2.0.2b15 Non-determinism with hinged objects

[tatt61880]

Glued object issue has been fixed. ([Fixed]2.0.2b14 Undo-ing changes vel/angvel)
However, I found same kind of issue with no glued object.

For reproducing this issue, I've prepared primitive scene below which has only 2 boxes and 1 hinge.

== How to reproduce bug ==
1. Load next scene.
[scene]63133[/scene]
2. Start simulation. (Starting simulation will just simulate 1 step and stop simulation automatically.)
3. Undo.
4. Start simulation again.
5. When you got same result, repeat 3. and 4. again and again.

I got next result,

When Sim.tick == 0,
pos(1) = 0.031603169 // When you downloaded the scene.
pos(1) = 0.031603109 // After undoing (1st time)
pos(1) = 0.031603049 // After undoing (2nd time)
pos(1) = 0.03160299
pos(1) = 0.03160293
pos(1) = 0.031602871
pos(1) = 0.031602811
pos(1) = 0.031602751
pos(1) = 0.031602692
pos(1) = 0.031602632
pos(1) = 0.031602573
pos(1) = 0.031602513
pos(1) = 0.031602453
pos(1) = 0.031602394
pos(1) = 0.031602334
pos(1) = 0.031602275
pos(1) = 0.031602215
pos(1) = 0.031602189
pos(1) = 0.031602189 // no-change anymore
pos(1) = 0.031602189
......

The list above is for Sim.tick == 0.
This means undoing changes position incorrectly.

This change is an arithmetic progression. I don't know this fact helps to solve this issue or not, though.

[Ivan]

I had this issue with a glued circle with pos = [239.75, 13.0].
My result was:

pos(1) = 239.75
pos(1) = 239.75002
pos(1) = 239.75005
pos(1) = 239.75008
pos(1) = 239.75009
pos(1) = 239.75012
pos(1) = 239.75015
pos(1) = 239.75018
pos(1) = 239.75021
pos(1) = 239.75024
pos(1) = 239.75027
pos(1) = 239.75031
pos(1) = 239.75034
etc.

The position didn't change for a square and a polygon.

[tatt61880]

I considered why this issue happens. My hypothesis is below.

Step1. When you load a scene, the scene variables will be put on a memory space. The memory space will be used for undo-point.
Step2. Simulation start. (No problem)
Step3. Undoing. It will load variables from the memory space.
Step4. After undoing, a normalization for variables overwrites variables on THE memory space.
Step5. Simulation start. (No problem?)
Step6. Undoing. It will load variables again but some variables ware overwritten by step4.

I don't know such normalization exist or not. But this can explain this issue.

Regards,
Tatt