I have noticed that Stockfish, Berserk, Ethereal all have a "ponder" option.
I thought it may be a clever alternative to ponderhit: set it to true at the start of the game and the engine won't stop thinking after sending bestmove.
However I tried:
That is not what the ponder option is supposed to do. The option only serves to notify the engine that it is playing a ponder game, so that it can adapt its time management based on the assumption that part of the time it spends thinking its clock will not be running. So that it can afford more thinking time per move for the same clock settings and number of moves to go.
An engine should only think it is pondering when it has received the command "go ponder" (and no ponderhit or stop yet).
After "go movetime" an engine should always stop searching and send a "bestmove" after the specified time.
Engines that behave differently are simply not UCI compliant.
Thanks.
I re-checked all 3 engines and it works like you described (I probably typed something wrong before).
I have also checked Stockfish's code and indeed it just increases optimumTime by 25%: https://github.com/official-stockfish/S ... n.cpp#L102
I have another question though: why it is not called UCI_ponder?
Are there any cases where a GUI should let the user set the value manually?
I guess that there also is a UCI v1 and v2, like there is a CECP v1 and v2. The first version of the protocol did not have the ucinewgame command, and the UCI_ prefixing was not yet conceived. But it already did have Ponder and Hash.
