I'll be reviewing it in a while. I'm not too sure about the memory layout it uses (__uni_char_data could really be compressed to use less memory for example), nor about the interface it exposes, but it does seem to work well. About is_grapheme_break though, isn't the implementation for legacy grapheme cluster rather than extended ones though? Dear Mathias, The uni car data was laid out in that way for speed and simplicity. Please feel free to do it better. The grapheme break worked off the character attributes and standard rules. If it needs changing or did not cover a later standard, then it should be updated. The principle would not change as it makes the string iterators quite straightforwards.