The possible problem that I can think about is that stockfish starts game without clear hash in match like the noomen match from fixed positions and previous game effect the next game.mcostalba wrote:Yes it does, otherwise it would be impossible to develop.Uri Blass wrote: The first question is if stockfish one cpu produce deterministic output.
I always test the cleanup/code rewrite patches benchmarking on a position's set, at fixed depth and with one thread only and checking if node count doesn't change, as example with the command line:
stockfish bench 100 1 13 default depth
Regarding the parameters, I start to think they are too much!!!!
To visually shrink them I am thinking of adding one morecalled "advanced options" (unchecked by default) that let Stockfish to show additional parameters only on request.
The point is that even with non deterministic behaviour in the first move if there is probability of 1/10 that stockfish play strange move then you probably can reproduce it.
If stockfish blunders in the first move of the game as happened based on the pgn then maybe the problem is that it starts games in noomen match without clear hash and in this case even with no bugs there is a problem in testing it because I do not like one program to learn from previous game when another program does not learn.
The problem may be in the gui that does not tell stockfish that it plays a new game but I think that it can be good if stockfish in default conditions clear hash automatically if it notice a new game even without command for new game and it is possible to do it simply by noticing that the moves that it gets from the gui are clearly different then the moves that it got last time so it cannot be the same game.
Uri
