from what I've seen, it seems that Jon's
workaround for the #Q0 issue, garbage in rollout output
(add the code "((cacheNode*)e)->ar[5]
= ((cacheNode*)e)->ar[6] = 0;" at the beginning of the
CacheAdd() function in cache.c) has
broken things even more.
Jon said the extra line solved the #Q0
issue, but it has introduced the new one (wrong values in
the MWC and equities at different plies):
one the extra line of code removed, I no longer see
strange MWC.
This explains why the "new"
bug was only in my builds (I was the only one with the extra line of
code).
I have rebuilt everything and put the
new distrib (GTK has been updated too, so reinstall everything)
on gnubg.org. Most likely, we will face
again the #Q0 bug in rollout output.