I'm upgrading some Qt (C++& QML) code from Qt4.8 to Qt5.1.
The Qt4.8 code is a trivial C++"QML viewer" app subclassing a QDeclarativeView, and a bunch of QML.
It's been easy enough to change this to use Qt5/QtQuick2's QQuickView except for one thing:
The Qt4.8 app has a method for printing to PDF:
void MyQMLViewer::printToPDF(const QString& filename) const { QPrinter printer(QPrinter::HighResolution); printer.setOutputFormat(QPrinter::PdfFormat); printer.setPageSize(QPrinter::A3); printer.setOutputFileName(filename); printer.setOrientation(QPrinter::Landscape); QPainter painter(&printer); render(&painter);}There were a few "environment" changes needed for Qt5.1 to get a QPrinter (ie add QT += printsupport to the project .pro file and #include <QtPrintSupport>), but there seems to be a more fundamental problem that QQuickView doesn't provide anything which is obviously compatible with the QGraphicsView/QPainter/QPaintDevice world of QPrinter (specifically, QQuickView has no render method, and all the drawing/painting/rendering-related methods it does have seem very tied up with OpenGL).
Any suggestions for how to best obtain high-quality PDF output from a QQuickView ?
(Note that I am not simply looking to screenshot the view; with QDeclarativeView, the code above generates PDFs with much better resolution even than the app fullscreened on my largest monitor).
I see the "QDeclarativeItem and QDeclarativeView" section of the "Porting QML Applications to Qt 5" guide does mention the loss of QGraphicsView-specific functionality, but doesn't offer any solutions (although it does mention workrounds for the case of items with custom QPainter-based rendering being bought into the new regime).
Update with some additional background info: an example of a PDF printed from QDeclarativeView using the above code can be found here. There's a png of the same view on a decent size monitor here. (This is actually the last slide in a series of slides; it's actually a gallery of the previous slides which bounces each slide onto the screen; if I had the time I'd look into the feasibility of the gallery being the only thing and transforming each scattered slide into view for a Prezi-style presentation; suspect QDeclarative isn't really performant enough though, which is one reason for wanting to get onto QtQuick2+Qt5.2's new scene graph stuff). Anyway, If you zoom the PDF up to 100% you'll see the text is... well it's better than anything a sanely sized image file will manage I think, although the sloping text baselines perhaps look a little uneven. There is also an issue with opacity values not being represented in the PDF (so the drop shadows and "bubbles" come out solid); another one of my motivations for trying for a QtQuick2 version was actually to see if translucent elements were dealt with any better. I assume the PDF just contains rasterized (or maybe vector outlines) of all the elements as utilities like "pdftotext" can't extract anything from it. I don't know enough about PDF tools to know how to inspect the internal structure of the thing but I assume there's some hierarchy there and the QML element tree is all laid out using a similar structure of nested transforms to the QML. Just for comparison and the sort of richness I'm potentially working towards here's a poster I did with LaTeX/Beamerposter; I find Beamerposter's rigid block structure rather limiting (and fiddly) compared with the possibilities QML seems to offer though. BTW, another thing on my wishlist/todolist is a QML element which can render LaTeX source, math and all, just to get the best of both worlds.
Update: Recent Qt blog post on all the backend changes in Qt5.8 has a comment linking to this Qt issue to use the new possibility of QPainter-rendered QtQuick scenes to render PDFs.
Update (2024): Qt have closed the open "Print QML scenes to PDF with the built-in software backend" issue ( QTBUG-55321 ) with a comment "We probably should have printing support of Qt Quick, but I don't think the software renderer is the best approach and we are unlikely to do so using it (the fact we haven't done so in the 8 years since we suggested this task should be proof enough)."