- Floating point inaccuracies. A simplified version of Patrick Rapin's "math.compare (x,y,delta)" [1] could be proposed as a workaround.