Numerical C Extensions Group (NCEG) Meeting Report

[Rex Jaeschke]

The following report is reprinted from Volume 1, Number 1 (June 1989)
of The Journal of C Language Translation, Copyright 1989 Rex
Jaeschke.  Permission is granted for duplication and distribution of
this report for the purposes of furthering the work of NCEG.

%---------------------------------------------------------------

                Numerical C Extensions Group Status


                          Rex Jaeschke
                          NCEG Convener

Introduction

When I conjured up the idea for an ad hoc group to define numerical
extensions to C earlier this year, I had no idea as to what the
reaction would be.  The evidence is now clear that this endeavor is
seen as being very worthwhile.  Not only have more than 90 people
asked to be added to the contact database, but 30 of them attended
the one-and-a-half day meeting at Cray Research on May 10--11.

The backgrounds of the attendees was diverse.  The supercomputing
industry was represented via Cray, Convex, Supercomputer Systems, and
Thinking Machines.  The IEEE community was well represented by Hough
(from Sun), Cody (from Argonne Labs), and Thomas (from Apple.) Other
organizations represented included Unisys, Microsoft, Digital
Equipment Corporation, H-P, CDC, IBM, Solborne, Farance, Inc.,
University of Minnesota, Intermetrics, and Information and Graphics
Systems.  The digital signal processing industry was represented by
Analog Devices, and LLNL, Army BRL, and Polaroid Corporation
represented the user community.  Dennis Ritchie from AT&T also
participated.

There was no real sentiment that we deliberately go against the
direction established by ANSI C.  In fact, quite the contrary. 
However, it was recognized that some of ANSI C's constraints may
impede our activities resulting in possible conflicts.  The whole
issue of errno and formatted I/O of NaNs and infinity are examples.

The Issues

The main purpose of the meeting was to identify and prioritize the 
principal technical issues. The group then voted on each topic 
indicating high or medium (or no) priority. The high priority votes 
were weighted twice as much as the medium, and the following list of 
priorities resulted.

		------------------------------------------
		          Main Numerical Issues
		------------------------------------------
		       Topic                 Priority
		------------------------------------------
		aliasing			29
		vectorization			27
		complex				27
		variably dim arrays		25
		IEEE issues			24
		exceptions/errno		24
		float/long double library 	23
		parallelization			22
		ANSI <math.h>			21
		array syntax			19
		extra math functions		17
		aggregate initializers		15
		inter-language issues		15
		wide accumulators		10
		math function precision		9
		non-zero-based arrays		8
		numerical representation	6
		new data types			4
		new operators			4
		function overloading		4
		------------------------------------------

Another topic, ``Arrays as first class objects'' had a high priority 
(21) but after considerable debate was dropped from the list since it 
was agreed its addition would likely cause great confusion to existing 
C programmers.

Formation of Subgroups

The bulk of the agenda time was then given to the top ten topics, each 
getting 20--30 minutes. For each of these topics, attendees 
volunteered to be the primary and alternate coordinator. (The minutes 
of the first meeting identify these people. In the interim, contact me 
for details.)

The intent is that the real technical work will go on between meetings 
and be coordinated by the leaders of each subgroup.  Then, at the 
following meeting, each subgroup will present the results of its work 
and make formal proposals as appropriate. This way, the committee can 
focus on the final, distilled issues rather than everyone getting 
involved at all levels. It will also significantly reduce the amount 
of paper in the mailings.

If you wish to participate in any of these subgroups it is your 
responsibility to contact the leaders and identify yourself, your 
concerns and how you can help. If your area of interest is not listed 
here, start your own subgroup and let me know.

Mailings and Submissions

Most of people interested in NCEG appear to have an e-mail address so 
that should make the subgroups' job much easier in coordinating 
various viewpoints and proposals. However, all formal distributions 
will be by paper mail. Since meetings are to be once every six months 
there will be two mailings between meetings. The first will occur 
within 4--6 weeks after a meeting and will contain minutes, new papers 
and other appropriate correspondence. The second will occur about 4--6 
weeks prior to the following meeting. The cut-off date for formal 
submissions for the September meeting is August 11.

Forward all correspondence to me (either by mail or via
uunet!aussie!rex) and I will assign it a document number. 
(Note that I do not have a troff formatter.) However, do that 
only if your paper is concerned with issues other than those being
handled by the subgroups.  For subgroup issues, forward papers to the
subgroup coordinators so they can include it in their submissions to
me.  The intent is to avoid excessive duplication of points and to
allow the short meeting time to be used more effectively.  The more
formal documents we have the slower it will go.

Tom MacDonald at Cray Research has agreed to do the mailings, at
least for the interim.  Frank Farance of Farance, Inc., has
volunteered to be the redactor of the group's working document. 
Thanks to Tom and Frank. (Thanks also to Randy Meyers from DEC, who 
acted as meeting secretary and to Cray for being meeting host.)

Formal Affiliation

There was general consensus that we become affiliated with a 
recognized standards organization. The final proposal was that we 
become a working group within X3J11. If we follow that route, it will 
result in our publishing a Technical Report, a non-binding report on 
our findings and recommendations. With suitable planning, we might be 
able to have that elevated to a Technical Bulletin and get it 
distributed with the ANSI Standard. Getting our extensions adopted as 
a standard is also possible, in the long term.  At this stage, I plan
to ask for agenda time at the next X3J11 meeting to discuss admitting
us as a work group.

In the interest of economy, the next two meetings are scheduled in the 
same location and week as those of ANSI C's X3J11. These NCEG meeting 
dates are September 19--20 (Salt Lake City, Utah), and March 7--8, 
1990 (New York City.)

%---------------------------------------------------------------

Rex

----------------------------------------------------------------------------
Rex Jaeschke     | C Users Journal     |  Journal of C Language Translation
(703) 860-0091   | DEC PROFESSIONAL    |1810 Michael Faraday Drive, Suite 101
uunet!aussie!rex | Programmers Journal |     Reston, Virginia 22090, USA
----------------------------------------------------------------------------