Add concept of bounding box and raycast to Object3D about three.js

[Mugen87]

The idea is to have bounding volumes on geometry level since that's where data which define the bounding box in local space. Logic like ray casting or culling makes sure the bounding volumes get properly transformed into other spaces for intersection tests if required.

There are two exceptions regarding this policy: Certain types of 3D objects like SkinnedMesh have bounding volumes on object level since the geometry data alone are not sufficient to compute correct bounding volumes.

[tripdragon]

I get that, however its counter to the available workflow game engines provide. In other engines you can add a bounds or volume component to the root and tweak as needed. And since three has the Box3.setfromObject, it seems it would be an obvious next step.

In your explainer the only time its optimized is if you add a cube geometry in the case of multiple meshes, and then setup management logic to resize as needed.

[tripdragon]

I read your reason again and you note SkinnedMesh as having this concept directly already. Would you not argue then the same reasoning is true that a multi meshed imported model does not have sufficient singular geometry of reference to build an adequate bounding volume so a full encompassing bounding box at object level would be a proper location?

One could go further and say imported models should have a subclass of Group() and handle some data caching and special functions.

Looking though the code at

three.js/src/objects/SkinnedMesh.js

Line 136 in d04539a

this._computeIntersections( raycaster, intersects, _ray );

Swapping out the digging through mesh function routine would be a proper place for it, and since all other classes rewrite their own raycast to handle current singular geometry bounds, then nothing should really suffer.