I think a reasonable compromise is to have "char" be a 32 bit type, whereas a "string" would be UTF8.