BoundBox too small

[Thomas_A]

I am surprised by the behavior of the following simple script

Code: Select all

obj = App.ActiveDocument.addObject("Part::Torus","Torus")
App.ActiveDocument.recompute()
obj.Shape.BoundBox

where the last command has the following output:

Code: Select all

BoundBox (-11.9818, -11.9954, -1.98542, 12, 11.9954, 1.98542)

Given that the default radii of the torus are 10 and 2, the bounding box must be at least of the size

Code: Select all

BoundBox (-12., -12., -2., 12., 12., 2.)

I would not care if it is bigger (this would just make it more inefficient) but it being too small is a huge problem for my code.

Is this intended behavior?

(I compiled FreeCAD from the current state of its git repository - so my version should be up-to-date.)

[Hannu]

FWIW:

Code: Select all

>>> ### Begin command Std_New
>>> App.newDocument("Unnamed")
>>> # App.setActiveDocument("Unnamed")
>>> # App.ActiveDocument=App.getDocument("Unnamed")
>>> # Gui.ActiveDocument=Gui.getDocument("Unnamed")
>>> Gui.activeDocument().activeView().viewDefaultOrientation()
>>> ### End command Std_New
>>> Gui.runCommand('Std_OrthographicCamera',1)
>>> obj = App.ActiveDocument.addObject("Part::Torus","Torus")
>>> App.ActiveDocument.recompute()
1
>>> obj.Shape.BoundBox
BoundBox (-12, -12, -1.98542, 12, 12, 1.98542)
>>> 

Now, how exact does things need to be?

Screenshot from 2020-02-22 09-14-02.png (21.89 KiB) Viewed 513 times

That is a 7.3 ‰ (per mille) difference; should we expect a "perfect" measurement?

Code: Select all

>>> round(1.98542,3)
1.985
>>> round(1.98542,2)
1.99
>>> round(1.98542,1)
2.0
>>> 

OS: Ubuntu 16.04.6 LTS (Unity/ubuntu)
Word size of OS: 64-bit
Word size of FreeCAD: 64-bit
Version: 0.19. <<--?? The latest daily, updated just an hour ago.
Build type: Release
Python version: 3.5.2
Qt version: 4.8.7
Coin version: 4.0.0a
OCC version: 7.3.0
Locale: English/UnitedStates (en_US)

[openBrain]

This is due to Shape tessellation settings.
FreeCAD shapes are BREP and so BoundBox may vary depending on tessellation settings (this is especially true on curved objects).
Look following snippet :

Code: Select all

>>> obj = App.ActiveDocument.addObject("Part::Torus","Torus")
>>> App.ActiveDocument.recompute()
1
>>> obj.Shape.BoundBox
BoundBox (-12, -12, -1.98542, 12, 12, 1.98542)
>>> obj.ViewObject.Deviation = 0.02
>>> obj.ViewObject.AngularDeflection = 1
>>> App.ActiveDocument.recompute()
0
>>> obj.Shape.BoundBox
BoundBox (-11.9983, -11.9996, -2, 12, 11.9996, 2)

I think in this case actually only Deviation plays a role but I'm not really a specialist.

[Thomas_A]

Thanks for the elaborations. I am new to FreeCAD and I now understand where the problem comes from.

That is a 7.3 ‰ (per mille) difference; should we expect a "perfect" measurement?

Perhaps my background in computer graphics is not appropriate here but in my conception, a bounding box contains the object completely. Of course, a 'perfect' bounding box would be one that coincides with the extremal points of the shape. I would not expect something like this.
BUT:
It's one thing for a bounding box to be too big and 'inefficient' but in my opinion, it is horribly wrong for the bounding box to be too small.

In my code, I generate points on the surface of the shape and then find out that they do not lie inside the bounding box?!
At least up to numeric precision (i.e., in the range of 1e-6), this should be the case. 7.3e-3 is 1000x times above this.

FreeCAD shapes are BREP and so BoundBox may vary depending on tessellation settings (this is especially true on curved objects).

I see. Is there a function that provides a conservative bounding box (i.e., a bounding box that might be too big but at least contains the object for sure)?

[openBrain]

Thanks for the elaborations. I am new to FreeCAD and I now understand where the problem comes from.

I see. Is there a function that provides a conservative bounding box (i.e., a bounding box that might be too big but at least contains the object for sure)?

Not a native function AFAIK. You can expand the returned bounding box with the deviation of the object.

[kisolre]

Dont have where to test right now but if a PD revolved torus is created what will be the bounding box?