Re: [gnugo-devel] semeai question

[Gunnar Farneback]

I wrote:
> Another conclusion of my analysis of the regression failures is that
> we could do much better if we had semeai analogues to the
> owl_does_defend() and owl_does_attack() functions. A
> find_more_semeai_moves() function in value_moves.c should also be
> beneficial.

I've put up an extension/revision of my previous patch
(gunnar_3_20.17) at
http://www.lysator.liu.se/~gunnar/gnugo/patches/gunnar_3_21.1

This patch implements an owl_analyze_semeai_after_move() function. By
restructuring the code in do_owl_analyze_semeai(), more specifically
breaking out a function semeai_trymove_and_recurse(), this could be
done with almost no additional code.

The new owl_analyze_semeai_after_move() function can be tested through
a new GTP command called analyze_semeai_after_move. I also took the
opportunity to rename the GTP command owl_analyze_semeai to only
analyze_semeai.

Another use of owl_analyze_semeai_after_move() implemented by the
patch is a revision of detect_tactical_blunder(). This was intended to
solve test case blunder:26, which it does and which is also the only
effect on regressions of this patch compared to gunnar_3_20.17.
Potentially this change could also have solved blunder:27 and a few
more test cases, were it not for the fact that the dragon amalgamation
would need to be redone after the blundering move before this test
could be effective.

/Gunnar