Re: [gnugo-devel] Semeai code redesign

[Arend Bayer]

I've posted a further revision of Gunnar's patch at devel.html.
I am positive that it is an improvement of .9c, although I don't have
regression results for exactly this version.

Apart from two trivial cleanups and an added comment/FIXME, this patch
 - revises the cutoff policy when we are out of nodes, to a radical
   "all semeais that run out of nodes are considered critical"-approach.
   This needs more improvement.
 - Reports to the higher levels whether semeai results should be considered
   uncertain (similar to owl results). This is only used in
   semeai_move_reasons() so far.
 - fixes a bug in find_more_owl_..._moves (plus some tracing improvements
   there).

I think this is still not ready for CVS.

Almost all FAILs in genmove tests seem to be due to uncertain semeai
results, that's why I think this is the remaining crucial point.

We should certainly try to run out of nodes less often. This can be
achieved by increasing semeai_variations or by _decreasing_ the
branch depths (I'd guess that SEMEAI_BRANCH_DEPTH is too high at 12).

Arend