As I stated in another reply. I don't actually want to count bits at all. Counting is just the simplest thing I could do with the bits in the test case. Turning a stream of bytes into a stream of bits can be quite useful.
But if you are reading a stream of bits, you don't want to read one bit at a time either, because that is pathologically slow. You want to read n buts at a time (where n varies each call, probably) at which point you're doing a very standard mask-and-shift with no magic...
