We have a 'divide by zero' situation in simgear/math/vector.cxx, but I'm
sure what the right fix is. Here's the offending routine:
// Given a point p, and a line through p0 with direction vector d,
// find the shortest distance (squared) from the point to the line
double sgdClosestPointToLineDistSquared( const sgdVec3 p, const sgdVec3 p0,
const sgdVec3 d ) {
sgdVec3 u, u1, v;
// u = p - p0
sgdSubVec3(u, p, p0);
// calculate the projection, u1, of u along d.
// u1 = ( dot_prod(u, d) / dot_prod(d, d) ) * d;
double ud = sgdScalarProductVec3(u, d);
double dd = sgdScalarProductVec3(d, d);
double tmp = ud / dd;
sgdScaleVec3(u1, d, tmp);;
// v = u - u1 = vector from closest point on line, p1, to the
// original point, p.
sgdSubVec3(v, u, u1);
return ( sgdScalarProductVec3(v, v) );
}
The problem is that the formula used here only works for nonzero
vectors (according to my old Linear Algebra textbook). Since
dot_prod(u,v) = dot_prod(v,u)
if u or d equal zero, we need to do something different. I'd suggest
something like:
...
double tmp;
// u = p - p0
sgdSubVec3(u, p, p0);
if (sgIsZeroVec3(u) || sgIsZeroVec3(d)) {
tmp = 0;
} else {
// calculate the projection, u1, of u along d.
// u1 = ( dot_prod(u, d) / dot_prod(d, d) ) * d;
double ud = sgdScalarProductVec3(u, d);
double dd = sgdScalarProductVec3(d, d);
tmp = ud / dd;
}
sgdScaleVec3(u1, d, tmp);
...
There's no such thing as sgIsZeroVec3() right now. This would
effectively fix the FPE, but is the vector math correct? Is setting
tmp=0 a proper fix?
Thanks
--
Cameron Moore
/ One cannot guess the real difficulties \
\ of a problem before having solved it. /
_______________________________________________
Flightgear-devel mailing list
[EMAIL PROTECTED]
http://mail.flightgear.org/mailman/listinfo/flightgear-devel
