Re: [gnugo-devel] 2 arraybound violations in 3.1.17

[Gunnar Farneback]

Teun wrote:
> Here are 2 more arraybound violations.
> The one in owl.c (no fix, assertion added) is triggered by owl test 149.

Broken pattern, fixed by the patch below. This kind of problem could
be avoided by better consistency checking in mkpat. I'll add this to
TODO in a later patch.

> The one in readconnect.c (with fix, please check) was triggered
> by connect.tst test 3.

The fix looks good.

> There are still more arraybound violations, more info will follow.

We're very interested!

> The rate at which arraybound violations are introduced/exposed in the program
> makes one wonder about the number of other less easy to detect bugs that is
> there ...

On the contrary I'd say that the arraybound violations are quite hard
to detect, unless you have special tools or happen to be lucky enough
to get a crash anyway. My feeling is that most other bugs either tend
to be caught by the compiler or turn up consistently in the
regressions.

With respect to playing strength I'm a lot more worried about
shortcomings of the algorithms and inappropriate patterns than bugs in
the code. With respect to stability the arraybound violations are at
least by far the dominant source of NNGS crashes.

/Gunnar

Index: patterns/owl_defendpats.db
===================================================================
RCS file: /cvsroot/gnugo/gnugo/patterns/owl_defendpats.db,v
retrieving revision 1.41
diff -u -r1.41 owl_defendpats.db
--- patterns/owl_defendpats.db  15 Dec 2001 14:35:27 -0000      1.41
+++ patterns/owl_defendpats.db  17 Dec 2001 16:47:03 -0000
@@ -4297,12 +4297,13 @@
 
 Pattern D1137
 # gf New pattern. (3.1.14)
+# gf Bugfix. (3.1.18)
 # See buzco:6
 
-?oo
-?..
-X*.
-O.?
+?oo?
+?..o
+X*.o
+O.??
 
 :8,B,value(65)