Well, that's about efficiency. If you want to be able to slice at grapheme boundary n, you have to iterate through the string counting up to n to find out wher

Well, that's about efficiency. If you want to be able to slice at grapheme boundary n, you have to iterate through the string counting up to n to find out where it is. Might as well just provide an iterator over graphemes in the first place. You can optimize slicing on codepoint boundaries by storing strings internally in a fixed-width representation, like Python does. But now you have to convert encodings all the time just to use UTF-8 for your I/O. And that still only gets you codepoints. It's not a great tradeoff to do it that way. It wastes memory and CPU time, and randomly accessing characters of strings is just not something you actually need to do often enough to optimize for it.