Print

Print


Steve O'Neill said:

>>I need data.
>
>Man, I wish that I had it.  I do have a presentation that was done in 1997
>(I beleive) that might provide some data.  But the reality is that to
>develop the amount and quality of data that would reinforce or refute these
>arguments doesn't exist because it would be too costly to produce, if not
>impossible.
>

Even good CMM Level 3 organizations should have this type of data
(unfortunately, my company, while being a Level 3 organization, started
looking at the data and found that it was not consistent, so we have to
start getting new data).  So it should not be "too costly" to produce the
data, if companies work with different languages on the same types of
projects.

I wonder if the Ada compiler vendors would like to provide some numbers?
ACT used Ada; Intermetrics used C to implement essentially the same
functionality (an Ada compiler front end).  I do not know the languages
used for others.

Steve Ziegler wrote a wonderful paper about defects in Ada and C.
Development cost should not be that difficult to come up with.  Of course
there are many differences in the products, but it would be two interesting
points.  Perhaps the vendors could get together to do an objective
comparison (of course, knowing some of the personalities involved, it might
take a few years for them to agree on anything :-) ).

>From your description of the two projects their approaches were very
>different.  Are you saying that the -standard practices- are that completely
>different depending on the language of choice?!

Yes and no.  Project A was perhaps more formal than Project C, and perhaps
the same people would be just as formal using any language.  That is the
problem with real-world examples.  It is impossible to get true apples to
apples comparisons.

>>The 65% and 20% were made up to make the end costs end up very close,
>
>Generating the data to fit the premise?
>
>If a program using incremental development takes 65% to get to the first
>detailed design they they are doing something DRASTICALLY wrong!
>

I see we are still not communicating well.  The 65% is -of the amount
budgeted for the first increment-.  The assumption (mainly because no one
seems to have any figures) is that integration costs will be much lower for
a project written in Ada than for other languages.

So the overrun will not be for the whole project, but only for the
increment in question.  Of course this will bring the total percentage
overrun down considerably, which is, of course, the point of doing
incremental development.  It also can cause projects to be canceled early
(which is another point both in favor and against incremental development.
It is good to catch a bad contractor early.  It is bad to see one bad
decision affect the entire project).

>>but they are not way off from what I heard back in the early 80s when I
>>started learning the language.
>
>If you heard these numbers in the early 80's then I doubt seriously if they
>address an incremental development paradigm - I doubt that it existed then
>and it surely wouldn't have been considered by the folks doing Ada at that
>time.
>

Maybe it did not have a nice name, but I used incremental development in
high school in the late 1960s, in college in the 1970s, and at work
starting in 1976.  I am somewhat surprised to hear it is considered a new
concept (not completely surprised because I have been surprised before with
people talking about supposedly new concepts, like Rate Monotonic
Scheduling, which was used on Apollo).  My guess is that Ada (the Countess)
used incremental development.

>>And I have never seen any data stating the opposite.
>
>Unfortunate, but probably true - but that doesn't mean that the old data is
>applicable.

>>and it does cost more up front, in my experience.
>
>My experience says that it costs a small amount <10% more up front but that
>by midway through the race my velocity vector is much longer and more on
>target than the C hackers.  The tortoise and hare come to mind here.
>
>>But, I would argue that this is not -standard practice- in Ada projects
>>(although, looking at some open-source Ada software, I might be open to
>>persuasion :-)
>
>Excuse me... but screw -standard practice-.  If it doesn't get you where you
>need to be then it is broken and needs to be fixed.
>

Ok.  So it would appear that all the teachers and all the textbooks are
wrong (it would not be the first time, of course).  In fact I will suggest
that Project A went overboard with modularization such that communication
with external devices went through several layers of software more than
necessary.

I have seen this elsewhere, when people come to me and say "Ada is slow."
I look at what they did, ask how they would have done it in C, and finally
ask why they did not do it that way in Ada.  It never occurred to them to
use allowable tricks (usually in the area of low-level device drivers where
those tricks are really needed).

So how do you teach "Use Ada's facilities; don't abuse them"?

>You clean up both the design and the code as you go.
>
>>In my experience, it is very difficult to go back and clean up >working
>>code.
>
>I guess you just need to know how to plan for it from the beginning.  I've
>done it many, many times.
>

Ah, but have you ever tried to get -others- to do it?  Have you tried to
convince management to add cost to a proposal to include it?  I generally
get a response "Do it right the first time".  Actually I get that response
when I try to convince myself.  I -want- the benefit of Ada's constructs
when integrating the software.  There is a lot of reading of other people's
code that goes on in that phase, so the maintenance benefits are also
integration benefits.  And it is so painful to convince anyone to go back
and change working code that I have found it better to make sure it is
"right" before accepting it into integration.

But doing it right is not cheap.

>>I said "detailed design review", not "preliminary design review".
>
>So, there was not preliminary review of the design?  There's a good plan.
>

There were many reviews.  The one where the problems came up was the
detailed design review.

>>Myths will not go away by calling them myths.  We need to prove it
>>("I don't have to; I am in the majority, say the C people." :-(  ).
>
>I doubt that there's anything that will sway them - much like with other
>supposed majorities.  You're (effectively) talking religion here.
>

I don't think so.  There have been a number of pendulum swings and
bandwagons (NASA jumped on Ada, and has since jumped to C and C++.  The
contractors jump towards the money).  Given a choice, most of my colleagues
will use a language they think helps their career (i.e. C++ and Java).
Roger Racine
Draper Laboratory, MS 31
555 Technology Sq.
Cambridge, MA 02139
617-258-2489