Re: [gnugo-devel] experimental_reading_1_33.1

[Arend Bayer]

On Fri, 19 Apr 2002, Daniel Bump wrote:

> Arend at one point commented that the tree based reading was
> very similar to the DFA, the principal difference being that
> symmetric patterns are built into the tree, and Tanguy and
> Trevor seemed to agree with this remark. So (it was suggested)

I think should correct that a little. The tree based pattern matching
does add a new idea: In run-time, (if I understand correctly) loads of
different paths are walked, and beforehand each pattern got inserted in the
path where it would most likely get eliminated as quickly as possible
(This may well be the essential point of the tree stuff.).
Furthermore, each pattern got inserted into this tree of possible paths
in each orientation.

> possibly the tree and dfa could be unified by giving the dfa
> an option not to condense symmetric patterns. This would make
> the dfa bigger but faster.

It might be that a DFA with all orientations of patterns built in might
make a better space/speed trade off than the tree-based pattern matching.
This can only be tried out.

Arend