Let moves whose safety depends on ko have influence strength

[gunnar]

This one-line patch,

--- engine/value_moves.c	23 Jan 2006 18:15:51 -0000	1.160
+++ engine/value_moves.c	23 Jan 2006 21:27:02 -0000
@@ -2443,5 +2443,5 @@
   mark_inessential_stones(OTHER_COLOR(color), safe_stones);
 
-  if (move[pos].move_safety == 1 && safe_move(pos, color) == WIN) {
+  if (move[pos].move_safety == 1 && safe_move(pos, color) != 0) {
     safe_stones[pos] = INFLUENCE_SAVED_STONE;
     strength[pos] = DEFAULT_STRENGTH;

lets stones have influence strength also when their safety depends on ko. Breakage:

endgame:880     FAIL T4 [K5]          Problematic.
trevorc:1000    PASS M11 [N9|M11]     Very good.
trevord:260     PASS Q2 [Q2]          Very good.
trevord:790     FAIL T13 [P14]        Accidental. Owl reading mistake.
strategy5:298   PASS D2 [D2|B7]       Good, but A5 still overvalued.
gunnar:76       PASS G17 [G17]        Good.
gunnar:90       PASS H3 [H3]          Very good. Primary target.
gunnar:93       PASS T13 [T13]        Improvement, but the testcase is probably bogus.
seki:403        FAIL B2 [C1|C2]       Accidental. A different move valuation problem remains.
6 PASS
3 FAIL
Total nodes: 1668473532 3296357 12523878 (-0.017% +0.0062% -0.024%)

The patch is not completely unproblematic as endgame:880 shows (T4 becomes overvalued). Probably some further tweaking is needed to fix that.

The remaining problem with seki:403 is related to penalty for ko-dependent owl move being applied although it should have been overridden by the non-ko semeai result.

[gunnar]

add semeai attack and defense codes in struct dragon_data2

[gunnar]

additional regression tests related to trevord:790

[gunnar]

The owl mistakes in trevord:790 are highlighted in the testcases that have been added in the revised patch gunnar_7_9.7b.

I tried to solve seki:403 without much success, or rather with too much side effects. The valuation of owl and semeai move reasons probably needs to be significantly rewritten. As a step towards this, gunnar_7_9.10 adds attack and defense codes for semeai moves in struct dragon_data2.

The endgame:880 failure is also left for another time.

[arend]

Just for the record, this patch also broke filllib:4.