[fixed] [sketcher] crash on decreasing spline knot multiplicity

[uwestoehr]

So why are the weights not uniform?

No idea how you managed to break it.

Good question. After several tries I found now out how:

I start from this file:

Spline-explanation2.FCStd

tesfile-weight-bugs

(22.52 KiB) Downloaded 13 times

depicted here:

initial

FreeCAD_pLcUWbzvej.png (28.19 KiB) Viewed 411 times

1. delete Constraint 3 (equality between first and third control point)
The shape of the spline is of course still the same, but the displayed distance between the knots changed:

after deleting constraint 3

FreeCAD_uritgwPluY.png (30.26 KiB) Viewed 411 times

2. increase now the radius of the third control point.
Now the shape changed correctly:

after weight increase

FreeCAD_PffTP2ZX2k.png (30.03 KiB) Viewed 411 times

3. select the circle of the first and third point and apply an equality constraint
Now the shape has not uniform weights, despite all radii are equal:

result

FreeCAD_bfuNULc9e6.png (28.37 KiB) Viewed 411 times

I think there are 2 bugs:
A: the distance between knots is incorrectly displayed after step 1
B. the weight is not reset whiley applying the equality constraint for the radiuses in step 3

[uwestoehr]

links that I like :
A Primer on Bézier Curves

This is amazing. Many thanks!

[wmayer]

It is line 83 of \src\Gui\ViewProvider.cpp:

I have tried it with a debug build on Windows and for me it (quite reliably) crashes inside the destructor of SbImage. The callstack is:

Code: Select all

 	ucrtbased.dll!00007ffd82d11f9b()	Unknown
 	ucrtbased.dll!00007ffd82d151c5()	Unknown
>	Coin4d.dll!operator delete(void * block) Line 38	C++
 	Coin4d.dll!operator delete(void * block, unsigned __int64 __formal) Line 32	C++
 	Coin4d.dll!SbImageP::`scalar deleting destructor'(unsigned int)	C++
 	Coin4d.dll!SbImage::~SbImage() Line 244	C++
 	Coin4d.dll!SbImage::`scalar deleting destructor'(unsigned int)	C++
 	Coin4d.dll!SoImage::~SoImage() Line 296	C++
 	SketcherGui_d.pyd!SoImage::`scalar deleting destructor'(unsigned int)	C++
 	Coin4d.dll!SoBase::destroy() Line 331	C++
 	Coin4d.dll!SoBase::unref() Line 555	C++
 	Coin4d.dll!SoBaseList::truncate(const int length) Line 172	C++
 	Coin4d.dll!SoChildList::truncate(const int length) Line 199	C++
 	Coin4d.dll!SoChildList::~SoChildList() Line 94	C++
 	Coin4d.dll!SoChildList::`scalar deleting destructor'(unsigned int)	C++
 	Coin4d.dll!SoGroup::~SoGroup() Line 281	C++
 	Coin4d.dll!SoSeparator::~SoSeparator() Line 469	C++
 	SketcherGui_d.pyd!SoSeparator::`scalar deleting destructor'(unsigned int)	C++
 	Coin4d.dll!SoBase::destroy() Line 331	C++
 	Coin4d.dll!SoBase::unref() Line 555	C++
 	Coin4d.dll!SoBaseList::remove(const int index) Line 136	C++
 	Coin4d.dll!SoChildList::remove(const int index) Line 171	C++
 	Coin4d.dll!SoGroup::removeChild(int childindex) Line 506	C++
 	FreeCADGui_d.dll!Gui::coinRemoveAllChildren(SoGroup * group) Line 83	C++
 	SketcherGui_d.pyd!SketcherGui::ViewProviderSketch::rebuildConstraintsVisual() Line 5330	C++
 	SketcherGui_d.pyd!SketcherGui::ViewProviderSketch::draw(bool temp, bool rebuildinformationlayer) Line 4324	C++
...

[wmayer]

In some cases it crashes inside GeomBSplineCurve::~GeomBSplineCurve when the Geom_BSplineCurve is cleaned up. As a test I modified GeomBSplineCurve::removeKnot() to do nothing and always returns true. This way the crash doesn't occur any more so that the root problem eventually could be caused by OCCT.

[uwestoehr]

In some cases it crashes inside GeomBSplineCurve::~GeomBSplineCurve when the Geom_BSplineCurve is cleaned up. As a test I modified GeomBSplineCurve::removeKnot() to do nothing and always returns true. This way the crash doesn't occur any more so that the root problem eventually could be caused by OCCT.

Many thanks for having a look. So we ca only file a bug report to OCC.

[uwestoehr]

I think there are 2 bugs:
A: the distance between knots is incorrectly displayed after step 1
B. the weight is not reset while applying the equality constraint for the radiuses in step 3

I thought I could fix this because both is the consequence of the same bug I think. I tried to debug where in the code the spline weights are calculated out of the control point circle radii but cannot find this code.

Can anybody please give me a hint where I have to search or in what case the weight calculation is done?

What I debugged so far is that after deleting the constraint, the weight of the affected point is set to 1 while all other weights are 4. After this first set to 1, the weight is subsequently set again, this time to 3.96. This is the final result and this is the cause why the knot distance changed.

[wmayer]

Many thanks for having a look. So we ca only file a bug report to OCC.

No. I have looked further into the issue and I was able to narrow the actual problem. Inside the function SketchObject::modifyBSplineKnotMultiplicity() at the very end a list of superfluous geometries is deleted and for some reason this causes the later crash.

If I comment out the line delGeometry no crash occurs any more. This explains why it doesn't crash if I modify GeomBSplineCurve::removeKnot to do nothing or my script above that doesn't enter edit mode of the sketch.
However, it doesn't explain the random crashes I have observed inside GeomBSplineCurve::~GeomBSplineCurve but at this point the program is in an undefined state anyway so that everything could happen.

Now it needs some further investigation to find out why the call of delGeometry is responsible for the crash. As said above on Windows it mostly crashes inside the destructor of SbImage and thus we have to find out how deleting the geometry is related to the undefined data in the SbImage class.

[wmayer]

I have implemented a function to delete several geometries in one step. This way it now crashes much less often and when it does then due to graphic driver issues or that the transaction objects got corrupted after a few undo/redo steps.

If the superfluous geometry isn't deleted at all I never got it to crash but it will be up to the user to remove this stuff manually.

[uwestoehr]

I have implemented a function to delete several geometries in one step. This way it now crashes much less often and when it does then due to graphic driver issues or that the transaction objects got corrupted after a few undo/redo steps.

Thank you. I get now indeed only random crashes after undoing a decrease multiplicity. With random I mean, that I cannot find a recipe to trigger the crash in every case.

OS: Windows 10 (10.0)
Word size of OS: 64-bit
Word size of FreeCAD: 64-bit
Version: 0.19.22932 (Git)
Build type: Release
Branch: master
Hash: 48365ecdbbde880ca4d1674cc0d8bae81d274d14
Python version: 3.6.8
Qt version: 5.12.6
Coin version: 4.0.0a
OCC version: 7.4.0.beta
Locale: German/Germany (de_DE)

[uwestoehr]

Can anybody please give me a hint where I have to search or in what case the weight calculation is done?

I go crazy, I debugged now several times, watched the changes in the variables but can still not find where the spline weights are calculated out of the given radiuses.
Thus any pointer to where to look in the code would help me a lot.