example attached where a fillet is created:
calculations for this are done in python.
[image: Screenshot 2023-07-22 at 9.41.49 PM.png]

On Sun, 23 Jul 2023 at 21:37, Sanjeev Prabhakar sprabhakar2006@gmail.com
wrote:

On 23.07.23 18:34, Sanjeev Prabhakar wrote:

That is great stuff, but in my view it would belong into the geometry
engine.

Maybe you can help with the calculation side of that?

ciao,
Torsten.

On Sun, Jul 23, 2023 at 09:37:10PM +0530, Sanjeev Prabhakar wrote:

I agree that a procedural language addition to openscad would be very
useful. Some things are much easier to express procedurally than the
way openscad would like you to formulate things.

For example building up an array element by element is awkward in
openscad, while "natural" in a procedural language.

It took me a long time to figure out why a module using a global
variable wouldn't instantiate in two different ways when I changed the
global variable between calls. (back then openscad didn't have the
warning that is generated nowadays).

I agree with the openscad folks that it is an advantage that an
openscad script is just "data": It cannot affect your computer beyond
bogging it down for a while while trying to render it. Just like a
"word processor documuent" should be.

How about allowing the python extension to be activated by a
commandline argument but also by using a different binary name.

pyopenscad, popenscad or maybe openpyscad. (That last one... might not
be such a good idea... ). And a different extension.

For those that are paranoid (and those that don't have python at
compile-time) it should remain possible to compile without python
support.

I'm hoping such a construction would satisfy both ends of this
discussion.

Roger.

--
** R.E.Wolff@BitWizard.nl ** https://www.BitWizard.nl/ ** +31-15-2049110 **
**    Delftechpark 11 2628 XJ  Delft, The Netherlands.  KVK: 27239233    **
f equals m times a. When your f is steady, and your m is going down
your a is going up.  -- Chris Hadfield about flying up the space shuttle.

Thanks

It's a little complicated to explain,  but the main thing is calculating
the intersection line between 2 solids.

Once you get the intersection line,  then it's not so difficult to create
any fillet.

On Sun, 23 Jul 2023, 22:16 Torsten Paul, Torsten.Paul@gmx.de wrote:

On Sun, Jul 23, 2023 at 06:45:41PM +0200, Torsten Paul wrote:

This specific case of "a fillet" is something that might be welcome in
openscad. But that is not the point that Sanjeev is trying to make.

There are manythings that are relatively easy to do in
python-enhanced-openscad, that are hard to do in vanilla openscad. Not
just fillets. His example might make sense to "Oh, but we want that to
be possible in plain openscad". You can't keep saying that for
everything that anybody can come up with.

Roger. 

--
** R.E.Wolff@BitWizard.nl ** https://www.BitWizard.nl/ ** +31-15-2049110 **
**    Delftechpark 11 2628 XJ  Delft, The Netherlands.  KVK: 27239233    **
f equals m times a. When your f is steady, and your m is going down
your a is going up.  -- Chris Hadfield about flying up the space shuttle.

On 23.07.23 18:59, Sanjeev Prabhakar wrote:

I have not looked at the details, but my understanding from some discussion
in the Manifold repository is there are IDs in the CSG output that detail to
which part the new faces belong. From that it should be possible to get the
information which vertices are that intersection line.

Maybe we can try some demo code separate from OpenSCAD to see if we can get
your algorithm into Manifold based C++ code.

ciao,
Torsten.

On Sun, Jul 23, 2023 at 10:29:34PM +0530, Sanjeev Prabhakar wrote:

I'm guessing that it can become complicated.
Try this:

fillet_radius = 3;
dist = 10;
size = 10;

fillet (radius= fillet_radius) {
cube ([size2+dist, size, size]);
cube ([size, size, 2size]);
translate ([size+dist, 0,0]) cube ([size, size, 2*size]);
}

Easy. But what if dist is reduced to below 2*fillet_radius ? You can
intersect the radii but then you'll get a sharp inside corner that the
fillet is supposed to be preventing.

And what if dist is reducted to below 1*fillet_radius ? Does the space
between the objects fill completely with fillet?

In the general case, this is quite hard.

Roger. 

--
** R.E.Wolff@BitWizard.nl ** https://www.BitWizard.nl/ ** +31-15-2049110 **
**    Delftechpark 11 2628 XJ  Delft, The Netherlands.  KVK: 27239233    **
f equals m times a. When your f is steady, and your m is going down
your a is going up.  -- Chris Hadfield about flying up the space shuttle.

On 23.07.23 18:59, Rogier Wolff wrote:

The related logic is already merged into OpenSCAD master and the
PR adding the actual Python interface is accepted in spirit, just
pending some detailed code review. So it's only a matter of finding
the time, no general open discussion.

https://github.com/openscad/openscad/pull/4702

Getting builds to enable this feature is going to be a challenge,
as I mentioned before, but we'll have to see if we can get some
help with that.

ciao,
Torsten.

I've been working on BOSL2 for a while with Revar, and we have 60k lines of
OpenSCAD code.  Based on that experience, I think that the problem with the
OpenSCAD language is NOT fundamentally that it is functional.  The benefit
of python is NOT fundamentally that it is procedural.  There are two basic
problems with OpenSCAD for doing more complicated user space processing.
One is the inability to gain access to geometry, so to do user space
design, everything from the ground up has to be done "outside" of the
OpenSCAD geometry engine.  The second problem is about efficiency.
OpenSCAD has no data structures.  Graphics algorithms assume access to
sophisticated and efficient data structures like a priority queue or a tree
or other kinds of things that are impossible to implement efficiently in
OpenSCAD.  So, yeah, you can implement the data structure, but instead of
being O(log N) or whatever it will be O(N^2).  So then trying to implement
that process that uses that data structure in user space becomes
intractable.  It's not clear to me that this limitation is fundamentally
about the language being functional.

Note that the filleting process displayed by Sanjeev is straight forward to
implement in principle.  You just find the intersection of two things and
put a fillet there.  Two complications arise for doing this in userspace in
OpenSCAD.  The first is representing the things, because again, you can't
use the geometry engine to do it.  Now maybe a solution to that problem is
coming?  The second complication is dealing with self-intersections of the
resulting polyhedron, because that's intractable in OpenSCAD.  Really it
seems like OpenSCAD should have a primitive that accepts a series of
profiles and links them together into a polyhedron, even if it's
self-intersecting.

Here's examples of doing the fillet problem entirely in OpenSCAD
userspace.  There are many examples at the link below.  I show a single one
in this message.

https://github.com/BelfrySCAD/BOSL2/wiki/rounding.scad#functionmodule-join_prism

[image: image.png]

On Sun, Jul 23, 2023 at 1:14 PM Sanjeev Prabhakar sprabhakar2006@gmail.com
wrote:

I am not sure 3D fillets are any use in FDM printing because of the layers.
I only use 2D fillets to get rounded tool paths.

On Sun, 23 Jul 2023, 20:08 Adrian Mariano, avm4@cornell.edu wrote:

Are you thinking that the layers break the propagation of stress through
the model, causing stress to concentrate at each layer, instead of
concentrating at a corner?  Have you observed that models which fail by
layers shearing do so and random places in the model?

On Sun, Jul 23, 2023 at 3:18 PM nop head nop.head@gmail.com wrote:

Adriano, thank you for your input
I appreciate your work with Revar on BOSL2 and I understand your reasoning
that BOSL2 could become obsolete,
but I am sure it will not. When openscad becomes more powerful, BOSL code
might be able to code with fewer lines, but  BOSL2 is still very important
to openscad.

Yes, I would appreciate OpenSCAD to have access to the internal vertices to
read them and work with the coordinates.

Unfortunately this is not as easy, because often a model is developed in
OpenCSG and there are no coordinates available in OpenSCG.
If you accept the fact that a full render has to be performed to get
coordinates of a submodule. It's just a matter of a PR.
If somebody guarantees,  that such a PR will actually be accepted/merged,
time to develop on such a PR would not be an issue.

My Main Problem with SCAD language is that variables cannot be reassigned,
once they are set -> could probably be solved with most other interpreter
languages.
The main benefit I see with python is not the opposite of the previous
statement.  Python is just any interpreter language choice. but its very
easy to learn and very easy to read.
PLUS: it has many accompanying libraries which will help you to do fancy
stuff with OpenSCAD.
for example this one:

http://guenther-sohler.net/openscad/examples/qrcode.txt
these few lines of code creates an qr code display object , which you can
scan  with your mobile. (Yes, i have seen openscad qr code libs, but this
one appears easier and has more options)

And your second statement : OpenSCAD has datastructures is also valid.
Polyhedra are composed of triangles which are stored as coordinates only!
It would be much better to compose triangles from
indices into a global vertex array. This is something I am right now
working on.(some ctests are not passed for my PR)
Information which could also be stored with the  "indexed " polygons is: a
lookup table from point -> which triangles use this point.

And Yes, I like your fillet and I personally appreciate , if Sanjeev's
knowledge could go into GeomtryEvaluator.cc

In the End, I'd like to see openscad become more powerful by being able to
change variables once they are set and to use external libs.
And yes, I know that I am aware that people might see this as a security
risk, but those people can stay safe and do not even need to bother when
not activating it.
The rest and the interested people instead can leverage the new power
coming from python.

Thank you for reading

On Sun, Jul 23, 2023 at 9:08 PM Adrian Mariano avm4@cornell.edu wrote:

The fact that prints always break along layer lines means adding a fillet
that is made of layers hardly affects the strength in my opinion.

On Sun, 23 Jul 2023, 20:24 Adrian Mariano, avm4@cornell.edu wrote:

nop head, this might be true with  an original print.
But if you e.g. handle ABS with acetone this is no longer the case.
In that case objects with  fillets will break far less than an inbound
sharp corner

cylinder(d=20,h=2);
cylinder(d=5,h=20);

the "edge" between the cylinder might very easily break, but you improve
the strength dramatically when you add a fillet instead.

On Sun, Jul 23, 2023 at 10:00 PM nop head nop.head@gmail.com wrote:

I'm confused...

On 2023-07-23 12:45 p.m., Guenther Sohler wrote:

I must have misunderstood your statement as it seems to me to say that
you're working on something that OpenSCAD's polyhedron command already
does: it takes a single list of vertices and each face is defined by a
list of vertex indices.

Could you clarify for me, please.

Steve

Sorry Steve,

I actually did not express myself very well.

you you create  cube in openscad, its stored internal as
6 faces, each containing 4 vertices and each vertices contains 3 values
representing x/y/z

the internal representation of openscad is actually triangles containing
coordinates instead of indices.

if you use openscad  polyhedron command in openscad of course you specity a
point list and a faces list containing indices.
but its immediately translated to this stupid coordinate-only format.

it would be much better if openscad object would interally contain:

• a list of vertices
• a list of polygons whereas each polygon is just a list of indices.

This is what i am currently working on and Marius appears to appreciate it,
but he wants a "Polygon Builder" in addition.

On Sun, Jul 23, 2023 at 10:15 PM Steve Lelievre <
steve.lelievre.canada@gmail.com> wrote:

Ah, now I understand.

Thanks

On Sun, Jul 23, 2023 at 13:23, Guenther Sohler guenther.sohler@gmail.com
wrote:

--
Cell +1 778 837 5771

On Sun, Jul 23, 2023 at 09:00:06PM +0100, nop head wrote:

In some cases, you can print it so that the largest load ends up in
the X or Y direction and then a fillet perpendicular to that ends up
being printed in one layer in the XY plane. This can benefit from the
fillet from strength.

That said: I think also a fillet in the other direction will help with
strength. I have seen objects delaminate exactly in the stresspoint
where a fillet would have helped.

And even if it doesn't help, it looks nice and less "openscad-like".

Roger. 

--
** R.E.Wolff@BitWizard.nl ** https://www.BitWizard.nl/ ** +31-15-2049110 **
**    Delftechpark 11 2628 XJ  Delft, The Netherlands.  KVK: 27239233    **
f equals m times a. When your f is steady, and your m is going down
your a is going up.  -- Chris Hadfield about flying up the space shuttle.

please see the explanation.
Although it seems quite complicated, but it takes around 1 sec to compute
and display the results
I hope this helps

On Sun, 23 Jul 2023 at 22:43, Sanjeev Prabhakar sprabhakar2006@gmail.com
wrote:

On Mon, Jul 24, 2023 at 10:01:22PM +0530, Sanjeev Prabhakar wrote:

Thanks for your nice explanation of the algorithm.

But what happens if I try to fillet this with radius of about 5:

cube ([40,40,10]);
translate ([0,20,12]) rotate ([0,90,0]) cylinder (d1=15, d2=0, h=40);

?

At x=0... the intersection line is not a line, but a plane. Or the
curve is not closed as you expect.

Near x=0, there is probably no problem.

But as x increases... ehh actually, directly from the start: the
bezier will partially lie INSIDE the orginal object. This means that
if you take that curve instead of the original object you'll have
taken material AWAY from the object. That is something that you don't
want to happen with a fillet.Think of adding a fillet as adding putty
to the sharp corner between the objects. This only ADDS to the
objects, never subtracts.

Roger. 

--
** R.E.Wolff@BitWizard.nl ** https://www.BitWizard.nl/ ** +31-15-2049110 **
**    Delftechpark 11 2628 XJ  Delft, The Netherlands.  KVK: 27239233    **
f equals m times a. When your f is steady, and your m is going down
your a is going up.  -- Chris Hadfield about flying up the space shuttle.

The approach explained in my earlier message will not work in such cases.

[image: Screenshot 2023-07-24 at 11.32.15 PM.png]

On Mon, 24 Jul 2023 at 22:35, Rogier Wolff R.E.Wolff@bitwizard.nl wrote:

On 24.07.23 20:06, Sanjeev Prabhakar wrote:

Thanks for the explanation, this is nicely visualized. I suppose
subdividing some of the surfaces might also be needed?

Are there solutions that are more general? For the experienced user
it's probably fine to give a couple of strategies and let them decide,
but for the casual user that could be difficult to select what to use
and when.

What do GUI CAD systems come up with for this example with the cone?

ciao,
Torsten.

On 7/24/23 12:32, Sanjeev Prabhakar wrote:

And this looks as if a set of tool paths could be defined in gcode from
it, RS-274-D IOW, respecting the machines spindle power & rpm ability
for depth of cut. # of loops and final finish accuracy determined by how
close the radius of the fillet matches the tool nose radius. If this can
be done in a useful time frame, the finished code to do this for
subtractive machining would be a good src of funds to bank for your
retirement.

Cheers, Gene Heskett.

"There are four boxes to be used in defense of liberty:
soap, ballot, jury, and ammo. Please use in that order."
-Ed Howdershelt (Author, 1940)
If we desire respect for the law, we must first make the law respectable.

• Louis D. Brandeis
  Genes Web page http://geneslinuxbox.net:6309/

Haha, thanks
retirement is nearby also

On Tue, 25 Jul 2023 at 03:03, gene heskett gheskett@shentel.net wrote:

rather than exactly matching the cone base with the cube it can be offset
to get the 3 lines and then shave off the extra cone.
This could be one solution I can think off
[image: Screenshot 2023-07-25 at 6.16.28 AM.png]

On Tue, 25 Jul 2023 at 00:36, Torsten Paul Torsten.Paul@gmx.de wrote:

On 7/24/23 20:49, Sanjeev Prabhakar wrote:

While I as I look to my 89th in a couple months will be retired for 23
years. I am a C.E.T. with an 8th grade education, and don't have the
math skills to do what you are doing. Rather precocious electronically
before I was a teenager, my 8th grade algebra teacher thought he was a
George Carlin and far more interested in what was in the girls panties
than he was in actually teaching the algebra class he was being paid to
do. So I quit school and went to work fixing what was then a brand new
tech, TV's at 14 yo. You would not be far wrong to say I grew up with a
hot soldering iron in one hand and a scope probe in the other by the
time I hit 17.  But real math? Nope...  What little I can do was taught
to me by a TI SR51 or a log-log-duplex Pickett slipstick. What you are
doing might as well be black magic to me.

But if you have an electronics problem, I can probably help.

[...]

Cheers, Gene Heskett.

"There are four boxes to be used in defense of liberty:
soap, ballot, jury, and ammo. Please use in that order."
-Ed Howdershelt (Author, 1940)
If we desire respect for the law, we must first make the law respectable.

• Louis D. Brandeis
  Genes Web page http://geneslinuxbox.net:6309/

Hi Gene,
You don't need to know the maths involved here,  there are functions
written to do all the maths and you need to just know how to use them.
Regards

On Tue, 25 Jul, 2023, 7:42 am gene heskett, gheskett@shentel.net wrote:

On Mon, Jul 24, 2023 at 11:36:35PM +0530, Sanjeev Prabhakar wrote:

Isn't fillet about the same as:

minkovski with a sphere of the fillet radius, Then invert, again
minkovski with that sphere and then invert again.

That would at least give a defined result for my "difficult case".

Roger. 

--
** R.E.Wolff@BitWizard.nl ** https://www.BitWizard.nl/ ** +31-15-2049110 **
**    Delftechpark 11 2628 XJ  Delft, The Netherlands.  KVK: 27239233    **
f equals m times a. When your f is steady, and your m is going down
your a is going up.  -- Chris Hadfield about flying up the space shuttle.

I have not used minkowski method in openscad, maybe someone else can comment

On Tue, 25 Jul 2023 at 14:34, Rogier Wolff R.E.Wolff@bitwizard.nl wrote:

Three minkowskis of any non-trivial shape takes forever. This example takes
more than 5 minutes:

[image: round.png]

On Tue, 25 Jul 2023 at 15:56, Sanjeev Prabhakar sprabhakar2006@gmail.com
wrote:

On 25.07.23 17:18, nop head wrote:

What's the code for this example?

ciao,
Torsten.

It is the round_3D test from NopSCADLib. I have distilled it out to this:

$fa = 6;
$fs = 0.25;
inf = 1e10;      // very big
eps = 1/128;    // small fudge factor to stop CSG barfing on coincident
faces.
layer_height    = 0.25;
teardrop_angle  = 45;

function r2sides(r) = $fn ? $fn : ceil(max(min(360/ $fa, r * 2 * PI / $fs),
5));    //! Replicates the OpenSCAD logic to calculate the number of sides
from the radius
function r2sides4n(r) = floor((r2sides(r) + 3) / 4) * 4;
//! Round up the number of sides to a multiple of 4 to ensure
points land on all axes

module circle4n(r, d = undef) {                    //! Circle with
multiple of 4 vertices
R = is_undef(d) ? r : d / 2;
circle(R, $fn = r2sides4n(R));
}

module extrude_if(h, center = true)                //! Extrudes 2D object
to 3D when h is nonzero, otherwise leaves it 2D
if(h)
linear_extrude(h, center = center, convexity = 5) // 3D
children();
else
children();                                // 2D

module teardrop(h, r, center = true, truncate = true, chamfer = 0,
chamfer_both_ends = true, plus = false) { //! For making horizontal holes
that don't need support material, set truncate = false to make
traditional RepRap teardrops that don't even need bridging
module teardrop_2d(r, truncate) {
er = layer_height / 2 - eps;    // Extrusion edge radius
R = plus ? r + er : r;          // Corrected radius
offset = plus ? -er : 0;        // Offset inwards
hull()
for(side = [0 : 1])
mirror([side, 0, 0])
intersection() {
hull()
translate([offset, 0]) {
circle4n(R);

                            if (teardrop_angle > 0) {
                                x = R * tan(teardrop_angle / 2);
                                if(truncate)
                                    translate([0, R / 2])
                                        square([2 * x, R], center =

true);
else
polygon([[0, 0], [eps, 0], [0, R +
x * tan(min(teardrop_angle, 90 - eps))]]);
}
}
translate([0, -2 * R])
square([R, 4 * R]);
}
}

render(convexity = 5)
    extrude_if(h, center)
        teardrop_2d(r, truncate);

}

module offset_3D(r, chamfer_base = false) { //! Offset 3D shape by
specified radius r, positive or negative.
module ball(r)
if(chamfer_base)
rotate_extrude()
intersection() {
rotate(180)
teardrop(0, r);

                translate([0, -r])
                    square([r, 2 * r]);
            }
    else
        sphere(r);

if(r > 0)
    minkowski() {
        children();

        ball(r);
    }
else
    if(r < 0)
        render() difference() {
            cube(inf / 2, center = true);

            minkowski() {
                difference() {
                    cube(inf, center = true);

                    children();
                }
                ball(-r);
            }
        }
    else
        children();

}

module round_3D(r, ir = undef, or = undef, chamfer_base = false) { //!
Round a 3D child single radius or separate inside and outside radii
IR = is_undef(ir) ? r : ir;
OR = is_undef(or) ? r : or;
offset_3D(OR, chamfer_base)
offset_3D(-OR -IR, chamfer_base)
offset_3D(IR, chamfer_base)
children();
}

module shape()
difference() {
square(40, center = true);

    square([20, 20], center = true);
}

module rounds() {
translate([50, 0])
round_3D(or = 4, ir = 2, chamfer_base = true, $fn = 16)
linear_extrude(40, center = true)
shape();
}

rounds();

On Tue, 25 Jul 2023 at 16:37, Torsten Paul Torsten.Paul@gmx.de wrote:

here is another approach for filleting such type of intersection, it was
suggested by someone earlier, I did some modification and it is super
efficient.
it took 0.4 sec to render.
But it works only for convex intersections.
fillet here is 5 mm
[image: Screenshot 2023-07-26 at 6.33.44 AM.png]

On Tue, 25 Jul 2023 at 20:25, Sanjeev Prabhakar sprabhakar2006@gmail.com
wrote:

On Wed, 2023-07-26 at 06:35 +0530, Sanjeev Prabhakar wrote:

On my machine...

F5: Total rendering time: 0:00:00.789
F6: Total rendering time: 0:00:00.290

Ubuntu MATE,
OpenSCAD version 2023.02.15.ai13654