Better faces

[elalish]

I'm replacing our CreateFaces algorithm. Our previous was based on connected components, but in the case of e.g. a high-res sphere, it would connect everything into one big face. So then we added an algorithm to check global tolerance of each face after the fact, and if that failed, it would just throw out all the faces and say each triangle was individual.

This new algorithm skips all that and instead finds faces by starting with the largest triangle and searching its neighbors until none are within tolerance. Then it gets the next largest unprocessed triangle and does it again, serially, until all triangles have been assigned to faces. It's a bit arbitrary, but works decently.

Ideally we can eventually follow this on with some improved edge collapsing code.

[fire]

This could be a way to attach an edge-length refiner with collapse or subdivide done per polygon.

[pca006132]

Is this causing the check in line 238 and 249 to always be false?

[elalish]

No, because they get reassigned on lines 240 and 253.

[fire]

can the random selection be more deterministic for like ai.. and import recreation / reproducibility

[elalish]

can the random selection be more deterministic for like ai.. and import recreation / reproducibility

Since this is based on sorting, it should be as deterministic as we can hope for.

[fire]

Can you see what happens when you try to merge https://github.com/user-attachments/files/17969298/node_3d.zip ?

[elalish]

Okay, I got bogged down on this one because I tried to solve too many things at once. The actual point of this PR, which is to create faces better and faster is working just fine. Then I tried to make it into a general purpose decimator, but got stuck on some Boolean regressions. Now I've undone that to get back to basics here. I'll do further work on the decimator in separate PRs.

[elalish]

@pca006132 This is odd - the CI is passing almost everywhere, but on three jobs it's failing, and not just a few tests, but a ton of them and even seg-faulting. On my mac everything is fine - even the address sanitizer isn't finding any undefined behavior. Do you have any idea what I did wrong?

[pca006132]

It seems that I can reproduce them locally with parallelization disabled. Will check.

[pca006132]

OK, for some reason Linux stdlib.h defines an int abs(int);, changing to std::abs fixes the issue.

[codecov]

Codecov Report

Attention: Patch coverage is 98.27586% with 1 line in your changes missing coverage. Please review.

Project coverage is 91.65%. Comparing base (447921c) to head (34062e6).
Report is 1 commits behind head on master.

Files with missing lines	Patch %	Lines
src/impl.cpp	98.24%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #1040      +/-   ##
==========================================
- Coverage   91.71%   91.65%   -0.06%     
==========================================
  Files          30       30              
  Lines        5951     5922      -29     
==========================================
- Hits         5458     5428      -30     
- Misses        493      494       +1     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[elalish]

Wow, thanks for the fix! I wonder what their abs was doing... One of these days I'll remember to always put std:: in front of everything.