So for example if we have a flash video already buffered and playing, or an autocomplete active with input, focus, and a dropdown visible (and maybe even loading something) but only have HTML of the full document and a copy of the HTML earlier, how can we merge them into the live DOM without the user's state being interrupted?
Ordinarily I'd just program a specific fix, specifying the right area to change given it's div name, manually but that is not a known variable for the situation at hand (I'm programming a Pythonic MVC framework with AJAX).
I want to change the smallest amount of nodes, and probably the deepest nodes that I can get away with.
It's ok to require ID's for some of the nodes (e.g: flash or autocomplete widget) but not possible to expect this for all nodes - so in some situations perhaps node position and types will be available to compare documents.
I understand this will not be a complete solution but in some cases it will be all that is required.