c backend

[Stephen Weeks]

> The problem is we have a bunch of Java code that uses its
> C FFI to make calls to C routines.  These routines don't
> yet exist, and I suggested that MLton could be used instead of
> C.  Having  C->SML hooks would be a start, but cumbersome,
> since we still need to write the C stubs that would be called
> from Java.  I was hoping to just use SML and have MLton produce
> the necessary C output.  It might make selling MLton here a 
> bit easier.

I'm not sure I understand.  But, the idea of the MLton C->SML
interface is that MLton would generate the .h file that corresponds to
the .o, so other code can really pretend the stuff was all written in
C.  In that case, I don't see what other C stubs that you would need
to write.

I should mention that all of this has the caveat that we aren't
shooting for data interoperability for our first whack -- we just plan
to support our usual simple types, like int, word, real, int array, ....