search

Matthies / RubiChess

Another chess engine
GNU General Public License v3.0
155 stars 17 forks source link

TB bug #397

Closed Matthies closed 1 year ago

Matthies commented 1 year ago

Rubi runs into a 3-fold in a TB6-win position. https://www.chess.com/computer-chess-championship#event=ccc-20-blitz-main&game=539 TB-bug.log

Matthies commented 1 year ago

Problem seems the logic: If repetition occured, only play moves with best DTZ. Exactly these moves are producing the repetitions. Probably test for each move if it produces a repetition and filter them out may be better solution.

Matthies commented 1 year ago

Another example here: https://www.chess.com/computer-chess-championship#event=ccc-20-blitz-main&game=804 TB-bug2.log

Matthies commented 1 year ago

A third example: https://www.chess.com/computer-chess-championship#event=ccc-20-blitz-main&game=935 This time Rubi ran into 50 moves draw, so repetitions may not be the problem. Common pattern seems that Rubi doesn't see resp. avoids pawn moves. TB-Bug3.txt

Matthies commented 1 year ago

This last 50 move draw could easily be reproduced with current master and it disclosed that the missing update of the halfmove counter (introduced with playMove which is used in the tb probing code) is the reason for wrong moves. So commit 4bdb8b5b083d39f07ed7028de9805bf1b57471f2 should fix this.