I had a similar design once, you can not debounce encoder outputs, you will miss reversals, just visualize "A" going hi>lo and a couple of uS later lo>hi. With Peters (elegant) solution, you will be very close; however your precision will be depending on the frequency of the timer interrupt. If your application is mission critical convert Peters code and burn it into a fast CPLD, that will give you reaction time in the low nS range. The problem with encoder processing is that there is always some slack in the mechanical linkage and when the stop is exactly at a border you can actually (in one case of 1.432 :-)) see the whole darn thing oscillate. Get the mechanical connetion to the encoder as rigid as humanly possible.

best of luck,

Erik Malund