A while back, I attempted to write a chess site that let you paste in chess games in PGN text format, and then step through them. What I found was pretty shocking: the PGN format looks simple, but is remarkably difficult for a computer to parse.

The reason is that the PGN format only specifies the piece type and destination square of each move. Ne4: a knight moves to e4, and it's up to the parser to figure out which knight. So to parse PGN, you have to teach your parser how the pieces move.

That by itself is not too bad, but it gets worse. What if both knights can move to e4? Then the format disambiguates. Nge4: the knight on the g file moves to square e4. But "can move" is a tricky notion in chess. This knight cannot move because it is pinned to the king. That pawn can capture en passant this turn, but not next turn. So to write a PGN parser, you must build a chess engine capable of analyzing a board sufficiently to determine legal moves.

One might hope that reasonable PGN text would be nice and disambiguate whenever two pieces attack the same square, and not require legal move analysis. Alas, it is not so. The "two knights attack a square but one is pinned" case is specifically called out as NOT requiring disambiguation in the spec at http://www.saremba.de/chessgml/standards/pgn/pgn-complete.ht... .

A further consequence is that it is not possible to make a "generic PGN" parser that can handle chess variants, like suicide chess. The rules of chess are baked into the format itself, and the parser must have knowledge of the rules used to play the game that it is presenting.