Perl operators that deal with paths suffer from The Unicode Bug. The path actually used is provided by the following sub:

sub path_actually_used {
   if (is_utf8($_[0]) {
      my $s = $_[0];
      utf8::encode($s);
      return $s;
   } else {
      return $_[0];
   }
}
[download]

That means that if you have encoded bytes in an upgraded string, Perl will get it wrong.

my $s = chr(9734);
mkdir($s);           # ok
utf8::encode($s);
mkdir($s);           # ok
utf8::upgrade($s);
mkdir($s);           # not ok
[download]

It's virtually impossible to get into that situation without a bug in your code because the encoding functions always return a downgraded string.

You've already identified the solution:

• If the path is a string of encoded text (i.e. UTF-8), passing it through utf8::downgrade($s) will ensure it's used correctly.
• If the path is a string of decoded text (i.e. Unicode Code Points), encoding it (e.g. using utf8::encode($s)) will ensure it's used correctly.