[_5ysi]

In the intro section, on extending to unicode being non-trivial...is the argument that if your input has some non-ascii data you probably have all non-ascii data and the above 127 bucket will be too full? Honest question.