Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!utgpu!water!watnot!watmath!clyde!rutgers!seismo!brl-adm!adm!dsill@NSWC-OAS.arpa
From: dsill@NSWC-OAS.arpa
Newsgroups: comp.lang.c
Subject: Re: C and rf (really parens)
Message-ID: <6823@brl-adm.ARPA>
Date: Fri, 10-Apr-87 11:55:42 EST
Article-I.D.: brl-adm.6823
Posted: Fri Apr 10 11:55:42 1987
Date-Received: Sat, 11-Apr-87 17:25:27 EST
Sender: news@brl-adm.ARPA
Lines: 28

Dan Hoey wrote:
>No, it says that for the purposes of C, the floating-point
>operations are ``commutative and associative'' and may be
>reordered.  If you want to object to K&R's redefinition of a term
>that has a well-defined mathematical meaning, object away, but
>that's what they did.

Well, I can't find anything in K&R that *explicitly* defines floating
point ops to be commutative and associative.  My point is this, since:
  - floating point operations in general are not commutative and
    associative,
  - a new C standard is being developed, and
  - everyone who uses C for numerical programming is bothered by the
    current approach,
shouldn't we be trying to decide what the *right* behavior is so it can
be considered for inclusion in the standard?  By "right" I mean the most
correct.  K&R allowed reordering expressions under the assumption
that the operations are commutative and associative and the reordering
would have *no effect on the result of the expression*.  However, there is
a BIG difference between real arithmetic and floating point arithmetic that
was entirely ignored by K&R.

I'm all for allowing the compiler to do its stuff, reorganizing and
optimizing away, but I also want to be able to predict the results of an
expression and/or program.

-Dave Sill
 dsill@nswc-oas.arpa