The circles in your "round" module are touching at single points, which is
not manifold.

If you modify it to only round one corner at a time, and rotate
appropriately at each position, then  it works fine.

module inside(){
polygon(points=[[0,0],[0,43],[43,43],[43,30],[48,30],[48,13],[43,13],[43,0]]);
}

module round(r,corner=0){
rotate(corner90) difference(){
square(2r+0.001,true);
translate([r,r]) circle(r);
}
}

module fillet(r,h){
difference() {
translate([0,0,h/2]) cube([2r + 0.001, 2r + 0.001, h], center
= true);
translate([r, r, 0]) cylinder(h,r,r);
translate([-r, r, 0]) cylinder(h,r,r);
translate([-r, -r, 0]) cylinder(h,r,r);
translate([r,- r, 0]) cylinder(h,r,r);
}
}

module roundinside(){
union(){
difference(){
inside();
translate([0,0])round(8,0);
translate([43,0])round(8,1);
translate([43,43])round(8,2);
translate([0,43])round(8,3);
translate([48,30])round(4,2);
translate([48,13])round(4,1);
}
translate([43,30])round(1.5,0);
translate([43,13])round(1.5,3);
}
}

module outside(){
offset(1.5) roundinside();
}

module wall(){
difference(){
outside();
roundinside();
}
}

$fn=80;

difference(){
union(){
linear_extrude(10) wall();
translate ([-8,17,0])cube([8,10,10]);
translate ([-1.5,17,0])fillet(1.2,10);
translate ([-1.5,27,0])fillet(1.2,10);
}
translate([-4.5, 10,5]) rotate ([-90,0,0]) cylinder (20,1.5,1.5);
translate([-20,22,0])cube([20,1,20]);
translate([-8,0,0])rotate([-90,0,0])fillet(3,50);
translate([-8,0,10])rotate([-90,0,0])fillet(3,50);
}

On Tue, Mar 2, 2021 at 7:59 AM Ray West raywest@raywest.com wrote:

On 02.03.2021 14:59, Ray West wrote:

Creating "circles" with discrete number of sides is in my opinion not
the best idea. I would have extruded polygons for this purpose since you
then create the required geometry explicitly. I think geometry
specification and mesh discretization is best considered different things.

In AngelCAD you can easily create a polygon from a radius and n sides
and create a "cylinder" by extrusion:

linear_extrude(polygon(r:100,np:8),100);

this is equivalent to OpenSCAD

linear_extrude(100) {
	   circle(100,$fn=8);
}

Carsten Arnholm

Em ter., 2 de mar. de 2021 às 14:29, Hans L thehans@gmail.com escreveu:

That doesn't explain everything. To simplify let us take the
following definition of fillet() which is equivalent to the original:

module fillet(r,h){
linear_extrude(h,convexity=4) round(r);
}

If we view the preview fillet(1.2,10) from the top with edges on and $fn=80
we get:
[image: wisker80.PNG]
So, the very tip of the fillet() base is formed by a reasonably fat
triangle and that should not be considered a non-manifold. However it is
rejected by F6 with an unusual error message when two disjoint fillets are
unioned with:

union(){
 //   linear_extrude(10,convexity=10)wall();
 //   translate ([-8,17,0])cube([8,10,10]);
    translate ([-1.5,17,0])fillet(1.2,10);
    translate ([-1.5,27,0])fillet(1.2,10);
 }

BTW, the error message of F6 is not the usual non-manifold warning but
rather:

Rendering Polygon Mesh using CGAL...

ERROR: The given mesh is not closed! Unable to convert to
CGAL_Nef_Polyhedron., location unknown <0,>

ERROR: The given mesh is not closed! Unable to convert to
CGAL_Nef_Polyhedron., location unknown <0,>

UI-WARNING: No top level geometry to render location unknown <0,>

that I had never seen before. I got those results with RC3.

Thanks for the replies, Although I could make a satisfactory print, i
was curious as to why, depending on the $fn value, it would often not
render. Thanks, Hans, for showing me the error of my ways. I very rarely
use openscad by extruding 2d, normally I use 3d solids, merely
adding/subtracting parts as required. Thanks for identifying the
problem, and correcting my code. I derived the 2d 'round', based on my
3d version ('fillet' in my code) which seems to give no problems for any
$fn value that I've tried. If  I alter my 'round' code, so the circles
do not touch, then hopefully that will work too. I was curious if there
was some reason, similar to the 'tip' in the document, explaining which
values of $fn were valid. It seems, that if it is not manifold, then
anything can happen.

On 02/03/2021 14:28, Hans L wrote:

A further investigation shows that the original fillet() and round()
modules produce in fact non manifolds, not because the circles and
cylinders are touching but because the square (in round()) and the cube (in
fillet()) are a bit bigger than they should be. Ray West's model works fine
and render successfully with the redefinitions:

module round(r){
difference(){
square(2r / + 0.001 */,true);
translate([r,r]) circle(r);
translate([-r,r]) circle(r);
translate([r,-r]) circle(r);
translate([-r,-r]) circle(r);
}
}

module fillet(r,h){
difference() {
translate([0,0,h/2]) cube([2r / + 0.001 / , 2r  /* + 0.001 */
, h], center = true);
translate([r, r, 0]) cylinder(h,r,r);
translate([-r, r, 0]) cylinder(h,r,r);
translate([-r, -r, 0]) cylinder(h,r,r);
translate([r,- r, 0]) cylinder(h,r,r);
}
}

With the redefinition of fillet() in my previous message I have solved part
of the trouble but round() was kept as the original and the
non-manifoldness was still present.

Em ter., 2 de mar. de 2021 às 17:59, Ronaldo Persiano rcmpersiano@gmail.com
escreveu:

For the sake of completeness,

Here's my modified round module, gives no errors -

module round(r){
   difference(){
       square(2*r,true);
          translate([r+0.001,r+0.001]) circle(r);
          translate([-r-0.001,r-0.001]) circle(r);
          translate([r+0.001,-r-0.001]) circle(r);
          translate([-r-0.001,-r-0.001]) circle(r);
   }
}

On 02/03/2021 18:36, Ray West wrote: