H. G. Muller wrote on Thu, Apr 13, 2023 06:14 PM UTC:
Implementing the rules was as easy as I thought;
it only required a custom script of 5 lines.
(But I did not implement resurrection of an e.p. captured Pawn.)
Unfortunately that does not mean the AI can actually play it.
The problem is that the AI does not truly recognize repetitions.
It only discourage reversing your previous move.
And it does not test for that on captures (which normally can never repeat a prior game state).
Even if it did, it would test the wrong thing, as repetitions here repeat the previous move of the same player,
rather than reversing it.
The consequense of this is that in the search it will at somepoint quickly encounter a position
where perpetual recapture is possible.
And since recaptures are always searched, no matter how deep the search already went,
this leads to an infinite regression that crashes the program.
I will have to think whether there is an easy solution to this.
Implementing the rules was as easy as I thought; it only required a custom script of 5 lines. (But I did not implement resurrection of an e.p. captured Pawn.)
Unfortunately that does not mean the AI can actually play it. The problem is that the AI does not truly recognize repetitions. It only discourage reversing your previous move. And it does not test for that on captures (which normally can never repeat a prior game state). Even if it did, it would test the wrong thing, as repetitions here repeat the previous move of the same player, rather than reversing it.
The consequense of this is that in the search it will at somepoint quickly encounter a position where perpetual recapture is possible. And since recaptures are always searched, no matter how deep the search already went, this leads to an infinite regression that crashes the program.
I will have to think whether there is an easy solution to this.