[MLton] Wrong behaviour of MLton.Random.alphaNumString?

[Jesper Louis Andersen]

I have the follwing little code snippet:

structure GenRandomFile = 
struct

(* Generate a n lines of random data, outputting it to file fname *)
fun genData (n,fname) =
  let val ostream = TextIO.openOut fname
      fun loop 0 = TextIO.closeOut ostream
        | loop n = 
             ( TextIO.output (ostream,
				  String.concat [
					MLton.Random.alphaNumString 76,
                                        "\n"
                                  ]
			     )
	     ; loop (n-1)
             )
  in
      (case n of
         NONE => loop 1000
       | SOME n => loop n
      )
  end

val _ = 
  case CommandLine.arguments () of
    [fname, n] => ( genData (Int.fromString n, fname)
	          ; OS.Process.success
        	  )
  | _          => OS.Process.failure
 
end

Which could be cleaner, nevermind that. running this with:

annah$ uname -a
NetBSD annah 1.6ZG NetBSD 1.6ZG (ANNAH) #0: Fri Dec 12 21:07:23 CET 2003  root@annah:/usr/src/sys/arch/i386/compile/ANNAH i386
annah$ mlton
MLton MLTONVERSION (built Sun Dec 14 02:41:24 2003 on annah)

./create-data test 10 gives the following output:
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This is not very random to me. So:

a) Is this the intended behaviour of alphaNumString? If yes I think
 we should explicitly note it only works up to a certain size
b) If it is not the intended behaviour, i would like to have input
 from other architectures than NetBSD, in order to narrow the bug down.
c) I tried to take a look at the code in question 
 (lib/mlton-stubs/random.sml). I find it rather wierd that after the
 first 6 characters have been set no more calls to rand () occurs.
 (Line 19 to 46). It seems that the random function works by wrapping 
 around the word, and currently I do not have the time to read the 
 algebra needed to get that little bugger sorted out.



-- 
j.