Re: [gnugo-devel] Bamboo connection : test case and patch

[Gunnar Farneback]

Stephane wrote:
> In the attached test game GNU Go looses 8 stones at move 172 by
> playing M16 instead of M15. Investigation (via the options -t -T)
> shows that it considers the left part of the upper black group
> to be dead despite the bambon joint. It seems to be an amalgamation
> problem since the two parts of the bamboo are not in the same dragon.

No, it's not really an amalgamation problem. Although it's not
unreasonable to have them in the same dragon, amalgamation of
tactically unstable worms tends to lead to complications. There are
indeed a couple of connection patterns to make such amalgamations but
those were only included to work around some owl problems and should
be removed sooner or later.

The real problem here is that the owl reading fails to find any
defense. Presumably this would be solved by Arend's escape
experiments. Another plausible solution is to find better heuristics
for when to test a tactical defense in do_owl_defend(), line 2615 of
owl.c.

/Gunnar