Re: [gnugo-devel] Reading patch

[Gunnar Farneback]

Nando wrote:
> The appended one-liner generates following breakage :
> 
> reading:177   PASS
> 13x13:31      FAIL   I believe M6 is correct.
> nando:110     PASS
> nando:111     PASS   (identical with previous test)

Nice.

> The impact on performance is negligible (< 0.01%)

I counter with another oneliner in reading.c. Here the cost is about
0.7% increase of reading nodes and the regression delta is

owl:179         FAIL 0 [3 R19]
blunder:23      PASS L2 [!F4|F5|F6]
arb:203         PASS T7 [T7]
arb:220         PASS F7 [!F6]
arend:3         PASS P6 [P6|Q5]

I haven't checked what's going on with owl:179 but the passes, in
particular blunder:23, are quite significant.

- allow conditional ko capture in attack4()

/Gunnar

Index: engine/reading.c
===================================================================
RCS file: /cvsroot/gnugo/gnugo/engine/reading.c,v
retrieving revision 1.106
diff -u -r1.106 reading.c
--- engine/reading.c    22 Feb 2003 16:02:25 -0000      1.106
+++ engine/reading.c    22 Feb 2003 20:17:53 -0000
@@ -3733,13 +3733,10 @@
         if (stackp >= branch_depth && k > 0)
           break;
         xpos = moves.pos[k];
-        /* Conditional ko capture is disabled
-        * because it seems too expensive.
-        * */
         if (komaster_trymove(xpos, other, "attack4-A", str,
                             komaster, kom_pos, &new_komaster,
                             &new_kom_pos, &ko_move,
-                            0 && stackp <= ko_depth && savecode == 0)) {
+                            stackp <= ko_depth && savecode == 0)) {
           if (!ko_move) {
            dcode = do_find_defense(str, NULL, new_komaster, new_kom_pos);
            if (dcode != WIN