reflecting differences between Linux and Cygwin in basis library

[Henry Cejtin]

In  general,  especially  where  POSIX  takes a stand, I think that the right
thing to do is to have almost all of MLton use the POSIX stand, and  to  then
implement  that  on  non-compilant  systems in extra files.  I did that a lot
with Scheme-48 and it was very nice.

In the specific example you give, I would provide  a  Cygwin  .h  file  which
defines the constant and a new C routine which checks for the new bogus value
and raises an exception.  In Scheme-48, when I could not  name  the  added  C
functions  what I wanted to, because I needed access to the native routine of
the same name to implement the `correct'  functionality,  I  used  #define's.
I.e., I would add
    #define real-posix-name real-posix-name-with-zulu-suffix
then in the C code to implement the work-around, I would initially do
    #undef  real-posix-name

This  was  some-what gross, but the point was that only broken systems needed
to look at the grossness.  On correctly functioning systems, the #define  was
not used and everything worked as documented by POSIX.

I would definitely not go the intersection route in general.