added the suggested solution, but not working currently

[aabdullah27182845]

Currently trying to resolve issue #2161 (Improve shading to properly display blind holes). The current resolution is not working as intended, but aiming to resolve this issue. This is why I have kept this as a draft pull request. Thanks!

[hannobraun]

Thank you, @aabdullah27182845! As I said in the issue, I'll be able to take a closer look next week.

[aabdullah27182845]

Thanks @hannobraun !

[aabdullah27182845]

For anyone else who's interested in looking deeper into this, I can list the following findings.

Adding a shader that depends on the distance from the initialised light point isn't the way to go about this, since we get interference on the shading of the surface of the cuboid of the surface, not just the hole.

With my limited knowledge on wgsl and limited knowledge on vector algebra, I think the way this could be solved is by calculating the normal vector to the plane of the cuboid, and then using that vector as our base and multiplying a darkening factor on top of that. Normal vectors always tend to be perpendicular to the plane, so I think that would be something reasonable we would use here.

Also, I'm in no way a computer graphics expert at all, I just thought this was a cool problem and wanted to solve it.

The picture above is what the normal vector is to the cuboid (perpendicular to the cuboid). I'm suggesting we solve that (I don't know how in the code, but usually they are solved using dot products equalling zero), and then use that vector to scale the shading.

[hannobraun]

The code already does what you're suggesting. It is computing a shading factor based on the angle to the normal here:

fornjot/crates/fj-viewer/src/graphics/shader.wgsl

Lines 42 to 45 in ef5830f

	let angle = acos(dot(light, -in.normal));
	let f_angle = angle / (pi * 0.75);
	let f_normal = max(1.0 - f_angle, 0.0);

This is visible in your screenshot. See the side walls of the holes.

That alone is not enough though, due to the problem tracked in #2161.

[hannobraun]

Closing due to inactivity.

@aabdullah27182845, feel free to reopen this pull request (or maybe open a new one) if you want to pick this up again in the future.