Greg Strong wrote on Sun, Apr 7, 2019 05:17 PM EDT:
Sorry it has taken so long to get back to you on this. A lot to chew on here.
First you mentioned engine-defined variants the GUI doesn't understand but still allows to be played. ChessV doesn't do this, and it's not high on my list of things to do, but there is certainly some value to it so it would be best if whatever we settle on supports this.
Then you mention my game variables concept and the more general concept of allowing engines to suppy additional options as things change (variant selected, other options selected, option buttons clicked ...) Certainly the more general concept has value, but I also think there is value in being able to specify that an option is a game-specific variable. This would make clear that, not only does the option only apply to a specific game, but that the option affects the rules of the variant being played and not just the internals of the engine AI (although it could certainly do that too.) This tells the GUI that, in order to play, both engines must support the same options, and both should be automatically set to the same values after the user selects them. And I think it makes sense that these options should be presented differently because they are an essential part of selecting what game is being played. In ChessV when you select Chess with Different Armies, the first thing you get is a dialog box to pick your armies with random armies pre-selected. (The same thing happens if you pick Fischer Random Chess - you get a dialog with a random position number selected, with a preview, but you can change it if you want to play a specific position.) I think game variables can offer a better user experience - if you don't care about, or even don't understand hashtable size, etc, who cares, but you better understand what the armies are. Picking the armies is part and parcel of picking what variant you want to play. And, also, some engines might play CwDA and offer extra armies that others do not.
Another thing I like about game variables it the potential they offer to allow an engine to play variants it wasn't specifically designed for. ChessV has some understanding of this now. If the user selects Corridor Chess, it is smart enough to offer as options any engine that supports both orthodox chess and the setboard option. Likewise, if it offers variant capablanca and has setboard, it will be offered as an option for Modern Carrera, but not for Schoolbook because of the flexible castling. (You did propose a clever workaround for this but I haven't got around to implementing it.) But engines that support flexible castling should offer the "Castling" option of "Flexible" in variant capablanca. Then, even if they don't know Schoolbook, Grotesque, or Ladorean, they can still play them, along with new variants that meet the same rules.
10x8 is popular, in the sense that there are lots and lots of variants, but almost all use one of a few castling rules, which I have named "None", "Standard", "Long", "Fexible", "Close-Rook", and "CRC". Likewise, on a 10x10, there are only a handful of castling rules, and a handful of rules for the pawn's multi-step moves that cover most games.
About a year ago you mentioned working towards a standard for defining variants so you don't have to configure Fairy-Max in one way and SjaakII in another. I think this is also worth pursuing - it would be awesome if we could acheive it, but this is a steep mountain. I respectfully suggest that this would make a good first step. This is the low-hanging fruit...
Sorry it has taken so long to get back to you on this. A lot to chew on here.
First you mentioned engine-defined variants the GUI doesn't understand but still allows to be played. ChessV doesn't do this, and it's not high on my list of things to do, but there is certainly some value to it so it would be best if whatever we settle on supports this.
Then you mention my game variables concept and the more general concept of allowing engines to suppy additional options as things change (variant selected, other options selected, option buttons clicked ...) Certainly the more general concept has value, but I also think there is value in being able to specify that an option is a game-specific variable. This would make clear that, not only does the option only apply to a specific game, but that the option affects the rules of the variant being played and not just the internals of the engine AI (although it could certainly do that too.) This tells the GUI that, in order to play, both engines must support the same options, and both should be automatically set to the same values after the user selects them. And I think it makes sense that these options should be presented differently because they are an essential part of selecting what game is being played. In ChessV when you select Chess with Different Armies, the first thing you get is a dialog box to pick your armies with random armies pre-selected. (The same thing happens if you pick Fischer Random Chess - you get a dialog with a random position number selected, with a preview, but you can change it if you want to play a specific position.) I think game variables can offer a better user experience - if you don't care about, or even don't understand hashtable size, etc, who cares, but you better understand what the armies are. Picking the armies is part and parcel of picking what variant you want to play. And, also, some engines might play CwDA and offer extra armies that others do not.
Another thing I like about game variables it the potential they offer to allow an engine to play variants it wasn't specifically designed for. ChessV has some understanding of this now. If the user selects Corridor Chess, it is smart enough to offer as options any engine that supports both orthodox chess and the setboard option. Likewise, if it offers variant capablanca and has setboard, it will be offered as an option for Modern Carrera, but not for Schoolbook because of the flexible castling. (You did propose a clever workaround for this but I haven't got around to implementing it.) But engines that support flexible castling should offer the "Castling" option of "Flexible" in variant capablanca. Then, even if they don't know Schoolbook, Grotesque, or Ladorean, they can still play them, along with new variants that meet the same rules.
10x8 is popular, in the sense that there are lots and lots of variants, but almost all use one of a few castling rules, which I have named "None", "Standard", "Long", "Fexible", "Close-Rook", and "CRC". Likewise, on a 10x10, there are only a handful of castling rules, and a handful of rules for the pawn's multi-step moves that cover most games.
About a year ago you mentioned working towards a standard for defining variants so you don't have to configure Fairy-Max in one way and SjaakII in another. I think this is also worth pursuing - it would be awesome if we could acheive it, but this is a steep mountain. I respectfully suggest that this would make a good first step. This is the low-hanging fruit...