id	summary	reporter	owner	description	type	status	milestone	component	version	severity	resolution	keywords	cc
6063	resize does not offset rectangles (etc.) correctly or crashes	dbfaken@…	Andrii Sydorchuk	"When using the 'corner_fill_arc' parameter to polygon_set_concept<T>::resize(), a minkowski sum using a polygonalized circle is used to get the offset shape.

A symptom of the problem is demonstrated by the following code:

{{{
 {
  polygon_set_data<int> ps1, ps2, ps3;
  ps1.insert(rectangle_data<int>(0, 0, 50, 50));
  ps2.insert(rectangle_data<int>(0, 0, 50, 50));
  ps3.insert(rectangle_data<int>(0, 0, 50, 50));
  std::cout << ""rect before resize: "" << ps1 << std::endl;
  ps1.resize(6, true, 0);
  ps2.resize(6, true, 5);
  ps3.resize(6, true, 6);
  rectangle_data<int> ps1_extents, ps2_extents, ps3_extents;
  extents(ps1_extents, ps1);
  extents(ps2_extents, ps2);
  extents(ps3_extents, ps3);
  std::cout << ""extents of resized rect with '0' (4) segments in circle: "" << ps1_extents << std::endl;
  std::cout << ""extents of resized rect with 5 segments in circle: "" << ps2_extents << std::endl;
  std::cout << ""extents of resized rect with 6 segments in circle: "" << ps3_extents << std::endl;
  }
}}}

If I put this at the end of the gtl_boost_unit_test.cpp, I get the following output:

{{{
rect before resize: Polygon Set Data { <0 0, 50 0>:1 <50 0, 50 50>:-1 <0 50, 50 50>:-1 <0 0, 0 50>:1 }
extents of resized rect with '0' (4) segments in circle: -5 54 -5 54
extents of resized rect with 5 segments in circle: -6 54 -6 55
extents of resized rect with 6 segments in circle: -6 55 -6 56
}}}


This shows that the extents of the offset shape vary as the number of segments does - which should not be the case!

The extents should all be: -6 56 -6 56

----

The problem (to my eyes) is that the polygonalized circle does not generally have the right shape to get proper offsets in axis-oriented directions, so the offset is basically multiplied by some factor of sqrt(2).
Attached is an image of the circle generated with num_circle_segments=16.  The radius of the circle - i.e. distance from center to each vertex - is 2.  The grid spacing is 1.  As can be seen, the top/bottom and left/right sides of the circle do not lie on the grid.
If the circle started with a vertex at(0,2) this would not occur and one would presumably get proper offsets in axis-aligned directions.
Alternately one could offset all the vertices slightly farther, but this would offset some vertices farther than desired.

I would guess the fix is to change the behaviour of make_arc(), which generates these points, but I'm not sure exactly what the original author's intent was - so maybe just using a new function would be best (after all, the complexity of make_arc is overkill when you're generating a simple full circle)


----

Another problem is that the behaviour of the 'corner_fill_arc' parameter is not really well described.  Using this gives much more than corner-filling behaviour - it gives different offsets in different directions.  I would suggest explaining it as it is - a minkowski sum with a polygonalized circle, with all that implies.

E.g. for a 45-degree segment the offset will differ from a 90-degree segment unless you pick the right value of num_circle_segments.

Maybe a more comprehensible solution would be to implement corner rounding by the intersection of the normal resize()d shape (as given by corner_fill_arc=false) with the resize()d shape given by corner_fill_arc=true with a larger 'resizing' (picked so that the minimum resizing equals the original resizing).

This would only affect corners, instead of the whole shape.
"	Bugs	assigned	To Be Determined	polygon	Boost 1.47.0	Problem