I'm a lazy guy, but I'd just implement algebraic notation and worry about the classical notation parsing later. The parsing isn't the hard part, it's the shorthand. Rf3 implies questions like "which rook?", and "QPxR" is sometimes shortenable to "PxR", but not always.

I've started a chess engine a few times (never really in Perl), and I always stopped earlier due to the thought that, really, I'd rather play humans :) Anyhow, it's a fun effort, full of lots of interesting questions like how to write efficient move validators, alpha/beta pruning, etc, etc, etc. I'm almost more than tempted to pick this up again, perhaps writing something a little less overplayed than chess, such as perhaps Abalone -- yeah, one ships with GNOME and/or KDE, but it could use a higher quality replacement that allows sidereal moves.

As a final thought, you may be able to extract some algorithms for PGN parsing from XBoard.

Good show for keeping up the motivation! I need more of that myself.