Add only a permitted increment

Wrapping a sum does not legalize an out-of-range increment.

• For width b, a defined increment n is in [0,2^(b−1)−1], including zero. Compute (s+n) modulo 2^b using ordinary nonnegative unbounded integer addition and modulus. Subtraction, negative increments and additions outside that range are not defined by this operation (§3.1). The helper reports an input error for these chosen local calls; the RFC does not prescribe an error packet.
• With eight bits, there are 256 labels and the largest defined increment is 127. Adding zero leaves a serial equal. Adding any defined nonzero increment makes the result greater than the starting serial, even when its printed integer label decreases (§§3.1,4.1,5.2).
• Published examples include 255+1=0,100+100=200,200+100=44. The last example first computes the ordinary sum 300, then reduces it modulo 256 to 44. Do not subtract 256 only when convenient or use signed host-integer ordering as the serial rule (§5.2).
• An eight-bit increment of 128 is undefined even if a machine can produce a wrapped numeric result. The limit concerns the size of one increment, not only whether the result fits. Several individually valid increments need a separate history analysis (§§3.1,4.2–4.4).