[gnugo-devel] RE: Symmetry corrections

[Portela Fernand]

Gunnar wrote:

> > > this function a little could at least reduce the noise.
> > 
> > Something like this (untested) patch?
>
> It's frightening how large regression delta this patch causes,
> presumably all of it accidental. In total 12 passes and 16 failures.

Yes, and unfortunately it doesn't completely prevent accidental changes. I
found a couple symmetry errors in the databases and tried to correct one of
them after applying the patch: still 6 (3+3) unexpected passes and fails.

I'm trying to enhance the patch by also sorting the move itself if both
pattern values and addresses match. I'll post the results once my box is
done with running the tests.

/nando