Cable model constraint matrix

[bakpaul]

Hello,

I was looking at the CableModel implementation, especially the buildConstraintMatrix implementation. I was surprised to see hthat the normal injected in the jacobian are not normalized. Fir instance, in this snippet, the two directions used to compute the last one are normalized but not the final one.

Deriv directionBeyond = previousPosition - currentPosition;
directionBeyond.normalize();

Deriv directionAhead  = currentPosition - nextPosition;
directionAhead.normalize();

Deriv slidingDirection = directionBeyond - directionAhead;
if (d_hasPullPoint.getValue())
{
    if (id_method == 0)
    {
        rowIterator.setCol(currentIndex, slidingDirection);
    }
    else 
    {
        for(unsigned int j=0; j<m_areaIndices[i].size(); j++)
            rowIterator.setCol(m_areaIndices[i][j], slidingDirection*m_ratios[i][j]);
    }
}           

Isn't this prone to numerical instabilities ? I understand that this norm allows to modulate the forces among the points, but how is that norm sound ? If the cable is straight this norm is equal to two, is this desirable ?

[hugtalbot]

Let's poke here @TanguyNav @EulalieCoevoet

[EulalieCoevoet]

Indeed!

[EulalieCoevoet]

Thanks Paul. Here's a PR #263.
Edit: we should not normalize this direction. If the cable is straight the "directionBeyond" and the "directionAhead" are opposite.