[gnugo-devel] Results from the value removal patch

[Inge Wallin]

Hi,

Just to give you a datapoint:

I tried the patch where pattern values >= 10 were removed
(inge_3_11.3) but with --experimental-connections turned on.  I did
'make test' on an unmodified 3.3.10 and a 3.3.10 with the patch.

The result was that there was one extra PASS and no extra FAILs for
the patched version, i.e. an improvement.  One strange thing was that
on the test 13x13:67 there was a FAIL that was created just by turning
on --experimental-connections.  With the patch, gnugo wanted to make
another move that was not B9, i.e. another FAIL on the same test.
However, when I ran the test by itself, by using "gnugo --mode gtp"
and just copying the GTP text of the test into the program, then gnugo
made the correct move!

This could be a cache problem or something else, I haven't
investigated it.  At this point I just wanted to inform you of the
fact. 

      -Inge