Re: [gnugo-devel] pattern-based reading spinoffs (27.7)

[Gunnar Farneback]

Dan wrote:
> The existing komaster code starts forbidding ko captures when someone
> illegally takes a ko, simulating a ko-threat/response/ko-capture.  If
> there are enough kos floating around, and if both players prefer
> taking a ko to filling a ko, this triggering illegal ko capture will
> never happen unless we implement superko (which Gunnar might be
> contemplating in view of his recent patch implementing a move stack).

No, I have no intention of implementing superko for use in reading and
I don't think it would be very useful either. What would be
interesting is to have an optional test for superko in the final step
of the move generation to avoid the blunder of playing an illegal move
in a tournament with superko rules. This is not on my agenda currently
but with my forthcoming undo patch this will become a much simpler
exercise than before since it adds move history to the board state.

> Should the aim of the komaster scheme be to absolutely limit
> the size of the tree or should it be to prune it just enough
> that if the players choose well then deep trees will be avoided?

I think the most important purpose is to enforce consistent reading of
groups living or dying in double ko, as exemplified in the docs.
Without the komaster scheme the outcome was arbitrarily decided by
where in the double ko capture sequences certain depth limits were
reached.

Second most important is to reduce the search trees in ko situations,
but it certainly shouldn't have to work against a broken move
ordering.

Third it would be nice if it could stop triple ko repetitions (all
legal with respect to the basic ko rule). This is out of scope for the
existing komaster schemes (except Trevor's new experiment) since they
only kick in when illegal ko captures are tried. Notice that this
isn't just a question of bad move ordering, there are valid positions
where repeating a triple ko is the only useful move, if allowed. I
think I know a good solution to this by extending the existing
komaster schemes but I have to implement and test it to know for sure.

/Gunnar