Don't set the point again when reinserting the same point

[mglisse]

Summary of Changes

In Triangulation and Delaunay_triangulation (but not Regular_triangulation with equal weights), if we insert a point and locate notices that there is already a vertex for this point, we call set_point to set this point again. I don't understand why we do that. In most cases it should be useless but harmless. To notice a difference, we would need a Point_d type for which 2 objects may represent the same position but differ in some way (probably include extra data like a color). But even in that case, it isn't obvious to me that using the color of the last copy of this point that gets inserted is better than the color of the first copy. If someone was using this mechanism to store the index of points, having a vertex that changes what point it represents in the middle of the construction could break some uses.

It would be nice if someone else could check the logic, not just trust me.

(I don't have any particular use for this, I only noticed it while copy-pasting insert so I could customize it)

Release Management

• Affected package(s): Triangulation
• License and copyright ownership: unchanged

[afabri]

What does the documentation say? How is that handled in 2D and 3D?

[lrineau]

The previous behavior is the one that is documented:

• in Triangulation::insert

  ON_VERTEX
  The point of the pVertex described by f is set to p.

• in Delaunay_triangulation::insert

  ON_VERTEX
  The position of the vertex v described by f is set to p. v is returned.

That behavior seems odd. If one want to change it, first check if that was mentioned during the review, and then eventually change the documentation and publish a breaking change.

[mglisse]

How is that handled in 2D and 3D?

In 3d and 2d:
https://doc.cgal.org/latest/Triangulation_3/classCGAL_1_1Triangulation__3.html#ad8d7a1aa3b310ba8d86ede726746fcb3
https://doc.cgal.org/latest/Triangulation_2/classCGAL_1_1Triangulation__2.html#a1025cd7e7226ccb44d82f0fb1d63ad4e

If point p coincides with an already existing vertex, this vertex is returned and the triangulation remains unchanged.

I am trying to change dD so it behaves likes 2D/3D.

[mglisse]

The previous behavior is the one that is documented:

Oops, I lost the reflex to look for the documentation in a completely different place than the code...

That behavior seems odd. If one want to change it, first check if that was mentioned during the review,

I searched for "insert" in the 3 rounds of reviews of Triangulation_d and didn't find anything related.

and then eventually change the documentation and publish a breaking change.

I set the PR to draft in the meantime.

[lrineau]

Looks good.

[sloriot]

Successfully tested in CGA-6.0-Ic-173