[S] Why MATHSOFT might not like an S+ compiler and why they perhaps should

Jens Oehlschlaegel (oehl@Psyres-Stuttgart.DE)
Thu, 1 Oct 1998 18:16:29 +0100 (MET)


Dear Charles Roosen,

I always thought it would be nice to have an S+ compiler and thought it
just would be a pitty not to have one. Now looking back to the last years
and thinking about what I expected S+ to be and what it turned out to be,
I now am convinced that a compiler is absolutely necessary in S+.

>From my point of view, a system like S+ (or R) should ideally serve three
functions:

(1) being an experimental environment for developing new procedures
(2) being an efficient enironment for statistical analysis, graphing etc.
(3) being a knowledge representation for statistical computing, for
learning and teaching

I think for combining the three a compiler is necessary: the S interpreter
was created to facilitate (1) the experimental part, where an interpreter
is sometimes but not always fine. For the efficient part often compiled
code is necessary, but as knowledge representation compiled code is
useless. I see five problems in S, which could be solved by an S+
compiler:

(1) Interpreted language is often too slow, even for experimental tasks
(2) Additional ressources are needed to convert (S+) code to C or Fortran
(3) Including C or Fortran code is a platform dependent task
(4) C or Fortran code is not as readable as S+ code
(5) compiled code substitutes source code which impairs S+ as a
knowledge representation

With language level 4 (S+5.0) it was shown how to make the
help system platform independent: by having help as part of each S+
object. This obviously increases S+'s value as a
knowledge base. Along similar lines, the above 5 problems could be
circumvented by having a compiler. Then a typical S+ object could consist
of

(1) S+ source code
(2) S+ compiled code
(3) documentation

Only exceptional S+ code should be replaced by source code and object code
of another language.

Of course such strategy would make all statistical algorithmical know how
absolutely transparent. I can well imagine that MATHSOFT were reluctant to
that. I appreciate that MATHSOFT maintains the environment, and find it in
a way understandable to keep secret something, e.g. the implementation of
the environment and of the click and go add ons. However if S+ were *the*
system for collecting (efficient) statistical algorithms, of course it
would not be adequate that any private company 'owns' such sources. It
looks like MATHSOFT has two very different choices in which direction to
develop their product. Of course I do not have any competence to know what
would be better for MATHSOFT on the long run, but I think I have explained
what kind of system I would pay for, if I had the choice.

I would appreciate if I could trigger an official answer from MATHSOFT
and suggest the academic community will carefully listen to it.

Best regards

Jens Oehlschlaegel

--
Dr. Jens Oehlschlaegel-Akiyoshi
Psychologist/Statistician (Ph.D.)

!! please note my new address and do not use the old one any longer !! !! oehl@geocities.com http://www.geocities.com/Tokyo/Bay/5663/ !!

----------------------------------------------------------------------- This message was distributed by s-news@wubios.wustl.edu. To unsubscribe send e-mail to s-news-request@wubios.wustl.edu with the BODY of the message: unsubscribe s-news