Qt 4.6 introduces many new features and improvements as well as bugfixes over the 4.5.x series. For more details, refer to the online documentation included in this distribution. The documentation is also available online: http://qt.nokia.com/doc/4.6 The Qt version 4.6 series is binary compatible with the 4.5.x series. Applications compiled for 4.5 will continue to run with 4.6. Some of the changes listed in this file include issue tracking numbers corresponding to tasks in the Task Tracker: http://qt.nokia.com/developer/task-tracker Each of these identifiers can be entered in the task tracker to obtain more information about a particular change. **************************************************************************** * General * **************************************************************************** - QtDBus * The minimum required version of the D-Bus reference library is now 0.93. **************************************************************************** * Library * **************************************************************************** * [245219] Added QXmlQuery::setFocus(const QString &focus); - QGraphicsBlurEffect * Since the 4.6 beta Qt::RenderHint has been moved to QGraphicsBlurEffect::BlurHint. - QVariant * Many optimisations * Added QVariant::toFloat() and QVariant::toReal() * Added QVariant(float) constructor - Qt::escape * now escape the double quote (") **************************************************************************** * Platform Specific Changes * **************************************************************************** - Significant external contribution from Milan Burda for planned removal of (non-unicode) Windows 9x/ME support. - QRegion is no longer a GDI object by default. This means it is no longer subject to gui-thread only nor does it potentially impact the 10.000 GDI object limit per process. By explicitly calling .handle() a GDI object will be created and memory managed by QRegion. The native handle is for reading out only. Any GDI calls made on the HRGN handle will not affect the QRegion. - [259221] QFileInfo::symLinkTarget() now supports NTFS symbolic links thanks to Konstantin Ritt (merge request 1217). - The reading code of QLocalSocket on Windows has been rewritten to improve reading performance. - On Windows CE the link time code geration has been disabled by default to be consistent with win32-msvc200x. **************************************************************************** * DirectFB * **************************************************************************** - When running Qt/DirectFB the DirectFB window id of a top level QWidget can be queried using property("_q_DirectFBWindowID") - Allow setting the background color of the primary layer in DirectFB with connect options. QWS_DISPLAY=directfb:bgcolor=red or QWS_DISPLAY=directfb:bgcolor=\#aabbccdd - Add support for rendering mouse cursor using a top level window when cursor support is not working in IDirectFBDisplayLayer. Define QT_DIRECTFB_WINDOW_AS_CURSOR to enable. - Add experimental support for optimized locking/unlocking of surfaces. Enable by defining QT_DIRECTFB_SUBSURFACE. - Fix a bug to conceptually allow using QDirectFBScreen as a subscreen of a proxy screen. Don't make QDirectFBScreen::instance() assume that QScreen::instance() is a QDirectFBScreen. - Optimize move/resize operations with DirectFB. With a properly configured DirectFB these operations are handled for us and we don't need to force a repaint of the window. - Support top level window opacity in DirectFB - Fix a bug when loading RGB32 images that would lead to rendering issues. - Fix a bug when painting with invalid pens/brushes. - Fix a bug that would show up when QWidget::scroll is called. - Implement support for loading pixmaps using IDirectFBImageProvider. - Disable color table support by default. This seems to work incorrectly on most boards and loading these images will generally require an image conversion anyway. - Fix a bug regarding QGraphicsProxyWidgets - Fix a crash when resizing windows - Compile with versions < 1.0. We still recommend using newer versions. - Considerable code cleanup - Various optimizations regarding painting, loading of images and creation of surfaces. - Allow forcibly falling back to raster engine for painting operations by exporting QT_DIRECTFB_FORCE_RASTER=1 - Reenable support for RGB32 primary surface. **************************************************************************** * Important Behavior Changes * **************************************************************************** - The experimental Direct3D paint engine has been removed. The reason for this is that Nokia focuses on OpenGL for desktop hardware accelerated rendering. - The default engine used to draw onto OpenGL buffers has changed in Qt 4.6. The QPaintEngine::OpenGL2 engine is now used as the default engine. This *may* cause compatibility problems for applications that use a mix of QPainter and native OpenGL calls to draw into a GL buffer. Use the QGL::setPreferredPaintEngine() function to enforce usage of the old GL paint engine. - When mixing OpenGL and QPainter calls you need to surround your custom OpenGL calls with QPainter::beginNativePainting() and QPainter::endNativePainting(). This is to ensure that the paint engine flushes any pending drawing and sets up the GL modelview/projection matrices properly before you can issue custom OpenGL calls, and to let the paint engine synchronize to the painter state before resuming regular QPainter based drawing. - Graphics View has undergone heavy optimization work, and as a result of this work, the following behavior changes were introduced. a) QStyleOptionGraphicsItem::exposedRect now contains the item's bounding rectangle, and QStyleOptionGraphicsItem::matrix is uninitialized by default. You can enable an exact exposed rectangle and a correct matrix by enabling the flag QGraphicsItem::ItemUsesExtendedStyleOptions. b) QStyleOptionGraphicsItem::levelOfDetails is obsoleted and its value is always initialized to 1. Instead you can call QStyleOptionGraphicsItem::levelOfDetailFromTransform(const QTransform &) to determine the level of detail. c) QGraphicsView no longer calls QGraphicsView::drawItems(), and in turn QGraphicsScene::drawItems(), by default. You can get the old behavior back by enabling QGraphicsView::IndirectPainting. d) QGraphicsItem no longer calls itemChange() for position and transformation changes. If you want to receive notifications for changes to the item's position and transformation, you can set the flag QGraphicsItem::ItemSendsGeometryChanges (which is enabled by default by QGraphicsWidget and QGraphicsProxyWidget). - QDesktopWidget on X11 no longer emits the resized(int) signal when screens are added or removed. This was not done on other platforms. Use the screenCountChanged signal instead - QUrl's parser is more strict when for hostnames in URLs. QUrl now enforces STD 3 rules: * each individual hostname section (between dots) must be at most 63 ASCII characters in length; * only letters, digits, and the hyphen character are allowed in the ASCII range; letters outside the ASCII range follow the normal IDN rules That means QUrl no longer accepts some URLs that were invalid before, but weren't interpreted as such. - The Unix configure-time check for STL is stricter now in Qt 4.6.0. This means some legacy STL implementations may fail to pass the test and, therefore, Qt will automatically disable STL support. This is a binary-compatible change: existing code will continue to work without being recompiled. However, it affects the source code, since some STL-compatibility API will not be enabled. Platforms affected by this change: * solaris-cc-* with the default (Cstd) C++ STL library recommendation: use -library=stlport4 See Sun Studio's documentation for the effects of this option - QVariant has now a new implicit constructor that takes a float. This means that code that assigned a float to a variant would create a variant with userType QMetaType::Float, instead of QVariant::Double. - QDataStream will now read and write all floating point numbers with the same precision, regardless of whether it's streamed into or out from a float or double. This is to maintain compatibility across platforms with a different default precision for qreal. The default is 64-bit precision for all floating point numbers, and this can be changed using the new function setFloatingPointPrecision(). Set Qt_4_5 as the version of the QDataStream to get the behavior of previous versions.