RE: [gnugo-devel] a question about conn.db

[Portela Fernand]

Gunnar wrote:

> It avoids some problems with the owl code and the move valuation. At
> least it did when they were written.

After some testing, I now understand the rationale behind these patterns.
Since the purpose is not so obvious, I would suggest to document it a bit
more.

> Try to take them out and see what happens now. Ideally they should not be
> needed.

Ideally, yes. In practice, they prove useful. It's just that they are
causing some problems in specific cases, like fragile double connections.
One of the test cases I submitted recently is a combination of one such
kosumi with a monkey jump. That's the reason why I got interested in these
patterns.

Removing them seems to cause more problems than keeping them. I think it is
ok, but we probably need to add some other patterns in conn.db with
connection inhibitors in order to prevent (hopefully rare) wrong and/or
problematic amalgamations.

/nando