Regression fixes by draqo

[arend]

This extracts the regression fixes out of all-in-one.zip in #148. (It supersedes reg_tests.diff there.)

[arend]

Supersedes reg_tests.diff in #148

[gunnar]

 # G13 is still sufficient to get a seki.
+#
+# After N6 the game could be:
+# N7 K12 G13 J13 N5 M3 L13 N4 H13 N6
+# and now white has to win extremely unforable 3 stage ko,
+# so this move could be acceptable - G13 and K12 leads to
+# the similar position - so the best move is N5, which
+# doesn't allow white making a ko.
 loadsgf games/kgs/tournament11/CrazyStone-GNU.sgf 87
 43 reg_genmove black
-#? [G13]
+#? [G13|K12|N6|N5]

I suspect this test is at the wrong move. Even with a tenuki white can't do better with the semeai than a bad ko which doesn't win the game. G13, J13, K12, N5, M3, N3, M2, K3, J5 all give seki without ko whereas N6 leaves the small ko aji mentioned above. At move 89 tenuki leaves white with a sufficiently good ko to win the game, N5, M3, N3, M2, K3, J5 give a clean seki and win for black, whereas J13, K12, and N6 suffice to win but leave a small ko aji.

[gunnar]

 loadsgf games/incident161.sgf 42
 28 owl_attack B8
 #? [1 C7]
+# after E9 C7 F9 there is a seki
 29 owl_defend B8
-#? [1 C7]
+#? [1 (C7|E9)]
 30 owl_attack D8
 #? [1 C7]
+# same as 29
 31 owl_defend D8
-#? [1 C7]
+#? [1 (C7|E9)]

The owl code is unlikely to understand the seki but fair enough, E9 works.

[gunnar]

--- regression/strategy2.tst	(revision 2379)
+++ regression/strategy2.tst	(working copy)
@@ -20,7 +20,7 @@
 
 loadsgf games/strategy14.sgf 125
 55 reg_genmove white
-#? [C12]*
+#? [C12|B13]

I don't like the shape of B13 but it looks like it works.

 # This is essentially a semeai problem
 loadsgf games/strategy22.sgf
 80 reg_genmove white
-#? [P4|Q4|Q3|S8]
+#? [P4|Q4|Q3|S8|R6]

R6 is fine.

[gunnar]

--- regression/arb.tst	(revision 2379)
+++ regression/arb.tst	(working copy)
@@ -66,7 +66,7 @@
 # Defend against a combination attack.
 loadsgf games/arb/game03.sgf 109
 232 reg_genmove black
-#? [M9|M10|L10|L11]
+#? [M9|M10|M11|L10|L11]

M11 is kind of ugly but it does work.

[gunnar]

--- regression/nngs4.tst	(revision 2379)
+++ regression/nngs4.tst	(working copy)
@@ -249,7 +249,7 @@
 #C13 looks correct here /evand
 loadsgf games/nngs/gnugo-3.3.10-niki-200210281349.sgf 144
 610 reg_genmove white
-#? [C13|D13]
+#? [C13|D13|B14]

I don't like B14. Looks like bad shape and leaves white a reverse sente move at D13 with good followup.