Remove nmpMinPly

[TarasVuk]

Passed STC:
LLR: 2.95 (-2.94,2.94) <-1.75,0.25>
Total: 108256 W: 28116 L: 27975 D: 52165
Ptnml(0-2): 394, 12286, 28608, 12465, 375
https://tests.stockfishchess.org/tests/view/66d5ffa99de3e7f9b33d14bd

Passed LTC:
LLR: 2.95 (-2.94,2.94) <-1.75,0.25>
Total: 184530 W: 46414 L: 46360 D: 91756
Ptnml(0-2): 152, 20203, 51499, 20261, 150
https://tests.stockfishchess.org/tests/view/66d6f6219de3e7f9b33d1562

bench: 1281840

[github-actions]

clang-format 18 needs to be run on this PR.
If you do not have clang-format installed, the maintainer will run it when merging.
For the exact version please see https://packages.ubuntu.com/noble/clang-format-18.

(execution 10816070910 / attempt 1)

[peregrineshahin]

It would be nice to understand if this is a looser form of NMP or a stricter one.. not so clear at first sight..

[MinetaS]

matetrack result:

Engine ID:     Stockfish dev-20240911-5a07e893
Total FENs:    6554
Found mates:   3161
Best mates:    2198

[xu-shawn]

It would be nice to understand if this is a looser form of NMP or a stricter one.. not so clear at first sight..

Should be looser since it removes the disabling of NMP on plies near the verification search root. So idealy this should be tested with the Zugzwang test suite https://github.com/official-stockfish/fishtest/wiki/Creating-my-first-test#useful-resources

[MinetaS]

Tested with Threads=1, Hash=1024 and go nodes 1000000000. Positions where PR took nodes >= 10M:

FEN	Bestmove	PR	master
k2N2K1/8/8/8/5R2/3n4/3p4/8 w - - 0 1	Rf7	11.04M nodes	51.69M nodes
8/6B1/p5p1/Pp4kp/1P5r/5P1Q/4q1PK/8 w - - 0 1	Qxh4+	726M nodes	63.26M nodes
n1QBq1k1/5p1p/5KP1/p7/8/8/8/8 w - - 0 1	Bc7	Not found	864M nodes
8/8/8/1B6/6p1/8/4KPpp/3N2kr w - - 0 1	Kd3/Ke3	Not found	60k nodes
8/8/8/1B6/6p1/8/3K1Ppp/3N2kr w - - 0 1	f4	731M nodes	41.95M nodes
8/8/8/2p5/1pp5/brpp4/1pprp2P/qnkbK3 w - - 0 1	h3	Not found, D245	1.76M nodes
1k3b1q/pP2p1p1/P1K1P1Pp/7P/2B5/8/8/8 w - - 0 1	Kd5/Bb5	Not found	Not found

[peregrineshahin]

Oops you have Not found, D245 an irreversible mistake here??
Could you please check if that is the case with master?

[peregrineshahin]

sorry I didn't see the master column, yeah this patch introduces a point of no return.

[MinetaS]

Oops you have Not found, D245 an irreversible mistake here??

For the specific position depth 245 was reached before 1B nodes and the bestmove was not found.

info depth 245 seldepth 22 multipv 1 score cp 0 nodes 1545651 nps 2851754 hashfull 1 tbhits 0 time 542 pv h2h4 a1a2 h4h5 a2a1 h5h6 a1a2 h6h7 a2a1 h7h8r a1a2 h8h1 a2a1
bestmove h2h4 ponder a1a2

[peregrineshahin]

very undesirable sitiuation, to keep falling in the same search heuristic trap over and over again, until reaching max depth, and halting the engine..

[MinetaS]

Well these positions are a small portion of the given epd. I would not make a conclusion with few test cases, still it needs analysis to figure out if this PR actually has regression on zugzwang positions.

[peregrineshahin]

Well these positions are a small portion of the given epd. I would not make a conclusion with few test cases, still it needs analysis to figure out if this PR actually has regression on zugzwang positions.

note here the conversation
https://discord.com/channels/435943710472011776/813919248455827515/1208072882178105364
It is the only example where we know that the problem is irrevirsible
No need for sample size weighting when you know exactly where the problem is from a billion of calls and repeated failure on this positions.
the counter argument should be to find a such position where the mistake can't be undone with master.

[peregrineshahin]

Well,I might be exaggerating but in principle I don't see any reason to have the verification search if not for irreversible mistakes.

[Disservin]

Since null move pr's pop up every now and then I propose we should find a solution on how we treat those, and what tests they have to pass in order to be considered for merging. We seem to have that discussion on a lot of these pr's with no clear path forward. If someone likes to propose something please open a discussion/issue.

I'll close this one in the meantime.