Re: [Bug-gnubg] MWC/Equity toggling

[Gordon Bower]

More on MWC/Equity stack overflow toggling.

I tried to reproduce the problem on my home computer (running Win98
instead of Win2K.) Both computers are using build 020813.

I didn't get a crash from a stack overflow this time. Instead I got a
nonsense window:

A cube decision came up, the hint window reported MWC of ~55% to take, 57%
to drop, 54% not to double. I toggled it to equity and it correctly
reported ~0.8 to take, 1.0 to drop, ~0.7 not to double. I toggled it back,
and got this:

Cubeless 0-ply MWC: 640870726362697620000000000000000000.00%

0.000 0.000 0.000 - 1.000 0.000 0.000

1. No double 10663823472330866<whole buncha zeros>.00%
2. Double, pass, 56.17%  -1066etc.00%
3. Double, take, -0.00%, -1066etc.00%

Proper cube action: too good to double, take
(1898563729699364200..... .0%)

This occurred when gnubg doubled, I dropped, and the tutor warned me I
shouldn't drop. My best guess as to what happened: when the hint window
first popped up, it correctly reported the position when the cube was
offered to me. After I toggled it, it appears to have scored the game *as
if I have already dropped* and encountered some sort of division-by-zero
trouble when P(game)=1, P(gammon)=0, P(backgammon)=0.

Armed with this theory, I played some more, and was able to reproduce the
problem: it *only* occurs if you attempt to make a bad drop, then ask for
a hint - attempting a bad double, or attempting to take when the tutor
says dropping is right, does not cause a problem.

Hopefully this will pin down the source of the problem enough to find out
if it happens under Linux too and know where to go from there.

GRB