[gnugo-devel] Semeai again

[Daniel Bump]

The version in CVS can now run the regressions compiled after
'configure --enable-experimental-semeai' without crashing.

The file regression/BREAKAGE shows the results (before Inge's
patch) with and without the experimental semeai. The two
unexpected passes without the experimental semeai should
be attributed to the change in pessimistic_compute_eyes()
which was discussed earlier. This is the patch compute_eyes_1_16.2.

With the experimental semeai there are 16 unexpected passes and
15 unexpected losses. These positions should be remembered as
cases where the semeai code makes a difference and which should
therefore be added to semeai.tst as semeai tests (in the format
of that file, not duplicating the genmove format of the existing
tests).

If anyone wants to help maintain regression/semeai.tst that
would be something someone could do.

The function that will be the focus of semeai development
is owl_analyze_semeai() in owl.c. But before working on
that, the file semeai.c needs to be rewritten. The old
analyze_semeai() shouldn't be taken down but may be
called when level < 8 or something like that. But the
file needs to be rewritten so that analyze_semeai()
and experimental_analyze_semeai() have the same format.
The code at the end of analyze_semeai() that generates 
move reasons should be moved to the loop in semeai(). After
these changes are made, we can start seriously hacking
on owl_analyze_semeai() and try to make it good.

Dan