Re: [gnugo-devel] Bug Game for gnugo 3.2 (with output)

[Arend Bayer]

On Mon, 10 Jun 2002, Gunnar Farneback wrote:

> Josh Stern wrote:
> > Output was:
> >
> > ***assertion failure:
> > value_moves.c:2550 - 0 near PASS***
>
> Has someone managed to reproduce either of Josh's two crashes? Just
> loading the position and generating a move doesn't suffice, at least
> not for me.
>
> The assertion above is slightly bizarre, by the way. It happens in
> this code:
>
>         default:
>           /* This means probably someone has introduced a new threat type
>            * without adding the corresponding case above.
>            */
>           gg_assert(0);
>           break;
>
> I'm pretty sure that this shouldn't be possible and if it were we
> should have caught it much erlier.

I couldn't reproduce it either. However, the patch below is obviously
necessary; it's indeed surprising but not too remote that we didn't
catch this earlier.

Arend

arend_3_4.2:

 - fix ALL_MOVE move reason to even value


Index: engine/move_reasons.h
===================================================================
RCS file: /cvsroot/gnugo/gnugo/engine/move_reasons.h,v
retrieving revision 1.19
diff -u -r1.19 move_reasons.h
--- engine/move_reasons.h       3 May 2002 17:57:58 -0000       1.19
+++ engine/move_reasons.h       10 Jun 2002 22:57:28 -0000
@@ -22,6 +22,7 @@
 /* values for move_reason.type */
 #define THREAT_BIT    1

+/* Only use even values for non-threat move reasons! */
 #define ATTACK_MOVE              2
 #define ATTACK_MOVE_GOOD_KO      4
 #define ATTACK_MOVE_BAD_KO       6
@@ -62,7 +63,7 @@
 #define ANTISUJI_MOVE           70

 #define EITHER_MOVE             100
-#define ALL_MOVE                101
+#define ALL_MOVE                102


 /* Bitmap values for move_reason.status */