feat(raycaster): Raycaster doesn't trigger objects when recursive is set to false (refs: v0.4.0)

[maissani]

Description:

Raycaster doesn't trigger objects when recursive is set to false

Environment:

• A-Frame Version: 0.4.0 (refs: 87c934a)
• Platform / Device: Tested on Chrome
• Reproducible Code Snippet or URL:
  ### Editor :
  https://ide.c9.io/maissani/testvr
  ### Application:
  https://testvr-maissani.c9users.io
  ### Files:
  https://preview.c9users.io/maissani/testvr
  ps: I introduced a console.log line 59310 in aframe-v4.js to reveal intersection value behavior

Expected Behavior and Issue

Raycaster with recursive true currently working properly.
Raycaster with recursive false is not working properly: it doesn't trigger on Objects.
It seems that

intersections = this.raycaster.intersectObjects(this.objects, data.recursive)

Doesn't return any intersection when set to false
This bug is probably related to Three.js or not

A simple workAround tho fix this is to set to true whatever ( and then rollback when it's fixed on three.js), and add a conditional check to get only the parrent object (only if he is the origin of intersection )

[donmccurdy]

Understand the confusion, but this is how THREE.Raycaster works. Parent THREE.Objects do not have geometry: if recursive:false is set the children are ignored, so there's nothing for the raycaster to hit. Neither A-Frame nor Three.js will be able to change this behavior, in all likelihood.

Your workaround is a good one, though, and perhaps a community component (shallow-raycaster?) should be created to allow it more conveniently.

[donmccurdy]

Ok, looking at this a bit more, here's a simpler Codepen demo. With no nesting whatsoever, and {recursive:false}, clicks on the box still don't go through because there's a wrapping THREE.Group around the mesh.

So, there is something to be fixed here. Either:

1. We remove the recursive property from the raycaster, because it should basically always be true.
2. Or, we try to implement shallow (1-level) raycasting in a way that avoids these wrapping elements.

[machenmusik]

(is the wrapping that is done always only one level deep?)

[donmccurdy]

(is the wrapping that is done always only one level deep?)

For stock geometries, yes. If there's a collada-model or something involved, there's also a good chance that the model introduces its own wrappers.