If you improve your algorithm to process directories as arrays of name components instead of copying back to strings, it will not be a problem any more and your algorithm should be faster, too.

As for the odd behavior, I suspect that it was needed for portability to some oddball half-broken commercial Unix back in the 5.005 era. :-)

Yeah, I thought about processing an array but I would still have to use File::Spec to split the file path and I could just feel it in my bones that there would still be some weird cross-platform issue lingering and I just want to go to bed. So I just did letter-by-letter comparison. Speeding it up will have be a task for another day.

$PM = "Perl Monk's";
$MCF = "Most Clueless Friar Abbot Bishop Pontiff Deacon Curate Priest Vicar";
$nysus = $PM . ' ' . $MCF;
Click here if you love Perl Monks

Conceptually directory structures are trees, and paths are lists of nodes that represent traversal⁽¹⁾. When you think of things that way instead of strings with slashes, life becomes easier, and common tools tend to just work. Pushing to the boundaries of your application as much representational decoration as possible, and internally working with more native data structure types is a principal that doesn't only apply to data serialization formats, it also applies to filesystem paths.

Your algorithm should be working with lists of directory names and file names, and converting to slash-delimited strings at the ingress and egress boundaries.