Re: [gnugo-devel]

[Gunnar Farneback]

Dan wrote:
> > ... this ``pair go'' approach will certainly complicate regression 
> > testing of the patch. (http://www.pairgo.or.jp/home.
> 
> Probably the movenum test should be disabled during regressions.

What kinds of complications are you expecting? As far as I know
movenum should have stable and well determined values between
regression runs or if specific test cases are run separately.

But I can see other problems with the pair go approach. For example
it can almost certainly be exploited by an opponent who knows about
it. An alternative which does not depend on the move number would be
preferable.

/Gunnar