[x3d-public] lights in switch nodes

[Michalis Kamburelis]

Note that we are touching two (independent) X3D specification rules
about lights in this thread:

(I just wanted to clarify that -- it doesn't contradict anything
anyone said, I think :) )

1. One rule is that lights that are not traversed (e.g. are in unused
Switch or LOD child) should not have any effect on the visible shapes.
I.e. these lights do not contribute to the visible shapes' color.

2. The other rule is that lights can be global or not global (X3D spec
"17.2.1.2 Scoping of lights"). This is controlled by the "global"
field, which is by default TRUE for PointLight and SpotLight (meaning
that they shine on all the geometry in the scene), and FALSE for
DirectionalLight (meaning it shines only at shape nodes that are
sibling nodes or lower).

    This is backward-compatible with VRML 97, however VRML 97 didn't
have a field "global" to change this. With X3D, you can just set
global=TRUE on all the lights, if you don't want scoped lights.

    The PointLight and SpotLight are by default global=TRUE, because
they are limited anyway by their "radius" field, so they will not
affect *every shape* on large scenes. The DirectionalLight is by
default global=FALSE, because it doesn't have a radius, and
DirectionalLight with global=FALSE would affect all the shapes in the
scene, which would not be optimal. While modern rendering methods do
not have the old fixed-function OpenGL limit of "8 lights per shape",
in practice you should still limit your lights-per-shape to have a
decent FPS, at least for classic "forward rendering" approach (as
opposed to "deferred rendering", which has different considerations).
I think it's reasonable for a renderer using "forward rendering" to
use only the first e.g. 8 lights (at this shape), even when using
shader rendering being able to in theory calculated 100 lights. It's
reasonable to have 100 lights on a large scene, but each particular
shape can probably be rendered accurately with at most 8 lights :)

    The VRML 97 Annotated Reference about DirectionalLight (
http://www.x-3-x.net/vrml/archive/annotatedVRML2/BOOK.HTM,
http://www.x-3-x.net/vrml/archive/annotatedVRML2/CH3.HTM#DirectionalLight
) gives a reasoning for it.

    The fact that Blender doesn't set global=TRUE on DirectionaLights
is just a bug in Blender exporter, IMHO :)  It is also fixed in Castle
Game Engine modified version of X3D exporter, see
https://castle-engine.io/creating_data_blender.php ,
https://github.com/castle-engine/castle-engine/wiki/Blender . If
anyone has some time, it should be reported to the Blender's devs.

Note that all the conditions above must be satisfied in order for the
light to shine (contribute to a color) on a given shape:

- The light node must be traversed (e.g. be in an active Switch or LOD child).
- The shape must be within the light node scope (if global=FALSE).
- The shape must be within the light node radius (for PointLight, SpotLight).

This is all written of course based on my understanding on VRML and
X3D specs :) The view3dscene and Castle Game Engine behave like this.

Some of my testcases:

- https://github.com/castle-engine/demo-models/blob/master/lights_materials/lights.x3dv
- https://github.com/castle-engine/demo-models/blob/master/lights_materials/directional_light_scope.wrl
- https://github.com/castle-engine/demo-models/blob/master/lights_materials/cubes_mesh_large.wrl

Regards,
Michalis