W3C home > Mailing lists > Public > www-xml-schema-comments@w3.org > April to June 2001

Re: float/double :conversion from lexical space to value space

From: Kohsuke KAWAGUCHI <kohsuke.kawaguchi@eng.sun.com>
Date: Fri, 13 Apr 2001 14:49:03 -0700
To: "Ashok Malhotra" <ashokma@microsoft.com>, www-xml-schema-comments@w3.org
Message-Id: <20010413142224.8486.KOHSUKE.KAWAGUCHI@eng.sun.com>

I really appreciate the time WG spent to discuss this issue, and I'm
glad to know that the change will be made.

However, I would be remiss if I say that the suggested change is enough.


In Clinger's paper, section.8 "OVERFLOW AND UNDERFLOW", page.7, he says

"In such cases, the computation may need to be repeated using some other
algorithm, depending on the policies that have been established for
handling overflow and underflow within the particular floating point
number system in question."

So practically it means nothing. Therefore the spec still has to state
explicitly that how those cases shall be handled. (Underflow is
currently not a problem, thanks to the current wording. But it may be
possible that it becomes a problem under the modified wording.)

That said, I beg WG to consider treating overflow/underflow as an error.
Several programming languages (including C and Java) treat them as
errors, whereas none treats them as +-INF as far as I know.

In my understanding, +-INF are intended to represent arithmetic errors
during the floating point value computation.

----
Kohsuke KAWAGUCHI
Phone: 650-786-0721
Received on Friday, 13 April 2001 17:49:00 GMT

This archive was generated by hypermail 2.2.0+W3C-0.50 : Sunday, 6 December 2009 18:12:50 GMT