I think my original idea was that separating the tree by device numbers would save time searching parts of the tree relating to irrelevant devices. For instance, if there are many cached inodes from one device and du is searching for an inode from another device. But I am prematurely optimising. The changes look all good to me
Thanks
