> Do you have a patch for these fixes, or commited to CVS already? Actually, I was waiting for confirmation if the fixes are correct. In particular, it turns out that (dx >> 1) on a negative value works correctly - ie divides by 2. So the difference is rounding. I don't know if the value should be rounded or not... Ah, what the heck. I'll commit it. Unless you tell me not to. Radek