[gnugo-devel] semeai patch

[Gunnar Farnebäck]

This short patch makes the semeai reading more reluctant to give up a
semeai ko if there still seems to be any potential for two eyes.
Breakage:

blunder:33      FAIL J1 [!H6|J1]
semeai:110      PASS 1 0 F3 [1 0 (PASS|F3|E1)]
semeai:111      PASS 1 0 F3 [1 0 (PASS|F3|E1)]
semeai:112      PASS 1 0 PASS [1 0 PASS]
semeai:113      PASS 1 0 PASS [1 0 PASS]
seki:112        FAIL C9 [B1]
4 PASS
2 FAIL
Total nodes: 1660993896 3061571 13212500 (+0.0011% +0.04% -0.0062%)
 
seki:112 is accidental. Now the semeai in the lower left is
correctly read but the move valuation is off, causing the
fail.

semeai:110-113 are all major improvements.

blunder:33 looks mostly accidental. Before the patch
owl_analyze_semeai_after_move white J1 K6 G7 (with
recompute_owl) says "1 1 H6" and after the patch it says "0 0
PASS". Correct is "1 0 H6".

- do_owl_analyze_semeai() revised

/Gunnar

Index: engine/owl.c
===================================================================
RCS file: /cvsroot/gnugo/gnugo/engine/owl.c,v
retrieving revision 1.222
diff -u -r1.222 owl.c
--- engine/owl.c        11 Sep 2004 01:01:03 -0000      1.222
+++ engine/owl.c        22 Sep 2004 15:39:55 -0000
@@ -1135,7 +1153,8 @@
       include_semeai_worms_in_eyespace = 1;
       if (!owl_estimate_life(owla, owlb, vital_defensive_moves,
                             &live_reasona, 0, &dummy_eyes,
-                            &eyemin_a, &eyemax_a)) {
+                            &eyemin_a, &eyemax_a)
+         && eyemax_a < 2) {
        include_semeai_worms_in_eyespace = 0;
        *resulta = 0;
        *resultb = 0;