Steps to Reproduce
Hello,
We were able to reproduce the crash using only assets shipped with the Engine.
The project has these relevant settings, which are just the default recommended to test the chaos mover:
[/Script/Engine.PhysicsSettings]
PhysicsPrediction=(bEnablePhysicsPrediction=True,bEnablePhysicsHistoryCapture=False,MaxSupportedLatencyPrediction=1000.000000,ResimulationSettings=(bEnableResimulationErrorPositionThreshold=True,ResimulationErrorPositionThreshold=10.000000,bEnableResimulationErrorRotationThreshold=True,ResimulationErrorRotationThreshold=4.000000,bEnableResimulationErrorLinearVelocityThreshold=False,ResimulationErrorLinearVelocityThreshold=5.000000,bEnableResimulationErrorAngularVelocityThreshold=False,ResimulationErrorAngularVelocityThreshold=2.000000),PhysicsReplicationLODSettings=(bEnablePhysicsReplicationLOD=False,MinimumBaseDistance=200.000000,BaseDistanceRadiusMultiplier=0.750000,BaseDistancesForResimulationMode=0.250000,BaseDistancesForFullPrediction=0.800000,TimeOverDistance=0.150000))
bTickPhysicsAsync=True
Open the map:
/MoverExamples/Maps/L_PhysicsSimulatedCharacter.L_PhysicsSimulatedCharacter
In the editor Settings set the “Run Under One Process” setting to false.
Launch multiple clients each on their own process as Play in Editor sessions:
Once they are loaded, before you take control most of them (90%) will crash. A “fatal Crash” window will show up for each client and they will close when “clicking” OK.
It does not happen if “Run Under One Process” is false. The editor doesn’t crash, and neither does the server.
By attaching a debugger while the client was loading we got a crash callstack which seems to imply that the simulation rewind is responsible for the crash. The Particle handle was non null, but the memory it points to appears to be invalid.
The engine is built from source using the 5.8 release.