Re: [gnugo-devel] GNU Go 3.5.5.

[Gunnar Farnebäck]

Arend wrote (on April 15):
> > I am a little worried about the fact that despite the
> > fact that there should have been no regression changes
> > since 3.5.5-pre-2 (yesterday), we now have an unexpected
> > fail in nicklas3:401.
> >
> > This must have been caused by either arend_5_5.14 or
> > arend_5_5.15. Is that possible?
> 
> Yes I got a FAIL in nicklas3:401 and a PASS in strategy5:286. You
> did not get the latter? (I still get it.) I didn't analyze them, but
> a priori this is not surprising.

I've tracked this down now. In retrospect our first response to this
mystery should have been to try

./regress.pike nicklas3:401 --valgrind

(or the equivalent
valgrind ../interface/gnugo -l games/nicklas/nicklas4.sgf -L 73)

The patch below solves the problem with nicklas3:401, which still
persisted in 3.5.7. I leave it to Arend to explain why the posted
patch arend_5_5.12 also contained this correction but never made it
into CVS together with the rest of the patch.

Regression delta:
nicklas3:401    PASS A5 [A5]
seki:303        PASS A2 [A2|C1|C2]
seki:305        PASS C2 [C2]
seki:905        FAIL C9 [C1]

- bugfix in do_owl_analyze_semeai()

/Gunnar

Index: engine/owl.c
===================================================================
RCS file: /cvsroot/gnugo/gnugo/engine/owl.c,v
retrieving revision 1.213
diff -u -r1.213 owl.c
--- engine/owl.c        25 May 2004 03:13:43 -0000      1.213
+++ engine/owl.c        28 May 2004 23:19:53 -0000
@@ -579,7 +579,7 @@
   if (stackp <= semeai_branch_depth && (hashflags & HASH_SEMEAI)
       && !pass && owl_phase
       && tt_get(&ttable, SEMEAI, apos, bpos, depth - stackp, NULL,
-               &value1, &value2, &xpos)) {
+               &value1, &value2, &xpos) == 2) {
     TRACE_CACHED_RESULT2(value1, value2, xpos);
     if (value1 != 0)
       *move = xpos;