Re: [gnugo-devel] Extremely bad mistake

[Gunnar Farneback]

SP Lee wrote:
> Yes, I'm still learning the basics of the code, but I'm sure the
> suicide reading is wrong in this case. I have made the variable
> verbose=3.

Ah, that might explain things. Could it be that you have been fooled
by the very incomplete RTRACE messages? I was thinking those were
merely useless but maybe they should be considered directly harmful.

Would anybody mind if I just remove them? (Historically these may have
been useful in the early development of the tactical reading but have
been superseded by more advanced techniques and have not been
maintained.)

> What I have traced is, from the moment gnugo was "considering attack
> of C19", until gnugo checked A16 in is_suicide in board.c, where I
> found board[41]=0 (A18 is empty). Of course, board[41] was 1 in the
> early analysis stage, but I don't know yet when it was changed to 0.
> Maybe somewhere it's not propagated through to the defense checking.
> The tracing message showed " A16 would be suicide".

No, this doesn't sound right. It would be too fundamental to have
passed unnoticed. I would rather suspect that the traces are
misleading.

/Gunnar