This is a Data.Vector question, hopefully here is a good place for those. Or would it be better to ask on stackoverflow? I want to generate a vector from another, where I interpolate between each point. So I need something like concatMap because the vector gets longer, but I need state because I need to remember the previous value (and I need to look at the next value to avoid overrunning, but that can be done with state and 'tails'). Stated imperatively, it would look like this: state = Nothing out = [] for (x, y) in signal: case state of Nothing -> out.append (x, y) Just (x0, y0) -> for (x, y) in interpolate ...: out.append (x, y) state = Just (x, y) The simplest option would be just doing the transformation on a list: `Vector.concat . snd . List.mapAccumL go Nothing . Vector.toList` But unless fusion is more powerful than I imagined, that builds a couple of lists and a bunch of intermediate vectors. I did try to read the core, but it's hard to follow when vector is involved. If they're large, it's probably no worse than a builder, but if they're small, well I'd want a builder to fill up each chunk efficiently... but there's no builder for Data.Vector (I did write one once, maybe I should pull it out and dust it off). unfoldr could also do it, but to emit multiple samples I'd need to make the state `Either ([a], state) state`, and some awkward plumbing, and constant tag checks. So it's looking like converting to a list and back is the most reasonable option... unless there's some way to express this that I'm missing?