Even more interesting...

[Stephen Weeks]

> I'll implement this with a conditional #define
> for the MLton runtime library. Then I will set this
> flag automatically for the Win32 platform - until
> a better solution for Win32 might be implemented.

Sounds good.

> > * Fix gc.c to not use weird smunmap calls.
> >   While not as easy as the above, this shouldn't be too bad.
> >   Whenver the GC uses the smunmap call to shrink a chunk of memory
> >   (either a stack or heap), you could instead allocate a new chunk of
> >   memory of the smaller size, copy over, and free the old memory.  Or,
> >   you could try using realloc, as Henry mentioned.
> 
> Wouldn't this give problems with other areas of memory pointing
> to these areas? So that one would have to forward all pointers?
> I don't think I would dare make such a modification to
> the GC when I didn't write it myself :)

You are right.  This proposal was a bad idea.