Mail Archives: djgpp/1998/03/26/13:28:37
In article <6fe13v$55q AT tuegate DOT tue DOT nl> you wrote:
> In article <35169C99 DOT 3B53 AT pobox DOT oleane DOT com>,
> Francois Charton <deef AT pobox DOT oleane DOT com> writes:
> > generating random looking maps). If you want a "true" (non replicable)
> > random series, you may want to use srand(time()) as a seed.
> As some people already have pointed out, this only works when your
> program runs at least for a single second.
I find it hard to imagine a program that runs through in less than one
second, but still has a real need for guaranteed different random
seeds each run :-) Such a program might be better transformed into a
centralized, long-running server process instead, using a single
normal pseudo-random number sequence.
> Does somebody know how to take care of getting a different value each
> time you run a program, no matter how often in a single second you start
> it ?
That target is absolutely impossible to reach, I think. Reason: time()
counts seconds (in DJGPP), and is 32 bits wide. The srandom argument
is also a 32 bit number. If you want to start your program more than
once per second, and insist on absolutely unique seeds per run, you'll
run out of numbers before time() reaches its limit of usability (some
time in the year 2037...). I.e. you can start that program more often
than there are different, unique seeds.
In the high energy physics experiment I'm in, we maintain an
experiment-wide seed value for that very reason: each time a
statistical simulation job is started, it requests a 'secure' seed
from the central database, and the central value is increased
automatically. All this just to avoid that two people start the same
simulation job, using time() as the seed, in exactly the same second,
and get the exact same results, by this coincidence.
--
Hans-Bernhard Broeker (broeker AT physik DOT rwth-aachen DOT de)
Even if all the snow were burnt, ashes would remain.
- Raw text -