Because the double precision approximation of 1.255 is slightly below 1.255 and the double precision approximation of 1.455 is slightly above 1.455. You introduce the error during the assignment, not during the manipulation. Ratazong's solution actually introduces additional errors into the process since now you have two approximated numbers, as well as additional maintenance complexity.

Why is your application so sensitive to this change? There are ways to handle it, but this is a consequence of the real hardware you are working with. Depending on your application, you could do all your math with integers, use Math::BigFloat, use Math::BigRat, or tweak your algorithm/order of operations.