| name | es2-coop-test |
| description | Run and interpret the Everspace 2 co-op regression tests — full session bring-up, the 19-check end-to-end verifier, the loadout provenance test and the combat authority test. Use after changing mod code to confirm co-op still works, or to diagnose which co-op feature regressed. Triggers: "verify coop", "run the coop tests", "did I break multiplayer", "test the loadout", "test combat routing". |
ES2 co-op regression testing
Every check runs against two live game instances. A full cycle is ~4 minutes.
Full cycle
scripts/coop-session.sh --kill # host(27100) + client(27101): load save, listen, connect (~3 min)
python3 scripts/verify.py # 19 checks, ~2 min
python3 scripts/verify.py --travel # 21 checks incl. a co-op location jump, ~4 min
scripts/coop-session.sh --kill --steam # same, but the host uses Steam P2P transport
verify.py prints PASS/FAIL per check and exits non-zero if any failed. What it covers:
roles (listen server / client), same world, player registry, client→host transform flow,
movement authority (repMove=0, i.e. no rubberbanding), world origin shifting disabled,
the host holding the client's ship loadout, the client's trigger driving the host-side weapon
component, mission progress reaching the client's own PlayerData, loot drops being mirrored,
death handling and kill attribution being armed, the client's own ship being fully armed locally
(no empty weapon slots — this is what breaks if ship data lands after PostInitializeComponents),
client-side damage suppression plus NPC fire mirroring, player capacity, and (with --travel) that
a jump takes both players along.
Targeted tests
- Loadout provenance:
python3 scripts/loadout-test.py coil_gun scatter_gun — has the client
rewrite its outgoing ship blob, forces the host to re-apply it, then prints both players' equipped
weapons as the host sees them. The client's pawn must show the rewritten weapon and the host's
must not.
- Steam transport:
python3 scripts/console.py 27100 steam self-checks the whole Steam path on
one machine — subsystem, SteamAPI init, socket subsystem, your own SteamID64 (the join address),
the resolved net driver and whether it is real P2P or silently degraded to UDP. Note the transport
must be selected BEFORE the first listen.
- NPC fire mirroring:
python3 scripts/console.py 27100 combat on both ports. The host should show
sent=N with a much larger deduped= (edge-detection working); the client should show a matching
applied=N with unresolved=0. unresolved climbing means NetGUIDs are not resolving on the client.
For a visual confirmation compare actors ProjectileBase 250 | grep -c Autocannon on both sides while
a fight is running — the counts should be in the same ballpark, not zero on the client.
- Combat authority:
python3 scripts/combat-test.py 15 — parks the client next to an isolated
turret, measures a no-fire window and then a firing window, and reports damage attributable to the
client's routed fire. Beware ambient NPC crossfire: always compare against the no-fire phase.
When something fails
python3 scripts/logof.py 27100 / 27101 gives the mod log for each instance; grep for
[coop] [net] [loadout] [travel] [combat] [authority].scripts/crash.sh symbolizes the newest crash (it reads the Proton prefix's
AppData/Local/ES2/Saved/Crashes, resolving ES2 frames via the PDB and mod frames via the
mod's own PDB).- Re-run a single check by hand with
python3 scripts/console.py <port> <command> (see es2-console).
Cautions
- Only two instances fit in this machine's RAM (~3 GB each) — do not launch a third.
- The host can die while idle near enemies and invalidate a test;
god 1 on both prevents that.
- To stage a fight, read an enemy position out of
actors ESPawn 40 and tp both players next to it,
then give it ~40 s — the counters only move while NPCs are actually engaging.
- Tests assume the default save;
--save NAME on coop-session.sh changes it.
