Re: [gnugo-devel] bigger TODO list

[Daniel Bump]

> Yes, I'd say you are. The tactical reading has always been of the
> belief that seki is life, or at least tactical safety. What my patch
> did was to improve the move generation for very small scale semeais
> and this may increase the chances of defending by constructing a seki.
> But to say that the tactical reading is or was lacking support for
> seki is off the mark.
> 
> The owl code, on the other hand, has a very limited understanding of
> seki.

The task of recognizing seki doesn't belong in reading.c but
in the semeai module.

The plan is that owl_analyze_semeai will soon be able to recognize
seki reliably. Currently I'm working on seki but postponing the
issue of ko. When we start to support ko, the type of the
functions owl_analyze_semeai and do_owl_analyze_semeai will change
from void to int.

Check out the examples in regression/semeai.tst. Many of these
positions are seki. The tests all pass but the last two examples
(based on the top position in semeai9.sgf) will fail if you replace
tactical_analyze_semeai by owl_analyze_semeai because of a
spurious amalgamation.

Dan