Ratings & Comments

I finally understand what is the problem here. There actually was an avelanche of bugs, some stupid (like accidentally masking off the color bit, or forgetting to mask the 'also burns' bit from the code of the promoted piece), and some more fundamental (introduced by the legal highlighting). The stupid ones are fixed now, so that black promotes to black pieces, and the promoted piece does show up on the purple squares.

The problem with the spurious burning is more fundamental, though: the NewClick input routine combined all moves with the same step to a single one for highlighting, making a list of all posible promotion choices for that move. This was fine when burning was only indicated by a different code for the promotion pieces, as it is when the pseudo-legal moves are first generated. Before the moves can be used they are 'scored', however, which also includes identifying which pieces get burned, and writing those into the move as extra locust squares. This used to be done after the user had selected the move and promotion pieces.

But for the legal highlighting it has to be done for all the moves of the selected piece, to make those in order to test legality even before the user selects one. But this interferes with combining the promotion choices when one of those would burn, and the other not. I am not sure yet how to solve this. It is conceivable that one of the promotion choices is legal, and the other not. (E.g. because you burn a piece blocking a slider attack on the King, or fail to burn a hopper mount.)

I suppose the 'middle ranks' are not middle enough. With odd ranks it only allows moves from the entral rank.

Anyway is the transferrer an interesting concept, or the joker should be enough even in this context?

Well, I dislike Jokers to begin with, so perhaps I am not the right person to comment on this. In principle you can make this as complex as you want: not imitate the piece that was moved, but for each possible type define the type the Joker should move as independently. This way you could also define partical imitation: e.g. if you want to disallowe.p. capture and double push when imitating Pawns, you could define a (no-participating) Pawn without these moves, and ilet the Joker imitate that when a Pawn is moves.

This could be implented by having the user define an array to map (moved) piece types onto (imitated) piece types. In the code you would then simply replace "imitated = movedPiece" by "imitated = transfer[movedPiece]". By default the transfer array would map every type on itself.

Oh, yes, thanks!

I suppose I wrote that because it defines three King types. But it appears to use the correct ones, so perhaps I was too pessimistic.

You mean, this can be done in the ID? I mean without you putting in extra work.

Since there are several bugs at once, I'll leave you be.

I might post new diagrams of my ideas in the future, for which you can weigh in as you see fit.

My only reservation is that introducing any form of burning makes it more a Tenjiku-Shogi variant than a Dai-Shogi variant. If it was up to me I would prefer to only use moves of a type that are already in Dai Shogi. A quite strong piece without solo-mating capability could be made by enhancing a Queen with linear two-leg moves like exist on the Horned Falcon and Soaring Eagle.

This logic kind of falls flat, since Tengu Dai Shogi does the same thing with hook moves and Lion Dog moves and is still considered a Dai Shogi variant. Queen+Lion could work for the power piece though (I'm no longer concerned about avoiding lone mating as long as there is still some difficulty in doing so).

I think I somewhat fixed it, now. The scoring routine, which has to be applied to the move exactly once before the latter can be performed, would interpret the promotion code (which would also indicate burning), and change it accordingly, as well as adding any burn victims to the move. For the purpose of legality testing I now always set the promotion code of the move to 0 first (meaning no promotion or burning), and set it back to the original value afterwards. That should always completely restore the move like the scoring routine had not been applied at all. (Which was the procedure before we did the legality testing.) So it has to be applied again on the move selected by the user. But by that time the user has also selected the desired promotion, so it will be known whether the move should burn or not, and the scoring routine will adapt the move accordingly.

This means the board state is now always modified as it should be for the chosen move. The legality testing could be off for moves that do burn, though (or kamikaze moves, which are also indicated by a special promotion code). As there will not be any burning during the legality test, the legal/illegal verdict might be wrong by failing to see a checking piece would be burned, or burning a piece discovered a check. But the destination will be highlighted anyway; just by the wrong symbol (grey cross vs normal). This would not prevent you from making the move, though. Instead of a move-entry bug it is just a highlighting bug.

To fix this I should enhance the procedure for legality testing on moves that always burn or self-destruct. Or perhaps on any move where there is no promotion with choice, as it would in principle also be possible to have type-dependent blocking. Only on burning moves the effect of 'scoring' the move is hard to revert; perhaps I should make a copy of the move for testing legality, and discard that copy afterwards. Promotion with choice is a more complex issue; all promotions are selected through the same highlighted destination square. But some of the choices could be legal, others not.

I also tried to reduce the annoying delay that occured on the first time a piece in the Diagram gets clicked. (The legality testing requires piece values to be defined, so rather than on opening the AI panel the time-consuming guestimating of piece values iss now done on the first click.) I did this by just running fewer iterations on that first click.

This article seems ready for a review. When this article is accepted I will delete the no longer needed 6 articles that were presenting these games separately.

Tenjiku Shogi also has no anti-trading rules. But I actually played that, and even after trading away the Fire Demons and neutralizing the danger of the jumping generals it remains an enormously tense game, because you can get back a Fire Demon through promotion. Which creates an immediately winning imbalance. The Water Buffalo's themselves are only average pieces (by Tenjuku standards), and are not easily traded. So equiping some less important pieces with decisive promotions might be an alternative to anti-trading for keeping a large game exciting.

It also helps that both the Fire Demon and Water Buffalo come in pairs, and are both sliders.

However, I discovered a new bug where any piece that makes a move to a non-highlighted square will burn/move/make pieces. The squares that get affected are a bit unpredictable though. This also appears to apply to diagrams where the spell parameter is not set, such as this one.

This should be fixed now too. It was a consequence of the partial revert of the legality patch, which moved the 'scoring' back to after the move was selected, instead of scoring each move in the move list. But illegal moves are not in the move list, and one is synthesized based on the origin and destination click when no move in the list that satisfies the clicks is left. I had forgotten to remove the scoring of that synthetic move, so that it was scored twice. The first scoring replaces the 'no promotion' code 0 by the encoding of the piece itself. But the second scoring then interprets the 'has moved' bit of this piece as the encoding of a burn. What exactly gets burned is then determined by the type number of the piece.

This article seems ready to be reviewed right now.

Thank you very much, I see what I can do :)