[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

RE: [gnugo-devel] Bug in owl code in 3.1.10

From: Gaertner, Greg
Subject: RE: [gnugo-devel] Bug in owl code in 3.1.10
Date: Mon, 15 Oct 2001 15:01:45 -0400

Further experimentation seems to indicate that --enable-dfa is the
   When I configure using --enable-dfa  I get
D5 cannot be attacked (160 variations)
D5 can be defended at j5 (32 variations)

If I build not using DFA I get the same thing as Dan got below. ie
D5 can be attacked at C8 (4 variations)
D5 cannot be defended (4 variations)

with  gnugo --quiet -l 3110bug.sgf -L63 --decide-dragon D5

        The reason I said the scoring was messed up is that the
estimated score
is  also to be based on the notion that this group was alive.  I don't
its a separate problem.


Greg Gaertner
Benchmark Performance Engineering
High Performance Servers 
Compaq Computer Corp.
(603) 884 - 0031

-----Original Message-----
From: Daniel Bump [mailto:address@hidden
Sent: Monday, October 15, 2001 2:12 PM
To: address@hidden
Subject: Re: [gnugo-devel] Bug in owl code in 3.1.10

Greg Gaertner wrote:

> The attatched sgf  file contains a game in which the owl code appears
> assess a group as alive, even though
> its either critical or  dead.
> gnugo -l 3110bug.sgf -Lxx -T     xx anything between 59 and 71
> illustrates the problem.

Not on my machine! I'm using the latest CVS version but there
have been no changes to the engine since the 3.1.10 release.

I assume you're talking about the D5 dragon.

gnugo -T shows this in cyan (dead) and we can confirm this as

address@hidden freedom]$ gnugo --quiet -l [file] -L63 --decide-dragon D5
finished examine_position
D5 can be attacked at C8 (4 variations)
D5 cannot be defended (4 variations)

Is it possible this is a platform dependent problem?

> 3.1.10 scoring is also messed up.  I assume its a side effect of
> this problem.

A statement like ``3.1.10 scoring is also messed up'' is not a
good bug report. Please give a reproducible example.


gnugo-devel mailing list

reply via email to

[Prev in Thread] Current Thread [Next in Thread]