Crash in semeai.c at --level 0

[arend]

gnugo -l crash.sgf --level 0 produces the following crash:

***assertion failure:
semeai.c:215 - board[semeai_defense_target] == (WHITE+BLACK-(board[dragon2[d1].origin])) near PASS***

 (variation 0)
   A B C D E F G H J K L M N O P Q R S T
19 . . . . . . . . . . . . . X X X X X X 19
18 . O . O X . . O . . O . O O O X X O X 18
17 . . . O X X X O . O . O X O X X O O X 17
16 . . O X X . . X O + O X X O . X O X X 16
15 . O O . X X X . X O X O X X X X O O O 15
14 . X X . . . . . X . . O X X . X O . . 14
13 . . X O . X . X O O O . O O X O O . . 13
12 . . . X X X . X X O . O O X X X X O O 12
11 . X O O O X O X . X O . . O X X O O X 11     WHITE (O) has captured 4 stones
10 . O . X O O O O X X X O O O O X . O X 10     BLACK (X) has captured 10 stones
 9 O . . . . X X X . . X O X O O X . O X 9
 8 O X X X X . X O X X . X X X O X X X X 8
 7 . O O O O X . O X . X O X . X O O X . 7
 6 . . O X O X . . O . X O . X X X O X O 6
 5 . O X X O X . . . . . O . X X O O O . 5
 4 . O . X X O O . . + . O X . X X O X . 4
 3 . . O . X O . . O . . O . X . X O O . 3
 2 . . . . . . . . . . O X X . O O . . . 2
 1 . . . . . . . . . . . O . . . . . . . 1
   A B C D E F G H J K L M N O P Q R S T

[arend]

I am a bit at a loss here. It seems this bug should happen whenever a dragon is involved in one semeai, which he loses regardless who moves first, but with uncertain semeai result. This should have hit us a lot more often.

[gunnar]

The reason why this haven't hit us more in the past is that the relevant assertion wasn't introduced until 3.7.8 (r2274). I'm less sure why we aren't seeing more of them now, although there have been a number of bug reports for these crashes.

As for the patch I'd prefer setting both best_defense and best_attack to -1 and keeping the assertion which is removed in the second hunk.

[arend]

I don't see the point of the 2nd assertion, since it's trivially satisfied due to the line above (in fact it will be optimized out by the compiler).

Whether setting best_defense = -1 or best_defense = 0 and defense_certain = -1 seems a cosmetic question to me.

[arend]

Sorry I misread the second assertion of course. Will post a revised patch.

[gunnar]

Yes, setting best_defense=-1 instead of defense_certain=-1 is cosmetic, but has a small advantage in being less subtle. No big deal though.

-      gg_assert(semeai_results_second[d1][d2] != -1);
+      gg_assert(semeai_results_second[d1][d2] != -1)

Maybe you should add the semicolon back in too. :-)

[arend]

semicolons...