[gnugo-devel] Test results, owl problem

[Daniel Bump]

I updated the test results, even though I'm not doing a release
tonight. The current version can be checked out with the CVS
tag rel-3-1-28-pre-3. Not much had changed in the test results
except for the new nngs1 tests. For details, see regression/BREAKAGE.

I remain of the opinion that we shouldn't do owl tuning until
after 3.2 but this weekend I got sucked into the owl code. I'll
mention a situation which is worth comment. See owl:257.

In the new file
regression/games/owl35.sgf an owl mistake is made when we run
decide-dragon on J3.

 7 . . . . . . . . . . . . . . . . . . . 7
 6 . . . . . . . . . . . . . . . . . . . 6
 5 . . . . . . . . O . . . . . . X . . . 5
 4 . . O + O . . . O X X . . . . + . . . 4
 3 . . . . . . . . X O . X . . . X . . . 3
 2 . . . . . . . . . . . . . . . . . . . 2
 1 . . . . . . . . . . . . . . . . . . . 1
   A B C D E F G H J K L M N O P Q R S T

GNU thinks that H3 captures J3. After

W:H3 B:J2 W:K2 B:H2 W:G2 B:G3 W:H4 the position
looks like this:

 7 . . . . . . . . . . . . . . . . . . . 7
 6 . . . . . . . . . . . . . . . . . . . 6
 5 . . . . . . . . O . . . . . . X . . . 5
 4 . . O + O . . O O X X . . . . + . . . 4
 3 . . . . . . X O X O . X . . . X . . . 3
 2 . . . . . . O X X O . . . . . . . . . 2
 1 . . . . . . . . . . . . . . . . . . . 1
   A B C D E F G H J K L M N O P Q R S T

Now X gives up because of this code, around line 2003 of owl.c:

     }
     if (probable_max < 2 && stackp > 2)
     move_cutoff = 99; /* Effectively disable vital moves. */
   }

This means that the lunch at G2 is not eaten. Taking
out this code improves things, but a further problem arises.
Instead of H4, W tries J1 and again B gives up. The pattern
D1336 would be matched, but the owl escape value at G4 is 0.
I tried a few more things to fix this problem but ended
up breaking more tests than I fixed.

Dan