Inspired by the recent question about 2d grids in Haskell, I’m wondering if it would be possible to create a two-dimensional zipper to keep track of a position in a list of lists. A one-dimensional zipper on a list allows us to really efficiently move locally in a large list (the common example being a text editor). But lets say we have a second dimension like this:
grid =
[[ 1, 2, 3, 4, 5]
,[ 6, 7, 8, 9,10]
,[11,12,13,14,15]
,[16,17,18,19,20]
,[21,22,23,24,25]]
Can we create some kind of zipper data structure to efficiently move not only left and right but up and down in the grid here? If so, what if we replace the list of lists with an infinite list of infinite lists, can we still get efficient movement?
Not quite, no. One of the key aspects of how zippers work is that they represent a location in a structure by a path used to reach it, plus extra fragments created along the way, with the end result that you can backtrack along that path and rebuild the structure as you go. The nature of the paths available through the data structure thus constrains the zipper.
Because locations are identified by paths, each distinct path represents a different location, so any data structure with multiple paths to the same value can’t be used with a zipper–for example, consider a cyclic list, or any other structure with looping paths.
Arbitrary movement in 2D space doesn’t really fit the above requirements, so we can deduce that a 2D zipper would necessarily be somewhat limited. Perhaps you’d start from the origin, walk a path through the structure, and then backtrack along that path some distance in order to reach other points, for example. This also implies that for any point in the structure, there are other points that can only be reached via the origin.
What you can do is build some notion of 2D distance into the data structure, so that as you follow a path down through the structure, the points “below” you are close to each other; the idea is to minimize the amount of backtracking needed on average to move a short distance in 2D space. This ends up being roughly the same approach needed to search 2D space by distance–nearest neighbor searches, efficient geometric intersection, that sort of thing–and can be done with the same kind of data structure, namely space partitioning to create a higher-dimensional search tree. Implementing a zipper for a quadtree, a kd-tree, or similar structures is straightforward, just like any other tree.