Properly handle negative volume meshes

[pca006132]

const {cube, sphere} = Manifold;
const foo = cube(3, true).subtract(cube(1, true));
const parts = foo.decompose();
console.log(parts.length);
const result = parts[1];

The result has infinite size (that we do not compute correctly), while not violating the manifoldness definition in https://github.com/elalish/manifold/wiki/Manifold-Library#manifoldness-definition I am not sure if this is expected.

[elalish]

The output I get from your script is:

2
Triangles: 12
Bounding Box: X = 1 mm, Y = 1 mm, Z = 1 mm
Genus: 0, Volume: -0 cm^3
Took 0.02 seconds

What is infinite? As far as I can tell it's just saying you have a negative volume cube, which is correct.

[pca006132]

Yeah I get what you mean. I just wasn't sure about whether our boolean ops work on these negative volumes (e.g. positive union negative is equivalent to difference of positive and positive).

Also, not sure if it would be a nice idea to relate these negative volumes to their bounding volume for decompose output.

[elalish]

I think they work - and I think they are consistent with Clipper's boolean in 2D. If we find any cases where they are not consistent, or otherwise confusing, let's discuss. But I don't think I want to allow the concept of unbounded solids into our floating-point library.

[pca006132]

I don't think it is working properly (at least not in the way I would expect):

const {cube, sphere} = Manifold;
const foo = cube(10, true).subtract(cube(5, true));
const parts = foo.decompose();
const result = cube(10).add(parts[1]);
console.log(result.decompose().length);

If we do subtract instead of add, result is just a cube(10) and the decomposition result still has length 2.

[elalish]

Huh, that's interesting. I'm going to have to think about that...

[elalish]

We may be able to shift the winding number calculation by 1 if the volume is negative. I'm curious if that'll cause other problems...

[pca006132]

btw should we convert this to an issue?

[elalish]

I dunno - I think the main problem is just that negative volumes are a bit counter-intuitive. And they render oddly (due to back-face culling, which becomes front-face culling). I still think it would be worth some some comparisons between 2D polygon booleans and 3D booleans. They should be consistent, but intuitive may be a bridge too far.

[pca006132]

what about the issue of boolean operation with negative volume? it does not work now (see above)

[elalish]

That's the thing - I'm not sure it's "not working" now. Can you make the equivalent polygons (negative area) and check what Clipper's Booleans do with them in 2D? I think it may be the same.

[pca006132]

I cannot make a negative area polygon in Clipper, our constructor calls union and removes the negative part. Also, it seems that the union algorithm in clipper correctly handles negative contour that overlaps with positive contour:

const c = CrossSection.ofPolygons([[[0, 0], [1, 0], [0, 1]], [[0.5, 0], [0.5, 1], [1.5, 0]]])
const result = c.extrude(1)

[elalish]

Fair enough, see #745