Of course I will comment

As soon as you start editing a path, it's bounding box might change. Suppose you drag the node outside the boundingbox, then the bounding box will grow. That makes it awkward to snap to that moving bounding box. Generally speaking, Inkscape only snaps to stationary objects. Nodes therefore snap to nodes/paths in other (stationary) objects, but also to the stationary part of the path that's currently being edited. The latter has not been implemented yet in the new node tool editor in v0.48 if I'm correct, but it does still work like that in v0.47
There are currently three classes of snap sources: bounding box nodes, path nodes, and others (such as guides, grids, text base lines, etc). Bounding box nodes won't snap to paths and path nodes, and vice versa. But the "other" class will snap to anything. For a detailed list see the second half of this page:
http://inkscape.modevia.com/doxygen/htm ... source.phpOK, sorry for getting too technical

. The reason for all this is that when anything would snap to anything, then you would be able for example to snap a node both to the path of a rectangle and to the bounding box of that same rectangle. For narrow strokes it would be nearly impossible to tell what you would have snapped to, and that's very inconvenient when snapping. People using snapping tend to be picky

. If the node snapped to the bounding box, then the stroke of the source path and the target path would not be aligned, i.e. the middle of the source path's stroke would be at the outside of the target path's stroke.
BTW, in this discussion I only considered visual bounding boxes, which are the default. Things are different when using geometric bounding boxes.
Does this make any sense

?