If you make just a small change to the Graphemes::next method then this breaks the tests in graphemes_test because the GraphemeCursor::next_boundary returns GraphemeIncomplete::PreContext(1). As far as I understand, this indicates that the cursor tries to look back at preceding grapheme clusters that it has already processed. I don't know if this behavior is intentional or required to process UTF-8 correctly, but it seems unnecessary because the tests do pass if you just use a fresh cursor instead and treat your current position as the start of the string. Some performance improvements might be possible if you do it that way.