i can't reproduce locally
#Previous position not being set after running a bench
somber badger
from the results it would appear that somehow the default position isn't startpos
but once again that doesn't happen to me so that's not possible
did you get the binary from the pre-releases on github?
what happens if as soon as opening the binary you type "d" instead of go?
blazing wagon
@cerulean violet didn’t you have something similar?
somber badger
i mean
jolly jungle
what
blazing wagon
can you type uci and give us the version string?
it might be default printed cant remember
somber badger
it should be
yeah the version string would be helpful
also the output of the d command
jolly jungle
you have 960 enabled, somehow
cerulean violet
<@716518045614604428> didn’t you have something similar?
I don't understand what's the issue or what's going on LOL
jolly jungle
bench
its the last position```cpp
// Chess 960
"setoption name UCI_Chess960 value true",
"bbqnnrkr/pppppppp/8/8/8/8/PPPPPPPP/BBQNNRKR w HFhf - 0 1 moves g2g3 d7d5 d2d4 c8h3 c1g5 e8d6 g5e7 f7f6",
"nqbnrkrb/pppppppp/8/8/8/8/PPPPPPPP/NQBNRKRB w KQkq - 0 1",
"setoption name UCI_Chess960 value false"
};
yep
can reproduce locally
if you bench it turns 960 on
for the future that's the type of information that is useful for the people trying to debug your stuff
so yeah you can "fix" this by not running bench each time i guess, while an actual fix gets committed
blazing wagon
you can also just do ./stockfish bench
blazing wagon
I don't understand what's the issue or what's going on LOL
xd i think you posted something similar #sf-dev with some duplicated output dunno
blazing wagon
I think we validate the net before each go
I can’t remember any real specific changes to that
Can you check master vs sf16?
jolly jungle
Stockfish 16 by the Stockfish developers (see AUTHORS file)
go depth 1
info string NNUE evaluation using nn-5af11540bbfe.nnue enabled
info depth 1 seldepth 1 multipv 1 score cp 2 nodes 20 nps 20000 hashfull 0 tbhits 0 time 1 pv g1f3
bestmove g1f3
go depth 2
info string NNUE evaluation using nn-5af11540bbfe.nnue enabled
info depth 1 seldepth 1 multipv 1 score cp 2 nodes 20 nps 20000 hashfull 0 tbhits 0 time 1 pv g1f3
info depth 2 seldepth 2 multipv 1 score cp 2 nodes 40 nps 40000 hashfull 0 tbhits 0 time 1 pv g1f3
bestmove g1f3
i don't use sf enough to know if it's standard
yeah ok it's just how it's always been
gotcha
cerulean violet
I would like to think so
since the net can be changed run time
but I still don't understand the main issue here
how is 960 enabled ? while he didn't enabled?
jolly jungle
its the last position```cpp
// Chess 960
"setoption name UCI_Chess960 value ...
he ran a bench, the last positions are 960 ones
cerulean violet
what's his issue to begin with?
is that he enabled chess960 but then lost the option after running bench?
jolly jungle
he executed stockfish, ran a bench, and then analyzed, expecting startpos to be used, but the last bench position was used
cerulean violet
okay that's a bug
jolly jungle
it has been like this for years tho (since sf 9)
hot brambleBOT
cerulean violet
I have seen bugs from SF1 still exist
does not mean it's not a bug
even if it's like that from SF9
last position in the bench command has no meaning
jolly jungle
I said that because the titled was saying it was an issue with the latest dev
cerulean violet
got it
cerulean violet
@blazing wagon I'm reading this back, do you know how can we clone/deep copy the states variable here
https://github.com/official-stockfish/Stockfish/blob/632f1c21cd271e7c4c242fdafa328a55ec63b9cb/src/uci.cpp#L147
such that we revert back the states variable after bench is done?
I think that would be a nice fix and a better expected behavior.
Correct me if it's the only thing that needs cloning
blazing wagon
Does clearing the deque not work and emplacing a new one back ?
cerulean violet
Well iiuc it works to solve the OP issue
I was thinking that someone will do something like
setoption name Threads value 4
position fen rnb1kbnr/ppp1pppp/8/3q4/8/5N2/PPPP1PPP/RNBQKB1R b KQkq - 1 3
bench
then after benching I would have liked that the old input he put still there, but I'm not sure if you are discussing the technical aspect or design aspect
blazing wagon
mh yeah okay not that straight forward
bench sets it's own options
youd probably better off by extracting bench completely from uci, to not edit the uci settings
cerulean violet
Yeah that would be cleaner
I was thinking about cloning states somehow
because that avoids setting somehow a new internal commands
but if it's not dependent on strings
then it would be indeed cleaner
blazing wagon
mh not quite, bench needs a lot of uci stuff to setup
cerulean violet
I wonder if the recent refactoring helps this issue
blazing wagon
well, startpos is set, not _previous_
wdym, what are your commands?
jolly jungle
wdym, what are your commands?
position startpos moves e2e4
bench
d
wtf ?
+---+---+---+---+---+---+---+---+
| n | q | b | n | r | k | r | b | 8
+---+---+---+---+---+---+---+---+
| p | p | p | p | p | p | p | p | 7
+---+---+---+---+---+---+---+---+
| | | | | | | | | 6
+---+---+---+---+---+---+---+---+
| | | | | | | | | 5
+---+---+---+---+---+---+---+---+
| | | | | | | | | 4
+---+---+---+---+---+---+---+---+
| | | | | | | | | 3
+---+---+---+---+---+---+---+---+
| P | P | P | P | P | P | P | P | 2
+---+---+---+---+---+---+---+---+
| N | Q | B | N | R | K | R | B | 1
+---+---+---+---+---+---+---+---+
a b c d e f g h
Fen: nqbnrkrb/pppppppp/8/8/8/8/PPPPPPPP/NQBNRKRB w GEge - 0 1
Key: A903849C3CC4940B
Checkers:
it's not the stat pos as you say
blazing wagon
yeah i think david was fooled by startpos chess960
jolly jungle
nah, I just did a quick check not looking too much into it and didn't notice it was 960
also whats with all the recent hate wtf
blazing wagon
yeah i got confused for a second
cerulean violet
wtf you can't even take a joke (removed the jokes)
unreal pollen
oh hi david
jolly jungle
hello