die.sml and the Rico bug

[Stephen Weeks]

I am revisiting Henry'y die.sml program, which was inspired by a
program he had written for Rico, and Henry claimed showed a segfault
in previous versions of MLton.  As a reminder, I have appended the
program to the end of this message.  I have tried the program on my
machine, which contains 1/2G RAM and 1G swap, using mlton-20011006,
20020115, and the currently checked in sources.  It appears to behave
correctly for all three versions.  That is, running

	die @MLton max-heap 769m -- <z.in

succeeds, but running

	die @MLton max-heap 768m -- <z.in

fails with an out of memory message (where z.in contains 170,000,000
bytes).

I am now confused what was the error Henry saw, since I am unable to
duplicate it.

--------------------------------------------------------------------------------
(*
 * Read in a file into keys.
 *)
fun readFile fd =
       let fun loop (res, size, used) =
                  let val n = Posix.IO.readArr (fd, { buf = res,
                                                      i = used,
                                                      sz = NONE } )
                  in if n = 0
			then res
                     else let val used' = used + n
                          in if used' = size
                                then let val size' = size + size
                                         val res' =
						Word8Array.array (size',
								  0w0)
                                     in Word8Array.copy { src = res,
							  si = 0,
							  len = SOME used',
							  dst = res',
							  di = 0 };
                                        loop (res', size', used')
                                     end
                                else loop (res, size, used')
                          end
                  end
       in loop (Word8Array.array (4096, 0w0),
                4096,
                0)
       end

fun main () =
       let val _ = readFile Posix.FileSys.stdin
       in ()
       end

val _ = main ()