Polygons are broken in some cases, can be fixed by adding 0.01 to coordinates.

[georgy-moshkin]

I can't understand why this is happening. Maybe there is some error in my code, but I do not see it. Here is how problem looks like:

But after replacing pp(2,16) = 0; with pp(2,16) = 0.01; it looks normal.
`

for s = 0:patch.N-1
pp(1,1) = 0; pp(2,1) = 0;
pp(1,2) = 0; pp(2,2) = len180/2-feed.width*3/2;

pp(1,3) = addLen-feed.width;    pp(2,3) = len180/2-feed.width*3/2;
pp(1,4) = addLen;               pp(2,4) = len180/2-feed.width*3/2+feed.width;

pp(1,5) = addLen;               pp(2,5) = len180/2-feed.width*3/2+feed.width*3-feed.width;
pp(1,6) = addLen-feed.width;    pp(2,6) = len180/2-feed.width*3/2+feed.width*3;

pp(1,7) = 0;                    pp(2,7) = len180/2-feed.width*3/2+feed.width*3;
pp(1,8) = 0;                    pp(2,8) = len180;

pp(1,9)  = -feed.width;         pp(2,9)  = len180;

pp(1,10) = -feed.width;             pp(2,10) = len180/2+feed.width/2+feed.width;
pp(1,11) = -feed.width+feed.width;  pp(2,11) = len180/2+feed.width/2;

pp(1,12) = -feed.width+addLen;      pp(2,12) = len180/2+feed.width/2;
pp(1,13) = -feed.width+addLen;      pp(2,13) = len180/2-feed.width/2;

pp(1,14) = -feed.width+feed.width;  pp(2,14) = len180/2-feed.width/2;
pp(1,15) = -feed.width;             pp(2,15) = len180/2-feed.width/2-feed.width;

pp(1,16) = -feed.width;             pp(2,16) = 0;

for k=1:16
  pp(1,k)=pp(1,k)+feed.width/2;
  pp(2,k)=pp(2,k)+offsY-len180-patch.length/2;
endfor

CSX = AddPolygon( CSX, 'network', 1, 'z', substrate.thickness, pp);

endfor

`

[BryceBuschman]

Hi @georgy-moshkin,

Have you tried to round your coordinates to 5 or 6 decimal places? I had a problem similar to this. In my cased it seemed like decimal truncation when building the polygon caused some points to fall on-top of each other/get too close to each other. Rounding to 6 decimal places fixed the problem. Good luck!

[Juleinn]

Had some trouble with this too, when using floating point for defining geometry floating-point errors would accumulate to the point where it would create close points causing cell size to be very small and thus computation time very large.

Something like

co = [[round(i[0], 5), 
  round(i[1], 5), 
  round(i[2], 5)] for i in local_verts]

works for me

Using fixed-point for geometry would be a good improvement I think (like in KiCAD where everything is nanometers as integers), but that's a significant amount of work to do

[georgy-moshkin]

Thanks for the suggestions. In my case, I fixed it by adding "pp=0;" in the end of the loop:

for n = 0:block.N-1
% some code fills pp(i,k) = ...
CSX = AddPolygon( CSX, 'top_metal', 1, 'z', substrate.thickness, pp);
pp=0; % erase pp() contents - THIS FIXES THE ISSUE
endfor;

I think this destroys 2-dimensional pp(i,k) array, so it can be recreated from the scratch on each iteration.
I haven't look into it afterwards and still don't understand what's going on here. In other words, I don't understand why result is being different, because I have fixed number of elements, a 2 x 16 array which is simply being overwritten with new values on each iteration.

[thliebig]

I think this is only a glitch in the visualization by AppCSXCAD? I think this is a bug in QCSXCAD and not in CSXCAD or openEMS. But I would need to confirm that...