sysinfo() kernel variance

[Henry Cejtin]

I  guess  that  I  would  be  happy  just  to be able to compile a MLton that
produced executables that could run on that machine, but it is an interesting
question.

When I was testing things out, if I run code compiled on a 2.4 kernel under a
2.2 kernel, mem_unit prints out as zero.  Looking at the kernel sources, this
can  be depended on.  Thus you can test for this in code compiled under a 2.4
system at run time.  This would still fail on a 2.2-kernel machine with  more
than  4  gig  of RAM or swap.  I.e.: if you are compiling on a 2.2 kernel (no
mem_unit) then act as if mem_unit is always 1.  If you are compiling on a 2.4
kernel, then at run-time check if mem_unit is 0.  If so, pretend it is 1 (and
don't try to grab any fields that were added in 2.4 kernels).

I don't think I would bother with the automake configure if this is the  only
thing  that needs testing, but I would call what ever does it configure and I
would include no Makefile, but only a Makefile.in which the configure program
copied to Makefile.  This way it feels like the standard configure.