Vector3::normalize() returns wrong result

[paexter]

Normalizing a vector vec with a very large component does not not lead to the expected result.

vec: Vector(0, 0, 3.40282e+38)
vec.length(): inf
vec.lengthInverted(): 0
vec.normalized(): Vector(0, 0, 0)

The expected result is vec.normalized(): Vector(0, 0, 1)

[pezcode]

You can probably mitigate the intermediate length overflow by using something like (vec/Math::abs(vec).max()).normalized()

[mosra]

(Sorry for the extremely late reply, only now going through the notification pile.)

Ah well, numerical issues. There's a tradeoff between doing the simplest possible thing (and being fast), or being immune to every possible corner case and ending up with a similar monstrosity as std::lerp(). I'd need to do some research, do you know about any "usual" way how this is being done robustly? I looked at Eigen's normalized() and it suffers from the same problem:

  RealScalar z = n.squaredNorm();
  // NOTE: after extensive benchmarking, this conditional does not impact performance, at least on recent x86 CPU
  if(z>RealScalar(0))
    return n / numext::sqrt(z);
  else
    return n;