[PyKDE] QPixmap problem under Windows?

[Erik Åldstedt Sund]

On Fri, Jul 22, 2005 at 10:42:59AM -0400, Andrew Bushnell wrote:
> 
> 
> Erik Åldstedt Sund wrote:
> 
> >Hi,
> >
> >We've got a problem which we hope some of you have at least seen
> >before, or maybe even know how to solve... When we run our unittest
> >suite under Windows we get the following warning/error messages from
> >Qt:
> >
> >	ASSERT: "src_dc && dst_dc" in kernel\qpaintdevice_win.cpp (388)
> >	...ASSERT: "src_dc && dst_dc" in kernel\qpaintdevice_win.cpp (388)
> >
> >Somewhere else in our tests it may print something like:
> >
> >	...ASSERT: "hBitmap" in kernel\qpixmap_win.cpp (117)
> >	ASSERT: "data->realAlphaBits" in kernel\qpixmap_win.cpp (118)
> >
> >and segfault. If we set up our test system to ignore the test that
> >segfaults it gets a little longer, but now the screen gets all messed
> >up and the computer hangs! Before it hangs it displays something like
> >the following two outputs:
> >
> >	.QPixmap::setMask: The pixmap and the mask must have the same size
> >	ASSERT: "src_dc && dst_dc" in kernel\qpaintdevice_win.cpp (388)
> >	QPainter::begin: Cannot paint null pixmap
> >	ASSERT: "src_dc && dst_dc" in kernel\qpaintdevice_win.cpp (388)
> >	QPainter::end: Missing begin() or begin() failed
> >	QPixmap::setMask: The pixmap and the mask must have the same size
> >	.QPixmap::setMask: The pixmap and the mask must have the same size
> >	ASSERT: "src_dc && dst_dc" in kernel\qpaintdevice_win.cpp (388)
> >	QPainter::begin: Cannot paint null pixmap
> >
> >	.QPainter::begin: Cannot paint null pixmap
> >	QPainter::begin: Cannot paint null pixmap
> >	QPainter::end: Missing begin() or begin() failed
> >	QPainter::begin: Cannot paint null pixmap
> >	ASSERT: "src_dc && dst_dc" in kernel\qpaintdevice_win.cpp (388)
> >	QPainter::end: Missing begin() or begin() failed
> >	QImage::convertDepth: Image is a null image
> >	QPixmap::convertFromImage: Cannot convert a null image
> >	QCursor: Cannot create bitmap cursor; invalid bitmap(s)
> >
> >This output is very strange as we've never seen anything like this
> >under Linux. Also the application itself seems to run very fine under
> >both platforms. To complicate things further:
> > * running each test-case one-by-one does not crash the computer
> >   (though some of the warning/error messages from Qt still appears on
> >   the screen).
> > * running one-by-one test in each testcase, we see no warning/error
> >   messages from Qt at all.
> >
> >If someone has seen anything like this before, or knows how to solve
> >this problem, we would be very grateful!
> >
> >
> >[ Some background: ]
> >
> >We are developing a cross-platform application using Python 2.3 and
> >PyQt 3.14.1 with Qt 3.3.4.
> >
> >We have written testcases that tests correct behaviour in
> >GUI-elements. When doing so we need to create a QApplication instance
> >before each test is run. This QApplication instance is then deleted
> >after the test is run in the following manner:
> >
> >	QTimer.singleShot(0, self.qapplication.quit)
> >	self.qapplication.exec_loop()
> >	del self.qapplication
> >
> >thus starting the GUI-thread, and quitting it first thing when it has
> >started, before deleting it.
> >
> >
> >Greetings,
> >Erik.
> >
> Erik,
> 
> It appears that you possibly are running out of resource handles or 
> something on your windows system. If you have not done so already, you 
> may want to look at using a cache for your QPixMaps and reuse object, 
> rather than create one each time. We had similar problems and when 
> turning to a caching mechanism, our problems disappeared. You should 
> also verify that the objects are getting cleaned up properly.

The QPixmap objects are allocated in code that we don't control (by
other Qt objects), so it's impossible for us to reuse objects through
a cache. 

Furthermore we have tried running the garbage collector (gc.collect())
and it seems to do a proper clean-up. After deleting the QApplication
object, the static call to QApplication.allWidgets() returns an empty
list. Also the static call to QObject.objectTrees() returns an empty
list. This should indicate that the objects are properly cleaned
up. Is there any other ways to look for unproperly cleaned-up objects?

Btw. thanks for the quick response!

Greetings,
Erik.

-- 
_______________________________________________________________________
Erik Åldstedt Sund                       <erikalds at scicraft dot org>