Re: [gnugo-devel] Patch: new cache implementation

[Heikki Levanto]

On Sat, Jul 05, 2003 at 01:38:43PM -0400, Inge Wallin wrote:
> Here is the first implementation of the new hash and cache scheme 

Interesting.

The regression differences worry me a bit.  You could add a verify-mode
that always returns "no result", when asked for a cached result, and on
storing checks the new result against the stored one (if any). If ever
attack(A1) on the same position gets different values, we know something
is wrong. Possibly a hash collision (how ever unlikely), or a differing
result due to hitting a node limit.  These should be easy to verify.

-Heikki

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