Stephen, this should be fixed now in SVN. This was indeed something wrong in a recent optimization in overview computation. This could cause integer overflow, that was triggered when dealing with a dataset where width * height > 2 billion. Thanks for reporting !