Calculations with DOUBLEs involve rounding errors due to the limited precision. There is nothing you can or should do about it but taking it into account. Using COSD "helps" for this specific case, but only by rounding the small result to zero by accident.
Reliable numerical methods do never rely on a certain result of a calculation. Using an interval is much saver:
if cos(x) == 0 % Bad, unpredictable results!
if abs(cos(x)) < 10 * eps % Save
See also:
