Re: [gnugo-devel] Tests inside SGF files

[Gunnar Farneback]

Trevor wrote:
> I find it very slow going to flip between my SGF viewer and the
> *.tst file.  Annotating the SGF file directly seems to make
> a great deal of sense.

I don't have much of a problem with this, but I guess it depends on
your editing environment and the relative time needed to analyze each
position.

> I'm considering enhancing regress.pl to take a marked up SGF test 
> file, and generate the results in much the same way that the *.tst
> files work now.  I'm not quite sure the best way to uniquely
> identify the problems, however - something that's very simple with
> the *.tst files.
> 
> Thoughts?

This may be easier to construct but it's no fun to maintain.
Evaluating patches to sgf files is almost hopeless. There's also the
identification problem you mention.

My suggestion is that you write a script which converts a game record
annotated with test cases to the *.tst format. That may give you a
comfortable way to produce test cases.

Dave wrote:
> Also, ISTR that some of the early autotest stuff was in the form
> of annotated sgf files. It was implemented as a --mode I think.
> But I think all that must have been removed now, in favour of the
> current scheme.

Yes, it was never really useful and was removed some time prior to
3.0. The only remaining trace is the --replay option, which still can
be used to test whether GNU Go plays the same or different moves
compared to a game record. Download 2.6 if you want to see the old
code and test files.

/Gunnar