It feels like a mismatch in expectations. It's fairly obvious the developers consider the persistent undo feature to be a convenience for short-term continuity if you are exiting and re-entering the editor over a single edit session, not a long-term storage of the history of the file. All of their decisions make perfect sense in that context, but it seems that expectation is not shared by some of their users, who considered it to be an important piece of data meant to be kept long-term. This feels a bit odd to me, given the tendency for undo systems to be fragile anyway, and the documented fragility of this persistence. The author of this article even had deliberate workarounds for some of this fragility which I feel would put me in 'should I even trust this feature?' mode. To me the main fix would be updating the documentation to scream even more loudly "Don't expect this to stay around!".

To me it's a bit like storing your important files in a ramfs and then complaining that your OS doesn't respect user data when it crashes. Sure, it's not exactly a documented behaviour of the system, but you're also not storing your data in a place the developers of the OS expected you to put important data.