Bug with real / tuples?

[Matthew Fluet]

> > Thanks for the bug report.  It is a bug in the native codegen; you can
> > compile with -native false and get the correct behavior.  I'm looking into
> > it.
> 
> Thanks a lot! :)

I found the bug; it was some bad logic associated with register allocation
for the fucom instruction.

Anyways, I've checked in the fix to our internal sources, so it will be
correct in the next release.  If you're comfortable enough with
rebuilding MLton from sources, you can grab
http://www.cs.cornell.edu/People/fluet/MLton/x86-allocate-registers.tgz
which replaces
mlton-20010806/src/mlton/codegen/x86-codegen/x86-allocate-registers.fun