[gnugo-devel] endgame move generator beginnings

[Paul Pogonyshev]

- endgame move generator beginnings implemented in endgame.c
- endgame() in ascii.c renamed ascii_endgame()

of course, endgame is not nearly that important as owl, connection
and tactical reading. still, a small improvement in endgame can
probably help to win one game of a million ;)

regression delta:

./regress.sh . endgame.tst
302 unexpected PASS!
801 unexpected PASS!
./regress.sh . golife.tst
10 unexpected PASS!
./regress.sh . trevorc.tst
1060 unexpected PASS!

all the passes look good. the first two were the target.

there are at least two more "hidden" passes in endgame:301 and
endgame:802, but they don't show up because gnugo mistakenly
overvalues other moves (the tests are the same as endgame:302
and endgame:802, but for other colors). in one case there is a
bogus strategical effect, in the other gnugo prefers to play a
completely pointless sente (0 points, but large followup).

at some point i had four more passes, but they all were for a
wrong reason (and, to some extent, for [!move] tests which can
really give confusing passes/fails sometimes), so i fixed them.

the patch adds yet another one move generator which is placed in
a new file endgame.c. currently, this file is pretty small, but
i think it was right to start a new one. the generator is called
from do_genmove() if there is no move larger than 6.0 points on
the board (same as for endgame.db patterns). it can be easily
turned off if something goes wrong.

it's quite possible that the implemented algorithm makes some
endgame patterns redundant.

as usually, we'll need to fix vc projects.

Paul

patch.tar.gz
Description: application/tgz