[Qt-jambi-interest] Instantiating 'QImage(byte[], ..)' in a loop crashes the Java-VM

[Gunnar Sletta]

> I'm experimenting with a tool to visualize raw image data (i.e 14bit grey
> values) using a QImage with 'Format_RGB32', following I convert the raw data to a
> byty[] array with 4 byte rgb data.
> To support playback I instanziate the images with
>       'image = new QImage(imgRgbBuf, 128, 128, QImage.Format.Format_RGB32);'

ouch... this function should not have been in the API. It cannot be used 
without causing crashes.

The problem is that the image buffer that you pass to QImage on the C++ 
side is a "unsigned char *" which has to remain valid throughout the 
lifetime of the original QImage. What happens to the byte[] buffer is 
that it is converted to a memory managed temporary unsighed char * and 
if GC kicks in between when you constructed the QImage and when you draw 
it, it will crash.

The right function to use is the overload that takes a QNativePointer, 
where you manually build up and preserve the databuffer until after the 
QImage is gone.

   QNativePointer data = new QNativePointer(QNativePointer.Int, width * 
height);
   // fill data with some pixels, the below should be
   // a green / blue gradient
   for (int y=0; y<height; ++y) {
       for (int x=0; x<width; ++x) {
           data.setIntAt(y * width + x,
                         0xff000000
                         | (x * 255 / width)
                         | ((y * 255 / height) << 8));
       }
   }
   QImage image =
     new QImage(data,
                width,
                height,
                width * 4, // bytes per line: 1 int == 4 bytes
                QImage.Format.Format_RGB32);

   // use the image for what you need it for..
   painter.drawImage(0, 0, image);

   // Make sure we data outlives the image
   image.dispose();

   data = null;

pixel manipulation with QImage is rather tedious because of the C++ / 
Java bridge, unfortunately.

I'll deprecate this function for the next release.

best regards,
Gunnar