Path: utzoo!utgpu!water!watmath!clyde!att!osu-cis!tut.cis.ohio-state.edu!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: <356@quintus.UUCP>
Date: 6 Sep 88 08:35:48 GMT
References: <535@nikhefh.hep.nl>
Sender: news@quintus.UUCP
Reply-To: ok@quintus.UUCP (Richard A. O'Keefe)
Organization: Quintus Computer Systems, Inc.
Lines: 23

In article <535@nikhefh.hep.nl> t68@nikhefh.hep.nl (Jos Vermaseren) writes:
>
>In the second edition of Kernighan and Ritchie it is explained that the
>standard response of a floating point operation (multiplication, sqrt etc)
>to an irregularity is returning special values.
>
>So
>	float x = -2.0;
>	z = sqrt(x);
>
>gives back some value. Amazingly enough many UNIX vendors like this philosophy,
>so even the UNIX fortran compilers often prefer it this way.

(1) See "matherr" in the System V documentation.  If you define a function
    int matherr(struct exception *x) it will be called when math library
    functions detect a problem.

(2) Returning funny values is exactly what IEEE floating point arithmetic
    is all about.  Note that the IEEE 754 standard requires AS THE DEFAULT
    that x+z and the like should return infinities or NaNs in various
    circumstances, rather than signalling an exception, and that IEEE 754
    provides no standard way of enabling exceptions (though it suggests
    that there should be one).