nx_sim_snapshot.nx

nx_sim_snapshot.nx -- generic simulation state save/restore. Foundation primitive for: - Rollback netcode (GGPO model, Cannon 2006): save last N frames, roll back when remote input correction arrives, re-simulate. - Turn-based replay (XCOM 2 / Into the Breach honestmode model): save state at turn boundary, replay deterministically. - VN save-anywhere (Ren'Py model): full-VM state snapshot for load/quicksave/rollback over the last ~128 statements. - Desync recovery (RTS lockstep): when frame_hash mismatches across clients, roll back to last known-good snapshot and re-simulate from authoritative input log. Two layers: 1. Flat byte-buffer save/restore (memcpy semantics). Caller owns the byte layout of their simulation state. 2. Ring buffer of N snapshots indexed by frame number with wraparound. Sized for rollback budget (default 8 frames at 60Hz = 133ms maximum prediction window per GGPO standard). Honest scope: this primitive does NOT serialize POINTERS. Caller is responsible for flattening pointer-graph state into a byte block before save and rebuilding pointers after restore. This matches how GGPO's `save_game_state` callback works. Source references (open): - GGPO Rollback SDK: https://github.com/pond3r/ggpo (MIT) - Tony Cannon "GGPO Inside Out": ggpo.net/file/Inside_GGPO.pdf - Skullgirls/Killer Instinct postmortems (GDC vault open) - Bernier "Latency Compensating Methods" (Valve PDF, open) genealogy_id: ggpo_2006_rollback + xcom_honestmode_replay + renpy_rollback_log lineage_id: simulation_state_persistence