Re: [gnugo-devel] Patch: new cache part 2

[Arend Bayer]

Gunnar wrote:
> Arend wrote:
> > > +  /* Return data.  Only set the result if remaining depth in the table
> > > +   * is big enough to be trusted.  The move can always be used for move
> > > +   * ordering if nothing else.
> > > +   */
> > > +  if ((unsigned) remaining_depth <= hn_get_remaining_depth(*node)) {
> > > +    if (result)
> > > +      *result = hn_get_result1(*node);
> >
> > That's a subtle change in behaviour that also needs discussion
> > (the old one only set the result if "==" in the above equation). I am not
> > against it, but one should be aware that it makes debugging more
> > difficult.
>
> Oh, this is dangerous. I would strongly suggest to continue
> considering the remaining depth as input data and throw it into the
> Zobrist hash. I just tested to change "<=" to "==" above and it does
> at least solve the failure of optics:1201.

I agree this needs dicussion, but I disagree with putting the remaining
depth into the Zobrist hash. If we want to do iterative deepening (and
I am pretty convinced that we should do it for owl), then we need to
be able to find identical positions with lower remaining depth.

Arend