LDrawLoader: Fail gracefully if an object could not be loaded, improve file compatibility

[gkjohnson]

cc @yomboprime

Description

• Changes "parse scope" defaults to use values that accommodate real work demo files
  • Change parse scope to default empty arrays for "faces", "lines" and "conditionalLines" since not all internal file definitions include an "!LDRAW_ORG" command where they were intialized.
  • Change default parse scope type of "type" to "Model" since not all internal file definitions include an "!LDRAW_ORG" command.
• Fail gracefully if a file could not be loaded instead of trying to finalize an "undefined" object.

Tested with the "Scorpion Temple" file here.

Some other notes for a future PR

• There's some weird smoothing on the scorpion claw that I'll try to track down at some point. Maybe smoothing is happening before all part edges are included?

• For some reason these pieces that are supposed to be a yellow color are rendering as black. Perhaps some metallic material settings need tweaking?

[gkjohnson]

For some reason these pieces that are supposed to be a yellow color are rendering as black. Perhaps some metallic material settings need tweaking?

It seems the material currently being set is a Phong Material because they're marked as "pearlescent" but for some reason the environment map is not being applied properly -- is there a bug with phong at the moment?

Here's what it looks like when using MeshStandard instead of MeshPhong material:

[yomboprime]

is there a bug with phong at the moment?

I remember there was some change about Phong's envmap that I thought it would affect this, but don't remember the change or where was it. :-/

[mrdoob]

It seems the material currently being set is a Phong Material because they're marked as "pearlescent" but for some reason the environment map is not being applied properly -- is there a bug with phong at the moment?

Phong material doesn't support IBL. Standard can't be used for these pieces too?

[yomboprime]

Phong material doesn't support IBL. Standard can't be used for these pieces too?

Yes, it can be used.

[gkjohnson]

My impression from the code was that phong was being used so the specular could be set to a complementary color in order to model something like SSS or iridescence for "pearlescent" materials. I think a glossy standard material would look just fine too, though. Happy to handle that in another PR when I get a chance.

[mrdoob]

Happy to handle that in another PR when I get a chance.

Sounds good!

[mrdoob]

Thanks!

[gkjohnson]

Regarding the smoothing issue I don't think it can be fixed (easily, at least). The smoothing algorithm works by looking at the vertices of the edges to see if they line up with the polygon vertices and if they do then that edge is determined to be "sharp" or unmerged. However in the scorpion claw piece the surface coloring is created from multiple meshes which breaks up the polygon edge -- and that isn't mimicked in the edge line geometry so there is no edge with matching vertices found in those cases 🤷 :

Addressing this would require more complex and more intensive searching to see if the polygon edge is a subset of one of these edge lines. Probably not worth it 😁

[gkjohnson]

Oo actually a relatively simple and fast(er) way to check this rather than brute force would to be to index all the edge lines by normalized direction, store the line vertices as projected distances, and then for each polygon edge iterate over all the parallel edges and see if the face edge falls entirely within any of the lines.

It would be a bit slower but I'd have to check by how much and it would look better in these cases 🤔. It would depend on how commonly there are a lot of parallel lines in a part. And worse case we could just run this approach when a piece is seen to have multiple materials on it. Or a face is sharing an edge with another face that has a different material.

Anyway -- notes for future self.

[yomboprime]

A thought: There will be a lot of lines parallel to the main axes (but i guess probably most of them will be sharp)

[gkjohnson]

At the level of the whole model there will be a lot of parallel lines on the main axes but normals are smoothed per piece now to improve performance so at that level I don't think there should be too many parallel edges (I'm hoping at least 😁). And it occurs to me that it would actually be best to index the lines by ray (a projected origin point and direction) meaning redundant edges would now lie on the same line rather than just parallel. Still a few things to think through.