Hi tokiop, thanks for your tests. The mixer test would be perfect if you would be allowed to use a fresnel function to blend the nodes. It cannot be a constant value (like it is in maxwell after version alpha) but a value which varies accordingly to the incidence angle.
To me the actual disney halo problem seems to be much simpler as it looks.
The specular disney shader, as i see it implemented in Lux, seems to me a simple reflection layer following a fresnel function on top of diffuse reflection.
The following tests are done in Lightwave which easily allows me to define a custom fresnel curve. No GI sorry!
I started with diffuse only
Then i added the fresnel reflection
Diffuse + fresnel reflection. Looks exactly like disney.
It is clear that there is an access of energy bouncing to the eye.
In reality, as closer we get to the tangent, less diffuse reflection reaches the eye because the specular reflection starts to rule and hides it. They cannot sum. The specular reflection, at tangent, hides the diffuse layer behind. For the same reason you cannot look through a glass if you look at it at very low incidence angle; reflection and trasparency are simply inversely proportional.
So i created an inverted fresnel curve and applied it to the diffuse
Diffuse - inverted fresnel
If now i add again the fresnel reflection i get someting which looks as smooth as the photographic samples.
It would be really cool to at least try in Lux to subtract the inverted fresnel to the diffuse channel.