various fixes in the prepare/init_transforms system

[exoexo-dev]

Objective

The prepare/init_transforms system has multiple issues:

• Returns early when !config.transform_to_position from loop so only the first item of the query is being processed
• is_rb: Has<C> is redundant in the presence of Added<C>
  • changed is_rb: Has<C> to is_rigid_body: Has<RigidBody>
• Where there is Position and/or Rotation and config.transform_to_position but no Transform, no new Transform is added
• When there is no Position and/or Rotation and config.transform_to_position the new_position/new_rotation ignore the current transform and only take the global transform

Solution

Added a test to cover this system and fixed the found issues. Also refactored out some of the deep nesting in the implementation. Note that the test does not yet cover GlobalTransforms and parents transforms.

---

Changelog

Fixed

Fixed various issues in the prepare/init_transforms system.

[Jondolf]

Thanks, looks good overall! I really like having this properly tested.

I left some small code quality suggestions, and fixes for the CI errors (mainly wrong precision for math types, ugh). The lint error seems unrelated, so you can probably ignore that.

[exoexo-dev]

Should we remove the comment // TODO: This system feels very overengineered. Try to clean it up? now?

[Jondolf]

Yeah, we can probably remove it. The system still isn't exactly pretty, but it's at least better and properly tested.

I think we might be able to make this and other component initialization logic a bit nicer once Bevy gets "required components" (a way to automatically insert components based on dependencies to other components, Cart is working on this), assuming it allows ECS world access. But we can experiment with that when they're actually released

[exoexo-dev]

All should be resolved, thx! Lmk if I can auto-squash the fixups

[Jondolf]

Thank you, looks good now! I generally tend to do squash merges like Bevy itself, so there's no need for you to squash things :)

(also I fixed the lint errors on the main branch so CI should now fully pass)