New fairly simple boolean edge case

[lehni]

This came out of your test suite @iconexperience, by using the slightly different code now for calculating the curve points: Sketch

var path1 = new Path('M387.6708283605882,409.85893821175773c43.34991911960765,43.42692935982444 7.025356174944989,-7.309686568753534 -21.280175804952762,21.350763849679026l42.68991000890696,42.16125690051729c-28.31035574280338,28.66533466571275 -77.0540871232289,-34.32700789309064 -63.873775729843544,-21.123281947448902z');
var path2 = new Path('M366.39065230143535,431.20970231882416c-7.258856574914432,7.349874368209157 -8.587656196327089,15.955888792558142 -8.65222959434982,20.669701880671994l59.99437111077259,0.8218485401191629c-0.06457342267412969,4.7138148876486525 -1.3933735123300153,13.319830673437139 -8.652230999115659,20.669705964951333z');

path1.strokeColor = 'red';
path2.strokeColor = 'green';
path1.strokeScaling = path2.strokeScaling = false;

var res = path1.unite(path2);

res.fillColor = 'yellow';
res.fillColor.alpha = 0.25;

[lehni]

It looks like not all intersections are found: Sketch

[iconexperience]

I am not sure. Two intersections in the center (both are crossings) could actually be correct.

[lehni]

Yeah you're right...

[lehni]

But something with these two found intersections is strange... It's very hard to see though. I can only zoom close enough on Firefox, but even there the glitches make an interpretation pretty much impossible: Sketch

[iconexperience]

The problem is that path 2 has a self intersection, which is first resolved when calling unite(). Then one of the two new sub-paths has only one crossing with path1, which cannot be correct. Here is the sketch

BTW, there is something very strange going on in the sketch. Why does the console show the point instead of a true/false value?

[iconexperience]

Not quite right, the problem is the other sub-path that actually has two intersections with path 1. Here is the simplified example with only two triangles:

var path1 = new Path({segments:[
    [366.3906525556354, 431.20970206143676], 
    [409.0805625645424, 473.37095896195405], 
    [345.20678683469885, 452.24767701450514]
    ], closed:true})
var path2 = new Path({segments:[
    [387.73560545337955, 452.2903284310746],
    [417.7327938178581, 452.7012527396153],
    [409.08056281874246, 473.37095870456665]
    ], closed:true});
path1.strokeColor = 'red';
path2.strokeColor = 'green';

var res = path1.unite(path2);
res.fillColor = 'yellow';

And here is the sketch

[iconexperience]

Am I doing something wrong or does this example actually work on the current develop path branch?

[lehni]

you mean branch? It doesn't work for me...

[iconexperience]

Yes, on the branch. I am a bit confused today.

[iconexperience]

You are correct, it fails on the current develop branch. I better finish for today.

[iconexperience]

As suspected in #959 removing the new edge case handling in CurveLocation.js lines 448ff fixes this issue. But the edge case handling is certainly there for reason.

@lehni Can you give an example of an edge case that CurveLocation.js lines 448ff was introduced for? I am having a difficult time trying to understand what this is actually trying to do.

[lehni]

I developed this code for #865. But many things have since changed, and I am not sure that we still need it. I am a bit surprised that the line intersection code produces a solution for perfectly parallel lines though, it shouldn't really. I'll look into this all a bit more now.

[lehni]

I can find only one failing boolean test, and this is actually not a failure, juts a rendering imprecision that results from a slightly different path topology that is actually more correct. So I think we can safely delete this code and adjust that one test. Well spotted, @iconexperience!

[lehni]

Here the sketch for that one case.

[lehni]

I think that the need for this workaround has disappeared with the recent improvements of precision in Curve evaluate()!