I'm sure you're right about infinite levels of undo being hard to implement in an imperative language. I do think functional languages have a lot going for them (even though I didn't think the original post made a very convincing case for functional languages).

I would be interested in hearing what advantage anybody thinks "infinite undo" would bring to a scientific simulation. I can sort of conjure up a scenario where you're doing a probabilistic optimization, and you want to be able to re-visit every single time step, in order to perturb the boundary conditions and watch how the system evolved differently. But the cost of storing all the meaningless "false" states that the solver generated as it searched around in the solution space for each time step, hardly seems worth it (compared to, say, some kind of way-pointing capability, which many scientific software programs have got already).