Author: Uri Blass
Date: 02:14:26 01/02/03
Go up one level in this thread
On January 02, 2003 at 05:00:06, Russell Reagan wrote: >I'm trying to nail down this bug in my move generator. I'm calculating perft for >positions and verifying them with other perft calculating engines. First I did >perft with only kings on the board, and things seemed fine. Then I added queens, >and things seemed fine. Then came the kings and rooks, and things seemed fine >when castling wasn't involved, but then I tried the following position: > >[D]r3k2r/8/8/8/8/8/8/R3K2R w KQkq - 0 1 > >Things stopped being fine. Other engines I tried got perft(4)=314346, but mine >gets perft(4)=314340, off by 6 nodes. Since there are *only* 26 legal moves in >the above position I set out to try each one, calculate perft, and find the >branch branch (or branches) containing the errors. > >The move Rb1 leads to a position that my engine incorrectly calculates perft 3. >So I had an incorrect perft 4, and I found one branch that has an incorrect >perft 3, so I'm on the right track (or so I thought). I tried each move from the >position after Rb1, and to my suprise not one move turned up an incorrect perft >1 or perft 2, which is what I was expecting. > >It seems like this shouldn't happen. Am I missing something? Or is this yet >another bug? > >Russell I suggest that you ask your program to give you a list of legal moves and numbers of perft n-1 in the same run that you calculate perft n. I did it when I was debugging my move generator. Your program in order to calculate perft 3 need to calculate perft 2 for every legal move and add the numbers so it can print for you a list of numbers for all the legal moves in the same run that it calculates perft 3 and you will find what number is not correct by trying it with other programs. You may have a bug that is result of do undo and in order to see in what position it happens you need to calculate perft 2 in the same run that you calculate perft 3. Uri
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.