Python DPolygon.sized() produces little triangles

Comments

• Vincent Lin

  February 2023

  Hello yanoff,

  I got the same result as you mentioned, but Region is ok. Not sure it works to you?

  Vincent
• dick_freebird

  February 2023

  I'm wondering whether this is "algorithmic" (and a bug) due to the
  high vertex count and long shrink dimension, causing "spurious
  projection" as the vertices might cross over each other when projected
  orthogonal to their flat? Or the single vertex becoming two, as the
  two flats' projections will differ?

  What happens (say) with a shrink-distance that is much less than the
  vertex-vertex distance? Just by way of debug. What happens if it's
  N=36 vertices instead of N=64 (or whatever)?
• Matthias

  February 2023

  Hello,

  no that's not a bug. Polygon#sized is not intended to do a full job. Instead you need to "merge" the polygon to remove the triangles (see https://www.klayout.de/doc-qt5/code/class_DPolygon.html#m_size).

  Background is: the size polygon in the general case will create 0 to n polygons - a negative sizing may render smaller, disconnected or even no polygon at all. So there is no polygon-to-polygon size function. The "Region" class is a collection of polygons, hence "sized" works here.

  There is a problem with "DPolygon": you first need to convert it to a integer-unit polygon before you can use it in the edge processor (as suggested in the documentation) or inside Region. If you have a minimum grid (database unit), you can use this transformation to create a Polygon:

  dbu = 0.001   # 1 nm resolution
  dpoly = ...   # your DPolygon object
  
  # convert the DPolygon to Polygon
  poly = RBA::CplxTrans::new(dbu).inverted * dpoly
  
  # do something with the polygon, e.g.
  region = RBA::Region::new
  region.insert(poly)
  # perform the size, assuming it will always produce a single polygon (CAUTION: this may not be true, see above)
  poly = region.sized(size_value)[0]
  
  # convert it back to DPolygon
  dpoly = RBA::CplxTrans::new(dbu) * poly
  

  Matthias
• yanoff

  February 2023

  Thanks for the advice!
  The problem was driven by trying to size() a DPolygon. The result was a shape with dimensions in nanometers instead of micrometers. I think another part of the problem was points snapping to the 1 nm grid points.

  I did find the method DPolygon.to_itype(dbu) to convert from DPolygon to Polygon. That seems more natural to me.
• Matthias

  February 2023 edited February 2023

  @yanoff Be careful: DPolygon.to_itype does not include the scaling of micrometers to database units. It will simply round to integers.

  It's only a convention, but DPolygon often represents a polygon in floating-point micrometer units while Polygon uses integer units which give you micrometers when multiplied by the database unit. Transformation between these spaces is done best using the CplxTrans objects as shown above.

  Just for your information.

  Matthias
• yanoff

  April 2023

  Both to_itype() and to_dtype() take an optional dbu argument. If this is provided, they do perform the scaling, don't they?
• Matthias

  April 2023

  Yes, that is supposed to provide the database unit you intend to have in integer space.

  So basically

  DPolygon = Polygon * dbu
  
  Polygon = DPolygon * (1/dbu)
  

  Matthias