From psmith@mozart.convex.com Thu Mar 19 17:52:37 1992
Received: from convex.convex.com by dkuug.dk via EUnet with SMTP (5.64+/8+bit/IDA-1.2.8)
	id AA22466; Thu, 19 Mar 92 17:52:37 +0100
Received: from mozart.convex.com by convex.convex.com (5.64/1.35)
	id AA01128; Thu, 19 Mar 92 10:51:33 -0600
Received: by mozart.convex.com (5.64/1.28)
	id AA15647; Thu, 19 Mar 92 10:51:32 -0600
Date: Thu, 19 Mar 92 10:51:32 -0600
From: psmith@mozart.convex.com (Presley Smith)
Message-Id: <9203191651.AA15647@mozart.convex.com>
To: Keith.Bierman@eng.sun.com, walt@netcom.com
Subject: Re:  (SC22WG5.75) Re: From market surveys to "design"
Cc: SC22WG5@dkuug.dk, psmith@mozart.convex.com
X-Charset: ASCII
X-Char-Esc: 29

Thanks Keith...good comments...


>Marketing prefers that I not discuss precise figures. I think it safe
>to say we have more than 100K users.

>Our articulated policy for years has been correctness (viz. correctly
>rounded base conversion "infinite" pi for trig arg reduction, etc.)
>portability (viz. easy to port *to*) and  then performance. During the
>last 3.5 years (my time at sun) I've told this personally to at least
>1000 users, and sales have steadily increased. I do not believe that
>*performance*uber*alles describes all Fortran programmers. I do
>believe that Presley is quite probably right in understanding the
>mind(s) of Convex's customers. Not all market segments share the same
>focus. 

Great... now I understand about your market and it's size.  Is that
100K users of Fortran or 100K machines sold?   We find that about
95% of our machines sold also have a Fortran compiler sold with them.


>Vendors do this sort of thing. I don't think we are prepared to share
>our results; competitive advantage and all that stuff. I don't think
>ANSI has funding to do anything resembling a good  job of it. So I am
>dubious of value of doing a volunteer effort. 

But we can talk to each other and share what is public information. 
I don't have to give you my customer list to give you an idea of what
the customers are saying and how they are using Fortran and about 
how many of them there are...

>Bad information is worse than "no" information.

Maybe... I'd reword that to say that some information is better than 
no information.  You and I and everyone has to judge the correctness
of any information you get at any time...


>Because someone's paycheck is cut by a vendor, that doesn't mean that
>the individual is voting as a compiler writer. I can't speak for
>others, but in my case I don't work for a compiler group. My
>background is in writing math libraries and writing application
>programs, and my biases are in the direction of making my old job(s)
>and the jobs of my current company's customers "easier" "better" "more
>productive". X3H5 has several folks from vendor shops who represent
>application departments, not compiler writers. So I am dubious of
>simple headcounts of the membership roster to determine "proper
>representation". 

If we had more real users on the committee that would be great.  I'd 
like to have people who have written lots of Fortran code validating 
each item added to the standard BEFORE it goes in... try applying it
to real code BEFORE it's voted in...

> ... more design up front

>The description given describes the WaterFall model of sw development.
>There seems to be a large movement *away* from it. The OO folks, the
>rapid prototyping folks, the evolutionary model folks (and yes, there
>are overlaps ;>) have all junked it. In my experience the WF model is
>virtually worthless. I readily acknowledge that experience varies, and
>is certainly domain bound. The WF model *may* be very good for
>standards; but I would not conclude that it is because it helps create
>good sw. I *think* history suggests it hasn't helped sw development in
>the aggregate.

I personally prefer just what I said above.  Each item that is going
to be added to the standard should be validated and reviewed before
being added.  I'm not in love with the WF model... pick one.  Rapid
prototyping or some combination... spell out up front how things will
be done...

> .... elitest snobbery in voting ....

>The undergraduate student who has 2 weeks of *even* Prof. L's course
>is not as qualified to judge the work of X3J3 as a programmer with 20
>years experience, responsibility for a large body of Fortran code, and
>etc. 

There are religious issues here.  The person with 20 years of experience
may not want change...  Fortran must change and evolve or it will die.
The issue it is to be shure that the right balance between change and 
the desire of people not to change is reached. 

There's also an economic reality...if it costs too much to change, people
will look for other alternatives.  

>I don't think there are simple answers to these questions. I think the
>debate is healthy and I agree with Presley that we have to act as
>adult as we can. If the committee cannot come to agreement *somehow* I
>do not think that there is a higher authority that can reasonably be
>appealed to (X3 has the *right* to overrule J3, but it cannot do so
>>*reasonably*. If they had the wherewithall to do the work, it would be
>their job ;>).

X3 is much more political than X3J3 (if that's possible!!).  When they
sense a problem with a committee coming to consensus, they are charted
with taking action.  When this committee gets their differences worked
out then the relationship with X3 will return to a rubber stamp effort.

>While I think the minority should be heard (and their arguments
>recorded for posterity), I would hope that in the future, that the
>minority will not take actions which delay work for years at a time.

Everyone can use the process.  When consensus is reached, then everyone
is a silent minority because no one got everything they wanted but 
everyone is signed up. 

One of the prime things that keyed X3 was NOT any technical argument, etc.

One of the prime issues discussed at X3 when Fortran 90 was made a separate
standard was that consensus was not being reached on X3J3.   It was pointed
out that the same 9 or so people had voted NO for multiple ballots on 
the standard at X3J3.   That was viewed that IF the committee was working
to obtain consensus, that number would be decreasing and/or the people
who were voting NO would have changed over time.   The fact the same 
group continued to vote NO indicated to the X3 guys that the committee
was deadlocked in coming to consensus.   That fact indicated to X3 that
those members of X3J3 had some problem(s) with Fortran 90 that the 
committee was not willing to address.    

If we want creditibility with X3, then we've got to obtain consensus...

>A standard in 1985 (even a suboptimal one) would have been better than
>one in 1992. Evolution *will* occur; we can either be part of the
>solution or part of the problem.

>Scott McNealy's dictim #N

>   1) Agree and commit
>   2) Disagree and commit
>   3) Get out of the way

Per Walt's mail, it seems that everyone was in agreement in 1985.  They
should have put out a standard then.

Thanks for the response.  Presley