BatchedMesh does not accurately compute bounding box when instancing geometry

[gfodor]

Description

There seems to be an obvious bug in BatchedMesh#computeBoundingBox. The routine BatchedMesh#computeBoundingSphere routine properly sets the for loop bounds to be 0 to drawInfo.length, but the BatchedMesh#computeBoundingBox routine sets the loop bounds to be 0 to geometryCount, which cuts the loop short if you have multiple instances of the same geometry.

https://github.com/mrdoob/three.js/blob/dev/src/objects/BatchedMesh.js#L308

The net result is you get an inaccurate bounding box if you instance the same geometry more than once in a batched mesh.

Reproduction steps

1. Add several instances of the same geometry to a batched mesh, at different transforms
2. Compute the bounding box
3. Notice the box is wrong

Code

No code or example, bug seems clear.

Live example

No code or example, bug seems clear.

Screenshots

No response

Version

r167

Device

No response

Browser

No response

OS

No response

[gfodor]

#29243