[MLton] MLton.Vector.create

[Henry Cejtin]

You  can't  quite  always  do what you want before calling the create without
using a ref cell and perhaps an extra option.  If I need to use some function
closed  over  sub  or update, and that is what the tabulator is going to use,
then I would have to use a ref cell carrying an option type and  fill  it  in
the first time that the tabulator was called.

I missed that you couldn't sub/update the last element in the tabulator ever.
Bad.

Funny, I think that you convinced me  that  the  currying  in  the  tabulator
should  stay.  The computation being staged could be useful, and also implies
that the sub and update functions don't vary between different invocations of
the tabulator.

Ah  yes, I was just thinking of the second version, where you pass sub/update
once and get back a tuple of the tabulator and a whacker.  I definitely  like
this  one much more.  It makes it clear that there is only one sub/update, it
being the same for all invocations of the tabulator and the whacker.  It lets
you  stage  things  in  a  perhaps  useful  way  (building  closures and data
structures once that are used both by the tabulator and the whacker).

I still am a bit disturbed by the complexity of the  restriction  on  calling
sub/update  from  the  tabulator, but with the whacker being separate, in the
case that the old code could handle one just wouldn't refer to sub or  update
at all in the tabulator, so it is syntactic.