(However, and as a limitation of the current implementation, using "\w" or "\W" inside a
"[...]" character class will still match with byte semantics.)
Well, logically, those are arguably the same (at least, have been since day two), so I'd call the current implementation "buggy" if it can't make those the same.