Re: [gnugo-devel] super-ko violation

[Gunnar Farnebäck]

Matthias wrote:
> The Gnugo-Bot I'm running on KGS managed yesterday to build the
> tripple ko (as black) you can find in the attached sgf.

Indeed, a genuine triple-ko.

> It seems, that Gnugo (3.6-pre1) violated the super-ko rule.
> Unfortunately I do not have the exact position, where this happened
> (some moves are missing at the end of the file saved on KGS). I
> reconstruced the moves starting with M1, and verified that gnugo
> would play in the position at the end of the attached file L6, which
> violates super-ko. KGS reported (in my log files) that the illegal
> move L6 was send from the engine, so I suppose this was indeed what
> happened.

Seems reasonable.

> I do not know whether this problem can be solved via gtp,

It cannot.

> but at least I wanted to let you know that this problem exists. And
> even if it is not possible to communicate the exact rules which are
> used, gnugo has to deal with a super-ko situation in one way or
> another.

There's a long-standing TODO item to implement superko detection. For
this particular position it may also be worth adding some test cases
about the moves leading up to the triple-ko.

> Even though I admit that this is not a very urgent thing to
> solve...

Well, the fact that we now have an example of a game leading to
triple-ko may increase the motivation to implement superko detection.

/Gunnar