So we can more easily get any errors from attempting to write the file.
It is possible to get them with QFile, by either doing .flush() or using
QIODevice::Unbuffered, but using the C API is a definite sure way. Plus,
since this is QFileSystemEngine, this avoids the possibility that QFile
may choose to use a different file engine than the native one, for some
reason. And it reduces overhead.
This allows us to more easily detect why the file creation failed and
therefore stop looping if the error wasn't EEXIST. That will avoid an
infinite loop in case the necessary directories exist but aren't
writable.
It's also moved above the renaming, such that the failure to populate
the info file prevents the renaming too. Both operations can have the
same likely errors, ENOSPC and EIO. The likelihood of EIO is very low,
for both; but for ENOSPC it's far more likely for writing the
file. Avoiding the ENOSPC error for the renaming is handled in a later
commit.
Change-Id: I9d43e5b91eb142d6945cfffd1786d417142ac728
Reviewed-by: Ahmad Samir <a.samirh78@gmail.com>
QStorageInfo is great, but rather expensive, so this introduces a faster
check by stat()ing the source file and $HOME, to see if they are the
same device, saving us two or three QStorageInfo constructions. That is
a necessary condition: if they aren't the same device, we know rename()
into $HOME/.local/share/Trash will fail.
But it's not a sufficient condition: they need to be the same mount
point and that's something only QStorageInfo will give us. Strictly
speaking, the only way to be sure that you can rename() into the trash
path is to, well, attempt it (as usual, something for a later commit).
Change-Id: I9d43e5b91eb142d6945cfffd1786c474cac25083
Reviewed-by: Ahmad Samir <a.samirh78@gmail.com>
This is not a security issue because we still use QIODevice::NewOnly
(O_EXCL) and loop again. But because we do so, we don't need to check
for existence with QFile::exists() in the first place.
Pick-to: 6.6
Change-Id: I9d43e5b91eb142d6945cfffd1786c98a39781517
Reviewed-by: Ahmad Samir <a.samirh78@gmail.com>
Make it receive the QSystemError so it can set the error condition
properly in case the suitable location for this input file can't be
found. This also includes the case when the input file does not exist in
the first place, which I moved into the function because upcoming
commits will imply this check anyway.
Change-Id: I9d43e5b91eb142d6945cfffd1786c6e59d3b0204
Reviewed-by: Ahmad Samir <a.samirh78@gmail.com>
Reviewed-by: Volker Hilsheimer <volker.hilsheimer@qt.io>
The m_children container isn't used at all, so remove it. Spotted by
Volker Hilsheimer.
Change-Id: I79db1f77c0e4caf8ebab1573a82e07396a6a806b
Reviewed-by: Christian Ehrlicher <ch.ehrlicher@gmx.de>
Instead, check the macro that we're about to use. This is also done in
qprocess_unix.cpp
Pick-to: 6.5 6.6
Change-Id: I8f3ce163ccc5408cac39fffd178d657b7594d07a
Reviewed-by: Marc Mutz <marc.mutz@qt.io>
Extract the information about the relation between invalid and valid
datetimes into a snippet, and include it in the documentation of
every relational operator.
Pick-to: 6.6 6.5 6.2 5.15
Change-Id: I61b239647efe928eb0758cfc5649b33ab4d06c7d
Reviewed-by: Edward Welbourne <edward.welbourne@qt.io>
We need to re-apply the application badge when the color scheme changes;
when a task bar button is being created for the fist time; or after Explorer
has crashed and re-started.
But we should only do that if the user has set an application badge
via our APIs. Otherwise we might end up clearing an existing badge
that was set via the native APIs directly.
Fixes: QTBUG-118117
Pick-to: 6.5 6.6
Change-Id: I1f1fecba44c118d4e3f7ef4119139c3ebd23f047
Reviewed-by: Oliver Wolff <oliver.wolff@qt.io>
The SWP_NOCOPYBITS flag helps suppress some jittering during resizes.
At the moment this is called even for plain moves with no window
resizing. Make sure that the window geometry has changed before applying
the SWP_NOCOPYBITS flag
Fixes: QTBUG-115992
Pick-to: 6.6 6.5
Change-Id: Ic0cb32d9eb3b557bf2b2ef5b6ba80d34e27c5c19
Reviewed-by: Oliver Wolff <oliver.wolff@qt.io>
Reviewed-by: Pavel Dubsky <pavel.dubsky@qt.io>
When the -executable parameter is specified, macdeployqt uses
@loader_path instead of @rpath. This case was not handled in
getBinaryDependencies() used for the code signing.
Fixes: QTBUG-118075
Pick-to: 6.6
Change-Id: Ie1e0d0781305e1849df9ec0d5fb1c3ce6713a62b
Reviewed-by: Tor Arne Vestbø <tor.arne.vestbo@qt.io>
Add two overloads to QNetworkAccessManager to support GET requests
with body.
Modify QNetworkReplyHttpImpl as well as these requests should not
be cached.
If the request is redirected it is possible that its type changes
from POST/PUT to GET and in this case the message body is deleted.
However, if a GET request has a body it should keep it after it has
been redirected - modify QNetworkReplyHttpImpl to keep the message
body after it has been redirected.
Fixes: QTBUG-112871
Change-Id: Ib01898638ed94238a98291870a5c51d56030868a
Reviewed-by: Edward Welbourne <edward.welbourne@qt.io>
Reviewed-by: Mårten Nordheim <marten.nordheim@qt.io>
Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org>
Reviewed-by: Timur Pocheptsov <timur.pocheptsov@qt.io>
Reviewed-by: Juha Vuolle <juha.vuolle@qt.io>
The m_edges container isn't changed after it's initialized in the
constructor; so make it const.
This amends commit 641bccce2a.
Change-Id: I387eb2562475bc4910700d48f67303b0a5f80ccd
Reviewed-by: Christian Ehrlicher <ch.ehrlicher@gmx.de>
`QVFuncList` and `QStartUpFuncList` are identical typdefs
(`QtCleanUpFunction` and `QtStartUpFunction` are identical typedefs):
typedef QList<QtCleanUpFunction> QVFuncList;
typedef QList<QtStartUpFunction> QStartUpFuncList;
So from the compiler's POV QVFuncList and QStartUpFuncList can be used
interchangeably, but from a code reader's POV, this is confusing.
Use IILE to make the local variable const.
This amends commits 9429226524 and
a887891271.
Pick-to: 6.6 6.5 6.2 5.15
Fixes: QTBUG-117242
Change-Id: I67f6af89027fe36a1915e815acd3c9446f7dcd5d
Reviewed-by: Marc Mutz <marc.mutz@qt.io>
Saves us from having to create a lambda functor object.
Change-Id: I5e790e693b57ae414ac6d6be84f18b76b3e8185c
Reviewed-by: Edward Welbourne <edward.welbourne@qt.io>
Mostly applying clang-format and clang-tidy fixits for:
- Redundant code (e.g. bool == true/false)
- Use member initializer list
- std::move for parameters taken by copy
Change-Id: I3b9a7f01db67291f889b42346a95c55ad74f054c
Reviewed-by: Marc Mutz <marc.mutz@qt.io>
Slightly improves performance in the new benchmark
Change-Id: I2d71143ff7bc1f32ebb172f20be1843dec123e6c
Reviewed-by: Allan Sandfeld Jensen <allan.jensen@qt.io>
When the Command key is pressed AppKit seems to do key equivalent
matching using a Latin/Roman interpretation of the current keyboard
layout.
For example, for a Greek layout, pressing Option+Command+C produces a
key event with chars="ç" and unmodchars="ψ", but AppKit still treats
this as a match for a key equivalent of Option+Command+C.
We can't do the same by just applying the modifiers to our key map,
as that too contains "ψ" for the Option+Command combination. What we
can do instead is take advantage of the fact that the Command modifier
layer in all/most keyboard layouts contains a Latin layer. We then
combine that with the modifiers of the event to produce the resulting
"Latin" key combination.
If the unmodified key is outside of Latin1, we also treat that as a
valid key combination, even if AppKit natively does not. For example,
for a Greek layout, we still want to support Option+Command+ψ as a key
combination, as it's unlikely to clash with the Latin key combination
we added above.
However, if the unmodified key is within Latin1, we skip it, to avoid
these types of conflicts. For example, in the same Greek layout, pressing
the key next to Tab will produce a Latin ';' symbol, but we've already
treated that as 'q', thanks to the Command modifier, so we skip the
potential Command+; key combination. This is also in line with what
AppKit natively does.
Fixes: QTBUG-96371
Fixes: QTBUG-79493
Task-number: QTBUG-112736
Change-Id: I30d678c1c7860642d3eed29c7757133ff74c6521
Reviewed-by: Timur Pocheptsov <timur.pocheptsov@qt.io>
An incoming key event with a set of modifiers can potentially match
a range of key combinations, depending on how the event's modifiers
are combined to produce "intermediate" representations of the event.
For example, given a normal US keyboard layout, the virtual key
23 combined with the Alt (⌥) and Shift (⇧) modifiers, can map
to the following key combinations:
- Alt+Shift+5 (Fully expressed combination)
- Alt+% (Shift consumed to produce %)
- Shift+∞ (Alt consumed to produce ∞)
- fi (Shift and Alt consumed to produce fi)
But in other cases the intermediate modifier combinations
produce the same key/symbol as other modifier combinations.
For example, pressing Alt (⌥) and Shift (⇧) with the 'c'
key on a US layout will produce:
- Alt+Shift+C (Fully expressed combination)
- Shift+Ç (Alt consumed to produce Ç)
- Ç (Shift and Alt consumed to produce Ç)
In this case, we don't want to reflect the standalone 'Ç',
as that has already been reflected in the more direct form
via Shift+Ç. Consuming the additional Shift modifier does
not produce any additional symbols.
The same can happen without the number of modifiers being
different, in case two modifiers produce the same symbol.
In this case we want to prioritize Command over Option
over Control over Shift.
There is similar logic in the Windows and XKB key mappers,
and the implementation in the Apple key mapper has been
adapted from the Windows key mapper.
Task-number: QTBUG-67200
Task-number: QTBUG-38137
Change-Id: I4f1aeebac78a5393f8da804b53cf588f7c802c1b
Reviewed-by: Timur Pocheptsov <timur.pocheptsov@qt.io>
We look for all Qt components when attempt configuring standalone
tests. If some 3rdparty dependencies are missing in user environtment
the configuring proccess will fail even if the dependency is not used
by the standalone tests. Avoid requiring the Qt components, so users
will see only warnings about the missing comonents or dependencies.
Fixes: QTBUG-117709
Pick-to: 6.6 6.5
Change-Id: Iecce40449cdf116f1a7c279ebb161f0f5c7f6a9f
Reviewed-by: Alexandru Croitor <alexandru.croitor@qt.io>
If the native method contains a jfloat parameter, I get
the warning/error:
Second argument to 'va_arg' is of promotable type 'JNITypeForArg<float>'
(aka 'float'); this va_arg has undefined behavior because arguments will
be promoted to 'double'
Change-Id: I8e8ee256b9bea01585b5f70554ba2fc537e2c94d
Reviewed-by: Volker Hilsheimer <volker.hilsheimer@qt.io>
Amend 1aba24a2ed and add check for the
EGL_DRM_MASTER_FD_EXT now used as older egl headers might not have it.
Pick-to: 6.6 6.5
Change-Id: I98b860d05396c24b8eb0e73172ac395c89da8628
Reviewed-by: Laszlo Agocs <laszlo.agocs@qt.io>
Reviewed-by: Alex Bu <alex.bu@qt.io>
It's explicitly undefined behavior to pass release/acq_rel
memory_order to load(), so don't.
This is private API, so no ChangeLog needed.
Reported-by: Fabian Kosmale <fabian.kosmale@qt.io>
Task-number: QTBUG-115107
Pick-to: 6.6 6.5
Change-Id: Iee119303d790c31937238ef92d900a25020e9713
Reviewed-by: Fabian Kosmale <fabian.kosmale@qt.io>
... by making it thread_local.
It is written and read by multiple threads at the same time, so it needs
to be protected. Since signal emission start and end happens in a single
thread, keep it thread_local rather than using an atomic.
Pick-to: 6.6 6.5
Change-Id: I98fc5438c512b45f936318be31a6fccbe5b66944
Reviewed-by: Marc Mutz <marc.mutz@qt.io>
... by making it thread_local.
As a natural (and welcome) side-effect, this makes output look sane in
multithreaded scenarios.
As for why it should be thread_local instead of an atomic:
Since signal emissions and slot invocations on one thread are not
necessarily correlated with another thread, they should not affect
one another's indentation level. As in, emitting QIODevice::readyRead
on a background thread should not make QEventLoop::aboutToBlock on the
main thread be indented. The only exception to this is BlockingQueued,
where one thread is directly tied to another (QTBUG-118145). But slot
invocations are anyway not currently printed for Queued connection
(see QTBUG-74099.)
Pick-to: 6.6 6.5
Change-Id: Iea1fc522d37626df14af419a3455a732729edf74
Reviewed-by: Marc Mutz <marc.mutz@qt.io>
Reviewed-by: Edward Welbourne <edward.welbourne@qt.io>
'.' in the regexp must be escaped.
e.g. "/qt/toolchain.cmake" matches "/qt.toolchain.cmake$"
Pick-to: 6.5 6.6
Change-Id: Ib6b66349e1619908a33b4a11d79f7ba19d0e8fdc
Reviewed-by: Joerg Bornemann <joerg.bornemann@qt.io>
It deoesn't like single quotes, which must be replaced with doubled-up
double quotes.
Pick-to: 6.2 6.5 6.6
Change-Id: I8e5f8047e72e4433926a9f4f2044407a4d823682
Reviewed-by: Edward Welbourne <edward.welbourne@qt.io>
Reviewed-by: Mårten Nordheim <marten.nordheim@qt.io>
Although used mostly in the same files, they're separate types, so
define them in separate places.
Pick-to: 6.6 6.5
Task-number: QTBUG-111228
Change-Id: I9e64b382ad48f9a74e432ccd49b6f5fcc9316da3
Reviewed-by: Ivan Solovev <ivan.solovev@qt.io>
QT_NO_OPENGL is defined in qtgui-config.h so we should include it before
checking the definition.
Fixes: QTBUG-115446
Pick-to: 6.5 6.6
Change-Id: I29b9d7d89fe4c079ca0cf767a1b1a63cc5621623
Reviewed-by: Alexandru Croitor <alexandru.croitor@qt.io>
We were using the .remove(Key) API on the map instead of
erase(iterator), so we were removing any reply of the same priority that
had not yet been popped from the queues.
Rewrote to drop loop and only work with iterators.
This issue was there since SPDY days, so not picking all the way back to
5.15, where HTTP2 anyway is not enabled by default.
As a drive-by, drop the #ifndef QT_NO_SSL, which was also there from
SPDY times, which was TLS-only.
Pick-to: 6.6 6.5 6.2
Fixes: QTBUG-116167
Change-Id: Id7e1eb311e009b86054c1fe3d049c760d711a18a
Reviewed-by: Edward Welbourne <edward.welbourne@qt.io>
Although used mostly in the same files, they're separate types, so
define them in separate places.
Pick-to: 6.6 6.5
Task-number: QTBUG-111228
Change-Id: I9e64b382ad48f9a74e432ccd49b6f5fcc9316da3
Reviewed-by: Ivan Solovev <ivan.solovev@qt.io>
Implements an iterator API and other standard member access functions
for sequential containers so that we can use ranged-for over an object
that is a jarray. Provides read-only access to individual elements
(which is mostly relevant for arrays of objects), or the entire data()
as a contiguous memory block (which is useful for arrays of primitive
types).
QJniObject call functions can return QJniArray<T> when the return type
is either explicitly QJniArray<T> or T[], or their Qt equivalent (e.g.
a jbyteArray can be taken or returned as a QByteArray). If the return
type is a jarray type, then a QJniObject is returned as before.
Arrays can be created from a Qt container through a constructor or the
generic fromData named constructor in the QJniArrayBase class, which
implements the generic logic.
Not documented as public API yet.
Added a compile-time test to verify that types are mapped correctly.
The function test coverage is added to the QJniObject auto-test, as
that already provides the Java test class with functions taking and
returning arrays of different types.
Change-Id: I0750fc4f4cce7314df3b10e122eafbcfd68297b6
Reviewed-by: Assam Boudjelthia <assam.boudjelthia@qt.io>
And some of them are not needed at all.
Pick-to: 6.6
Change-Id: Ia4778c7016573eff3eefc2f6838e458008161da6
Reviewed-by: Christian Strømme <christian.stromme@qt.io>
Make the QT_D3D_NO_FLIP env.var. have an effect again.
This env.var. has the following effects:
- SwapEffect is set to DXGI_SWAP_EFFECT_DISCARD
- Scaling is set to DXGI_SCALING_STRETCH (no other option with the
blitting legacy model)
- Alpha works without having to deal with DirectComposition, the dcomp
code path is therefore skipped completely in legacy mode
- Requesting a HDR mode behaves incorrectly (there's an unwanted
conversion to SDR or something like that)
- Different window resizing artifacts. Instead of the big black/white
bars, that is typical with the modern, efficient flip swapchains in
non-Qt applications as well, there is a bit of shimmering on the
right side esp. when resizing on the left side. The option of using
the legacy is model provided mainly for users where this is
important.
- Reduced performance due the using the old blitting model, although
that probably won't be visible for many typical Qt applications on
desktop PCs.
Only for D3D11, because D3D12 does not support non-flip swapchains.
Note: this is incompatible with QT_QPA_DISABLE_REDIRECTION_SURFACE.
The reason to reintroduce this option is to provide a way, even if
just as a developer-focused environment variable, to get a behavior
that is identical to other frameworks and non-Qt applications that
still use D3D11 with the legacy swapchain modes in their rendering
engines. This applies first and foremost to window resizing, where
the visual artifacts common with flip model swapchains may be
misunderstood to be caused by Qt. Having a way to opt-in to the
legacy model allows avoiding/clarifying Apples-to-Oranges
comparisons.
Pick-to: 6.6 6.5
Change-Id: I04e46f71a96fa56cace38703e0e9b93b43bfebc7
Reviewed-by: Christian Strømme <christian.stromme@qt.io>
Previously we had some inline c-string literals. But since the parameter
for those is const-ref QString it has to actually allocate the storage
and convert the string to UTF-16.
By putting it as a function that returns a QString constructed
with u""_s, we instead create a cheap non-owning QString that just
refers to the string somewhere in memory.
As a drive-by: move other string-literals into functions as well.
Change-Id: I2f2ca5b979cfa772665fa83689837f991b0c656d
Reviewed-by: Marc Mutz <marc.mutz@qt.io>
Configure time executables don't adopt new cmake flags if they were
changed. Cache all flags that were used when building configure time
executables and consider changes when decide rebuilding them.
Pick-to: 6.6 6.5
Change-Id: Ifba77833f362c790120f0ab1f808a855327bc49f
Reviewed-by: Alexandru Croitor <alexandru.croitor@qt.io>
Reviewed-by: Amir Masoud Abdol <amir.abdol@qt.io>
There's nothing wrong with device labels starting with a dot. Whether
udev would encode those as \x2e is unknown, but we may as well not tempt
fate in case it has changed or changes in the future.
Also including QDir::System in case udev places the actual device nodes
in /dev/disks/by-label instead of a symlink.
As a nice and intentional side-effect, QDirIterator no longer performs a
stat() in each of the entries, removing the double stat'ing that started
happening with the previous commit.
Pick-to: 6.6
Change-Id: I9d43e5b91eb142d6945cfffd1787681b4cf2bb58
Reviewed-by: Ahmad Samir <a.samirh78@gmail.com>
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
In addition to what parseMountInfo() filtered, we also filter entries
with zero total bytes (other than the root filesystem). This avoids
creating yet another QStorageInfoPrivate that may not be used, but most
importantly it avoids calling root() for that check, which would call
parseMountInfo() again.
Pick-to: 6.6
Change-Id: I9d43e5b91eb142d6945cfffd1787538dd3f285d0
Reviewed-by: Qt CI Bot <qt_ci_bot@qt-project.org>
Reviewed-by: Ahmad Samir <a.samirh78@gmail.com>
Reviewed-by: Edward Welbourne <edward.welbourne@qt.io>
Mine has 41 lines, of which 22 are returned by parseMountInfo with
filtering. That meant the file was parsed once to get the listing, then
22 times more to create a QStorageInfo for each entry. Now
QStorageInfo::mountedVolumes() opens the file and parses it only once.
Pick-to: 6.6
Change-Id: I9d43e5b91eb142d6945cfffd178752ef6c2122f2
Reviewed-by: Edward Welbourne <edward.welbourne@qt.io>
Reviewed-by: Ahmad Samir <a.samirh78@gmail.com>
Ensure that this environment variable is respected, as ASAN errors
(e.g. memory leaks) can cause the selftests to fail since they expect
no errors in their output.
Fix an incorrect function name in a comment, while we're at it.
Fixes: QTBUG-118041
Pick-to: 6.5 6.6
Change-Id: I07502101aabd8743df898ae8fe4a693c4733c4af
Reviewed-by: Edward Welbourne <edward.welbourne@qt.io>
