Since I have a long way to go before becoming a good programmer, I sometimes refer to Code Complete, The Mythical Man-Month and the like to keep me on the right track.
I think I have reached that point, of throwing away the first one built, with the Blunder PGN to FEN chess translation component I have been programming for the past month.
To be honest with myself, I foresaw these design problems back when I originally did the design. I knew I would have to deal with many of the things I am dealing with now way back when I was doing the original design (although not totally - checking that a piece is pinned to the king is more important than I thought it would be, if I thought of it all). The thing is, designing the program with all of that in mind would be "boring". It would be too abstract initially, it wouldn't DO anything until quite a lot of the program was coded. The way I programmed this, it worked right off the bat - at least with the first PGN I used as a basis. It translated the first ply of the first move correctly, and then the next ply of the first move, then the first ply of the second move and so on. After that all worked, I tried another PGN. As I sought to get it working for my various PGNs, I added more and more functionality to the program.
The method functionality I need now seems rather abstract, or at least more abstract than the functionality I have now. "Check to see if piece (rook or queen) is pinned to king horizontally", "Check to see if piece (bishop or queen) is pinned to king diagonally", and so on. Things are a little more abstract than I'd like, but if I try to keep things very specific, I will have much, much more coding to do.
The program currently does over 95% of PGNs correctly, but there are too many possible corner cases to deal with. The functionality that deals with plies (half-moves), which is most of the program, has to be rewritten.
The main thing I focused on with the initial design was the data structures. I did change things around a bit, especially the Board class, which is my half-way class between the translation of the PGN to FEN. I also realized while programming that I needed a Move class. When functionality got to where over nine out of ten PGNs parsed, I wanted to do PGN files that had multiple games within it - and thus a Game class was created as well.
One nice thing is, aside from the edge cases I have to redesign for, my PGN to FEN converter has some aspects that are superior to the two other converters I've found out there - Lutz Tautenhahn's PGN-to-FEN converter and 7th Sun Green Light Chess's pgn2fen.exe program for Windows (or Linux, with WINE). Tautenhan's program I tested out more - I saw two problems - one, castling ability which is disabled due to a rook move is re-enabled if the rook moves back to the square. I'm fairly sure this is not legal with FIDE rules. Secondly, if a pawn move results in pawn promotion, Tautenhan's converter does not reset the half-move clock due to the pawn move, but in fact increments it. I believe this is not the case with FIDE rules, but am less sure. As far as the Green Light chess converter, I have not looked at it as much as Tautenhan's, but I do know it does not mark en passant squares in the FEN.
Blunder's converter marks en passant squares, disables castling availability properly, and resets the halfmove clock on all pawn moves - even pawn promotions, which I believe is the correct behavior. Now I just have to redesign and abstract the methods that deal with converting a ply to a new configuration for my Board object. Which is most of the methodology for the program. I might tinker a little more with the data structures, perhaps making them a but more robust.