geogram
geogram copied to clipboard
Improve the halfedges API
Summary
This PR extends the halfedges API without breaking existing functions. The only modified files are src/lib/geogram/mesh/mesh_halfedges.h
/.cpp
.
Motivation
I need to move on surface meshes and the halfedges API of Geogram allowed me to implement the operators I wanted. But I had several issues:
- this part of the library was not very documented (mentioned in the wiki , just the doxygen comments and almost untouched since the switch to GitHub)
- the absence of some getters (origin/extremity vertex indices, right facet) results in a lot of redundancy in my code
- the API is too low level (the user needs to know where are
H.facet
andH.corner
in relation toH
, the meaning of next and previous is unclear) - I ended up creating a child class with custom methods and tweaking things until functions were doing was I expected
Because my code relies more and more on halfedges, I plunge into the original class to understand what was already done and to add what I needed. I found that 'next' and 'prev' don't seem to follow the same convention around facets and vertices...
Notes on existing functions
Could be helpful in the wiki
New functions
-
halfedge_vertex_index_from()
: because before the user could access the 3D coordinates of the origin but not the vertex index -
halfedge_vertex_index_to()
: because before the user could access the 3D coordinates of the extremity but not the vertex index -
halfedge_facet_left()
which is basicallyH.facet
but now the user doesn't need to know the underlying structure -
halfedge_facet_right()
, to improve readability of the user's code -
move_clockwise_around_facet()
which is more explicit thanmove_to_prev_around_facet()
-
move_counterclockwise_around_facet()
which is more explicit thanmove_to_next_around_facet()
-
move_clockwise_around_vertex()
which is more explicit thanmove_to_next_around_vertex()
-
move_counterclockwise_around_vertex()
which is more explicit thanmove_to_prev_around_vertex()
I hope those modifications are in line with your vision for Geogram.
Let me know if something should be tweaked. I particular in-code documentation about 'next' and 'prev' whose meaning I must have missed.
I can share the Excalidraw file if it is useful.
Here is a small app (can be appended to the PR) that showcase this behavior:
-
move_to_prev_around_facet()
→ clockwise -
move_to_next_around_facet()
→ counterclockwise -
move_to_prev_around_vertex()
→ counterclockwise -
move_to_next_around_vertex()
→ clockwise
Maybe prev/next make more sense around facets, but I think (counter)clockwise makes more sense around vertices.
PLOT TWIST : I still don't understand everything but it seems like I was wrong about the halfedges representation.
-
H.facet
→ ~left facet~ right facet -
move_to_prev_around_facet()
→ ~clockwise~ counterclockwise -
move_to_next_around_facet()
→ ~counterclockwise~ clockwise -
move_to_prev_around_vertex()
→ ~counterclockwise~ clockwise -
move_to_next_around_vertex()
→ ~clockwise~ counterclockwise
Incoming fix
@BrunoLevy my modifications are now ready for review
Addition:
- ASCII representation of a halfedge with the underlying
facet
andcorner
- comments about the
facet
andcorner
fields - getters for the vertex index at the origin/extremity (there were only getters for their coordinates)
- getters for the facet index at the left/right, for better abstraction
- (counter)clockwise movement around a facet (can useful when the halfedge direction is not important)
- (counter)clockwise movement around a vertex, more natural than next and previous (better abstraction)
- allow to ignore borders when moving around a vertex, to avoid to set/unset facet regions when the user need to parse all facets around a vertex, while using facet regions
There were inconsistencies in the facet normals direction of the meshes I tested (some inward, some outward).
This leads us to a design choice:
- Should Geogram expect outward normals? → make it explicit & provide (counter)clockwise movements
- Should Geogram expect consistent normals directions and works with both cases? → provide adaptive (counter)clockwise movements and either construct a
MeshHalfedges
with an additional argument, or makeMeshHalfedges
detect facet normals direction - Should Geogram not expect a particular normals direction? → only add the comments, getters for vertex indices & 2nd argument of
*_around_vertex()
Hi there ! Wow, impressive work and documentation ! It's fantastic ! (sorry for beeing not reactive at all, I was completely swamped in many projects + administrative tasks). About the names of the things, we do not trust orientation, this is why the names of the functions are not "clockwise" and "counterclockwise" because depending on the input mesh, they may be swapped ! And also, consider a non-closed surface, then it depends from which side you look at the surface. Anyway, at any moment if you want to discuss we can plan a videochat (will be easier than exchanging written messages, especially if we need to draw things).
Now this PR doesn't assume any orientation (3rd design choice of my last message). No more move_(counter)clockwise_around_facet()
, move_(counter)clockwise_around_vertex()
.
In-code documentation
- Explain what
MeshHalfedges
,MeshHalfedges::Halfedge
andMeshHalfedges::Halfedge.corner
are -
MeshHalfedges::Halfedge.facet
is described as the supporting facet (no more left or right) - ASCII representation that explicitly assume counterclockwise vertices order (outward normal)
Allow to check if facet regions are defined
🆕 is_using_facet_region()
To be asserted at the beginning of functions receiving a MeshHalfedges
and expecting facet regions
Allow to ignore borders when facet regions are used
with a new argument for move_to_*_around_vertex()
Useful when we need to keep track of all the facets between two border segments
Allow sub-classes to access class attributes
private
→ protected
Getters for the index of the halfedge's origin/extremity vertices
Currently we can have their 3D coordinates (halfedge_vertex_from()
and halfedge_vertex_to()
) but not their index.
Because I'm used to the "point" naming from mesh.vertices.point()
for coordinates, I wanted to rename existing functions:
-
halfedge_vertex_from()
→halfedge_point_from()
which returns avec3
-
halfedge_vertex_to()
→halfedge_point_to()
which returns avec3
- 🆕
halfedge_vertex_from()
which returns anindex_t
- 🆕
halfedge_vertex_to()
which returns anindex_t
But inside, the coordinates getters use Geom::mesh_vertex()
from mesh_geometry.h
, so it seems "vertex" sometimes means the index, sometimes the coordinates.
So, I've chosen:
-
halfedge_vertex_from()
which returns avec3
(unchanged) -
halfedge_vertex_to()
which returns avec3
(unchanged) - 🆕
halfedge_vertex_index_from()
which returns anindex_t
- 🆕
halfedge_vertex_index_to()
which returns anindex_t
Getters for both facets
- 🆕
halfedge_facet_main()
= supporting facet =H.facet
- 🆕
halfedge_facet_secondary()
= facet at the other side of H
Getters for facet corners
- 🆕
halfedge_bottom_main_corner()
= facet corner of the main facet which is on the origin vertex - 🆕
halfedge_bottom_secondary_corner()
= facet corner of the secondary facet which is on the origin vertex - 🆕
halfedge_top_main_corner()
= facet corner of the main facet which is on the extremity vertex - 🆕
halfedge_top_secondary_corner()
= facet corner of the secondary facet which is on the extremity vertex