At 07:36 AM 12/30/98 -0800, Robert C. Leif, Ph.D. wrote:
> Congress should require that DoD and other Government agencies analyze the
> results of their previous software practices and create a database for
> monitoring throughout their lifecycle all future software projects.
Good idea. I have an on-line copy of "Department of Defense Guide for
Managing Information Technology (IT)as an Investment and Measuring
getting projects to follow it is another matter. You might also look at
the Software Program Managers Network (http://www.spmn.com). The control
panel there is a very interesting idea, and it's time may have come. The
little books are wonderful for dropping on the desks of project managers
that just don't get it. ;-)
> I like your data. However, it is still anecdotal. It is possible to do a
> cross-over study. However, as we agree, it can not be blind.
Unfortunately, MITRE is in a special situation here. MITRE, on behalf
of the DoD has access to lots of real data, but with very stiff
restrictions on sharing that data with ANYONE. So while I can share data
collected before I came to MITRE, and there is some data that MITRE has
published with the permission of the government (mostly by Victor Basili
and others relating to NASA work for a center now part of the University of
Maryland), a lot of good data you want sits on shelves gathering dust.
However, good data is out there for the taking from other sources. For
example, the Ada Europe conference in 1990 had several good papers on
exactly this subject. (Proceedings published as Ada: Experiences and
prospects by the Cambridge University Press.) To quote from one article
there (An Ada Case Study in Cellular Telephony Testing Tools by Harr
Doscher of Motorola): As can be seen from Figure 5, the M5000 has had 96%
fewer post ship date defects reported than the C effort....productivity on
the M5000, measured in lines of code generated per day normalized to lines
of assembly language, was roughly twice that of projects done in C.
So again to sum up, you should expect some productivity gains from
using Ada 83 during development, but the startling results come after the
software is delivered. Ada 95 provides much more productivity, and we
haven't seen any signs of quality decreases, but that may be due to
programmers not yet using all the new (to Ada) features of Ada 95.
Why expect decreases in quality? Because in every--anecdotal,
agreed--case I know of where people tried to use a binding to an existing
class library in other than small projects, they have complained of quality
problems, not just with the library, and switched to a less OO approach. I
am hoping to do a small study next year on what an Ada class library should
look like to avoid these problems.
Robert I. Eachus
function Message (Text: in Clever_Ideas) return Better_Ideas is...