Syncing child transforms produces jittery results, which causes further issues

[AnthonyTornetta]

When moving at high speeds, the difference is quite apparent as seen here:

The non parent-child relationship is on the left & the weird parent/child pair is on the right.

The camera follows the buggy child in the right pair, which is why the left pair keeps going up as the right pair slows down and the child eventually phases through the parent.
In this example, the top cube is traveling 1m/s slower than the bottom cube.

I assume the error is located somewhere in writeback_rigid_bodies. Maybe something wrong with the parent transform being a frame out of sync?

Code for the example in the gif: https://gist.github.com/AnthonyTornetta/2a6a4573126071c961eabe3c95aca247

Update:
After more testing, this is also apparent at low speeds. Even with it only moving 2m/s, the jittering is noticeable & they interact poorly with each other.

Video at low speed (2m/s) (ignore my jittery recording software):
You can see the left pair behave properly and the right parent/child pair react incorrectly.

[Aceeri]

What is the expected behavior here? A bit confused on what should happen.

Tbh I feel like this is a case where we'd ideally be able to opt out of propagation of transforms

[AnthonyTornetta]

From what I saw in the code I believe this is the expected behavior, though this is really a dimforge question. I assumed that objects should behave like they have like fixed joints that can move around as they please. While cartoony, it makes sense they would behave that way since a child will normally move with its parent in bevy.

I wrote up a fix for this in my fork of bevy_rapier /~https://github.com/AnthonyTornetta/bevy_rapier and found that this is because while it does move the child with its parent, it doesn't sync their velocities. So rapier thinks a 200m/s thing is hitting a -1m/s object, even though it should think the object is 199m/s because it's a child of the 200m/s object.