[2D Physics] RigidBody2D breaks when on a StaticBody2D and Engine.time_scale is set to 0

[Stronkkey]

Godot version

4.1.1 stable

System information

Godot v4.1.1.stable unknown - Artix Linux - Vulkan (Compatibility) - Mesa Intel(R) HD Graphics 4000 (IVB GT2) () - Intel(R) Core(TM) i5-3320M CPU @ 2.60GHz (4 Threads)

Issue description

Setting Engine.time_scale to 0 while a non-sleeping RigidBody is on a StaticBody sets all Node2D properties from the RigidBody to nan. Note that this does not happend with SceneTree.paused.

Steps to reproduce

1. Run the MRP
2. Press space bar when the RigidBody is on top of the StaticBody. This should set Engine.time_scale to 0
3. The RigidBody should disappear and have their Node2D properties set to nan

Minimal reproduction project

crash.zip

[Stronkkey]

Forgot to mention that only one RigidBody breaks per StaticBody. This means that if two RigidBodies are on a StaticBody, only one breaks when pausing.

[AThousandShips]

See also:

• SoftDynamicBody3D disappears when time_scale is set to 0.0 #63446
• Physics is not deterministic when using time scale. #24334
• Time scale affects PinJoint2D's softness #5418

[vpellen]

I think I ran into this myself when trying to do time-freeze shenanigans.
My best guess is a division-by-zero error somewhere within collision resolution, probably to do with calculating relative velocities.
If you're after a practical solution, I found that setting time_scale to the smallest nonzero float possible seems to prevent the NaNs from showing up and any movement that was still there effectively got lost in rounding process.