Approach move kos (was: Re: [gnugo-devel] Good semeai results!)

[Arend Bayer]

Gunnar wrote:
> > Delivery-Date: Mon Jan 28 16:44:54 2002
> > Date: Fri, 18 Jan 2002 12:32:37 +0100 (CET)
> 
> Wow, 10 days delay. Wonder what kind of sightseeing this message and
> another one that just arrived have been doing.
Well, too bad for the message, it seems to have spent silently 10 days on
my machine, and suddenly it turned up again --- I must have misconfigured
s.th with the automatic sendmail-triggered dialup.

> > > I'm not sure how the semeai version is in terms of speed.
> > > It also needs to be taught about kos.
> > 
> > Might this be a good moment to extend the ko scheme? After all, most
> > (maybe all?) approach kos result from semeais.
> 
> Do you have any proposal for how to extend the ko scheme?
I haven't tried to exactly understand the komasterscheme yet, in particular
I do not know what happens for multiple kos, and what I am writing here
is completely ad hoc.
I think for a single ko there is no need to change the komaster scheme.
Look at the following lines extracted from an arbitrary attack function
in reading.c (starting line 3258):

    if (komaster_trymove(xpos, other, "attack3-A", str,
                         komaster, kom_pos, &new_komaster, &new_kom_pos,
                         &ko_move, stackp <= ko_depth && savecode == 0)) {
(Here an attack move is tried.)
      if (!ko_move) {
(...)
      }
      else {
        if (do_find_defense(str, NULL, new_komaster, new_kom_pos) != WIN
            && do_attack(str, NULL, new_komaster, new_kom_pos) != 0) {
          savemove = xpos;
          savecode = KO_B;
        }
      }
      popgo();
    }

I do not know what it means to pass the (new_komaster, new_kom_pos) to
do_attack (if we try two moves for the same color in a row I do not see
where there could be any ko issues, but that might because I do not
understand the double ko issues yet).
If we would call do_attack with empty komaster, then I think savecode
should be
- KO_B only if do_attack reports a WIN
- bad double ko if it reports KO_A
- one step approach move ko if it reports KO_B
- two step approach move ko if it reports one step approach move ko
- ...

Does that make sense?

Arend