\documentclass[nocolor,memo]{j3}
\renewcommand{\hdate}{13 February 2007}
\renewcommand{\vers}{J3/07-113r2}
\usepackage{lineno}
\usepackage{longtable}
\usepackage{xr}
\externaldocument{007}
\begin{document}
\input txtest
\vspace{-10pt}
\begin{tabbing}
\phantom{Reference: }\=\kill\\
To: \>J3\\
Subject: \hspace*{0.25in}\>RRSPACING (UTI 092)\\
From: \>Van Snyder\\
\end{tabbing}
\section{Introduction}
UTI 092 argues that RRSPACING(X) should be an IEEE NaN if X is IEEE Inf.
On the other hand, however, RRSPACING(X) can be written as
ABS(FRACTION(X)) * RADIX(X) / EPSILON(X). RADIX and EPSILON are inquiry
functions, so their result values don't depend upon their argument values,
while the result value of FRACTION(X) is defined to be "the same value as
X" if X is an IEEE infinity. I would be really surprised if ABS(Inf) or
ABS(-Inf) were not Inf, so RRSPACING($\pm$Inf) ought to be +Inf.
Malcolm argues that the definition of FRACTION is defective: IEEE
infinity is not a sensible result for an IEEE infinity argument; the
result ought to be NaN. I suppose it's OK to return the same NaN if the
argument is NaN. Rather than make RRSPACING consistent with the
defective FRACTION, we probably need an interp to correct FRACTION. Or
could we just announce an incompatibility in 1.6.3?
J3 has decided not to tackle this issue until interp 42 is tackled.
\edits{07-007}
\stdref{420:10+}[Editor: Delete UTI 092.]
\label{lastpage}
\end{document}