I may be missing some details, but at first glance, this solution seems strange because:

• hidden elements are not affected by the transformation, so should they really be counted and allocated as TransData?
• With this patch, Nurb-type curves still do not count hidden elements. Usually the code for CU_NURBS reflects that for CU_BEZIER, so it might be good to describe why they got "different".

(My impression is that td++; and that tail++; should be removed).

@Germano Cavalcante (mano-wii) Yes, in fact my first solution was to just remove the

else if (is_prop_edit && head != tail) {
  tail->flag |= TD_NOTCONNECTED;
  td++;
  tail++;
}

part. I thought this would work just fine as well. The problem is that now we cannot set the TD_NOTCONNECTED flag, which calc_distanceCurveVerts uses. Just removing td++; and tail++; also won't work because the calc_distanceCurveVerts function expects the hidden handle to have that flag set. But I think I would agree that a better solution is to just remove the else if and rewrite the calc_distanceCurveVerts function so that we don't need the hidden handles anymore.

Reanalyzing the code, I don't think TD_NOTCONNECTED is making any difference because, if I'm not mistaken, hidden elements are not considered selected and therefore are not added to the queue in calc_distanceCurveVerts.
I actually think that this TD_NOTCONNECTED flag and the td->dist member are amenable to future removal as td->rdist with the value FLT_MAX is enough to indicate what is not affected proportional edit and the Connected Only option apparently cannot be changed during the operation.

If you don't mind, I plan to edit this patch to focus on an alternative solution. (I still think td++; and that tail++; should be removed).