Thing is, getting substrings is an operation one often performs in a loop – e.g. in order to split a string by a delimiter. If extracting the substring takes linear time, and the number of substrings you extract depends on the length of the string, then you end up quadratic, which can be really painful.
Anyway, in your average code, if you just “need a substring” (and are dealing with UTF-8 encoded strings), you should be using byte indices instead of codepoint indices. After all, they’re both equally useless if you need a semantic notion of “character”, so you may as well go with the metric that’s more efficient. A Scheme interpreter may be a special case due to the requirements of the Scheme standard, but, well, special cases are special and don’t justify adding easily-misused functionality to the standard library. And even then, you’d be better off using a non-UTF-8 string type throughout your interpreter (not just converting at the boundaries as you’re suggesting).
