Sketch. Max possible fillet fails when multiple faces are present.

[ajegorovs]

Hi.

I want to put known max fillet on a rectangle edge. It works correctly on a single rectangle. But fails if there are multiple 'rectangle instances'.

Looks like it does not matter how they were created. Ive tried cutting one bigger rectangle in two via subtract mode, pushing multiple points, drawing segments and assembling manually.

One curious remark is that even when single point is pushed, if its not (0,0) max fillet on a rectangle breaks.

import cadquery as cq
#from ocp_vscode import (reset_show, show_object, Camera)
#reset_show()

W = 1               # big rectangle width. for cutting
L = 2               # rectangle height/length
gap         = W/5   # gap between rectangle

W_1         = (W - gap)/2   # split rectangle width
r           = W_1/2         # half-width = fillet radius


# push one point
sk = cq.Sketch().push([(0,0)]).rect(W_1, L).reset() # fillet is done
sk = cq.Sketch().push([(1,0)]).rect(W_1, L).reset() # fails on fillet
sk.vertices('>Y').fillet(r).reset()   # fail
#show_object(sk , name= '1pt' , grid = True, reset_camera = Camera.KEEP)

# 2 faces via subtract
sk    = cq.Sketch().rect(W, L).reset()
sk.rect(gap, 1.5*L, mode='s').reset()   # cut vertical gap via 'subtract'
sk.vertices('>Y').fillet(r).reset()     # fail
#show_object(sk , name= 'bool' , grid = True, reset_camera = Camera.KEEP)

# cut out two slots
sk= cq.Sketch().rect(2*W, 1.05*L)
sk.push([(-0.5,0), (0.5,0)]).rect(W_1,L,mode='s', tag = 'aa')   # cut slots, remember verts
sk.vertices(tag='aa').fillet(r) # fail
#show_object(sk , name= 'slots' , grid = True, reset_camera = Camera.KEEP)

What are the options of fixing this behavior? As a short-time solution, maybe iterating/mapping fillet onto separate faces? doing r *= 0.9999 and 'cleaning' close-by points?
Yes, it is possible to reduce radius slightly and then everything goes through. But i would prefer a clean, high-level solution.

Last two entries of a trace:

File c:\Repos\venvs\cadquery\venv\Lib\site-packages\cadquery\sketch.py:616, in <genexpr>(.0)
    609 """
    610 Add a fillet based on current selection.
    611 """
    613 f2v = self._matchFacesToVertices()
    615 self._faces = Compound.makeCompound(
--> [616](file:///C:/Repos/venvs/cadquery/venv/Lib/site-packages/cadquery/sketch.py:616)     k.fillet2D(d, v) if v else k for k, v in f2v.items()
    617 )
    619 return self

File c:\Repos\venvs\cadquery\venv\Lib\site-packages\cadquery\occ_impl\shapes.py:3472, in Face.fillet2D(self, radius, vertices)
   3468     fillet_builder.AddFillet(v.wrapped, radius)
   3470 fillet_builder.Build()
-> [3472](file:///C:/Repos/venvs/cadquery/venv/Lib/site-packages/cadquery/occ_impl/shapes.py:3472) return self.__class__(fillet_builder.Shape())

StdFail_NotDone: BRep_API: command not done

[lorenzncode]

What are the options of fixing this behavior?

1. Manually specify a reduced fillet radius by epsilon value

As you mentioned, reducing the radius works.

from cadquery import Sketch

W = 0.4
L = 2
r = W/2
gap = W/2

eps = 1e-4

sk = Sketch().rarray(W + gap, 10, 3, 1).rect(W, L).reset()
sk.vertices('>Y').fillet(r - eps)

2. Create your own fillet function. Call with invoke.

The wrapper might make multiple fillet attempts or perhaps one attempt with reduced radius value.

Here is an example that makes multiple attempts starting from the given value to 80%.

from cadquery import Sketch
import numpy as np

W = 0.4
L = 2
r = W/2
gap = W/2


def fillet_try(r):
    """Quick example for discussion only.
    Replace with say log spacing to gradually increase step size after each try, handle errors.
    """

    def callback(s):
        for v in np.linspace(r, r*0.8, 8):
            try:
                s = s.fillet(v)
                print(f"fillet value {v:0.2f} OK")
                break
            except:
                print(f"fillet value {v:0.2f} failed!")
        s.reset()
        return s

    return callback


sk = Sketch().rarray(W + gap, 10, 3, 1).rect(W, L).reset()

# start with fillet value too large to demonstrate fail attempts
sk.vertices('>Y').invoke(fillet_try(r + 0.2*r)) 

3. Perhaps fillet2D can be changed to detect failure and retry with reduced radius automatically

See experimental branch here:
https://github.com/lorenzncode/cadquery/blob/a7eda61aeed845eb6af9c1d94e336a0b26800bc7/cadquery/occ_impl/shapes.py#L3478

[ajegorovs]

@lorenzncode, i really like the approach with .invoke(), ive never used it before. It certainly opens up new possibilities in my workflow. IMO reducing radius is a band-aid solution. I did some thinking and realized that

1. not having a continuous arc is not as important for my task- i was afraid, that when offsetting face outward, segment splitting two fillet arcs would be amplified. This is not the case, it stays the same size.

But there is a 'counter argument'. Having a single continuous arc might (is?) better for CNC applications due absent curvature discontinuities (EDIT: wrong fillets come in tangent. Just more segments.) IDK though, just speculating...

2. Maybe the fact that you can apply full 180 fillet to a rectangle is a bug-feature, and we should not expect this from other shapes. I assume it happens on kernel level as the error is "StdFail_NotDone: BRep_API: command not done" ?
   Lets assume its a feature. Whats disturbing, is the issue with

Sketch().push([(1,0)]).rect(W, L).reset().vertices().fillet(W/2)    # 1. fail
Sketch().push([(0,0)]).rect(W, L).reset().vertices().fillet(W/2)    # 2. pass
Sketch().rarray(1, 1, 1, 1).rect(W, L).reset().vertices().fillet(W/2)    # 3. pass <<<
Sketch().segment((0,0),(W,0)).segment((W, L)).segment((0,L)).close().assemble().vertices().fillet(W/2) 4. pass

Which might signal of problems with .push(). .rarray() 'pushes' Locations, so im not sure why #1 fails.