Selective undo and redo let users disable or re-enable a specific historical operation while attempting to preserve unrelated later work. This is particularly valuable in interaction-heavy and collaborative editing environments, where mistakes may be discovered only after substantial later work has accumulated and where discarded edits are costly to reconstruct manually. Achieving this is significantly harder when operations are dependency-constrained: whether they remain valid after a toggle may depend on complex semantic conditions, and naively replaying long history suffixes after every toggle is too expensive for interactive use. To this end, we present a dependency-sensitive change-propagation algorithm that preserves a replay-style correctness target without full replay after every toggle. Our approach decomposes editor state into fine-grained slots with reducer-backed semantics, evaluates operation legality against prefix state, and propagates the effect of a toggle only through the affected neighborhood of later operations and state changes. We implement the mechanism in the scene editor of an in-house game IDE built on a heavily modified version of the Godot Engine. Across differential testing against two exact replay baselines and a validity-aware fuzzing campaign comprising 45 runs and 450,000 recorded actions, our change-propagation pipeline matched the reference semantics on all evaluated workloads. It substantially reduced toggle latency and semantic work relative to both baselines, including a reduction of about 194× in prerequisite evaluations. These results show that replay-style selective undo/redo can be made practical for operations with hard dependency constraints.

@inproceedings{paperbot12463,
  title = {A Selective Undo/Redo Algorithm for Dependency-Constrained Operation Histories Based on Change Propagation},
  author = {Xiaoxiong Fan and Hou Tam and Hanxi Zhu and Song-Hai Zhang},
  booktitle = {ASE},
  year = {2026},
  doi = {10.1145/3832783.3837482}
}