A bug in Crafty 20.14

[ilari]

Already posted about this at http://wbforum.vpittlik.org/viewtopic.php?t=6873

In short, the problem is that Crafty can't force a mate in this easily won endgame without tablebases: 5R1N/K7/8/3k4/8/8/8/8 w - - 0 131
If I give it enough time it _may_ win by sacrificing the knight to avoid a draw by 50 moves.

I've confirmed this with both the 64-bit Ubuntu build (from the multiverse repository) and the win32 build from Robert Hyatt's ftp site.

[jwes]

Already posted about this at http://wbforum.vpittlik.org/viewtopic.php?t=6873

In short, the problem is that Crafty can't force a mate in this easily won endgame without tablebases: 5R1N/K7/8/3k4/8/8/8/8 w - - 0 131
If I give it enough time it _may_ win by sacrificing the knight to avoid a draw by 50 moves.

I've confirmed this with both the 64-bit Ubuntu build (from the multiverse repository) and the win32 build from Robert Hyatt's ftp site.

On my computer with 768m hash it takes about 3 minutes to find the mate, but it seems singularly clueless until it sees the mate (it keeps wanting to move the knight back to h8).

[bob]

Already posted about this at http://wbforum.vpittlik.org/viewtopic.php?t=6873

In short, the problem is that Crafty can't force a mate in this easily won endgame without tablebases: 5R1N/K7/8/3k4/8/8/8/8 w - - 0 131
If I give it enough time it _may_ win by sacrificing the knight to avoid a draw by 50 moves.

I've confirmed this with both the 64-bit Ubuntu build (from the multiverse repository) and the win32 build from Robert Hyatt's ftp site.

On my computer with 768m hash it takes about 3 minutes to find the mate, but it seems singularly clueless until it sees the mate (it keeps wanting to move the knight back to h8).

There is a gross bug in EvaluateMate() which is in evaluate.c... If you want to fix it, look at the top of EvaluateMate() which has this:

int mate_score = DrawScore(1);

change it to this:

int mate_score = 0;

Not sure what is wrong there and how it happened, but a change was made somewhere along the way and the basic idea is that in a KBN vs K ending crafty plays it just like it should. For any other ending without pawns, it has no idea about driving the king to the edge of the score, which is a killer. The above change fixes it...

I tried KQ vs KR, which is a trivial ending for a computer (without tablebases) and it had no idea at all. Which led me to the bug...

[Jim Ablett]

Crafty 20.14 Intel compiler 10 PGO windows build with bugfix:

http://myfreefilehosting.com/f/14ef30b726_0.33MB

Jim.

[BBauer]

I can access your site,
but how can I download a file?
regards
Bernhard

[Alexander Schmidt]

I can access your site,
but how can I download a file?
regards
Bernhard

Click on: "Click Here To Download"

Alex

[Jim Ablett]

Hi Bernhard,

I can access your site,
but how can I download a file? Surprised

There is some problems with the download servers/providers I've
been using for my homepage download links, that's why I'm using
an alternative for the Crafty link.

Hopefully fixed soon:)

Jim.

[beachknight]

Hello Jim,

Could you compile for mt=2 for me?

TIA,

[Jim Ablett]

Crafty 20.14 Intel 10 PGO compiled with /DSMP +

/DCPUS=2

/DCPUS=2 (CoreDuo only)

/DCPUS=4

/DCPUS=8


http://www.2xupload.de/file/662e6812c46 ... u.zip.html
http://www.2xupload.de/file/994049b6bd4 ... u.zip.html
http://www.2xupload.de/file/a055d20e6b2 ... u.zip.html
http://www.2xupload.de/file/60fa270c65d ... o.zip.html


Jim.

[Dirt]

Crafty 20.14 Intel compiler 10 PGO windows build with bugfix:

http://myfreefilehosting.com/f/14ef30b726_0.33MB

Jim.

The 'Click Here To Download' link there just times out for me.