(j3.2006) (SC22WG5.3524) [Fwd: Fwd: [Numeric-interest] IEEE 754R revision effort]

Van Snyder Van.Snyder
Mon Feb 25 15:43:55 EST 2008


This is a letter from David Hough concerning IEEE 754r that I received
via IFIP WG 2.5.

Van
========================================================================
Bo Einarsson
Mathematics Department
Link?pings universitet
SE-581 83 Link?ping, SWEDEN
------ Forwarded message -------


From: David Hough 754R work <754r at ucbtest.org>
To: numeric-interest at ucbtest.org
Date: Fri, 22 Feb 2008 11:27:58 -0800 (PST)


The IEEE 754R revision effort has entered its eighth year.
All the active participants are pretty tired and would like to wrap
things
up as quickly as possible.

However I see no value in publishing in haste and repenting at leisure
while
trying to explain specifications that are ambiguous
or obscure.    I am particularly concerned about issues around
expression evaluation
and the mapping of programming language expressions into operations of
the draft
standard.    754's unspecification in this area has led to 20 years of
misunderstandings,
particularly around higher intermediate precision.

But there can be opposition to rewriting ambiguous text, even
rewriting which
is not supposed to change the specification,  since a clarification
might meet
one person's presuppositions and not another's, and hence might
precipitate further
review cycles.

It would help me determine whether I am on the right track or jousting
at windmills if
some individuals from the technical computing application writing, 
mathematical software libraries, language design, and language
implementation
communities would take a fresh look at the troublesome areas
and say what they think.     I have already read the comments from 
persons on the IEEE 754R sponsor ballot group, on which chip and
system implementors
are well represented.

I have written a long evaluation of the current draft 1.6.0 covering
high-level
issues:

 http://754r.ucbtest.org/msc-ballots/ballot5.html

which also has some low-level nitpicking, and also a list of lost
causes - things that
seem suboptimal but unfixable.  

I am currently most concerned with draft 1.6.0 clause 10 and clause
5.4.1.
Some of my commentaries are in the form of proposed strikeouts in
green and
proposed additions in red:

 http://754r.ucbtest.org/msc-ballots/clause10.pdf
 http://754r.ucbtest.org/msc-ballots/formatof.pdf

but they may be hard to interpret without the context of a whole
draft.
Draft 1.6.0 isn't publicly accessible, so if you need it,
you must request a review copy from Bob Davis (bob at scsi.com). 
An older draft 1.5.0 is available at 

 http://754r.ucbtest.org/drafts/archive/2007-10-05.pdf

which has a number of differences from 1.6.0, but not many in these
particular clauses.
_______________________________________________
Numeric-interest mailing list
Numeric-interest at ucbtest.org
http://mailman.oakapple.net/mailman/listinfo/numeric-interest




More information about the J3 mailing list