🚀 go-pugleaf

Gary Cattelino (gary@st.unocal.com) wrote:
: I've been using gnuchess-4.0.pl60/xboard-2.0.pl25 as a chess
: learning tool.  I like to use the backward/forward feature of
: xboard to try out different move sequences and see how they
: play out.  When I'm done playing a sequence, I select "Machine Black"
: and continue the game.  I've noticed that when I play like this,
: gnuchess sometimes plays very badly.  It selects incorrect moves that
: allow it to be checkmated or gives away pieces.  I've saved positions
: and loaded them back in to see if gnuchess would make the same
: bad moves and it doesn't!  It plays much better from the saved
: position than it did the first time around.  What's causing this!?
: Why does the use of forward/backword cause gnuchess to select such
: bad moves?  I thought position evaluation was a deterministic thing.
: Is gnuchess selecting random moves?
: 

I've recently submitted a fix for this bug.  It isn't some interesting
defect in the algorithm (sorry, Kris Kugel).  Rather, gnuchess is
failing to initialize a flag used in detecting whether the user has
typed something to interrupt gnuchess's search.  So occasionally it
will stop searching instantly after starting, resulting in a random
move.  This seems to happen only when you use xboard and click in
commands fast.  At any rate, it will be fixed in the next patchlevel.

	--Tim