Path: utzoo!utgpu!news-server.csri.toronto.edu!clyde.concordia.ca!uunet!cs.utexas.edu!uwm.edu!bionet!agate!darkstar!saturn.ucsc.edu!sidney
From: sidney@saturn.ucsc.edu (Sidney Markowitz )
Newsgroups: comp.lang.c++
Subject: Re: Problems of Zortech C++ v2.1
Message-ID: <5286@darkstar.ucsc.edu>
Date: 19 Jul 90 20:29:57 GMT
References: <82035@tut.cis.ohio-state.edu> <31714@cup.portal.com> <1331@proto.COM>
Sender: usenet@darkstar.ucsc.edu
Distribution: na
Organization: University of California, Santa Cruz
Lines: 22

In article <1331@proto.COM> joe@proto.COM (Joe Huffman) writes:
>Okay, first off from page 34 of E & S: "It follows, contrary to popular
>misconception, that the safety of any conversion from an integral type
>to a floating type is implementation dependent."  

Reading the previous lines in that paragraph on page 34 reveals that
"it follows" from the minimum magnitude values defined in sections
3.2.1 and 3.2.2 which allow an implementation to support integral
values with more digits than the precision of some floating type
values. So you can have a 10 digit long but only 6 digits of precision
in a float and still have a conforming implementation. The very next
sentence on page 34 specifies that the implementation dependence is
when the integral value is in the range that can be represented as the
floating type, but not represented exactly. That means that if you
have an integral value that has no more digits than the precision of
the floating type, the conversion is absolutely safe.

-- sidney markowitz <sidney@saturn.ucsc.edu>

Disclaimer: I guess I should let on when there's any hint of
competitor-bashing, that I work for Borland, but I am not officially
speaking for Borland.