RE: [gnugo-devel] RE: Persistent cache

[Portela Fernand]

> I'm sure this can and does happen. It's possible that it's a
> serious problem but in the absence of evidence to the contrary
> I would tend to assume that it doesn't happen often enough to
> justify expanding the active area.

In any case I won't be proposing an expansion and I never thought about it.
I was rather thinking of a selective cache entry invalidation based on
criterions which keep to be defined. And it doesn't look like I'll be able
to propose something before long, I still have much to learn about the
engine.

Anyway, before closing this topic on my side, I'll add the couple lines
which lead me to think that the problem is at least serious.

I made Gnu Go replay a random game twice, once with persistent cache,
another time without (completely disabled). Here is the result (the game
has 267 moves)

--- with_cache.txt      2002-11-05 22:03:06.000000000 +0100
+++ without_cache.txt   2002-11-05 22:03:42.000000000 +0100

-Move 69 (black): GNU Go plays D11 (25.56) - Game move F18 (13.15)
+Move 69 (black): GNU Go plays C10 (28.88) - Game move F18 (14.08)

-Move 75 (black): GNU Go plays D11 (24.58) - Game move N10
+Move 75 (black): GNU Go plays P13 (24.67) - Game move N10

-Move 79 (black): GNU Go plays D11 (24.26) - Game move L17 (2.31)
+Move 79 (black): GNU Go plays P13 (27.90) - Game move L17 (2.33)

-Move 81 (black): GNU Go plays D11 (24.22) - Game move H15 (1.47)
+Move 81 (black): GNU Go plays P13 (34.75) - Game move H15 (1.47)

-Move 113 (black): GNU Go plays E19 (20.02) - Game move F13 (14.52)
+Move 113 (black): GNU Go plays H8 (20.08) - Game move F13 (15.40)

-Move 117 (black): GNU Go plays D15 (21.87) - Game move D13 (2.69)
+Move 117 (black): GNU Go plays D15 (24.26) - Game move D13 (2.69)

-Move 137 (black): GNU Go plays L18 (34.18) - Game move L18 (34.18)
+Move 137 (black): GNU Go plays L18 (41.21) - Game move L18 (41.21)

-Move 159 (black): GNU Go plays O11 (13.16) - Game move S13 (2.45)
+Move 159 (black): GNU Go plays O11 (13.16) - Game move S13 (0.23)

-Move 191 (black): GNU Go plays F8 (26.20) - Game move J18
+Move 191 (black): GNU Go plays G6 (20.11) - Game move J18

The differences in the 31 first moves were omitted. Total 9 noticeable
differences, leading to 6 (actually 4) diverging decisions. 

Opinions might differ here, and I'm sorry to insist, but I really don't
like that.

/nando