Re: [gnugo-devel] pattern-based reading spinoffs (27.7)

[Trevor Morris]

>This issue came up fairly recently in connection with the patch
>komaster_1_22.1. Gunnar proposed around this time to switch to
>one of the other schemes, which I did not want to do. I think
>Gunnar's final word on this issue is:
>
>http://mail.gnu.org/pipermail/gnugo-devel/2002-January/001123.html
Yes, I followed that discussion, which is what got me tinkering 
with komaster in the first place.


>Your scheme seems to use only 4 values, since you have an
>
>  ASSERT1(komaster >=0 && komaster <=3, 0);
>
>It seems to me that any scheme that uses only 2 bits is 
>going to suffer from incomplete information. We'd consider
>switching to another scheme but we wouldn't want to do it lightly.
>I think we'd have better luck if we could get one more bit for
>komaster.

Yes, I have no doubt that my scheme suffers!  However, it does
gain a state over the other schemes because it doesn't use
a PASS state.  PASS is implied if kom_pos == NO_MOVE.

Of course, another full bit couldn't hurt.

-Trevor