Serialization test fails due to projectiles

[historic_bruno]

Apparently projectiles are using model (VisualActor) data that influences the simulation via a projectileID. Models shouldn't influence the simulation because they aren't synchronization safe, but this test in particular is failing because it creates a secondary simulation to compare the predicted simulation state with the actual one. This secondary simulation has no visual components, which would cause e.g. CCmpProjectileManager::LaunchProjectile to silently return 0 for an ID, which wouldn't match up with the actual projectileID.

It's very easy to reproduce this:

• set CSimulation2Impl::m_EnableSerializationTest to true and rebuild
• create a test scenario with a ranged unit
• attack a target, and the test fails

I'm attaching the oos logs anyway, and here is the diff between the debug state logs:

--- debug.after.a	2012-09-24 17:32:36.730352300 -0400
+++ debug.after.b	2012-09-24 17:32:27.636832200 -0400
@@ -69430,7 +69430,7 @@
           "x": 0.9986754490469063,
           "z": 0.051452380615089145
         },
-        "projectileId": 1
+        "projectileId": 0
       }
     ],
     "8": [
@@ -69451,7 +69451,7 @@
           "x": -0.9993504712481683,
           "z": 0.03603658718668677
         },
-        "projectileId": 2
+        "projectileId": 0
       }
     ]
   }

[historic_bruno]

I'm thinking this is also the cause of OOS errors during rejoin (to reproduce, start a multiplayer game with no AIs but have some ranged unit attacks going on while the client disconnects and rejoins). Either the projectile code needs to be redesigned or the projectileID needs to be a non-synced variable somehow.

[Jonathan Waller]

In 12865:

Network synchronise the projectile id to prevent out of sync errors. Fixes #1681