Re: [gnugo-devel] eye value patch

[Gunnar Farneback]

Arend wrote:
> I had just a look at the owl failure in this position (at first not
> realizing that it might be related to your plans):
> 
> |XXXX.
> |OOOX.        (O to move)
> |.O.OX        (GNU Go sees no way to live)
> |.aOXX
> +----
> 
> In trying to understand this failure, I came up with the patch below to
> evaluate_diagonal_intersection, which changes the topological eye valuation
> of a from 3.0 (half eye) to 2.75 (half eye, can be made false only with
> ko) as it should be.
> However, the existing graph matching can't use this improved information
> yet, so I am not proposing this patch for CVS. It might however be worth
> trying out together with your planned chanes.

I don't plan to handle ko contingent margins, or for that matter ko
contingent eye values in general, at this time. The solution to the
former problem would probably be to call the graph matching twice.
Once for the eye shape obtained if the ko is lost and once for the eye
shape obtained if the ko is won. This is of course not particularly
difficult to do, but it's not clear how to represent the result in a
useful way.

/Gunnar