compiling MLton whole-program with SML/NJ

[Stephen Weeks]

Today, I tried using SML/NJ 110.37 to compile MLton whole (102k
lines).  On a 733 MhZ machine with .5G, SML/NJ struggled mightily with
lots of swapping, and finally finished.  Here are the compile times

real time:  6 hours 42 min
user + sys: 3 hours 59 min

It generated a 38M heap file.  The usual size is about 35M, so it was
only slightly bloated.  Although, I had expected the size to shrink,
due to inlining functors and other optimizations.

I'll try later to get some timings and see if it's any faster than the
separately compiled version.