[gnugo-devel] More about dinners

[bump]

In another message:

http://mail.gnu.org/archive/html/gnugo-devel/2003-02/msg00098.html

I explained the concept of a dragon dinner. Basically a
dinner of a dragon is an adjacent worm which is tactically
alive but owl dead, such that the life of the dragon ultimately
depends on eating the dinner.

Perhaps the dinner should be a dragon, not a worm. I think
perhaps it is better to take a worm, not a dragon. The
dinner worm is contained in some dragon, an I'll call
that the dinner dragon.

I think some of GNU Go's current problems can be traced
to forgetting to eat the dinner. For example, if either
a liberty is filled of a dragon with a dinner, or if
a stone is added to a dead dragon that is a dinner,
the dinner dragon can be marked thrashing. Also, 
whatever piece of code marks the dragon as dinner
can also be asked to provide a move that attacks it,
and a move reason can be generated.

We've already noted the need to do better in attacking
thrashing dragons. I'd add this to the list of ideas in:

http://mail.gnu.org/archive/html/gnugo-devel/2003-01/msg00141.html

See other messages in that thread.

There are two functions that should be called on to
recognize dragon dinners. One is owl_defend, and I
started work on implementing this. I ran into trouble
with the Zobrist hashing. I think the str2 field
in the read result can be used to store the dinner
but I didn't get this working yet.

The other function that can be called on to recognize
dragon dinners is new_semeai and there is no issue
with hashing.

Dan