Re: Ada for Boeing 777

[Ken & Virginia Garlington <redhawk@flash.net>]

Vance Christiaanse wrote (with editing):
> 
> In article <Pine.SUN.3.91.960219111334.8550B-100000@L10SUN8>, Frank
> Petranka <fpetran@l10server.nswc.navy.mil> wrote:
> 
> > As someone who has used both Ada and C, I was suprised and disappointed
> > that code written in Ada produced no fewer problems than code written in
> > C.
> 
> Two observations:
> 
> 1) Published accounts of software development efforts tend to be highly
>    political documents.

That may be, but from indirect information I've received, this article is
fairly accurate for the 777.

> 
> 2) The quotes you provided didn't mention any actual statistics and the
>    wording suggests that the Ada problems may not have been related to
>    the language itself, but to the compiler and tools.

This is definitely an issue for embedded systems, particularly in aerospace.
Many of these systems don't get to use the most popular processors/OSs for
a variety of reasons. As a result, the compilers are not widely used, and
there are more latent bugs.

Some other observations:

3) My personal experience is that our switching from JOVIAL to Ada did not,
by itself, cause large drops in the number of problems in our software. Of course,
the operational defect rate of the F-16 flight control software (JOVIAL), so
far as I know, is zero. As a result, it's going to be hard to beat that!

What is happening (at least, I think it's happening) is that we are able to
at least maintain that rate as the complexity of our applications explodes. Ada
is a small part of maintaining that rate. It wasn't clear to me, but if the article
was claiming that the number of problems in the newer (presumably more complex)
Ada 777 systems was comparable to the (presumably simpler) systems written in
other langauges, then that's an encouraging result.

4) A lot of the defects in software have nothing to do with software coding or
design, particularly with the sort of software development process used in
aerospace. They tend to be introduced at the requirements level. So, it's not
surprising that large differences between languages don't occur in that
environment. In a SEI Level I environment, it probably makes a bigger difference.

5) I'm seeing a lot of companies going to code-generation tools and other
techniques to support domain-specific architecture development. I think this
trend is also going to make the language choice less important in the future.
For example, reusing an often-used "C" routine in a similar environment can
often (but not always) be more reliable that writing an Ada routine from scratch.