H. G. Muller wrote on Sun, May 3, 2020 03:22 PM UTC:
I think the problem here is that there are actually two different forms of 'attacked'. Perhaps it is better to make that explicit by giving them different names, rather than call them the same, and specify under which conditions one definition applies, and under which another. It is hard to foresee whether there will occur cases which would like to use one definition, which we inadvertantly assigned to another. For instance, your criterion of outcome determining doesn't seemt to apply to Tai Shogi: capturing a royal can end the game, but it is not legal for an Emperor (which is a universal leaper, and thus always has every piece in its reach) to do it when the royal was protected. So normally a protected royal is not considered attacked by the opponent's Emperor. So this is a case where legality of the actual capture does matter.
In chess-programmers jargon on uses the qualification 'pseudo-legal' to indicate that rules for checking should not be taken into account. Making the way pieces move dependent on the checking rules indeed introduces recursion in the definition. But it cannot be excluded that this is exactly what the designer of a variant intends. I already mentioned 'Recursive Chess', which is FIDE except that it is not legal to expose your King to capture even for capturing the opponent's King. The recursion is innocent, because it always terminates: there are only two Kings that can be captured. The difference with FIDE can be formulated in terms of an after-move: in FIDE there is no after-move after King capture, so the player whose King is captured can never retaliate; he has already lost. In recursive Chess King capture grants one after-move, and if that also captures a King, the player that made the second capture wins! (Making the first King capture illegal, as it is forbidden to expose yourself to an immediate loss. So you would not be in check by pinned pieces in Recursive Chess.) This is similar to King baring in Shatranj, where the bared player gets an after-move to counter-bare, and salvage a draw that way.
So I think it is better to explicitly distinguish (legal) attack from 'pseudo-legal attack'. The FIDE checking rule could be formulated as "it is not allowed to expose your King to pseudo-legal attack". (IMO any pseudo-legal attack on a King would also be a legal attack, because in FIDE you get no after-move to retaliate after King capture. But not everyone might agree with that interpretation, or even be aware of it, so it is always better to explicitly add the 'pseudo-legal' qualifier.) Similar, in the move rules of Fusion Chess it would be proper to stress that it are pseudo-legal attacks that have consequences for splitting.
To keep the definition of 'attack' simple, we could just take out the word *legal* from the definition I proposed, and add the sentence:
We can distinguish *legal* attacks from *pseudo-legal* attacks, depending on the nature of the involved capture.
where both links would point to the entry for 'legal' in the glossary, with the definition I proposed.
You have a good point about the piece-type-dependent nature of squares being attacked. Chu Shogi forbids Lions to capture each other from a distance, but only when they are protected (i.e. recapture is possible). I have always seen this as a rule similar to the FIDE checking rule, where you cannot expose a Lion to recapture, rather than a King. (Chu Shogi has no checking rule for its King.) So Ln x protected Ln would still be pseudo-legal. There has indeed been discussion whether that rule should be interpreted recursively or not, in cases where more than two Lions are involved. For the Janggi Cannon CxC is always forbidden, though, so it cannot be explained as it not being legal to expose your Cannon. The Ultima Chameleon is another piece that captures different victims with different moves.
The problem with 'attacked' for a square is that it is not obvious what you should imagine to be on it. Most likely the issue of an attacked square comes up in connection with describinbg the move of some piece, and that would make it natural to assume a piece of that type on it. That can still be different from imagining an actual piece moves there, as the disappearence of that piece from its old location could affect whether captures to the square are possible. E.g. a lame Dababba on d1 would not attack f1 when the King is still on e1. Can white castle O-O? I would say not; if the King moves to g1 in two steps, after the first step the Dababba can capture him. So he would be moving through check. If there had been a Cannon on d1, it would have attacked f1 before the castling, but no longer after the first King step. (But of course you would end up in check once you would put the Rook on the other side of the King.)
I think the problem here is that there are actually two different forms of 'attacked'. Perhaps it is better to make that explicit by giving them different names, rather than call them the same, and specify under which conditions one definition applies, and under which another. It is hard to foresee whether there will occur cases which would like to use one definition, which we inadvertantly assigned to another. For instance, your criterion of outcome determining doesn't seemt to apply to Tai Shogi: capturing a royal can end the game, but it is not legal for an Emperor (which is a universal leaper, and thus always has every piece in its reach) to do it when the royal was protected. So normally a protected royal is not considered attacked by the opponent's Emperor. So this is a case where legality of the actual capture does matter.
In chess-programmers jargon on uses the qualification 'pseudo-legal' to indicate that rules for checking should not be taken into account. Making the way pieces move dependent on the checking rules indeed introduces recursion in the definition. But it cannot be excluded that this is exactly what the designer of a variant intends. I already mentioned 'Recursive Chess', which is FIDE except that it is not legal to expose your King to capture even for capturing the opponent's King. The recursion is innocent, because it always terminates: there are only two Kings that can be captured. The difference with FIDE can be formulated in terms of an after-move: in FIDE there is no after-move after King capture, so the player whose King is captured can never retaliate; he has already lost. In recursive Chess King capture grants one after-move, and if that also captures a King, the player that made the second capture wins! (Making the first King capture illegal, as it is forbidden to expose yourself to an immediate loss. So you would not be in check by pinned pieces in Recursive Chess.) This is similar to King baring in Shatranj, where the bared player gets an after-move to counter-bare, and salvage a draw that way.
So I think it is better to explicitly distinguish (legal) attack from 'pseudo-legal attack'. The FIDE checking rule could be formulated as "it is not allowed to expose your King to pseudo-legal attack". (IMO any pseudo-legal attack on a King would also be a legal attack, because in FIDE you get no after-move to retaliate after King capture. But not everyone might agree with that interpretation, or even be aware of it, so it is always better to explicitly add the 'pseudo-legal' qualifier.) Similar, in the move rules of Fusion Chess it would be proper to stress that it are pseudo-legal attacks that have consequences for splitting.
To keep the definition of 'attack' simple, we could just take out the word *legal* from the definition I proposed, and add the sentence:
where both links would point to the entry for 'legal' in the glossary, with the definition I proposed.
You have a good point about the piece-type-dependent nature of squares being attacked. Chu Shogi forbids Lions to capture each other from a distance, but only when they are protected (i.e. recapture is possible). I have always seen this as a rule similar to the FIDE checking rule, where you cannot expose a Lion to recapture, rather than a King. (Chu Shogi has no checking rule for its King.) So Ln x protected Ln would still be pseudo-legal. There has indeed been discussion whether that rule should be interpreted recursively or not, in cases where more than two Lions are involved. For the Janggi Cannon CxC is always forbidden, though, so it cannot be explained as it not being legal to expose your Cannon. The Ultima Chameleon is another piece that captures different victims with different moves.
The problem with 'attacked' for a square is that it is not obvious what you should imagine to be on it. Most likely the issue of an attacked square comes up in connection with describinbg the move of some piece, and that would make it natural to assume a piece of that type on it. That can still be different from imagining an actual piece moves there, as the disappearence of that piece from its old location could affect whether captures to the square are possible. E.g. a lame Dababba on d1 would not attack f1 when the King is still on e1. Can white castle O-O? I would say not; if the King moves to g1 in two steps, after the first step the Dababba can capture him. So he would be moving through check. If there had been a Cannon on d1, it would have attacked f1 before the castling, but no longer after the first King step. (But of course you would end up in check once you would put the Rook on the other side of the King.)