Desync when cancelling a gravship launch (CloseSessionAt uses Find.CurrentMap)

[cmlee119]

Summary

Cancelling a gravship launch confirmation desyncs clients. Reproduced on vanilla Multiplayer + Odyssey with no third-party mods (async time + multifaction).

Reproduce

• Async time + multifaction game with multiple maps.
• Host presses the gravship launch button on the gravship map; the confirmation dialog appears on all peers.
• A client's camera is on a different map than the gravship map.
• Host cancels the launch.
• The client desyncs. On the client, only the gravship map stays paused while the other maps keep ticking; the host has all maps paused as expected.

Not 100% on its own, but reliable under a little load (e.g. several maps running at 3x). Observed messages: Map instances don't match and Wrong random state on map N.

Cause

PatchGravshipPreLaunchCancel.Postfix closes the session with:

GravshipTravelUtils.CloseSessionAt(Find.CurrentMap.Tile);

Cancelling is a synced command, so this runs on every peer — but Find.CurrentMap is the map each peer's camera happens to be on, i.e. local UI state, not synchronized simulation state. A peer whose camera is on another map calls CloseSessionAt with the wrong tile and never closes the GravshipTravelSession, so that map stays paused only for them. Since GravshipTravelSession.IsCurrentlyPausing(map) => map == Map gates whether the map ticks, the per-map tick counts diverge → desync.

Of the four CloseSessionAt call sites in GravshipTravelSessionPatches.cs, this is the only one using Find.CurrentMap; the others pass deterministic tiles (curTile / landingTile / takeoffTile).

Present on current master (4a3be27) and dev.

Evidence

Desync report captured with local_metadata.txt listing only Harmony + Prepatcher + DLCs (including Odyssey) + Multiplayer — no third-party mods — and Map instances don't match at the moment of cancel. Comparing the host/client saves at the desync, only the gravship map had diverged: that map alone was off by tens of ticks and thousands of rand iterations relative to the others, which were in sync. (Report/saves available on request.)

A PR fixing this follows.

---

Investigated and written with the help of Claude (Anthropic).