Re: [gnugo-devel] Backspace crashes

[Arend Bayer]

I've tried to analyze these backspace crashes a bit, but didn't get very
far...

> Dan wrote:
> > I just looked at two examples of this:
> > 
> > http://mail.gnu.org/pipermail/gnugo-devel/2001-November/000422.html
> > http://mail.gnu.org/pipermail/gnugo-devel/2001-October/000364.html
> 
> More examples:
> 
> http://mail.gnu.org/pipermail/gnugo-devel/2001-October/000051.html
> http://mail.gnu.org/pipermail/gnugo-devel/2001-November/000559.html

>From the sgf-files and the ascii board dumps we got, I'd conclude that
the following happened:

GNU Go passed at some point in the game. Later, a move was taken back.
This toggles GNU Go to replay the sgf-tree. Here, it could not parse
its own pass, and that is the point where the assertion failure was toggled.

(In all the sgf-files, the node with GNU Go to move is later in the game
than the position shown by the ascii board printed by the assertion
failure.)

However, this didn't help me in identifying or reproducing the bug.

Does anyone esle have a guess why this happened? Btw, any reason not
to include the random seed in the bug output as below?

Arend


Index: printutils.c
===================================================================
RCS file: /cvsroot/gnugo/gnugo/engine/printutils.c,v
retrieving revision 1.11
diff -u -r1.11 printutils.c
--- printutils.c        19 Dec 2001 07:51:18 -0000      1.11
+++ printutils.c        17 Jan 2002 19:20:15 -0000
@@ -274,9 +274,9 @@
 #endif
 
   fprintf(stderr, "\n\
-gnugo %s: You stepped on a bug.\n\
+gnugo %s (Seed %f): You stepped on a bug.\n\
 Please save this game as an sgf file \
-and mail it to address@hidden", gg_version());
+and mail it to address@hidden", gg_version(), seed);
 
   abort();  /* cause core dump */
 }