Merge lp:~osomon/unity-2d/bigendian into lp:unity-2d/3.0

[Olivier Tilloy]

Fix the byte order of images on Big Endian architectures (like PowerPC).
Original algorithm by Paul J. Wells, reworked for efficiency.

See the discussion in bug #758782 for details.

The original solution was proposed by Paul J. Wells, I merely re-wrote it to be more efficient. This branch was tested by Paul who confirmed it fixes the issue on his G5.

Ideally it should be tested on another big endian piece of hardware to double-confirm it fixes the issue, and of course it should be checked against regressions on little endian platforms (i386, amd64, …).

[Ugo Riboni (uriboni)]

I double checked that it doesn't cause any regression on amd64 but can't test on any ppc machine.
Let's merge it so that it will get more widespread testing, including on ppc.

review: Approve

=== modified file 'launcher/UnityApplications/iconimageprovider.cpp'
--- launcher/UnityApplications/iconimageprovider.cpp	2011-04-28 13:09:25 +0000
+++ launcher/UnityApplications/iconimageprovider.cpp	2011-05-07 08:35:57 +0000
@@ -139,7 +139,26 @@
                    gdk_pixbuf_get_height(pixbuf),
                    gdk_pixbuf_get_rowstride(pixbuf),
                    QImage::Format_ARGB32);
+
+#if Q_BYTE_ORDER == Q_LITTLE_ENDIAN
+    /* ABGR → ARGB */
     QImage swappedImage = image.rgbSwapped();
+#else
+    /* ABGR → BGRA */
+    /* Reference: https://bugs.launchpad.net/unity-2d/+bug/758782 */
+    QImage swappedImage(image.size(), image.format());
+    for (int i = 0; i < swappedImage.height(); ++i) {
+        QRgb* p = (QRgb*) image.constScanLine(i);
+        QRgb* q = (QRgb*) swappedImage.scanLine(i);
+        QRgb* end = p + image.width();
+        while (p < end) {
+            *q = qRgba(qAlpha(*p), qRed(*p), qGreen(*p), qBlue(*p));
+            p++;
+            q++;
+        }
+    }
+#endif
+
     g_object_unref(pixbuf);
 
     if (size) {