[BUG] Arc using width and thickness doesn't attach cleanly

[MacksMind]

Describe the bug An arc defined using width= and thickness= has a strait side along the X axis. Attaching to a square and using in difference() reveals an imperfection in the attachment. A similar imperfection is seen when using translation to position the arc. Both are sensitive to the arc parameters, and the imperfection appears and disappears as parameters change.

Code To Reproduce Bug

include <BOSL2/std.scad>

w_list = [117, 120, 125, 130, 145, 150, 185];
h1 = 80;
h2 = 20;
z = 10;
r = 5;

module toast(w, join=false) {
  square([w,h1])
  if(join)
    attach(BACK, FRONT) arc(n=30, width=w, thickness=h2);
  else
    back(h1/2) arc(n=30, width=w, thickness=h2);
}

for(i = [0: len(w_list)-1]) {
  right(i*200) {
    fwd(h2 + 10) {
      linear_extrude(height = z)
      difference() {
        offset(r=r) toast(w_list[i], true);
        toast(w_list[i], true);
      }
    }

    back(h1 + 10) {
      linear_extrude(height = z)
      difference() {
        offset(r=r) toast(w_list[i], false);
        toast(w_list[i], false);
      }
    }
  }
};

Expected behavior A clean outline with no internal artifacts.

Screenshots

Additional context Outlines with internal artifacts produce GCAL errors.

• OpenSCAD Version: 2024.05.20 (openscad/openscad@8a890a9e1)
• Other libraries used: BOSL2 (f9e6e67)

[adrianVmariano]

This appears to happen only with the dev version of OpenSCAD. It also appears that it's just OpenSCAD doing it's thing where you've got two shapes that exactly share a boundary and it gets confused. I haven't seen that happen in 2D before, but I also don't tend construct geometry that way. If you use attach() and give $overlap=0.01 then the extra junk goes away.

It's interesting that this problem doesn't occur in the stable version. I guess you could post an OpenSCAD issue on the OpenSCAD github and see what they say.

[adrianVmariano]

I poked around a bit more and realized that sometimes the arc has an endpoint like [x,1e-15]. It appears that this is the cause of the issue. The stable OpenSCAD must be handling such points differently than the dev version. This sensitivity seems a bit unfortunate. But in this case I have a fix where I ensure that arc endpoints fall exactly where they were specified.

[MacksMind]

@adrianVmariano Thanks!