Re: [gnugo-devel] cosmic patch

[Heikki Levanto]

On Sun, Aug 10, 2003 at 10:59:31AM -0700, address@hidden wrote:

> I don't see how this is relevant. Of course changing the random seed
> can give different answers and this is as it should be.
> 
> But the function gtp_reg_genmove explicitly sets the random seed to 0.
> So both regression runs were done with the same random seed 0.

I am somewhat disturbed by this. I think it would pay off to verify if
this really is a hash collision issue, or something else. It could be
some distant buffer overflow, or other nastyness...

If we verify that it is a hash collision problem, then we can decide to
live with it, or to do something about it. Since we have decided that
our regressions are deterministic, we should probably fix the varying
hash values anyway.

-Heikki

-- 
Heikki Levanto  LSD - Levanto Software Development   <address@hidden>