Re: [gnugo-devel] crash

[Gunnar Farneback]

Evan wrote:
> I just found a crash in current cvs.  The following GTP stream reproduces
> it:
> 
> boardsize 9
> play black G6
> play white E6
> play black E7
> play white F6
> play black F7
> play white G5
> genmove black

This is extremely bad. The crash happens when reading the constraint
for the small avalanche joseki, pattern JK68 in komoku.db. It turns
out that the computed vertices used in the constraint come out all
wrong. It looks like the constraint callback might be called with the
wrong transformation number from do_corner_matchpat(), or at least
with a transformation number which isn't consistent with the pattern
that has been matched.

The appended patch (not for cvs) to komoku.sgf adds a debug output
which shows how the constraint vertices have been mapped, which should
simplify debugging.

It may also be useful to have this alternative GTP script which
doesn't crash but shows (with the debug output) that the problem
exists for all board sizes.

boardsize 19
play black C4
play white E4
play black E3
play white D4
play black D3
play white C5
showboard
genmove black

/Gunnar

Index: patterns/komoku.sgf
===================================================================
RCS file: /cvsroot/gnugo/gnugo/patterns/komoku.sgf,v
retrieving revision 1.16
diff -u -r1.16 komoku.sgf
--- patterns/komoku.sgf 21 Jul 2003 16:33:19 -0000      1.16
+++ patterns/komoku.sgf 17 Nov 2003 01:03:55 -0000
@@ -100,7 +100,7 @@
 
 (;B[nd]MA[ki]
 
LB[qc:A][rd:B][qb:C][re:D][qf:E][rf:F][nc:G][oe:H][pe:I][qg:J][pf:K][of:L][pg:M]
-C[; xplay_attack(A,B,C,D,E,F,G,H,I,J,K,L,M,M)
+C[; gprintf("%d %1m %1m %1m %1m %1m %1m %1m %1m %1m %1m %1m %1m 
%1m\\n",trans,A,B,C,D,E,F,G,H,I,J,K,L,M), 
xplay_attack(A,B,C,D,E,F,G,H,I,J,K,L,M,M)
 ];W[qc]MA[ki]C[U];B[rd]
 MA[ki];W[qb]MA[ki];B[re]MA[ki];W[pf]MA[ki]C[U]
 (;B[qf]MA[ki]C[U];W[nc]MA[ki]C[U];B[pe]MA[ki];W[oe]MA[ki];B[of]MA[ki];