Is like I understand seeing the Arena "F4" log or I´m missing something?
If you pay atention to the image below, the last information about engine2 moves Arena gives to engine1 was the last to one engine2´s move: a4b4
Arena never informed to engine1 the last move from engine2: Rb4-a4!
Is that possible?
May be is not necesary..... but I think the minimum information an engine deservs to know after play a game, is the last move from its opponent in that game!!
Yes it's an Arena problem. Doesn't send the last move after it's got a result. Possibly to give authors a jump as they might find their engine pondering, on move and the clock running to time out..after a game end.
I think that it's correct to never send the last move, if it is a 3 fold repetition. As seen in the log, Arena sends the result and this could be enough.
If it would sends the move, then the engine would starts thinking and that is a waste of time.
stegemma wrote:I think that it's correct to never send the last move, if it is a 3 fold repetition. As seen in the log, Arena sends the result and this could be enough.
If it would sends the move, then the engine would starts thinking and that is a waste of time.
For this particular case, that is arguably true since FIDE rules say you can claim a draw if you are able to repeat (and demonstrate to the arbiter how), not that the game ends in a draw if you do.
This could translate to the engine sending the move, followed by a result claim. The opponent then gets the result notification from the arbiter.
In general though, the GUI should really send all moves until the conclusion of the game. The engine may not do anything with this information, but if it has some sort of learning enabled it could be important.
Either way, an engine should always process "result" messages from the GUI and consider the current game completed when it receives one.
It is a bit sloppy, but in general not harmful. There is no obligation to send 'result' commands; it is just a service for engines that want to do learning, which GUIs might or might not provide. The learning usually considers statistics on opening lines, and it is questionable whether the move causing the repeat would be important for that.
However, XBoard/WinBoard does send the move after adjudication, after putting the engine in 'force mode' first, so that it will not start thinking.
stegemma wrote:I think that it's correct to never send the last move, if it is a 3 fold repetition. As seen in the log, Arena sends the result and this could be enough.
If it would sends the move, then the engine would starts thinking and that is a waste of time.
For this particular case, that is arguably true since FIDE rules say you can claim a draw if you are able to repeat (and demonstrate to the arbiter how), not that the game ends in a draw if you do.
This could translate to the engine sending the move, followed by a result claim. The opponent then gets the result notification from the arbiter.
In general though, the GUI should really send all moves until the conclusion of the game. The engine may not do anything with this information, but if it has some sort of learning enabled it could be important.
Either way, an engine should always process "result" messages from the GUI and consider the current game completed when it receives one.
