Spurious CAG "area is not closed!" error about csg.js HOT 12 CLOSED

@fischman thanks
I think there was a small bug introduced last release for 2D format exports. I think this is fixed in the pending 0.5.2 release, and will try.

from csg.js.

From @fischman on October 26, 2016 15:14

FWIW I get the same failure on openjscad.org and locally on e16c68a141437321d75b7401d48a851de2b48929

from csg.js.

@fischman Sadly, it looks to be a issue with the CSG.js library. I'm just starting to work with this library, and will log this as a bug against CSG.js.

from csg.js.

@fischman So, I took a look at this today.

Internally, the CAG sides are converted into CSG walls, and then CSG.unionSub() is called to do the work. The resulting CSG is then converted back to CSG sides. Deep inside CSG.unionSub() there's a problem, as the returned CSG (and therefore the CAG) are not properly closed, i.e. polygons are not clean.

This is going to take some time give that I'm now dealing with 3D math.

from csg.js.

This issue was moved to z3dev/hello-world#3

from csg.js.

Ooops! The tools automatically closed this. Reopening.

from csg.js.

@fischman This issue does not exist in the latest version of JSCAD. Can you confirm?

I also added a test in the V2 test suite for this. Hopefully, this will not reappear.

from csg.js.

Original report still repros on openjscad.org.
I don't have a v2 checkout to test with, if that's what you meant.

from csg.js.

@fischman i have been looking at how to improve the boolean operations for 2D geometry in V2. I have indentifed some issues with points returned from the boolean operations, where there are slight differences in the floating point numbers. In order to prevent these issues, I have added some logic to round the values, and produce points that 'snap' to EPS boundries.

    let x = Math.round(v3[0] / EPS) * EPS + 0 // no more -0
    let y = Math.round(v3[1] / EPS) * EPS + 0 // no more -0
    return vec2.fromValues(x, y)

This definitely helps for anything that I can think about.

However, this issue still fails. The union of CAG c and CAG d is not closed. So, I have some questions...

Do you know that CAG c and CAG d (above) actually have two outlines? If you follow the sides given, you will find two outlines in each of the CAGs.

Having said that... I suspect that those points came from another boolean operation. So, my next question is... how were those set of points created?

I haven't given up but wanted some feedback.

from csg.js.

@z3dev load https://openjscad.org/#https://gist.githubusercontent.com/fischman/2db39c62a5b7a73d28f2657d6af2e00d/raw/af6cba2efc3044befc3fc6232f217b65cc56a35b/flattener.jscad uncheck "label triangles" and change model to "heart", click Update, wait a few seconds/minutes depending on how fast your CPU is, then click "Generate DXF". Wait many minutes. End result is:
Uncaught Error: Area is not closed!
Line: 9103,col: 9
(https://openjscad.org/dist/index.js)

The input geometries are from https://github.com/gilboonet/manual_flattener/blob/master/aplat_manuel_base.jscad
so it may be that there is duplication there and it may be that my flattening logic introduces unintentional duplication.

But AFAICT the union of two closed CAGs can never not be a closed CAG, so the specifics of the geometries seems irrelevant, no? Either one of |c| or |d| in the original report should report as not being a closed CAG, or their union should not report as not being a closed CAG.

from csg.js.

So, it seems that the floating point math inside the booleans trees has a limitiation. Through the splitting of polygons, there are slight differences in the vertices created. Such as:

[-49.3186188, -15.8015509, -1.0000000] vs [-49.3186150, -15.8015509, -1.0000000]

This is expected because we are working with geometry represented by imprecised numbers. Sigh.

And I now know that by changing the EPS value, these values can be 'snapped' back together again. The appropriate value for this issue is 1e-4.

There has been a lot of discussion about having a different value for EPS (currently 1e-5). And some folks have suggested that the EPS should be calculated based on the geometry, as designs use different scales. #93

I'm not sure this solves anything but I just wanted to keep this information somewhere.

from csg.js.

I'm back. There was a recent change to the logic for 2D union/intersect/subtract, which basically snaps the points to a dynamically calculated EPS; jscad/OpenJSCAD.org#535. As a result, this issue can now be closed.

Here are the initial 2D geometries.

Here is the result of the union.

from csg.js.