Regression: resolveCrossing() moves intersections

[iconexperience]

In the following example the intersections are moved during resolveCrossing() and produces a wrong result.

var path = new paper.Path({segments:[
        [100, 200],
        [300, 200],
        [300, 250],
        [150, 150],
        [150, 200],
        [350, 200],
        [350, 300],
        [100, 300]
    ], closed:true});
var res = path.resolveCrossings();
res.fullySelected = true;

Expected result:

Acutal result:

This behavior was introducded in fbb0c59 , which was introduced to fix a similar issue.

This certainly has something to do with the overlaps, but I do not fully understand what is going on.

[iconexperience]

I do not think that fbb0c59 caused this issue, it merely brought it to light.

Some testing showed that after the first call to divideLocations() in resolveCrossings() the _time values of the crossings in the middle are not correct. Before first call to divideLocations() the _time values at the intersections are as follows (crossings in the middle are at index 1 and 6):

0: 0.32635182233306964
1: 0.5841272008905138
2: 0
3: 0.5
4: 0.5
5: 0
6: 0.4158727991094863
7: 0.6736481776669302

After first call to divideLocations():

0: 0
1: 0.5841272008905138
2: 0
3: 0.5
4: 0.5
5: 0
6: 0.4158727991094863
7: 0

The correct values at index 1 and 6 would be 0.5.

A simple fix would be to run through all intersections after the first call to divideLocations() and simply call getTime() or getCurve() on each intersection. This will update the _time property to the correct value.

[iconexperience]

Created pull request #1204 as a fix for this issue.

[lehni]

I think there should be a better solution to this... I want to think about it a bit.

[lehni]

@iconexperience this was a tough one to crack! But with this new code, no recalculation of curve-time should ever be required, and rescaling is correctly handled if dividing happens before and after a given location. I hope it was worth the effort for performance...

[iconexperience]

@lehni
There are some problems with the new code. Here is an example (which is quite and awful path):

var p = new paper.Path({segments:[
        [256.22272245372784, 322.2911416479855, 1.3846994306732086e-7, -1.0032863428932615e-10, -0.07145907403264573, 0.000051770563516129187],
        [256.28274925468855, 322.29109816063755, 33.69934566220388, -0.024413795440409558, -0.020001069211389222, 0.000014489955333374382],
        [256.222722318658, 322.29114164808345, 0.020004442897743502, -0.000014492788977804594, -1.3846999991073972e-7, 1.0032863428932615e-10]
    ], closed:true});
p.resolveCrossings();

The variable rescaleLocs in divideLocations() is not initialized. This may be easy to fix.

[lehni]

Did you try the latest state of the develop branch? I already fixed this, I think.

[iconexperience]

@lehni Sorry, your are correct. I tested agains an old version. This has already been fixed.