Re: [gnugo-devel] Regression test suites

[Darren Cook]

> As for the super KO task on the tactical list, would using
> (hashes/message digests) to encode the board position of a given move be
> adequate.
> ...
> The reason to use a sound, cryptographic hash algorithm is that the:

You really need to compare the pros and cons to zobrist hashing; it has
the same three advantages you've given, but is very quick to update as
you move up and down the search tree.

Recently on the compgo mailing list I suggested keeping two zobrist
hashes, one of 96 bits, one of 32 bits. If two 96 bits hashes match, you
then assert the 32 bit hash matches. If the assert ever fails then your
96 bits was not big enough. Don Dailey (I think) pointed out that simply
using a 128 bit hash, then a bit mask for the checking portion is
probably better.

Darren