Right now we user a bread-first/depth-first hybrid tree walk to sequentially assign IDs to nodes. This is super unstable, adding a single node having the potential to invalidate the majority of IDs in the map. Something the rustc folks are considering is hashing (a subset) of a given node and using that hash as the ID instead for their IDs. This produces a lot more stable IDs, we should consider this!