or use features that I don't use and so don't remember to prevent backtracking.

Only, it won't work here, because If I did /a(?>.*?)b/, it would never match, because once the .*? ate a "b", it would never give it back.

print 'xxxxaxxxxbxxxx' =~ /a(?>.*?)b/ ? 'match' : 'fail';
[download]

So yours is the better solution... you don't need the "?", because /^[^a]*?a/ and /^[^a]*a/ will match the exact same substring. I have no idea which is faster, or whether it even makes a difference, so I'll just leave it in.

So, my code adapted to produce your pattern:

$string = 'abcd';
$regex = '^' . join '', map "[^$_]*$_", map quotemeta, split //, $stri
+ng;
# embed capture to get offset:
$regex =~ s/(?<=\*)/(/; $regex =~ s/$/)/;
$sequence = 'xxxxxaxxxxxxbxxxxxcxxxxdxxxxx';
if($sequence =~ /$regex/s) {
    printf "I found a match, at offset %d and with total length %d\n",
+ $-[1], $+[1]-$-[1];
}
[download]

Instead of m/a(?>.*?)b/ you can use m/a(?>.*?b)/ (b inside the cut assertion)

Afaict regex inside the cut assertions can backtrack, but once the engine left the assertion it will not try to match different text with it.

But the fact remains that you have to know quite some details regular expressions to get that working, so the approach with negated char classes is preferable.