Add_polygon seems very slow

dacastror

Hi guys, I want to ask if there is something faster than add_polygon, I see much difference when I commented the lines 15 and 16. with this pair of lines becomes 130 times slower!
seems that the command "add_poligon" does something very complicated or is very inefficient or I'm using it in the wrong way
in the following code I make a rectangle subdivided;

t = Time.now
lados_x = 100
lados_y = 100
mesh = Geom;;PolygonMesh.new((lados_x+1)*(lados_y+1),lados_x*lados_y)
l1 = []
for i in 0..lados_x
   l1[i] = Geom;;Point3d.new(i,0,0)
end
l2 = []
for j in 1..lados_y
   for k in 0..lados_x
      l2[k] = Geom;;Point3d.new(k,j,0)
   end
   for i in 0..lados_x-1
      #mesh.add_polygon l1[i], l1[i+1], l2[i+1]
      #mesh.add_polygon l1[i], l2[i+1], l2[i]
   end
   l1 = l2.clone
end

=begin
ent=Sketchup.active_model.active_entities
grupo = ent.add_group
entities = grupo.entities
entities.fill_from_mesh mesh, true, 0
=end

Time.now-t

Does anyone have any suggestions?

(google translator)

danielbowring

The problem is you are using Point3d arguments - this means each call to add_polygon needs to call add_point three times. You can add all the points beforehand and then use their index to add the polygons. This should add a significant speed boost.

Here's the code:

t = Time.now
lados_x = 100
lados_y = 100
mesh = Geom;;PolygonMesh.new((lados_x+1)*(lados_y+1),lados_x*lados_y)
l1 = []
for i in 0..lados_x
   l1[i] = mesh.add_point(Geom;;Point3d.new(i,0,0))
end
l2 = []
for j in 1..lados_y
   for k in 0..lados_x
      l2[k] = mesh.add_point(Geom;;Point3d.new(k,j,0))
   end
   for i in 0..lados_x-1
      mesh.add_polygon l1[i], l1[i+1], l2[i+1]
      mesh.add_polygon l1[i], l2[i+1], l2[i]
   end
   l1 = l2.clone
end


ent=Sketchup.active_model.active_entities
grupo = ent.add_group
entities = grupo.entities
entities.fill_from_mesh mesh, true, 0


puts Time.now-t

dacastror

!woooow this reduces the time to less than one-third!

thank you very much Daniel!

thomthom

Also, if you ever need PolygonMesh.point_index http://www.sketchup.com/intl/en/developer/docs/ourdoc/polygonmesh#point_index - don't use it. It's slow. Instead, store the point and index you get from PolygonMesh.add_point into a hash and it'll be much much faster.

And, if you are adding the geometry to an Entities collection that has no other entities in it, use Entities.fill_from_mesh as it's faster.

dacastror

Thanks for the tips Thom, this will be very useful at some time

I have a question, how can get the vertices of this surface?

thomthom

Ah, very good question. I've been struggling with that myself. Trying to compare and map based on position - but floating precision might cause problems.

thomthom

Ok - here is a bare bone test:
https://gist.github.com/thomthom/5160068

Dan Rathbun

@thomthom said:

Ah, very good question. I've been struggling with that myself. Trying to compare and map based on position - but floating precision might cause problems.

In your code you have:
%(#008000)[# Because Geom::Point3d objects cannot be compared ...]
.. BUT the API overrides the ==() method for the Geom::Point3d class, that WILL test the objects co-ordinates within SketchUp's internal tolerance.

Also...
Sketchup::Vertex.position returns a Geom::Point3d
Geom::Point3d.x ( .y and .z) all return Length class objects, ...
and ALL of the comparison methods for the Length class have been overridden to use SketchUp's internal tolerance.

BTW, the Sketchup::Vertex class could use a wrapper:

class Sketchup;;Vertex
  def ==(other)
    return false unless other.is_a?(self.class)
    return false unless other.position.is_a?(Geom;;Point3d)
    self.position == other.position
  end
end

Dan Rathbun

@thomthom said:

` %(#008000)[# What what if you generate a mesh, then add it to SketchUp - is there a

chance that SketchUp slightly adjusts the points within it's tolerance range

in order to merge vertices etc?]`

Should the general rule of thumb be to use arrays of Length objects to build Geom::Point3d objects, rather than arrays of Float objects ??

thomthom

@dan rathbun said:

@thomthom said:

Ah, very good question. I've been struggling with that myself. Trying to compare and map based on position - but floating precision might cause problems.

In your code you have:
%(#008000)[# Because Geom::Point3d objects cannot be compared ...]
.. BUT the API overrides the ==() method for the Geom::Point3d class, that WILL test the objects co-ordinates within SketchUp's internal tolerance.

@dan rathbun said:

@thomthom said:

` %(#008000)[# What what if you generate a mesh, then add it to SketchUp - is there a

chance that SketchUp slightly adjusts the points within it's tolerance range

in order to merge vertices etc?]`

Should the general rule of thumb be to use arrays of Length objects to build Geom::Point3d objects, rather than arrays of Float objects ??

Ah yea - I meant to clean up that comment. It is misleading. What I actually had in mind was what happens when you use Geom::Point3d as a key in a hash - if you create a Point3d object with the exact same values the hash is different.

dacastror

thank you very much for the replies dan and Thomthom, I'll take a look at the code Thomthom experimental, to see if I can use some of this code

(google translator)