Path: utzoo!attcan!uunet!husc6!mailrus!cornell!uw-beaver!teknowledge-vaxc!sri-unix!quintus!ok
From: ok@quintus.UUCP (Richard A. O'Keefe)
Newsgroups: comp.lang.fortran
Subject: Re: Fortran vs C for computations
Message-ID: <362@quintus.UUCP>
Date: 7 Sep 88 09:51:02 GMT
References: <1443@valhalla.ee.rochester.edu> <3256@lanl.gov>
Reply-To: ok@quintus.UUCP (Richard A. O'Keefe)
Organization: Quintus Computer Systems, Inc.
Lines: 38

In article <3256@lanl.gov> jlg@lanl.gov (Jim Giles) writes:
>From article <1443@valhalla.ee.rochester.edu>, by davis@galaxy.ee.rochester.edu (Al Davis):
[[ in response to a statement that Fortran is better than C for numerical
   work because sqrt() can be trapped in Fortran but not in C,
   that it *can* be done in C, with matherr().
]]
>Anything C can do portably with library routines, Fortran can also do
>portably with library routines.

I would like to see a more convincing proof of this.

>Just because you haven't seen a Fortran
>compiler that does X, doesn't mean X can't be done with Fortran.

But the topic was not what *could* be done, but what *HAS* been done.
There are numerous Fortran processors which provide means of trapping
or ignoring floating-point exceptions.  Trouble is, they don't do it
the same way.
[I checked the IBM SAA FORTRAN Reference manual, but evidently MVS Fortran
and OS/2 Fortran do not have a common form of numeric exception handling.]


- (pre-ANSI) C has matherr() and signal(SIGFPE, ...), though the draft
  ANSI standard doesn't contain matherr().
- ADA [LRM 11.1, 4.5.7] doesn't require that an implementation detect
  or signal numeric errors, but it _does_ provide a standardised way of
  catching them if they are detected and signalled.
- COBOL has had the "ON OVERFLOW user-action" clause for catching
  fixed-point overflow for decades.
- PL/I has the ([NO]{OVERFLOW,UNDERFLOW,...}): condition prefixes and
  the ON {OVERFLOW,UNDERFLOW,...} statement.

How come _the_ language for floating-point calculations doesn't have
a portable way of catching or suppressing floating-point exceptions?

Bitwise operations and a WHILE loops are covered in a supplementary
MIL-STD for Fortran; is there also some supplementary standard which
supplements Fortran with this facility?