Arpad Rusz wrote: The max. number of captures for one side in a legal position (T.Marlow, W.Cross, 1967)
[D]r1n1n1b1/1P1P1P1P/1N1N1N2/2RnQrRq/2pKp3/3BNQbQ/k7/4Bq2 w - - 0 1
(74 white captures)
This position is interesting because is able to blow up your qsearch() !
Try it on your favorite engine
Position is not ever going to matter in a game so IMO not worthy of repair. We've had a similar discussion in the past with similar position. There are literally millions of positions that need to be addressed first IMO There was no need to be rude to Uri Marco, because he never means to criticize, this is just Uri.
Anyway I was posting here mainly because think I might have just found a very tiny improvement for the quiescence search by double pass filtering but it is not going to solve this artificial position. Is there any interest from the Stockfish team to try it out because we have to beat the cloud... Are you all by yourself Marco? Joona has not posted in a while and neither has Tord in the last two weeks...
Eelco
Debugging is twice as hard as writing the code in the first
place. Therefore, if you write the code as cleverly as possible, you
are, by definition, not smart enough to debug it.
-- Brian W. Kernighan
Eelco de Groot wrote:IMO There was no need to be rude to Uri Marco, because he never means to criticize, this is just Uri.
And I am just myself !
I didn't mean to be rude with him, it is just that I really hate sentences like "it is like this, it is like that" even more when totally untested and even even more when the possible test would just take few minutes: so it is not that I wanted to criticize Uri, it is just that I've read an escalation of things that I don't like, one right after the other in a quick shotgun fire It just happened to be Uri that wrote that, but if it is was someone else the reaction would have been the same, perhaps even worst because with Uri we have exchanged many interesting posts in the past and I hope to continue like that also in the future.
BTW if Uri has no problems to write "this and that" I think he should have no problems in accepting a strong reply, I have no problems to read strong sentences against me or what I have done....it is just that I don't keep quiet
sluijten wrote:How about captures & promotions, what is the maximum number for any position?
You get quite a few from this, which I call "King in a Can":
[d]rnbq1bnr/PPPPPPPP/8/3ppp2/3pkp2/3NpN2/4K3/8 w - -
I assumed that you are restricting captures to promotion captures.
If you assume 255 as maximum for any move type from any legal board position, I don't think you could go wrong.
218 is the largest number of legal moves of any type from one position that I have seen.
Of course, examples are not proofs of any kind. It is possible that more than 256 are possible. It's just that nobody has ever proven it one way or the other.
Arpad Rusz wrote: The max. number of captures for one side in a legal position (T.Marlow, W.Cross, 1967)
[D]r1n1n1b1/1P1P1P1P/1N1N1N2/2RnQrRq/2pKp3/3BNQbQ/k7/4Bq2 w - - 0 1
(74 white captures)
This position is interesting because is able to blow up your qsearch() !
Try it on your favorite engine
Position is not ever going to matter in a game so IMO not worthy of repair. We've had a similar discussion in the past with similar position. There are literally millions of positions that need to be addressed first IMO There was no need to be rude to Uri Marco, because he never means to criticize, this is just Uri.
Anyway I was posting here mainly because think I might have just found a very tiny improvement for the quiescence search by double pass filtering but it is not going to solve this artificial position. Is there any interest from the Stockfish team to try it out because we have to beat the cloud... Are you all by yourself Marco? Joona has not posted in a while and neither has Tord in the last two weeks...
Eelco
Thanks for your support.
Note that I disagree with the idea that
"Position is not ever going to matter in a game so IMO not worthy of repair."
I think that it is a good idea to care about the worst case
and I believe that the position is only a symptom of a problem that can also happen in some positions of practical games(in less extreme cases) even if in games the qsearch does not explode but only has too many nodes that cause the program to be slower in achieving a big depth.
Dann Corbit wrote:Sure, but I think it would be difficult to generate a lot more promotion choices than that setup.
You can certainly add a few.
[D]rnbqqbnr/PPPPPPPP/8/3ppp2/3pkp2/3N1N2/4K3/8 w - - 0 1
If the goal is to test your move generator, then no problem. If the goal is to try to blow up your search, change the two white Knights to Rooks, as the current position is Mate in 1 (Nf2).
[D]rnbqqbnr/PPPPPPPP/8/3ppp2/3pkp2/3R1R2/4K3/8 w - - 0 1
I let Stockfish think for three minutes (on a very slow machine, relatively) and it still couldn't find mate, although the score is +51.75.