From bug-octave-request at bevo dot che dot wisc dot edu  Thu Oct 12 05:22:10 1995
Subject: FP exception when dividing by small complex number
From: John Eaton <jwe at bevo dot che dot wisc dot edu>
To: Eyal Doron <doron at goofy dot mpi-hd dot mpg dot de>
Cc: bug-octave at che dot utexas dot edu
Date: Thu, 12 Oct 1995 05:20:30 -0500

Eyal Doron <doron at goofy dot mpi-hd dot mpg dot de> wrote:

: Octave generates a divide-by-zero (which is a FP exception on an Alpha)
: when dividing by a complex number whose elements are smaller than
: sqrt(realmin). I've been told this also happens on other machines
: (except there it returns Inf or NaN).
: 
: Repeat-By:
: ---------
: 
: octave:22> 1/(1e-200*(1+I))
: warning: Attempted division by zero.
: error: floating point exception -- trying to continue

Like all floating point problems on the Alpha, this will only really
be fixed once gcc has complete support for IEEE floating point
operations, or once Octave can be compiled with another C++ compiler
that does.  I don't plan to work on the latter, and I have no idea
when gcc will be fixed (I've asked in bug-gcc if anyone is working on
this, but got no reply).

On machines where IEEE FP operations do work, you can get suppress the
warning message with

  warn_divide_by_zero = "false";

(try that with Matlab! :-).

Thanks,

jwe