Going through all of my NURBS regression tests, this fails with a file named wrong2.slvs

I'll try more variations. Going over every curve seemed to be the problem. Still appreciate feedback, this is looking kind of ulgy IMHO but still somewhat contained.

I need to debug this... not enough stack in my brain to think about it "on paper" (the phone screen) :-)

@ruevs If you look at a simplified test using triangle mesh:

I think what's happening but not certain: The two surfaces from the colinear sketch lines get merged into one surface but I don't think the lines get merged, so it has 6 trim curves. Then when it does union with the large box, the long edge is split at the perpendicular surfaces of the small box, but not at the point where the 2 shorter trim curves meet, so then it fails to assemble everything correctly. Splitting a line at every curves end vertex caused the additional split, but since the 2 coincident rectangles merged into one, splitting at that surface corners is not enough. have not confirmed but it seems plausible.

I suspect going over all the curves was killing cache performance in addition to that being a complex test (nearest point to curve). Every curve was being checked against every curve and then every surface in the other shell. Maybe I'll try this surface bounding box test and then check the ends of that surface trim curves instead of the surface corners.

Updated with MakeMaxMin, changed the comment, pulled the OMP out into a different PR.

@ruevs I changed it back to using curve end verticies. It's faster and fixes wrong2.slvs as well. Turns out the performance problem was solved by using bounding boxes prior to doing math like ClosestPointTo().

Should it stay two commits or should I squash them?

@phkahler I apologize for the impertinence but I pushed a small cosmetic improvement to the code. NFC.

The performance is great.

• failed.lathe.slvs.zip works
• adapter-plate.tar.gz works
• wrong.zip still broken when you shorten the big extrusion - as before
• wrong2.zip works
• coincident4.zip works

@phkahler I apologize for the impertinence but I pushed a small cosmetic improvement to the code.

No problem at all. BTW how do you push changes to a PR like that?

Going to merge now ;-)

I think that if you check the box "allow maintainers to change" or something when creating the pull request then it gives rights to them to push to your branch.

So I have a remote for you:
phkahler https://github.com/phkahler/solvespace

So:
git fetch --all
git checkout phkahler vertex-split
change stuff
git add.... whatever
git commit --amend
git push --force