Re: [gnugo-devel] Persistent cache strangeness.

[Daniel Bump]

> The following GTP gives different answers:
> 
> loadsgf games/trevor/trevor_26.sgf 39
> gg_genmove black
> loadsgf games/trevor/trevor_26.sgf 39
> gg_genmove black
> 
> The problem can be avoided by clearing the owl persistent cache when
> calling loadsgf, though the -t -t traces are still different, even
> after that.  This should not be the case, right?

Why do you say the problem has to do with the persistent
cache, and how did you check that clearing the cache fixes
the problem?

I think the main problem is not the persistent cache but
the scheme in evaluate_position(). If the persistent
cache is also a part of the problem then I really don't
understand what is happening. But I'd like to hear why
you think it is.

If this is a persistent cache problem, there is a true
bug lying around here.

Perhaps the problem can be fixed by having genmove clear
the parameters worms_examined and so forth. The caching
scheme would still give a speedup in owl tests such as
ld_owl but would hopefully fix this problem with 
repeated calls to genmove.

Dan