While researching a random character generator for a routine to generate a new password when someone forgets theirs... I came across the following code (I ported to FPC) which reads random numbers from the linux device urandom (/dev/urandom).

//////////////////////////////////////////////////////////////
// Application: devrandom
//      Author: G.E. Ozz Nixon Jr.
// ===========================================================
// Code ported from: http://www.linuxquestions.org/questions/programming-9/shell-script-random-variable-4088/
//
// RANGES: ./devrandom 0 999999999
//////////////////////////////////////////////////////////////

{$mode DELPHI} // for "raise"

uses
   sysutils, // for format, and exception
   baseunix;

var
   fd:longint;
   result:longint;
   min,max:longint;
   number:int64;
   err:longint;

begin
   if paramcount=2 then begin
      val(paramstr(1),min,err);
      val(paramstr(2),max,err);
   end
   else if paramcount=1 then begin
      min := 0;
      val(paramstr(1),max,err);
   end
   else begin
      system.writeln('Usage: devrandom [[min] max]');
      halt(2);
   end;

   fd := fpopen('/dev/urandom', O_RDONLY);
   if fd<0 then begin
      raise exception.create('');
      halt(1);
   end;
   result := fpread(fd, @number, sizeof(number));
   if result<0 then begin
      raise exception.create('');
      halt(1);
   end;
   fpclose(fd); // was missing in the C code!
   number := number shr 1;
   number := number mod (max - min + 1) + min;
   system.writeln(format('%d', [number]));
   halt(0);
end.

     In the above code the application is opening a Linux device as a file and reading sizeof(number) bytes into the memory address reserved by number. Finally it does a shr to drop the negative bit, then using math it does a mod to make sure the result is between min and max. One of the flaws to this concept is the use of mod affects the quality of the random. The modulo operation finds the remainder of the division. Whereas, random is working bit by bit to truly generate a unique bit pattern (byte).

     Doing a shr to drop the sign also changes the value luckily we could simply say number := abs(number); in pascal to simply do shl 1, shr 1 ... while forcing the bit to be zero. Another possible solution, which I prefer is to use bitwise AND and drop the "sign" bit.

   number := number and $7FFFFFFF; // this works for all languages

G.E. Ozz Nixon Jr.

Brian Ellixson
FreePascal User