Description
The validity criterion on corners is currently
The incident boundaries of a corner must not be associated to the same axes, they must be associable in pairs, or make a πππ trio
Turns out, some corners are wrongly classified as valid. Below is a "pinching corner" with two X and two Z incident boundaries. Under the prescribed axis alignments, the red triangle chart will be crushed and the polycube will not conform to the geometry.
Here is the STEP file for this model valence-4_pinching.step.gz
This configuration appears in several test models: MAMBO B59, S25, M3...
A better validity criterion would be
The incident boundaries of a corner must be associable in pairs or make a πππ trio, and two successive boundaries must not be associated to the same axe
Note that invalid corners are currently "fixed" by inserting a chart on and around where the invalid corner was located. This will not be a great fix on pinching corners. We should distinguish pinching corners and high valence corners (where incident boundaries are associated to the same axis) when we process invalid corners.
Extending the adjacent blue chart no longer captures some feature edges, but this leads to a better polycube:
