The problem was introduced in cd2afafb where we removed some code that
was meant to adjust the header's offset upon row removal.
The problem with this is that visualIndexAt() is likely to return -1 in
QHeaderView::paintEvent, which in turn will lead to calling paintSection
for each and every section.
Task-number: QTBUG-18551
Reviewed-by: Thierry
(cherry picked from commit d814e378987348ce2123d083b01ea6fb6c3e6bbf)
Add support for QDir::NoDot and QDir::NoDotDot for setFilter in
QFileSystemModel.
Task-number: QTBUG-14760
Reviewed-by: Frederik
(cherry picked from commit b60d82fd56897b1a1d3cc730172f71c27a497ede)
Sometimes during initialization the QAccessibleItemRow will
still be in an invalid state.
Reviewed-by: Jan-Arve
(cherry picked from commit 90b4cf4b1aa0f70a62118e200e76dc1dc57985cc)
Buttons would not report their text when hidden, which is inconsistent.
Reviewed-by: Jan-Arve
(cherry picked from commit 1897ca20a343121422b354a7910814ddd37abd17)
This fixes a regression that was created by
75e478abdf336bbdc1b00e2ca4f5293d5455a0cb.
That broke accessibility on 64 bit windows, since lParam can both be
0x00000000fffffffc and 0xfffffffffffffffc.
However, MSDN explicitly says that lParam should be casted to a DWORD,
which would result in (an unsigned) 0xfffffffc in both cases. This can
then be compared to OBJID_CLIENT (defined to ((LONG)0xFFFFFFFC).
Reviewed-by: Prasanth Ullattil
(cherry picked from commit 504941bc50234c225f162192491815bc4d6c38cf)
This was probably caused by the fact that the only spinbox
visible in the main control panel has an inactive frame border.
In XCode 4, however the spin buttons are generally attached
to an active lineedit frame, so we change the default for 4.8.
Reviewed-by: gabriel
(cherry picked from commit 6c9d808c5726893e9aa673ca8b0cbebae67f641c)
If the macro is set, we should not use cocoa exceptions either, as
this causes compile failures
Rev-By: jbache
(cherry picked from commit cdb5729d8e1ffc4a00b52d6d4bbee34a9820a193)
navigate would not return the right index in the parent if the current
widget was not the visible one.
Reviewed-by: Jan-Arve
(cherry picked from commit fdeeaa9d61efea9cca783a1d4098ae505df24390)
Return Check/Uncheck for checkable buttons.
Partially revive the buttons unit test.
Reviewed-by: Jan-Arve
(cherry picked from commit 6040eeebfb1ab3be3906295c373033cd5b5d9dc3)
Return app name instead of window title for root accessibility object.
Return Window as accessible type for the main window.
Reviewed-by: Jan-Arve
(cherry picked from commit 9a5b0d7a579572cd7e7faf869ab1a6684800f592)
Reviewed-by: Gabriel de Dietrich <gabriel.dietrich-de@nokia.com>
Merge-request: 980
(cherry picked from commit e3cd651d92a9e550fe52360d1be6ae41d0f2ab85)
* 'master' of git://scm.dev.nokia.troll.no/qt/qtbase-staging:
Add internal documentation for QUnifiedToolbarSurface.
Fix the autotest condition.
Change the repaint() call to an update().
Fix the update() autotest for raster.
Change the repaint() to an update().
Set the default graphics system to raster.
Revert "Switch the default graphics system to raster on Mac."
Fix an race condition in the auto test.
Fix an race condition in the auto test.
Fix a race condition when the main window is being destructed.
Switch the default graphics system to raster on Mac.
This document is aimed for developers. This is why
it is directly written in the header file. This is
not part of the public API.
Reviewed-by: Richard Moe Gustavsen
(cherry picked from commit 891b6ac7236d21b69bdb54b00051422cee004059)
The previous preprocessor directive was aimed to
exclude Mac OS X.
With the raster engine, the behavior is unified
and we don't need to have a separate path for
Mac OS X/Cocoa.
The new condition excludes only Mac OS X with a
graphics system other than raster or Carbon.
Reviewed-by: Jiang Jiang
(cherry picked from commit 2c6af885d959f90b801c74dc5d389a720dc9fd1d)
In show_sys(), if we directly call repaint() this
will triggers too many UpdateRequest events.
This fixes the qwidget autotest
"compatibilityChildInsertedEvents".
Reviewed-by: Richard Moe Gustavsen
(cherry picked from commit 747962e6ec20a59b7e2ed67c5cd685258f199a86)
With the CoreGraphics engine, we expect the test
to fail with update(), but with the raster
engine the behavior is the same across platforms.
Hence we don't need a special case for Mac OS X
with the raster engine.
Reviewed-by: Samuel Rødal
(cherry picked from commit 75d2387fbf005b022437855ab6433790372639f8)
Forcing to repaint might cause a recursive
repaint. Since there is no apparent reason to
directly repaint, we just call update().
Reviewed-by: Richard Moe Gustavsen
(cherry picked from commit 244fedd484b022881b906b1bc794d5af19d02843)
This reverts commit a5d40fd3814ab7c8e865912c03a918bfd5994998.
We have to fix the regressions due to the raster engine before putting
it by default.
(cherry picked from commit 3197fe2af911673c6291db0102e90a0d7f6ae926)
Deleting the page in the wizard without removing it
first leads to a crash when the wizard tries to
access a deleted page.
Reviewed-by: Samuel Rødal
(cherry picked from commit 4024a08239c3e69bb2e0ca045ccbdf3fc900f675)
Deleting the page in the wizard without removing it
first leads to a crash when the wizard tries to
access a deleted page.
Reviewed-by: jasplin
(cherry picked from commit 3bff1637cd49617d334c1be63c20e008fac93be1)
During the destructor of QWidget, we delete the
layout. If the layout is not set to 0 afterwards,
a check on the layout might turn true, but any access
will end with a segfault.
Reviewed-by: João Abecasis
(cherry picked from commit abc5a632942c23496d75c49b3b0b4a674cdafdf8)
Fix a bug in QVector::append(), it should use the capacity for new size,
when it is implicit shared and capacity is bigger than the new size.
Autotest included.
Task-number: QTBUG-11763
Reviewed-by: joao
Reviewed-by: Olivier Goffart
The old check would fail for an installed copy of Qt. Instead, base
the check on using QTDIR if it present, otherwise use the parent of
the bin directory, and then check that "mkspecs" is present.
Reviewed-by: Olivier Goffart
Animation drivers were changed to used signals instead of virtual
functions, but the default animation driver was left behind.
Reviewed-by: Michael Brasser
* 'master' of git://scm.dev.nokia.troll.no/qt/qtbase-earth-staging:
Skip test on MacOS due to problems with corewlan plugin
Fix QNetworkConfigurationManager usage outside main thread first
Update QTBUG-17223 for Qt 4.8
fix tst_qnetworkreply::httpProxyCommands autotest
Send User-Agent from the network request in http proxy CONNECT command
Add autotests for configuration dependent network proxies
Fix QNetworkReplyImpl error handling
Enable per network configuration proxy settings in QNetworkAccessManager
Allow a network configuration to be included in a proxy query
Fix error handling in write for socks socket engine
QNetworkConfigurationManager creates the engines loaded from plugins
as objects in the main thread.
If a QNetworkConfigurationManager instance is created in a worker thread
without any instance previously existing in the main thread, then it
is uninitialised until the main thread has run.
This causes allConfigurations() to return an empty list if called
immediately after instantiation, for example.
This fix initialises the plugins using blocking queued connections,
which causes the worker thread to block until the initialisation function
has been called in the context of the main thread.
Deadlock is possible if the main thread is for some reason waiting on the
worker thread, but it will not deadlock on QNetworkConfigurationManager's
mutex.
If this is a problem for an application, it should use
QNetworkConfigurationManager from the main thread first to preload the
plugins.
Task-number: QTBUG-18795
Task-number: QTBUG-18799
Reviewed-by: Cristiano Di Flora
In Qt 4.7, http network requests are assigned to http connection
channels before connecting the channel.
In Qt 4.8, channels are connected "blind" as this gives a performance
improvement in certain circumstances.
On the assumption that User-Agent should be the same for all the
requests being sent to the server in a given burst, we use the first
queued request to set the user agent for a http proxy.
Task-number: QTBUG-17223
Reviewed-by: Markus Goetz
Reviewed-by: Martin Petersson
Due to architecture changes in Qt 4.8, clearing the proxy before the
request is complete causes the http connection to not use any proxy.
The issue is that the proxy isn't resolved until after the bearer has
been started (which is correct in the general case, as system proxy
is unknown until that time).
Also increased the test's timeout from 1 second to 15, as starting a
bearer can be slow.
Reviewed-by: Markus Goetz
Some proxies can discriminate based on the User-Agent when sent a
CONNECT command for establishing a HTTPS connection.
With this change, if the User-Agent header is set in the QNetworkRequest
then it will be passed to the http socket engine for use in the connect
command sent to the proxy.
As before, "Mozilla/5.0" will be used by default when no user agent
has been set.
Task-number: QTBUG-17223
Reviewed-by: Markus Goetz
1. test that systemProxyForQuery returns something for all configs
2. test that QNetworkAccessManager uses the settings for the
configuration it was started with.
Task-number: QTBUG-18618
Reviewed-by: Peter Hartmann
The backend was never started when compiled without bearer management,
now it is.
Now emits the error signal in case of startup errors which would leave
the state machine hanging. Previously it just printed a warning.
Reviewed-by: Peter Hartmann
