Re: [gnugo-devel] remaining patches

[bump]

> I am a bit doubtful about this patch. Wouldn't this code rather belong
> in the relevant defense function? I am also not sure whether komaster
> issues should get considered here.

For the record, here's the delta for evan_3_13.4:

./regress.sh . reading.tst 
55 unexpected PASS!
./regress.sh . ld_owl.tst 
182 unexpected FAIL: Correct '2 S1', got '1 S1'
./regress.sh . semeai.tst 
31 unexpected PASS!

It is true that the tactical reading code claims that
the attack only produces ko. But this is a characteristic
of the tactical reading code. It does not recognize X as
unconditionally dead in this position:

OOOOOO
OXXXXOO
OX.X.XO
-------

We've known this, and accepted it for a long time. There
was a thread about this long ago and we decided (as I
recall) not to try to change this behavior. It is not
very much different from the fact that the reading code
does not recognize X as dead in this position:

OOOOOOOOOO
O........O
O.XXXXXX.O
O........O
OOOOOOOOOO

Therefore I'm not sure reading:55 is a correct test with
our conventions.

In the position of reading:55, GNU Go plays the correct
move at E19 with either color to move. Moreover, the
owl code gets the answer right here.

$ gnugo -l owl03.sgf --quiet --decide-string D19
D19 can be attacked with ko (good) at E19 (81 variations)
D19 can be defended at D18 (2 variations)

$ gnugo -l owl03.sgf --quiet --decide-dragon D19
D19 can be attacked at E19 (6 variations)
D19 can be defended at E19 (2 variations)

In any case I won't use this patch for 3.3.13.

Dan