[gnugo-devel] dangerous antitesuji

[max-d]

consider this position

(;RU[Japanese]SZ[19]HA[0]KM[5.5]
;W[bi]
;B[bj]
;W[bm]
;B[bn]
;B[bo]
;B[bp]
;B[bq]
;B[br]
;W[bs]
;W[ci]
;W[cj]
;B[ck]
;B[cl]
;W[cm]
;B[cn]
;W[co]
;W[cp]
;W[cq]
;W[cr]
;W[df]
;B[dj]
;W[dm]
;B[dn]
;W[do]
;B[ek]
;W[el]
;W[em]
;B[en]
;W[eo]
;B[fk]
;B[fm]
;W[fn]
;W[fo]
;B[gm]
;W[aj]
;B[al]
;W[bl]
;B[bk]
)


Adding the following pattern into the Owl defensive pattern database
solves the problem ( at least GNUgo does know that  K7 defends B8 even if it
does not play here because ,i think,it does not see that saving B8 kills
B6).
I think the pattern maybe a lot more general and i discover that such "s"
property are not only needed for moves which can "trivially" been captured
as it is said in GNUgo Doc.

Pattern D1392

XXOo
O.X*
----

:8,s,value(85)

(i don't know exactly about values )


2/ Now edit the position with a black stone at F8
(add ;B[fl] at the end of the sgf ) .
K7 no longer works ,GNUgo knows this, but if we still play here it will
say black K9 is  antitesuji according to

pattern AS3

OOX       don't play potentially risky gote move
*X.
---

:8,sOX

OdA
*Bc
---

;!weak(A) && !weak(B) && oplay_attack(*,c,*)

>antisuji(*);
>backfill_helper(*,c,d);

md.