We have to use the 4.4.3 compiler toolchain for producing binaries
for ARMv5 on Android, since the 4.7 toolchain has a regression for
this architecture. The regression has been fixed, but the fix has
not been released yet, so until it has, we need to use the 4.4.3.
However, the 4.4.3 toolchain has a different bug, which breaks
compilation in qtimageformats with a message about redefinition of
uint. This works around that bug for Android builds. When the
patched compiler has been released, we can revert this.
Task-number: QTBUG-30921
Change-Id: I620c186c6e932413a4de1dd331fbf4b9401f2e72
Reviewed-by: Paul Olav Tvete <paul.tvete@digia.com>
Change-Id: Ida382a80dba882bbeb920756adc0c16321efe37e
Reviewed-by: Oswald Buddenhagen <oswald.buddenhagen@digia.com>
Reviewed-by: Stephen Kelly <stephen.kelly@kdab.com>
Change-Id: I98145fb3149dd8f7aba60d305da98c148aa9c705
Reviewed-by: Volker Krause <volker.krause@kdab.com>
Reviewed-by: Stephen Kelly <stephen.kelly@kdab.com>
If a process dies before all output is read into the internal buffer
of QProcess, we might lose data. Therefore we must drain the output
pipes like we already do in the synchronous wait functions.
Task-number: QTBUG-30843
Change-Id: I8bbc5265275c9ebd33218ba600267ae87d93ed61
Reviewed-by: Oswald Buddenhagen <oswald.buddenhagen@digia.com>
Reviewed-by: Friedemann Kleint <Friedemann.Kleint@digia.com>
If \wrapper appears in a \class comment or a
\qmltype comment, qdoc will not print warnings
when it finds public members of the class or the
QML type that are not documented.
The \wrapper command is added to several
opengl classes. This reduces the number of
qdoc warnings by several thousands.
Task-number: QTBUG-30755
Change-Id: Iba1eebc1590ccf54100e40fe91423240c1b3d09d
Reviewed-by: Topi Reiniö <topi.reinio@digia.com>
Reviewed-by: Venugopal Shivashankar <venugopal.shivashankar@digia.com>
Reviewed-by: Martin Smith <martin.smith@digia.com>
Reviewed-by: Jerome Pasion <jerome.pasion@digia.com>
QApplicationPrivate::translateTouchToMouse always sets
buttons() to Qt::LeftButton for synthesised events. This is wrong
for a mouse release, since 'button' should in that case be
Qt::LeftButton, and 'buttons' should be Qt::NoButton (since no buttons
are actually being pressed).
This caused problems for QGraphicsView, which refuses to
release any mouse grab set on a QGraphicsItem if at least one
button is being pressed (which was always true).
This resulted in broken drag behavior on touch platforms.
Change-Id: Iefe63cd753f9f8bb04278fd04a4d728e3deda25e
Reviewed-by: Samuel Rødal <samuel.rodal@digia.com>
Reviewed-by: Tor Arne Vestbø <tor.arne.vestbo@digia.com>
Reviewed-by: Frederik Gladhorn <frederik.gladhorn@digia.com>
Reviewed-by: Shawn Rutledge <shawn.rutledge@digia.com>
When QGraphicsView has sceneInteractionAllowed == false (e.g
when dragMode == QGraphicsView::ScrollHandDrag), all touch
events are accepted. This is wrong, and will stop mouse synthesising
from happening on touch platforms. This in turn will make
ScrollHandDrag not work (since no mouse events will come through).
This patch will call QEvent::ignore() if the touch event isn't send
to the scene, which will cause a mouse event to be synthesised.
Note that according to http://doc.qt.digia.com/qq/qq11-events.html
the correct approach would probably be to just return 'false', rather
than calling QEvent::ignore(). But this logic is not followed
consistently elsewhere (e.g in QApplication::notify), so I choose
to follow what the code actually expects for this bugfix.
Change-Id: Ida777647134c41661bab156d7b164ebd882a6bb1
Reviewed-by: Shawn Rutledge <shawn.rutledge@digia.com>
One reason is when we are using XInput 2.2, because touch devices
will then generate touch events only. For the other X11 scenarios,
X11 does its own mouse emulation.
QPlatformIntegration::styleHint() wasn't overridden yet. The remaining
hints are TODO for now.
Change-Id: I2e444a00a18b33ed840ebfa8d8218655c2c39aad
Reviewed-by: Samuel Rødal <samuel.rodal@digia.com>
It's possible to override event() to receive the QCloseEvent and
do event->ignore() to prevent the window from closing.
Task-number: QTBUG-31019
Change-Id: I9abed47fca02a002b78727f98d678a824854adfc
Reviewed-by: Samuel Rødal <samuel.rodal@digia.com>
Reviewed-by: Stephen Kelly <stephen.kelly@kdab.com>
-"qt-sql" removed in qtdoc
-no longer needed because "Data Storage" and Qt SQL pages already serve
as the main introduction to SQL support in Qt
Change-Id: I7384ae04ab4c09ecc9d76668e4e7f836095066c1
Reviewed-by: Topi Reiniö <topi.reinio@digia.com>
Reviewed-by: Martin Smith <martin.smith@digia.com>
Reviewed-by: Leena Miettinen <riitta-leena.miettinen@digia.com>
defaultFramebufferObject() may be called from anywhere, at any point,
not just makeCurrent(). One example is the glyph-cache, which uses it
to re-bind the default FBO after generating the texture cache.
If the default FBO had already been created, but the render-buffer was
out of sync with the window size, we would end up in the resize code
without the correct context current, and without the render-buffer's
owning FBO bound. This caused "Failed to make complete framebuffer
object 8cd7" warnings at runtime.
We now make the context current and bind the FBO, even though it might
already be bound and the context current from makeCurrent(), or when
initially creating the FBO. For the future we should move the whole
resize logic out of defaultFramebufferObject() and call it from
makeCurrent(), or possibly [EAGLView layoutSubviews]. That's a
higher impact change though, which we reserve for the 'dev' branch.
Change-Id: I50ea949c12a02ad1af6ec9fdc3215d5da85b324f
Reviewed-by: Richard Moe Gustavsen <richard.gustavsen@digia.com>
OpenGL ES2 doesn't support NPOT textures in combination with GL_REPEAT,
so for OpenGL ES2 we use a custom program that emulates repeat by taking
the fractional part of the texture coordinates.
This is not enough though, as merely setting GL_TEXTURE_WRAP_x to GL_REPEAT
with a NPOT texture is an error in some implementations, so we have to
guard the call to updateTextureFilter() in updateBrushTexture() with
a check for OpenGL ES2 and use GL_CLAMP_TO_EDGE instead.
This fixes missing/black backgrounds in the diagramscene example on iOS.
Change-Id: I5020090b5f17faeb06dcab9dc0292459e021af30
Reviewed-by: Richard Moe Gustavsen <richard.gustavsen@digia.com>
Reviewed-by: Samuel Rødal <samuel.rodal@digia.com>
Even though the intentions of this warning were good, the warning
was a bit harsh. In addition, in certain circumstances
(like the autotest demonstrates) we could end up calling object()
on an interface where the object was in the destructor. This
could happen because: *after* we got the destroyed() signal, the
widget would still notify the accessibility framework of a FocusOut
event.
Since the code even called object() from isValid(), we could not even
(as a defensive measure to circumvent this issue) check the isValid()
of an interface without getting this warning (duh).
So - for isValid(), the warning is not needed at
all, since the caller will of course check the result of isValid() and
act accordingly.
As for the result of object(), it should always be a pointer, but it
might point to a partially destroyed object. To detect this, you simply
check isValid() first:
if (iface->isValid())
doStuff(iface->object());
Change-Id: I206307fe618806133d8c6bc338c412d0009d7181
Reviewed-by: Frederik Gladhorn <frederik.gladhorn@digia.com>
Take into account the possibility that the target platform plugin does not
support platform services.
Change-Id: I48e7fac2e1230a9a7d450414044d23ed26b334be
Reviewed-by: Konstantin Ritt <ritt.ks@gmail.com>
At the moment if d->m_data == systemData() it calls
systemLocale()->query but forgets about the standalone part
so you get the wrong data
This patch introduces the new enums so that backends can implement
properly the standaloneMonthName feature properly. At the moment the Windows
and Mac ones still return the monthName, the Unix and Blackberry ones return
the data we store in months_data
Change-Id: Idc5a50b04ab1f914f16c7385be1dca2e027feae3
Reviewed-by: Lars Knoll <lars.knoll@digia.com>
Reviewed-by: Rafael Roquetto <rafael.roquetto@kdab.com>
Reviewed-by: Mehdi Fekari <mfekari@blackberry.com>
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
For bundling Qt, we need two things:
1. We need to build a regular .jar file out of the Java files,
so that they can be built into the app package. Dexing the
classes first (i.e. compiling the JVM bytecode to Dalvik
bytecode) is required for loading the .jar file at run-time,
but cannot be used for building it into the app, so we need
two different paths.
2. We need to specify which extra files have to be bundled for
each module (this is primarily for plugins and imports). This
is because there is no static dependency on these files, so
it cannot be detected during deployment.
Task-number: QTBUG-30751
Change-Id: I733603ee5d1c64bd7c5b9357eb5d993b9d0298f7
Reviewed-by: Paul Olav Tvete <paul.tvete@digia.com>
Reviewed-by: Oswald Buddenhagen <oswald.buddenhagen@digia.com>
Add the enablers so that Qt Creator (or another
deployment tool) can add a specification in the app's
meta data of which libraries are bundled and the
Java code required to extract plugins and imports into
the required directory structure inside the app's
data directory.
This is intended to be an alternative to using Ministro
for deployment, and the mechanism of extracting
libraries on first startup is a work-around for the
requirement in Qt of having this directory structure.
For Qt 5.2, the approach should be changed to load
plugins directly from the app's lib directory and
the other files in imports will be bundled as qrcs
in the native plugins.
Task-number: QTBUG-30751
Change-Id: Ibdb3a672548b4802f9bf3ecd05fc194426ac30e7
Reviewed-by: Paul Olav Tvete <paul.tvete@digia.com>
KeccakF-1600-opt32.c:497:5: error: comparison between signed and unsigned integer expressions [-Werror=sign-compare]
harfbuzz-thai.c:264:49: error: comparison is always false due to limited range of data type [-Werror=type-limits]
These warnings are caused by "char" defaulting to unsigned on ARM. In
particular, the second warning was introduced by commit
785e95ef0a, which is not upstream...
qbenchmarkvalgrind.cpp:224:5: error: variable ‘_qzz_res’ set but not used [-Werror=unused-but-set-variable]
This one was fixed for x86-64 in 7b54571ec2 but not
for the other platforms.
KeccakF-1600-opt32.c:250:5: error: comparison between signed and unsigned integer expressions [-Werror=sign-compare]
This one is wasn't caught before because it applies only to big-endian
code.
Change-Id: Ice33b639e55d95140cbf912bb81b6f508ed3744a
Reviewed-by: Richard J. Moore <rich@kde.org>
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
51ee309a79 introduced this check, but it
was supposed to be >= (it's available in 4.8.1, not after 4.8.1)
Change-Id: Id993b128de5c3500684833aea8ef556b31aac5f2
Reviewed-by: Olivier Goffart <ogoffart@woboq.com>
The check in [QIOSOrientationListener orientationChanged] ensured we
never reported the two unsupported orientations through QPA, but we
were reporting back the orientation through QIOSScreen::orientation()
as well, and that didn't have a guard for -1. This resulted in crashes
in client code that assumed the range of QScreen::orientation() was
defined by the enum, such as the paintedwindow example.
The listener now ignores the two unsupported orientations, which leaves
us at the previous orientation. For the conversion function, we still
have to support all UIDeviceOrientations, so we fall back to portrait
for the two unsupported orientations. In the future we should consider
caching the previous value explicitly, or fall back to the interface
orientation.
Change-Id: Ic19d0ce86b4ddea250ea927d5e8664396b2b68fd
Reviewed-by: Richard Moe Gustavsen <richard.gustavsen@digia.com>
Internally iOS double-buffers its rendering using copy instead of flipping,
so we reported that our context was single-buffered so that clients could
take advantage of the unchanged buffer. This failed when clients (such as
Qt itself) then assumed that calling swapBufferes() was not needed.
We now properly report that we're double-buffered, and we'll have to find
another way to report the way double-buffering works if that's still an
optimization we'd like to provide to clients.
Change-Id: Id2e4faa68ed3b837ad01d6f22b2927fc9c9769c2
Reviewed-by: Richard Moe Gustavsen <richard.gustavsen@digia.com>
Reviewed-by: Samuel Rødal <samuel.rodal@digia.com>
Since 39a052c664, QAccessible::State is no
logner an enum that moc understand.
moc currently silently ignores it the Q_ENUMS
Change-Id: Iecc30ad57055fc9ccaa33e9e9c400d96997d0902
Reviewed-by: Jan Arve Sæther <jan-arve.saether@digia.com>
Reviewed-by: Frederik Gladhorn <frederik.gladhorn@digia.com>
QSharedDataPointer does not actually need a class derived from
QSharedData. All it needs is a member called "ref".
Change-Id: I2f7fe4cc143478ef7ef64681eada16e2d4c2e63a
Reviewed-by: Lars Knoll <lars.knoll@digia.com>
Reviewed-by: Jędrzej Nowacki <jedrzej.nowacki@digia.com>
It's not used anywhere, so we don't need to cache the locale data
index. We already have the pointer to the QLocaleData anyway.
This saves us a few roundtrips calculating the index from the data
pointer only to get the data pointer again.
Change-Id: I6905d20a382ddcb9fb04cc886a17499b467f905a
Reviewed-by: Robin Burchell <robin+qt@viroteck.net>
Reviewed-by: Lars Knoll <lars.knoll@digia.com>
Reviewed-by: Jędrzej Nowacki <jedrzej.nowacki@digia.com>
Direct benefit is that the code between the two QLocale constructors
taking language, country and (maybe) scripts is merged.
This will also allow us to cache the QLocale::c() result.
Change-Id: Ia46c5a37764dc287bfcd3a52a022ac413c53a582
Reviewed-by: Robin Burchell <robin+qt@viroteck.net>
Reviewed-by: Jędrzej Nowacki <jedrzej.nowacki@digia.com>
Reviewed-by: Lars Knoll <lars.knoll@digia.com>
We have to call DetachCurrentThread() for each time we call
AttachCurrentThread(). Fortunately we have this convenience
class that we prepared earlier.
Task-number: QTBUG-30847
Change-Id: I5ffb94b336d3787a3bae197bab22b91770d58848
Reviewed-by: Eskil Abrahamsen Blomfeldt <eskil.abrahamsen-blomfeldt@digia.com>
The back button would be non-responsive for 5 seconds after
hiding the software keyboard. This is a minimal change that
does not look into why we need to have a 5 second delay in
the first place.
Task-number: QTBUG-30752
Change-Id: Ied514b77650cea7accc37a03efef2ce861090f65
Reviewed-by: Eskil Abrahamsen Blomfeldt <eskil.abrahamsen-blomfeldt@digia.com>
Most of the logic was still following the 10.6 UI guidelines, and had
not yet been upgraded to the 10.7 new button look. We tried to keep
10.6 compatibility were possible, and improve 'small' and 'mini' Aqua
sizes support.
Change-Id: I64139f24cccd095e9349b27a987395210b55c586
Reviewed-by: Jens Bache-Wiig <jens.bache-wiig@digia.com>
The Config class is further modified to make
use of the current directory information it
stores with each configuration variable.
Task-number: QTBUG-30725
Change-Id: I34c845e6c05d7868266324f1d54e56f94d709f95
Reviewed-by: Jerome Pasion <jerome.pasion@digia.com>
The Config class is modified to build a single
multimap containing a record for each variable
found in each .qdocconf file. Each reacord
contains not only the name and value of the
variable, but also its location in the qdocconf
file it was read from and the path to that file.
This single multimap replaces 3 maps in the
Config class.
Task-number: QTBUG-30725
Change-Id: I049a69790f943b24c014a24b55b2b39725a1b56f
Reviewed-by: Jerome Pasion <jerome.pasion@digia.com>
When the control is read only. This is documented but not implemented.
Add functions to zoomIn and zoomOut.
Task-number: QTBUG-30845
Change-Id: I692b5f8cc5791498d34d35ea3dafa18b6e5d3e65
Reviewed-by: Frederik Gladhorn <frederik.gladhorn@digia.com>
One can zoom in/out text inside a QTextEdit when the widget is
read-only or when using the zoomIn/zoomOut functions.
Zooming inside a HTML document only works if the font-style is not set
to a fixed size inside the document though. Adding this information
to the documentation.
Change-Id: I66a62da53827e1ce3241ba16b91e86926b97c297
Reviewed-by: Jerome Pasion <jerome.pasion@digia.com>
Reviewed-by: Frederik Gladhorn <frederik.gladhorn@digia.com>
