Greg Strong wrote on Fri, Nov 15, 2019 12:21 AM UTC:
We might miscommunicate, because I am thinking in terms of a pseudo-legal-move search, while you might want to determine move legality before searching them.
A reasonable distinction - from the point of view of a pure engine, it might not matter if a move is technically legal if you return -INFINITY for those moves that aren't. For a GUI (that is, either Game Courier, or ChessV at the root), we need to know what is legal vs. illegal, although you do address this (quoted later.)
IMO independently trying to establish legality of moves in a search is just a waste of time, as moves are legal more often than not. Just do the search, and when the move is illegal this will discover it soon enough, and will return -INFINITY. But most of the time there is nothing to discover. As long as an illegal move will get a score lower or equal than anything else, the search will work fine.
What you say makes perfect sense. It does. And yet, Stockfish does not work this way. It does not allow a king to be captured and evaluate it badly - rather, it considers that move to be illegal, even in a deep search. I've looked at a ton of open source chess engines, and what you desribe is not how they work. In truth, I don't really "understand" chess engines well at all - I sort of understand, but have a very, very hard time visualizing it. You clearly understand these things on a level that I do not. I'm really just a sophisticated copy-cat who models my open-source engine after the code of others. What you describe is a nice simplificaiton, but you are far off the beaten path and I am reluctant to follow you down this road, especially given that ChessV doesn't just play Chess, it plays over a hundred games, and it does it with absolutely NO "special cases" for any given game jammed in the middle of anything. None. It does what it does by means of a very sophisticated architecture of message-passing between Game, Piece, PieceType, and Rule classes glued together dynamically at run-time. Not to brag, but there is nothing else in the world like it. I don't want to tear apart what works on promise of what might be a 0.01% improvment in a typical game, nor what might make an unlikely corner-case of Marseillais more playable.
If you want to test legality of an input move, just do a sufficiently deep search on it (1 ply + QS in normal chess, 2 ply + QS in Marseillais), and reject it if the score is -INFINITY. You don't need any special code for that.
Ok, here you might be right. As a GUI, I really only need to know what is legal at the root, and again, what you say makes sense. But I am afraid to follow you. For one thing, ChessV must, must, must do the right thing in every chess variant which it comes across. I stipulate that as an immutable parameter of its design (a parameter which I have temporarily broken by implementing Makruk without the counting rules.) Is it always the case that an evaluation of -INFINITY (game lost) is the same as a move that is not legal? And 1 ply + QS in normal chess, 2 ply + QS in Marseillais? Doesn't this already indicate that we don't really know and we are doing an approximation? Is this univerally applicable to all variants it supports and may support? For Progressive Chess, does it need to be X ply + QS? And what the heck does QS even mean in terms of a game like Marseillas where there are no quiescent positions?
Please don't misintrepret - I am not doubting you. Your engines are awesome and there are many different games where you have the best engine in existance and, in some cases, the only engine. I'm not doubting you. I'm doubting MYSELF. I'm just not willing to make a radical change when I cannot fathom the consequences.
All that said, at some point I will have a sophisticated test suite of dozens of positions in dozens of different games so I can test architectural changes with some level of confidence. For example, I really like the alternative to mate-distance-pruning you suggested on another forum that is like a universal application of it that prefers the shortest route to any advantage. It seems genious, and I want to use it. But I'm afraid, given that Stockfish doesn't use it ... And I have no robust test suite to prove it doesn't break anything...
A reasonable distinction - from the point of view of a pure engine, it might not matter if a move is technically legal if you return -INFINITY for those moves that aren't. For a GUI (that is, either Game Courier, or ChessV at the root), we need to know what is legal vs. illegal, although you do address this (quoted later.)
What you say makes perfect sense. It does. And yet, Stockfish does not work this way. It does not allow a king to be captured and evaluate it badly - rather, it considers that move to be illegal, even in a deep search. I've looked at a ton of open source chess engines, and what you desribe is not how they work. In truth, I don't really "understand" chess engines well at all - I sort of understand, but have a very, very hard time visualizing it. You clearly understand these things on a level that I do not. I'm really just a sophisticated copy-cat who models my open-source engine after the code of others. What you describe is a nice simplificaiton, but you are far off the beaten path and I am reluctant to follow you down this road, especially given that ChessV doesn't just play Chess, it plays over a hundred games, and it does it with absolutely NO "special cases" for any given game jammed in the middle of anything. None. It does what it does by means of a very sophisticated architecture of message-passing between Game, Piece, PieceType, and Rule classes glued together dynamically at run-time. Not to brag, but there is nothing else in the world like it. I don't want to tear apart what works on promise of what might be a 0.01% improvment in a typical game, nor what might make an unlikely corner-case of Marseillais more playable.
Ok, here you might be right. As a GUI, I really only need to know what is legal at the root, and again, what you say makes sense. But I am afraid to follow you. For one thing, ChessV must, must, must do the right thing in every chess variant which it comes across. I stipulate that as an immutable parameter of its design (a parameter which I have temporarily broken by implementing Makruk without the counting rules.) Is it always the case that an evaluation of -INFINITY (game lost) is the same as a move that is not legal? And 1 ply + QS in normal chess, 2 ply + QS in Marseillais? Doesn't this already indicate that we don't really know and we are doing an approximation? Is this univerally applicable to all variants it supports and may support? For Progressive Chess, does it need to be X ply + QS? And what the heck does QS even mean in terms of a game like Marseillas where there are no quiescent positions?
Please don't misintrepret - I am not doubting you. Your engines are awesome and there are many different games where you have the best engine in existance and, in some cases, the only engine. I'm not doubting you. I'm doubting MYSELF. I'm just not willing to make a radical change when I cannot fathom the consequences.
All that said, at some point I will have a sophisticated test suite of dozens of positions in dozens of different games so I can test architectural changes with some level of confidence. For example, I really like the alternative to mate-distance-pruning you suggested on another forum that is like a universal application of it that prefers the shortest route to any advantage. It seems genious, and I want to use it. But I'm afraid, given that Stockfish doesn't use it ... And I have no robust test suite to prove it doesn't break anything...