| Commit message (Collapse) | Author | Age | Files | Lines |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The use of QWinOverlappedIoNotifier in QWindowsPipeReader restricts us
in the following ways:
- The handle that gets assigned to QWinOverlappedIoNotifier is forever
tied to an I/O completion port.
- Other notification mechanisms like I/O completion routines of
WriteFileEx do not work with such a handle.
- No other QWinOverlappedIoNotifier can be registered for this handle.
To achieve the ultimate goal of making QWindowsPipeWriter thread-free
(to fix QTBUG-23378 and QTBUG-38185) we remove the usage of
QWinOverlappedIoNotifier from QWindowsPipeReader and use the
ReadFileEx API instead.
This has the additional advantage of removing the need for any thread
synchronization, as the I/O completion routine runs in the thread that
ReadFileEx was called on, leading to simpler and faster code.
Change-Id: I05c983e1f1e49d7dd27e3b77a47f87cae9c3f4c6
Reviewed-by: Oswald Buddenhagen <oswald.buddenhagen@theqtcompany.com>
|
|
|
|
|
|
| |
Task-number: QTBUG-51859
Change-Id: Id8bbcc9f0503ab2742e8da7f3b5de03fd46714b2
Reviewed-by: Eskil Abrahamsen Blomfeldt <eskil.abrahamsen-blomfeldt@theqtcompany.com>
|
|
|
|
|
|
|
|
|
|
| |
When an item is rendered into a QPixmap sent to the QDrag
implementation, make sure it's size is scaled with the current window's
devicePixelRatio, so it does not appear blurry on high-dpi screens.
Change-Id: Idf38c0993e8529aff7107ff1ac412de9cf10f311
Task-number: QTBUG-46068
Reviewed-by: Shawn Rutledge <shawn.rutledge@theqtcompany.com>
|
|
|
|
|
|
|
|
|
|
| |
CSIDL_APPDATA should be used instead of CSIDL_LOCAL_APPDATA
on Windows CE. Amends 910f719bd111813f37278b67d07f9d12cb03a4ff .
Task-number: QTBUG-50570
Change-Id: I0cc310ef5fe3fbaefae9c84dd9db8cf48ff48499
Reviewed-by: Tobias Koenig <tobias.koenig@kdab.com>
Reviewed-by: Andreas Holzammer <andreas.holzammer@kdab.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
While Cocoa requires an NSMenu to be coupled to an NSMenuItem
(just as Qt requires a QMenu to be coupled to a QAction), making
that a hard coupling comes with some limitations. This is because
Cocoa won't allow the NSMenu object to be simultaneously coupled
to more than one NSMenuItem and, similarly, an NSMenuItem can
only be added to a single parent NSMenu. Therefore, it becomes
difficult to share one QMenu between two different QMenuBars in
different windows, or to use a QMenu as context menu while being
accessible from the menu bar.
Previous solutions to circumvent those limitations were less than
ideal (see 119882714f87ffeb6945fdb2d02997ae125ff50c for the
QMenuBar shared QMenu issue). Other workarounds that relied on
that hard coupling, like 996054f5e65bc676aaea0743c2eacec51918e4aa,
also added gratuitous complexity.
In this patch, we break that hard NSMenuItem-NSMenu coupling, and
we replace it with a temporary, looser coupling. As a consequence,
* QCocoaMenu only contains and manages a NSMenu instance,
removing the previously used NSMenuItem. It gets a temporarily
attached NSMenuItem instead.
* QCocoaMenuItem gains a safe pointer to its QCocoaMenu property
removing the necessity containingMenuItem() in QCocoaMenu.
* QCocoaMenuBar manages its own NSMenuItems.
With this setup, we bind the NSMenu to its parent NSMenuItem at the
last moment. In QCocoaMenuBar, when we call updateMenuBarImmediately().
In QCocoaMenu, we use the delegate's -[QCocoaMenuDelegate menu:
updateItem:atIndex:shouldCancel:] method which is called when Cocoa
is about to display the NSMenu.
Note: There's still one use case we don't support, which is sharing
a toplevel QMenuBar menu. This is because Cocoa's menu bar requires
each of its menu items to have a submenu assigned, and therefore we
can't rely on that last moment assignment.
Task-number: QTBUG-34160
Task-number: QTBUG-31342
Task-number: QTBUG-41587
Change-Id: I92bdb444c680789c78e43fe0b585dc6661770281
Reviewed-by: Timur Pocheptsov <timur.pocheptsov@theqtcompany.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The current implementation makes the window too big when
a QPixmap with a DPR != 1 is set. Circumvent the problem
by using a QRasterWindow.
Task-number: QTBUG-46068
Task-number: QTBUG-50938
Change-Id: I0fca91f571937250c740f1400bd60286330fb595
Reviewed-by: Błażej Szczygieł <spaz16@wp.pl>
Reviewed-by: Alexander Volkov <a.volkov@rusbitech.ru>
Reviewed-by: Shawn Rutledge <shawn.rutledge@theqtcompany.com>
|
|
|
|
|
|
|
| |
Task-number: QTBUG-46615
Change-Id: I7f75bc7da35b9330753130338a06feb49533061c
Reviewed-by: Andrew Knight <andrew.knight@intopalo.com>
Reviewed-by: Maurice Kalinowski <maurice.kalinowski@theqtcompany.com>
|
|
|
|
|
|
|
|
| |
Merged in the upstream version. The remaining diff to clean 1.6.20 is
archived in the qtpatches.diff file.
Change-Id: I56f557bfe04ac1aa0e2c090826bbb144ae93cbb7
Reviewed-by: Eskil Abrahamsen Blomfeldt <eskil.abrahamsen-blomfeldt@theqtcompany.com>
|
|
|
|
|
| |
Change-Id: I38412a119d2a91685b3fd2e4a459d33a60b154b0
Reviewed-by: Topi Reiniö <topi.reinio@theqtcompany.com>
|
|
|
|
|
|
|
|
|
| |
Makes the 96DPI attribute check avoid undefined behavior by using
QCoreApplication::instance() directly, instead of calling through
qApp, which performs an invalid cast to QGuiApplication.
Change-Id: Ib86e7d2461b462a2d623f1364414f7d4d2293f22
Reviewed-by: Marc Mutz <marc.mutz@kdab.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Previously WinRT was using the UTC backend which fails on all platforms
for some QDateTime autotests related to timezone items. Hence switch to
the Windows implementation for WinRT as well.
However, the windows backend does query the registry heavily, which is
not supported on WinRT. Instead use the API version provided by the SDK.
Long-term we might want to switch to this version on desktop windows as
well, as direct registry access would not be required and we could
harmonize the codepaths for both platforms.
Change-Id: I620b614e9994aa77b531e5c34c9be1da7e272a30
Reviewed-by: Oliver Wolff <oliver.wolff@theqtcompany.com>
|
|
|
|
|
|
|
|
|
|
|
| |
If only one timer is processed in there it is possible that a reoccuring
timer which has a very low timeout blocks all the other timers from
being triggered. This high frequency timer might be the only one to
be triggered in every processEvents call.
Task-number: QTBUG-51888
Change-Id: I8a0026d1e8519171ab60d1b47c494a15d30328b3
Reviewed-by: Maurice Kalinowski <maurice.kalinowski@theqtcompany.com>
|
|
|
|
|
|
|
| |
There is no apparent reason why the handle should be duplicated.
Change-Id: I8ff2cde2f050934ed0dd9ab2d39a1b1efa327a17
Reviewed-by: Oswald Buddenhagen <oswald.buddenhagen@theqtcompany.com>
|
|
|
|
|
| |
Change-Id: I01e4b69b138fd19fc7e67751d93adebc1326b2f9
Reviewed-by: Orgad Shaneh <orgads@gmail.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
QRect::center() should be defined for any
QRect(x1,y1,x2,x2), INT_MIN <= x1, x2, y1, y2 <= INT_MAX
because the average of two signed integers is always
representable as a signed integer.
But not when it's calculated as (x1+x2)/2, since that
expression overflows when x1 > INT_MAX - x2.
Instead of playing games with Hacker's Delight-style
expressions, or use Google's patented algorithm, which
requires two divisions, take advantage of the fact that
int is not intmax_t and perform the calculation in the
qint64 domain. The cast back to int is always well-
defined since, as mentioned, the result is always
representable in an int.
Fix a test-case that expected a nonsensical result due
to overflow.
[ChangeLog][QtCore][QRect] Fixed integer overflow in
center(). This fixes the result for some corner-cases
like a 1x1 rectangle at (INT_MIN, INT_MIN), for which
the previous implementation could return anything
(due to invoking undefined behavior), but commonly
returned (0, 0).
Change-Id: I1a885ca6dff770327dd31655c3eb473fcfeb8878
Reviewed-by: Lars Knoll <lars.knoll@theqtcompany.com>
|
|
|
|
|
|
|
|
|
|
| |
QCocoaBackingstore::toImage() can only be Q_DECL_OVERRIDE
if QPlatformBackingStore::toImage() is present, which
it isn’t for NO_OPENGL builds.
Change-Id: Ib116f40fd26defb29a8d520d3e3fb104d8da8d57
Task-number: QTBUG-51694
Reviewed-by: Timur Pocheptsov <timur.pocheptsov@theqtcompany.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This reverts commit 90de48493be283b9afb249f6a0fd8dbd8958517d.
The call isn't necessary, but invokes undefined behavior.
It invokes undefined behavior because deleteTLSysExtra() is called
from deleteExtra(), which is called from ~QWidgetPrivate(), which is
called from ~QObject(). Thus, by the time we call q->windowType()
within setWinId(), q is no longer a QWidget, but only a QObject, and
calling a QWidget member function then is UB.
UBSan confirms:
qwidget_p.h:300:5: runtime error: downcast of address 0x2afdd4053620 which does not point to an object of type 'QWidget' (the Q_Q macro)
0x2afdd4053620: note: object is of type 'QObject'
qwidget.cpp:1712:93: runtime error: member call on address 0x2afdd4053620 which does not point to an object of type 'QWidget'
0x2afdd4053620: note: object is of type 'QObject'
It is also unnecessary:
deleteTLSysExtra() is called from two places: QWidget::destroy() and
deleteExtra(). deleteExtra() is only called from ~QWidgetPrivate()
which is only called from ~QObject() called by ~QWidget(), which,
however, already calls QWidget::destroy(). QWidget::destroy(), in
turn, unconditionally (for non-desktop widgets, at least) calls
setWinId(0) itself.
So fix the UB by removing the call without replacement.
Conflicts:
src/gui/kernel/qwidget_qpa.cpp
Change-Id: Ib3a8cc9d28a096183f1d3dfd1941ea5fdc6a4aac
Reviewed-by: Friedemann Kleint <Friedemann.Kleint@theqtcompany.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Found by UBSan:
qrawfont.cpp:618:60: runtime error: load of misaligned address 0x2acee92a5569 for type 'const quint32', which requires 4 byte alignment
Fix by using MAKE_TAG(), like everywhere else, instead
of a load through a type-punned and misaligned pointer.
Change-Id: I52b88ca05a57f7d8c5e5bce953384de49514079b
Reviewed-by: Konstantin Ritt <ritt.ks@gmail.com>
Reviewed-by: Lars Knoll <lars.knoll@theqtcompany.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The existing code derived a helper class from QSqlResult and
overloaded two protected functions as public ones so the test
could call them after casting QSqlResults to that helper class.
Both the cast (which is a C-style cast, but with combined
static_cast and const_cast semanics) and the following member
function call are undefined behavior.
Fix by making the test class a friend of QSqlResult, and
dropping the casts.
Change-Id: I09de2e2b46976d01cfce25892aec6ad36881d3eb
Reviewed-by: Mark Brand <mabrand@mabrand.nl>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Left-shifts of negative values are undefined in C++. In particular,
they don't behave arithmetically.
Reported by UBSan:
qcosmeticstroker.cpp: 72:15: runtime error: left shift of negative value -14/-19/-32/-33/-34/-37/-38/-63/-64/-192/-384/-1280
qcosmeticstroker.cpp:444:20: runtime error: left shift of negative value -64
qcosmeticstroker.cpp:451:26: runtime error: left shift of negative value -1
qcosmeticstroker.cpp:483:26: runtime error: left shift of negative value -1
qcosmeticstroker.cpp:762:20: runtime error: left shift of negative value -64
qcosmeticstroker.cpp:774:26: runtime error: left shift of negative value -1
qcosmeticstroker.cpp:813:47: runtime error: left shift of negative value -1
qcosmeticstroker.cpp:839:20: runtime error: left shift of negative value -64
qcosmeticstroker.cpp:851:26: runtime error: left shift of negative value -1
qcosmeticstroker.cpp:889:47: runtime error: left shift of negative value -1
qcosmeticstroker.cpp:932:27: runtime error: left shift of negative value -64
qcosmeticstroker.cpp:995:27: runtime error: left shift of negative value -3/-64
Fix by using ordinary multiplication instead, because negative
left-hand-side values don't look like they are an error.
Change-Id: Icbebd41f6ddd3dca4abd385585fc0f82064fe8b6
Reviewed-by: Allan Sandfeld Jensen <allan.jensen@theqtcompany.com>
|
|
|
|
|
|
|
|
|
|
| |
Show the widget when its initial size is 0 and the layout changes the
size during showing.
Task-number: QTBUG-51788
Change-Id: I3251ac27328f9715ff13d96e1b82fbf824d9e79d
Reviewed-by: Dmitry Shachnev <mitya57@gmail.com>
Reviewed-by: Shawn Rutledge <shawn.rutledge@theqtcompany.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Maintain virtual siblings list on screen deletion.
QCocoaIntegration::updateScreens() has a loop which
will delete all non-current QScreen objects using
QPlatformIntegration::destroyScreen().
destroyScreen() vill eventually call QWindowPrivate::
setTopLevelScreen() which accesses the virtual siblings
list for the deleted screen.
This can cause a stale pointer access if the virtual
screen list is not up to date, especially when disconnecting
two screens at the same time.
Change-Id: Ia6b9d01edf8e5eea25b64604a2b3b28b173125f7
Task-number: QTBUG-48275
Reviewed-by: Timur Pocheptsov <timur.pocheptsov@theqtcompany.com>
Reviewed-by: Morten Johan Sørvig <morten.sorvig@theqtcompany.com>
|
|
|
|
|
|
|
|
|
|
|
| |
There could be cases (mostly when compiled on old systems, since modern
openssl versions don't include such insecure ciphers) in which
defaultCiphers included a cipher that wasn't in the supported ciphers
list. With this patch we make sure that defaultCiphers is a subset of
supportedCiphers
Change-Id: I545ea21f5fd3a6ed13b366cdd56a1393233f9fc9
Reviewed-by: Richard J. Moore <rich@kde.org>
|
|
|
|
|
|
|
|
|
| |
The SNI extension must not include the trailing dot, even though this
is legitimate for the host header.
Task-number: QTBUG-51821
Change-Id: Ib7a7d8b1f8f98bc99ae745b03d2b97e507adefaf
Reviewed-by: Daniel Molkentin (ownCloud) <danimo@owncloud.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Found by UBSan:
qnetworkrequest.cpp:1016:19: runtime error: load of value 4294967295, which is not a valid value for type 'KnownHeaders'
KnownHeaders does not contain a failure state, and no negative
values. -1 is therefore not a valid value for an object of type
KnownHeaders, so loading one is considered UB.
Fix by returning the result of parseHeaderName() as an int,
only casting to KnownHeaders after checking for the failure
case.
Change-Id: I6b165fe2b15c747344a9b2750bb753582c5bcbeb
Reviewed-by: Richard J. Moore <rich@kde.org>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Ideally all native NSWindows and NSViews owned by
QCocoaWindow should be deallocated during the QCocoaWindow
destructor. In reality this does not always happen
since Cocoa is free to hold references to the views
after Qt releases its reference.
We can help Cocoa clean up:
- Clear the first responder for the NSWindow under
the ~QCocoaWndow() autoreleasepool.
- Use an autoreleasepool to clean up temp objects
from [NSWindow orderFront:] immediately.
Together this makes the QNSView lifetime be contained
by the QCocoaWindow lifetime, at least for simple
QWindow usage. It also fixes the observed memory leak
reported in QTBUG-51766
Change-Id: Idd224f54ebd6f61f274461a204ff30c666b22768
Task-number: QTBUG-51766
Reviewed-by: Timur Pocheptsov <timur.pocheptsov@theqtcompany.com>
|
|
|
|
|
|
|
|
|
|
|
| |
In addition to handling the pure rotation enforce a size change as well.
This way content is redrawn for the correct orientation.
It was done for Windows Phone 8.1 already, we only need to extent this
to Windows 10.
Task-number: QTBUG-50336
Change-Id: I6b3b964f44b631757ea856331c50f53c39ed9ec3
Reviewed-by: Oliver Wolff <oliver.wolff@theqtcompany.com>
|
|
|
|
|
|
| |
Task-number: QTBUG-38114
Change-Id: Id653487a03ca2920c46cf16e45f28677a69fa570
Reviewed-by: Oliver Wolff <oliver.wolff@theqtcompany.com>
|
|
|
|
|
|
| |
Task-number: QTBUG-36061
Change-Id: Ibde65735d861af4e1ef768e9e4314d30fed534a1
Reviewed-by: Oliver Wolff <oliver.wolff@theqtcompany.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Suppose a client connects while the QLocalServer is still in the loop
that calls addListener. The connection would SetEvent(eventHandle),
but every call to ConnectNamedPipe would ResetEvent(eventHandle).
Thus, the connection is never detected by the notifier on eventHandle.
Callers of addListener must check the connection state of every
listener to make sure that no client connected while setting up
listeners.
Task-number: QTBUG-49254
Change-Id: Ia961927ea76973708e6e3f73510695eb5d6a0e4c
Reviewed-by: Oswald Buddenhagen <oswald.buddenhagen@theqtcompany.com>
|
|
|
|
|
|
|
|
|
|
| |
QNetworkAccessManager::createRequest() had three checks relevant only
to GET and HEAD requests; rather than testing for this in each of the
cases, test for it once and skip all three if it fails. Tidied up the
residue of conditionals in the process.
Change-Id: I7baee8067a03afdc7cb0a77f1a50759dc4233843
Reviewed-by: Timur Pocheptsov <timur.pocheptsov@theqtcompany.com>
|
|
|
|
|
|
|
|
|
| |
QUrl::setScheme() parses and canonicalises the scheme, so that
scheme() always returns a lower-case string anyway; no need to
.toLower() it.
Change-Id: Ied00814b63f159386a42552dcf06346ee56f9f97
Reviewed-by: Timur Pocheptsov <timur.pocheptsov@theqtcompany.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Sometimes, in the .cpp, the declaration has the argument name in comments
because it is not used (instead of using Q_UNUSED). The old qdoc could
parse that, but once clang is used, these comments are not seen anymore.
So add the argument names to the headers. This is also good for things
like auto completion, which uses only the header to know what the
argument name is.
I grepped for " */)" and made sure all the functions that are documented
have the right arguments. I also added the name to all the function around
for consistency.
Change-Id: I1aaa37e25a1985f7f51653f047a1ac2633242b56
Reviewed-by: Marc Mutz <marc.mutz@kdab.com>
Reviewed-by: Martin Smith <martin.smith@theqtcompany.com>
|
|
|
|
|
|
|
|
| |
If Xft enabled font antialiasing, QFont::NoAntialias would have no
effect as it would be overridden.
Change-Id: I4dae264bc6674ae81f181cc9ce85851174d42544
Reviewed-by: Eskil Abrahamsen Blomfeldt <eskil.abrahamsen-blomfeldt@theqtcompany.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Found by UBSan:
qmimemagicrule.cpp:166:53: runtime error: load of misaligned address 0x00000124bcb9 for type 'const short unsigned int', which requires 2 byte alignment
qmimemagicrule.cpp:166:53: runtime error: load of misaligned address 0x00000124bcb9 for type 'const unsigned int', which requires 4 byte alignment
Fix by using new qUnalignedLoad<T>() instead of a
load through a type-punned pointer and misaligned
pointer.
Change-Id: I6b876f1ce7e01369fbb25a51263d1ad04be07d52
Reviewed-by: David Faure <david.faure@kdab.com>
|
|
|
|
|
|
|
| |
Remove placeholder formatting and add noquote.
Change-Id: I4a89f88778caf007ce42bbf57edfb514fe76bcdb
Reviewed-by: Kai Koehne <kai.koehne@theqtcompany.com>
|
|
|
|
|
|
|
|
|
|
|
|
| |
Found by UBSan:
qjsonparser.cpp:741:30: runtime error: store to misaligned address 0x0000019b1e94 for type 'quint64', which requires 8 byte alignment
Fix by using the qToLittleEndian() overload that can
store to misaligned memory.
Change-Id: Ib84bd30b13c68f7fdb8870c9fbbfac15cff0112d
Reviewed-by: Olivier Goffart (Woboq GmbH) <ogoffart@woboq.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This is quite an unlikely scenario, but not impossible.
It could be that the wheel widget is destroyed during
an update phase event. In that case, wheel_widget would
be a dangling pointer for any subsequent wheel event.
We protect against this with a QPointer.
However, that would mean that if the next wheel event
were to be an end phase event, that event would be lost.
So we go through the usual code path, except that we won't
set wheel_widget in the case of an end phase event.
Change-Id: I59a912b845dcc249e1edc60b4dc28bf308d807d9
Reviewed-by: Shawn Rutledge <shawn.rutledge@theqtcompany.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This warning is triggered when we try to apply the Q_DECL_HIDDEN
attribute to a class in an unnamed namespace. Such classes are
already not exported.
qobjectdefs.h:175:108: warning: ‘visibility’ attribute ignored [-Wattributes]
qobjectdefs.h:198:108: warning: ‘visibility’ attribute ignored [-Wattributes]
Added a test on gadgets (and QObjects) in unnamed namespaces,
because qtbase currently does not contain such Q_GADGETs.
Done-with: Thiago Macieira <thiago.macieira@intel.com>
Change-Id: Ic747cc2ab45e4dc6bb70ffff1438c747b05c5672
Reviewed-by: Lars Knoll <lars.knoll@theqtcompany.com>
Reviewed-by: Olivier Goffart (Woboq GmbH) <ogoffart@woboq.com>
|
|
|
|
|
|
|
|
|
|
| |
Some embedded servers use LF to mark the end of an individual header,
but use CRLF to mark the end of all the headers. The GoPro WiFi
interface does this, as an example.
Change-Id: I227ab73622c84f439a6cf8703d020393c4d8bf69
Reviewed-by: Marc Mutz <marc.mutz@kdab.com>
Reviewed-by: Markus Goetz (Woboq GmbH) <markus@woboq.com>
|
|
|
|
|
|
| |
Task-number: QTBUG-51617
Change-Id: Ic1d258c56165947ff821b1bf4d044bcf29b41a3b
Reviewed-by: Laszlo Agocs <laszlo.agocs@theqtcompany.com>
|
|
|
|
|
|
|
|
| |
Instead of just QVariant(enum_value).
Task-number: QTBUG-49383
Change-Id: Id57c65b68d4328816046bc35301dc6afba47b727
Reviewed-by: Friedemann Kleint <Friedemann.Kleint@theqtcompany.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Previously, the macro Q_OS_WIN64 was checked, causing warnings:
tools\qstring.cpp(6183): warning C4311: 'reinterpret_cast': pointer truncation from 'void *' to 'unsigned long'
tools\qstring.cpp(6183): warning C4302: 'reinterpret_cast': truncation from 'void *' to 'unsigned long'
when compiling WinRT/64bit, where it is not defined.
Change-Id: Ib9d8405108c85170aba18b13f9c64083136bc5ee
Reviewed-by: Maurice Kalinowski <maurice.kalinowski@theqtcompany.com>
Reviewed-by: Marc Mutz <marc.mutz@kdab.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The code infers from the presence of an address in a
QHash<QGesture *, ...> that the address belongs to a
QGesture. So far that is fine enough.
But in order to perform the lookup, it static_cast<>s
the QObject* argument to a QGesture* for the QHash::
contains() call. Even though the pointer is not
dereferenced, the cast is UB. Says UBSan:
qgesturemanager.cpp:558:73: runtime error: downcast of address 0x2ab83364f3a0 which does not point to an object of type 'QGesture'
0x2ab83364f3a0: note: object is of type 'QDBusConnectionManager'
which is a particularly hideous error message because
of the constantly-changing completely-unrelated actual
type in the second line of the message:
52 QDBusConnectionManager
19 QSocketNotifier
14 QFusionStyle
13 QAction
6 QApplication
3 QGraphicsWidget
1 Window
1 TestRunnable
1 RectWidget
1 QTimer
1 QSingleShotTimer
1 QOffscreenSurface
1 QGraphicsProxyWidget
1 QDefaultAnimationDriver
1 QDBusPendingCallWatcherHelper
This error is also _very_ common, triggered 116 times
in a single run of make -C tests/auto check.
Fix by using qobject_cast first and then doing the
lookup only when the cast succeeded.
Depending on the performance of qobject_cast<>, this
may actually perform better, too.
Change-Id: I884ec7d885711acc3c1d004ce93c628268d8fc18
Reviewed-by: Olivier Goffart (Woboq GmbH) <ogoffart@woboq.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Found by UBSan:
src/corelib/tools/qstring.cpp:587:42: runtime error: load of misaligned address 0x2acbf4b7551b for type 'const long long int', which requires 8 byte alignment
src/corelib/json/qjson_p.h:405:30: runtime error: store to misaligned address 0x0000019b1e52 for type 'quint64', which requires 8 byte alignment
src/corelib/tools/qhash.cpp:116:27: runtime error: load of misaligned address 0x2b8f9ce80e85 for type 'const qlonglong', which requires 8 byte alignment
src/corelib/tools/qhash.cpp:133:26: runtime error: load of misaligned address 0x2b8f9ce80e8d for type 'const ushort', which requires 2 byte alignment
Fix by memcpy()ing into a local variable. Wrap this trick in
template functions in qsimd_p.h. These are marked as always-
inline and use __builtin_memcpy() where available in an
attempt to avoid the memcpy() function call overhead in debug
builds.
While this looks prohibitively expensive, from the pov of the
C++ abstract machine, it is 100% equivalent, except for the
absence of undefined behavior. In one case, the cast produces
a local temporary which is then copied into the function, and
in the other case, that local variable comes from return value
of qUnalignedLoad().
Consequently, GCC compiles these two versions into identical
assembler code (only verfied for ucstrncmp, but there's no
reason to believe that it wouldn't hold for the other cases,
too).
Task-number: QTBUG-51651
Change-Id: Ia50d4a1d7580b6f803e0895c9f3d89c7da37840c
Reviewed-by: Olivier Goffart (Woboq GmbH) <ogoffart@woboq.com>
Reviewed-by: Allan Sandfeld Jensen <allan.jensen@theqtcompany.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
QCoreApplication ctor
As reported by ubsan:
src/gui/kernel/qplatformintegration.cpp:463:10: runtime error: downcast of address 0x7ffdc2942490 which does not point to an object of type 'QGuiApplication'
0x7ffdc2942490: note: object is of type 'QCoreApplication'
src/gui/kernel/qplatformintegration.cpp:466:14: runtime error: downcast of address 0x7ffdc2942490 which does not point to an object of type 'QGuiApplication'
0x7ffdc2942490: note: object is of type 'QCoreApplication'
src/gui/kernel/qplatformintegration.cpp:466:43: runtime error: member call on address 0x7ffdc2942490 which does not point to an object of type 'QGuiApplication'
0x7ffdc2942490: note: object is of type 'QCoreApplication'
to name just a few which are reported when running gui and widget
auto-tests; there're definitely more where these came from.
This is caused by QCoreApplication::init() being called from the
QCoreApplication ctor, calling virtual functions on Q*AppPrivate,
which happen to attempt, in this case, to emit QGuiApp signals.
At that point in time, the QGuiApplication ctor has not entered
the constructor body, ergo the object is still a QCoreApplication,
and calling the signal, as a member function on the derived class,
invokes UB.
Fix by cleaning up the wild mix of initialization functions used in
this hierarchy. The cleanup restores the
1. Q*ApplicationPrivate::Q*ApplicationPrivate()
2. Q*ApplicationPrivate::init(), calling each base class'
init() as the first thing
two-stage construction pattern commonly used elsewhere in Qt to make
sure that the public class' object is fully constructed by the time
each level's Private::init() is called.
Change-Id: I290402b3232315d7ed687c97e740bfbdbd3ecd1a
Reviewed-by: Lars Knoll <lars.knoll@theqtcompany.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Found by UBSan:
src/gui/painting/qcosmeticstroker.cpp:150:55: runtime error: index -1 out of bounds for type 'QT_FT_Span_ [255]'
src/gui/painting/qcosmeticstroker.cpp:150:99: runtime error: index -1 out of bounds for type 'QT_FT_Span_ [255]'
src/gui/painting/qcosmeticstroker.cpp:151:55: runtime error: index -1 out of bounds for type 'QT_FT_Span_ [255]'
That code path makes no sense if no span has been populated
yet, so skip the whole block if current_span == 0.
Change-Id: I832b989e89c118dc48ab5add3a28bb44c1936a76
Reviewed-by: Allan Sandfeld Jensen <allan.jensen@theqtcompany.com>
|
|
|
|
|
| |
Change-Id: Ic900bf000cec52b3ebf0fd0fc61f42252f3200e6
Reviewed-by: Timur Pocheptsov <timur.pocheptsov@theqtcompany.com>
|
|
|
|
|
|
|
|
|
| |
An if with no side-effects in its test and an empty body is a no-op.
An else block with nothing but a no-op in it is a no-op.
A no-op without even pedagogic value is just a distraction.
Change-Id: I224831a325e6b770d0a99d726d96f73da4b8c11f
Reviewed-by: Timur Pocheptsov <timur.pocheptsov@theqtcompany.com>
|
|
|
|
|
|
|
|
|
|
| |
Three checks of the same #if managed to save repetition of (if I felt
charitable) three shared lines, compared to combining the three into
one, which leaves the code easier to read (and obviates the need for
one of the "shared" lines). Split a long line while moving it.
Change-Id: I762d10ae1df1224c749206b8eb490bafd7ea4900
Reviewed-by: Timur Pocheptsov <timur.pocheptsov@theqtcompany.com>
|