Re: [gnugo-devel] Persistent cache

[bump]

> Well, I haven't got any result so far, mainly because I got side-
> tracked by a more important issue, which I must report I think.
> 
> I was testing a variation of Dan's patch when I noticed following :
> although I had tested under debugger that nngs3:230 was solved by my
> patch, it still failed when run in the regression batch. When you look
> at the 2 patches Dan sent, a couple noticeable things are the use of
> owl escape values in the 2nd version and the nngs3:230 PASS
> disappearance. My patch was also based on owl escape values.
> 
> I think this is also related to Dan's post a couple weeks ago :
> http://mail.gnu.org/pipermail/gnugo-devel/2002-October/003157.html
> 
> Looking at the persistent cache code, it seems to me that a change in
> the escape values within the active area can indeed get unnoticed in
> some cases and consequently generate such discrepancies. IMHO, fixing
> this is a quite high priority, but I have no clue how.

I sent a comment about this to the list earlier this morning but it 
seems to have disappeared into the sometime black hole that is mail.gnu.org.

Gunnar looked at the example of strategy5:225 and decided that
the problem was that the persistent cache was not big enough.

As to what can be done about it, either:

(1) Expand the persistent cache, or

(2) Have gtp_loadsgf clear the cache.

I am not in favor of (1) because it would impact performance,
and I'm not convinced this is a very serious problem in game
situations.

Dan