Author: Robert Hyatt
Date: 06:55:45 11/06/99
Go up one level in this thread
On November 06, 1999 at 00:52:32, Bruce Moreland wrote: >On November 06, 1999 at 00:16:20, Robert Hyatt wrote: > >>If you only knew 1/2 as much as you _think_ you know. I could easily post >>the end of the log file so you could see why it hung, but that isn't the >>point. > >You have the log for that game? Why did it hang? > >bruce I have three 'modes' in EGTB endings: (1) if the position is a mate, I stop the search quickly once I find a move that mates at least one move sooner than the move from the last real move played in the game. Pretty normal. (2) if the position is "mated" it used to do a 60 ply search every time, even though it is forced. This produced great big log files, and was _not_ what I had intended. But it was the way it worked. (3) swindle mode where if the side on move is ahead, it will search to avoid making stupid moves that draw, even though the root position is drawn. The bug had to do with mode (3), and problem (2). I fixed (2) but managed to screw up swindle mode, so that in rare cases, the search would stop in a way that crafty thought meant that it was time to read in a move, but the move had been read already. So it just stopped, and did a read(). And nothing was coming in as the move was already read. It hung again, but I was watching, killed it, and found the problem. Vs "ban" I wasn't on at the time and didn't see it. That is why I run with the -reuseChessProgram false as it is possible for this kind of bug to hang it permanently. With the reuse option as above, when a game ends, xboard actually 'kills' the process and starts a brand new one. Which kills the hung engine and starts Crafty again. It was a 'simple fix' that didn't quite work as this code is way too complex now, with "swindle mode" plus the mode "I don't have all the promotion databases, but I do have the "p" database, so don't screw it up..." This plus the ability to find a move to ponder by searching for the other side (puzzle mode) leaves a lot of potential for breaking something, since timing can be critical when reading moves, aborting the search, etc...
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.