Re: [gnugo-devel] regression patch

[Gunnar Farneback]

Arend wrote:
> I've put a regression patch in CVS (it's linked at devel.html). It adds
> two tests; the owl test looks actually like an interesting eye valuation
> problem. It is a seki position, eye shape .XXX. ; GNU Go would analyze
> it correctly if the inner worm would be alive (no outside liberties) or
> dead (two or more outside liberties), but gets it wrong with one outside
> liberties: the three stones are considered lunch, and it generates a
> move to eat it, destroying its own eye space.

I've seen similar problems before and they are extremely annoying. I
think the solution is to eliminate the distinction between lunches and
eyespaces, which are today analyzed completely differently. There are
probably some complications with this but in most cases it should
allow significantly better analysis of the eyevalue of lunches.

Observant people may have noticed that the new eyes.db includes lots
of eyespaces which can't occur today because they involve lunches,
e.g. pattern 6937. In other words we are well prepared for a paradigm
shift in this area.

/Gunnar