RE: [gnugo-devel] reading problem

[Portela Fernand]

Arend wrote:

> Hi Nando,

Hi

> >    * Comment: This is currently not a problem since no reduced depth
> >    * reading is performed.
> >    */
> >
> > Well, about the above comment, I would say it isn't true. There's a
> >
> >   decrease_depth_values();
>
> I don't think this should lead to corrupted cache entries. We do store
> "depth - stackp" in the cache data, and if this doesn't match, the cache
> entry is ignored.

If I'm not mistaken, this applies to the both reading caches (so, code in
cache.ch and persistent.ch), but not for the less sophisticated one
implemented in safe_move(), from which I quoted the FIXME/comment.

>       (Actually, I think we should change this policy: If it
>       doesn't match, the old result should at least be used as a hint,
>       i.e. should be the first move to try. And if the old result was
>       read out at higher depth settings, we might as well just believe
>       the result.)

Looks reasonable to me.

> To find out which pattern caused the difference you can try "-d0x1"
> (generates A LOT OF debug output), and make a diff between the two
> output files. Maybe you can then track it down further.

I'll try.

> Unfortunately, these things are a little hard to debug.

Hmm, this sounds like a sweet euphemism to me. ;)

/nando