Update: Oops, I had the terms reversed on the regex, it should be $i =~ $x .... I've corrected it, but it doesn't really change the results.

For single characters without repeats, such as any permutation of 'abc', this seems the shortest, most readable method:

@x = glob('{a,b,c}'x3); # for testing: all permutations, plus many mor
+e

$x = '(?:([a-c])(?!.*\1)){3}'; # regex

for $i (@x) {
    $i =~ $x and say $i; # was $x =~ $i
}
[download]

which outputs:

abc
acb
bac
bca
cab
cba
[download]

For single character, with limited repeats, such as any permutation of 'aabbcc', a slightly different approach is needed. I tried the following, but it matches almost everything. I suspect there's some unexpected behavior in the backtracking?

@x = glob('{a,b,c}'x6);
$x = '(?:a()|a()|b()|b()|c()|c()){6}\1\2\3\4\5\6'; # limited repeats a
+llowed

for $i (@x) {
    $i =~ $x and say $i; # was $x =~ $i
}
[download]

which outputs 540 of the 728 strings given (at least 1 of each char is present):

aaaabc
aaaacb
aaabac
aaabbc
aaabca
...
cccbba
cccbca
ccccab
ccccba
[download]

Is there some other magic to DWIM?

For nonrepeating, multichar permutations, even when individual chars are shared between tokens, this approach works:

@x = glob('{abc,bcd,cde}'x3);
# permutations of 'abc/bcd/cde'
$x = '(?:abc()|bcd()|cde()){3}\1\2\3';

for $i (@x) {
    $i =~ $x and say $i; # was $x =~ $i
}
[download]

which outputs:

abcbcdcde
abccdebcd
bcdabccde
bcdcdeabc
cdeabcbcd
cdebcdabc
[download]

I found this here on Stack Overflow.