double quotes)
2) put the hexadecimal value of the 2048-bit group in a `reference.txt`
file (c&p from [2])
3) compare the output of openssl asn1parse with the reference. For
instance like this:
$ diff <(openssl asn1parse < encoded.txt | grep -m 1 INTEGER | perl -pe 's/.*://; s/\n//') <(perl -0777 -pe 's/\s//g' reference.txt) && echo OK
OK
[1] https://datatracker.ietf.org/doc/html/rfc8247#section-2.4
[2] https://datatracker.ietf.org/doc/html/rfc3526#section-3
[ChangeLog][QtNetwork][QSslDiffieHellmanParameters] The default
Diffie-Hellman parameters are now using the 2048-bit MODP group from
RFC 3526.
Pick-to: 6.6 6.5 6.2 5.15
Change-Id: I47133cd78ba0e954b8f93a3da09fa2c760c9f7a8
Reviewed-by: Timur Pocheptsov <timur.pocheptsov@qt.io>
This minimizes any multi-threading / file-locking issues as the file is
closed once the contents are read.
This change assumes /proc/self/mountinfo is available on Linux systems,
and doesn't fallback to setmntent(). It's been around since at least
Linux Kernel 2.4.0.
This requires exporting qstrntoll() for the unittests (using
QT_AUTOTEST_EXPORT and wrapping the those unittests in "#ifdef
QT_BUILD_INTERNAL"), otherwise linking fails.
Fixes: QTBUG-77059
Change-Id: I0363258a9979ea6dadfe5e36c02534ffbd3386c5
Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org>
Reviewed-by: Axel Spoerl <axel.spoerl@qt.io>
Some shortcomings and unexpected problems are not unlikely.
The basic feature, with 2 views going to a texture array's
0 and 1 elements, seems to be working with macOS, also with
MSAA. Instanced drawing has not been verified. (relevant
because layered rendering works via instancing in Metal
and the QRhi backend has to adjust the instance count
in every draw call)
Fixes: QTBUG-114774
Change-Id: I3655e0d2c658b88c4cd6b52a32f94134324e4ac9
Reviewed-by: Andy Nichols <andy.nichols@qt.io>
Just handling attachments with a render buffer set and only one
attachment is not what other backends do. They support all attachments
and also the cases of multisample (2D) textures and multisample texture
arrays.
By extension this allows multisample 2D texture arrays. (GL 3.2+/GLES
3.1+) This was previously not working probably since the correct texture
target was never used.
Pick-to: 6.6
Change-Id: Ibe929faaf86824a596f9794240d1becc51f68e43
Reviewed-by: Andy Nichols <andy.nichols@qt.io>
Follow what got documented in the previous patch.
Task-number: QTBUG-114790
Change-Id: I3e8b3028639c3e4edd39eb0f9446f7eae5f96f3b
Reviewed-by: Andy Nichols <andy.nichols@qt.io>
