[gnugo-devel] Goal dragon cuts (was: Pattern A1120)

[Portela Fernand]

Gunnar wrote:

> Let's hear your ideas. Speed is probably the critical issue.

Maybe not. We could let the code continue to assume that
connectivity is ok unless some events happen. For instance I
was thinking of :

- adding B class to some patterns in owl_attackpats.db, which
  would update the same_dragon field of the moves list.
- in do_owl_attack(), watch for events :
  * moves[k].same_dragon has some specific value
  * the tried move captures something in the goal
- locate the 2 (or more?) splitted groups
- choose the new target (see below)
- updating the owl data structure doesn't look easy and practical
  so building a new local_owl_data struct seems unavoidable. But
  we might have cache issues here. I have to check this.
- try to owl_defend the new goal
- restore the owl data structure

Then I think we should just define a policy to decide which goal
becomes the target and which one(s) are just tails. Criterions
could be size of the groups and how much original stones each
group has (while waiting for adding more complex ones).

A source of complexity here are (as usual) kos. I'm not too sure
how to deal with this. Ideas ?

/nando