Re: [Bug-gnubg] FEATURE REQUEST: Modifications to epc display

[Joern Thyssen]

On Thu, Apr 22, 2004 at 11:35:46AM +0200, Øystein Johansen wrote
> >===== Original Message From Joern Thyssen <address@hidden> =====
> >The race net doesn't know anything about epcs; it's actually a one-sided
> >rollout. However, I've removed that since calling the OSR code would
> >reset the seed for the MersenneTwister RNG.
> 
> BTW: Can we have a stack of random seeds?
> 
> Each time we want a OSR we push a seed into the stack, and when the OSR is 
> done we pop one off.
> 
> Is this possible? Is this smart?

I wouldn't use stack. Instead, everywhere we use a RNG we keep a record
of private information for the RNG. For example, currently the M-T RNG
uses a global array long mt[624]. We could put this array (and a few
other variables used for M-T) into a struct which we pass on every call
to genrand. This way it's possible to use several instances of the M-T
RNG. The same is possible BBS, ISAAC, and MD5. For ANSI and BSD we've no
access to the interval variables for the RNG unless we pull out the code
from libc, modify it and put it into gnubg. For the manual dice and
random.org, dice from file there are no issues. 

Joern