distance calculation can be wrong for rings with collinear segments

[stfn <p1433025@…>]

While unit-testing our code using boost::geometry, I stumbled upon a weird distance calculation: For two rings that are perfect 2d boxes and share the orientation and one segment length, their distance is calculated as zero although they are clearly not intersecting:

​https://wandbox.org/permlink/Z9DMLo52IdhZRCFb

I made a rough sketch of the situation: ​https://www.desmos.com/calculator/7jbmeu9jqq.

---

Note the line

// slightly change the values
vehicleRing[2] = Point2D(vehicleRing[2].x() + 1e-10, vehicleRing[2].y() + 1e-10);

.. changing vehicleRing[0] or vehicleRing[1] also results in zero distance, whereas slightly moving vehicleRing[2] results in the expected distance.

---

Sorry for being lazy and not finding a simpler test-case (resp. simpler numbers), I just copied the values I generated in my own testing pipeline.

[stfn <p1433025@…>]

updated example: ​https://wandbox.org/permlink/837NbUIZ7fNNOKEF

[stfn <p1433025@…>]

Replying to stfn <p1433025@…>:

updated example: ​https://wandbox.org/permlink/837NbUIZ7fNNOKEF

damn! this one: ​https://wandbox.org/permlink/p51mDlCIKOjOTHcW