Sean Price wrote in a message to Darin McBride:

 DM> What are you seeding?  Try using time(NULL), for an example.

 SP> I just want to seed the standard rand() function.

Be careful: srand() takes an unsigned int, while time() returns a time_t. 
These are different sizes on many compilers.  To handle this portably, do
something like this:

   #include 
   #include      // for UINT_MAX
   #include 

   srand((unsigned int)(time(NULL) % UINT_MAX));

Note that this is NOT necessarily guaranteed to give you a uniform
distribution of inputs to srand(), although it will in any practical
implementation where time() returns a 32-bit number and srand() takes an
int which is either a 16-bit or 32-bit number.

If you need to guarantee uniform distribution of inputs to srand(), the
proper solution is to apply a hash (such as the ELF hash I discussed in a
prior message to you) and a prime modulus less than INT_MAX before feeding
the seed to srand().

Note especially that the random number generators provided with most
compilers are awful.  Borland's and Watcom's are particularly bad, since
even their 32-bit OS/2 compilers only return 16 bits of significance from
rand().  ANSI C guarantees that RAND_MAX is at least 32767, and Watcom
gives you the bare minimum.  Just because rand() returns a 32-bit int does
not necessarily imply that the most significant 16 bits are ever non-zero.

If you have any kind of serious need for a random number generator, you
should use one of the widely accepted ones from the published literature. 
As I mentioned earlier, Bryan Flamig's "Practical Algorithms in
C++" has a very thorough, readable, and up to date treatment of this
subject.
 
-- Mike


---