[gnugo-devel] arend_3_7.1: bugfix in adjusted_worm_attack_value()

[Arend Bayer]

The bug about negative attack_either move values was a simple typo in
adjusted_worm_attack_value(), the relevant hunk of the patch starts at
line 1048.

Apart from that I have moved the " > 0.0" check (negative adjusted worm
attack values could still happen) inside adjusted_worm_attack_value(),
removed two lines of dead code and added a FIXME.

I am running regressions now.

Arend

 - bug fix in adjusted_worm_attack_value()
 - FIXME added regarding several EITHER_MOVE reasons

Index: engine/value_moves.c
===================================================================
RCS file: /cvsroot/gnugo/gnugo/engine/value_moves.c,v
retrieving revision 1.47
diff -u -r1.47 value_moves.c
--- engine/value_moves.c        11 Aug 2002 12:13:02 -0000      1.47
+++ engine/value_moves.c        14 Aug 2002 10:57:52 -0000
@@ -1026,7 +1026,6 @@
 static float
 adjusted_worm_attack_value(int pos, int ww)
 {
-  int color;
   int num_adj;
   int adjs[MAXCHAIN];
   int has_live_neighbor = 0;
@@ -1035,7 +1034,6 @@
   float adjustment_down = 0.0;
   int s;

-  color = OTHER_COLOR(board[ww]);
   num_adj = chainlinks(ww, adjs);
   for (s = 0; s < num_adj; s++) {
     int adj = adjs[s];
@@ -1048,7 +1046,7 @@
       adjustment_up = 2*dragon[adj].effective_size;

     if (worm[adj].attack_codes[0] != 0
-       && !does_defend(pos, ww)
+       && !does_defend(pos, adj)
        && 2*worm[adj].effective_size > adjustment_down)
       adjustment_down = 2*worm[adj].effective_size;
   }
@@ -1057,7 +1055,14 @@
     adjusted_value += adjustment_up;
   adjusted_value -= adjustment_down;

-  return adjusted_value;
+  /* It can happen that the adjustment down was larger than the effective
+   * size we started with. In this case we simply return 0.0. (This means
+   * we ignore the respective EITHER_MOVE reason.)
+   */
+  if (adjusted_value > 0.0)
+    return adjusted_value;
+  else
+    return 0.0;
 }


@@ -1909,6 +1914,9 @@

       case EITHER_MOVE:
        /* FIXME: Generalize this to more types of threats. */
+       /* FIXME: We need a policy if a move has several EITHER_MOVE
+        * reasons. Most likely not all of them can be achieved.
+        */
        worm1 = either_data[move_reasons[r].what].what1;
        worm2 = either_data[move_reasons[r].what].what2;
        aa = worms[worm1];
@@ -1932,13 +1940,7 @@
        TRACE("  %1m: %f - either attacks %1m (%f) or attacks %1m (%f)\n",
              pos, this_value, aa, aa_value, bb, bb_value);

-       /* FIXME: The restriction to this_value > 0
-        * should not be necessary. See century-2002:30 and nngs2:160.
-        * Debugging this example might show a deeper problem
-        * with EITHER_MOVE reasons.
-        */
-       if (this_value > 0)
-         tot_value += this_value;
+       tot_value += this_value;
        break;

       case ALL_MOVE: