Ah, OK. Well the system is the same as you used to generate passwords: once you have an array of valid code points for the characters, you can choose 10 random codepoints from that array, like
$chosencodepoint = $codepoints[int rand @codepoints];
and construct the string, for example with:
$string = pack 'U*', @chosencodepoints;
(pack('U', $int) is pretty much equivalent to chr($int) except it also guarantees the output will be turned into UTF-8, and most of all: with the star this can easily join multiple characters without an explicit join.)
My point was: when printing it out, Perl will convert it to valid UTF-8. Don't worry about that.
An in case you want no duplicates, you can repeat the process for each character until you find no duplicates. With so many characters to choose from that is virtually guaranteed to be faster than shuffling the whole array (with the Fisher Yates shuffle) and next picking the first 10 code points. Or, you can make custom version of Fisher-Yates that stops shuffling after 10 iterations. | 