Symbolic: Status Report 2007.04.23
Work continues on getting the cognitive searcher ready for its debut on FICS. This should happen by the end of this week assuming that I can get Feisty Fawn (Ubuntu 7.04 Linux) up and running on the 3 GHz P4 box by then.
A few bugs were found in the Lisp code. In one of them, Symbolic's hint production was returning a correct move but sometimes without rank/file disambiguation markings. This is only a problem when the move is sent back to the command processor via ASCII text and the receiver can't match the move against the current position.
A more serious bug occurred when the search tree was many, many ply in depth. I had added a feature to the cognitive search where the predicted variation, taken from the search tree, was automatically extended at the end of the search to include moves from tablebases and other sources. Well, this worked just fine for most cases but crashed when testing a mate in 124 in the KQPKQ endgame. It turns out that the ChessLisp interpreter's recursive descent evaluator hit the runtime stack space limit. Increasing the available space via a linker directive should help, but there's some problem getting the linker to recognize the appropriate directive. The problem only occurred with trees well over a hundred ply deep, rare but still possible with the PV extension. I've re-written a the few whole tree scanning Lisp routines to operate in iterative vs recursive mode and this has solved the problem.
There is also a minor bug with the threefold draw detection issuing false negatives in some places. I'm still working on this.
Symbolic: Status Report 2007.04.23
Moderators: hgm, Rebel, chrisw