[MLton] filedes = int (was: Stack size?)

[Stephen Weeks]

> However, more complicated FFI scenarios, say using NLFFI, might
> require building a C-struct with a file_desc component, at which
> point we would like to have the C-rep of a file_desc in hand.  So,
> there would likely still be a need for exposing the representation.

True.  I don't object to having a wrapper/unwrapper.  But it seems
like the signature approach is strictly better than the datatype
approach.  One can expose the wrappers if one wants, and one can also
use the FFI if one wants.  Of course one is explcitly choosing to
expose the type to the FFI which could bite one later, but that's more
a design issue than a technique issue.