Fehler im Konturfräsen V17 ?

[mlampert]

Yeah I noticed that - there's definitely something wrong.

[wandererfan]

Look at the image I added on the ticket. Besides the extra internal line (which shouldn't be there) theres also a small spurious segment in the top corner. This wreaked havoc with that part contour.

At first glance there's nothing in findShapeOutline() that uses Vector2d or BoundBox2d. The outline in Taschen.fcstd is still wrong (should be rectangle), but I'm not sure the fix for issue #2858 broke it.

Can you post the file that generated the snapshots, please?

wf

[sliptonic]

attached.

I select the Pocket002 object and run this macro

Code: Select all

import TechDraw
from FreeCAD import Vector

sel = Gui.Selection.getSelectionEx()[0]
obj = sel.Object.Shape

contourwire = TechDraw.findShapeOutline(obj,1, Vector(0,0,1))
myObject = FreeCAD.ActiveDocument.addObject("Part::Feature","myObject")
myObject.Shape = contourwire

[wandererfan]

Nasty, nasty bug. Involves comparison of bounding boxes and this OCC issue: http://opencascade.blogspot.ca/2009/04/ ... g-box.html

Doesn't seem to be strictly deterministic. I got this from ProfileTortureTest.fcstd:

ProfileTortureTest Outline

OutlineWire0.png (2.62 KiB) Viewed 1750 times

But this from Taschen.fcstd:

Taschen Outline

TaschenOutlineWire0.png (2.53 KiB) Viewed 1750 times

Anyway, the fix is to not use the bounding boxes to determine relative sizes, but to use the actual enclosed area of the wires.
MIght be a little slower, but should be accurate.

PR will be along shortly.

[wmayer]

The ticket issue #0002858 was opened by wandererfan but I am not sure if there was real problem with the old behaviour. However, it looks like there is now a problem with the new behaviour.

So, for me it's OK to revert the changes and introduce a method isEqual(Vector2d, double) where a tolerance can be passed. IIRC for example the gp_Pnt class of OCC also does an exact check of the coordinates.

[sliptonic]

I like that.
The new behavior is causing a problem in several of the Path operations that don't use the TechDraw findShapeOutline function. I've been trying to isolate exactly where but it's eluding me so far.

[wandererfan]

The ticket issue #0002858 was opened by wandererfan but I am not sure if there was real problem with the old behaviour. However, it looks like there is now a problem with the new behaviour.

So, for me it's OK to revert the changes and introduce a method isEqual(Vector2d, double) where a tolerance can be passed. IIRC for example the gp_Pnt class of OCC also does an exact check of the coordinates.

This is the change that breaks findShapeOutline:
in Base/Vector3D.h
- static inline float_type epsilon() { return FLT_EPSILON; }
+ static inline float_type epsilon() { return DBL_EPSILON; }

FLT_EPSILON 1E-5
static const Standard_Real Precision_Confusion = 1.e-7;
DBL_EPSILON 1E-9
Our test for zero is finer than our calculations.

As far as I can tell Path/CAM doesn't use the FreeCAD Vector2d, so this is probably causing the trouble there too.

[wmayer]

This is the change that breaks findShapeOutline:
in Base/Vector3D.h
- static inline float_type epsilon() { return FLT_EPSILON; }
+ static inline float_type epsilon() { return DBL_EPSILON; }

Using FLT_EPSILON for a Vector3d which internally uses double (not float) is too rough. So, I will revert the above changes and introduce an isEqual method where you can pass an arbitrary tolerance. This way we will be consistent to OCC's gp_Pnt class.

[wandererfan]

Comparison with tolerance is a good addition to Vector2d and Vector3d.

[sliptonic]

This is the change that breaks findShapeOutline:
in Base/Vector3D.h
- static inline float_type epsilon() { return FLT_EPSILON; }
+ static inline float_type epsilon() { return DBL_EPSILON; }

Using FLT_EPSILON for a Vector3d which internally uses double (not float) is too rough. So, I will revert the above changes and introduce an isEqual method where you can pass an arbitrary tolerance. This way we will be consistent to OCC's gp_Pnt class.

Has this been reverted yet? Should I create an issue to track it?