[log in to unmask] said:

  > If an 'experiment' trying to compare a C and an Ada team resulted
  > in the C team creating a new CASM (C Ada Subset Macro) language
  > that was better than either C or Ada, then we'd surely like to
  > hear about it.  If it was better than C but worse than Ada (or
  > the opposite) then it should prehaps replace whichever it beat,
  > and if all three were in a dead heat then it's evidence of 'no
  > difference' - at least for that application domain and set of
  > programmers.

   The result was definitely Ada better than Ada in C which was much
better than "ordinary" C.

   The one that killed me though, and I still remember it well, was
when the support people had to give their evaluation.  (At this point
six subsystems have been out in the field in several thousand
machines, three subsystems in Ada, two in C, one in Pascal.
Everything else, except for a few FORTRAN subroutines, was in
assembler.) Pages and pages of charts on more bugs in C than Pascal,
but Pascal bugs and assembler bugs took longer to find, etc., etc.

   "What about Ada maintenance costs."  (Asked by the VP who headed
the division.)

   "We don't have any data."

   "Why not."

   "No STARS (bug reports) have been closed."

   "How many STARS are open on the subsystems written in Ada?"

   "None."

   Sound familiar?  (And it wasn't stupidity.  This guy knew was Ada
would do to the size of the maintenance staff, but there were too many
people around who knew the answers for him to lie.)

 >    Clearly such an experiment, especially with a 'Western Electric
 >  effect', and across multiple apps and programmer sets, is not
 >  something to be done definitively by a few folks, even at a large
 >  company.

   I assume this is the same as the Hawthorne effect?  (Based on a
stud which showed that almost any change improved productivity, if it
showed management interest in productivity.)

 > Even non-experimental collection and analysis of data on
 > past projects is a big job.  But if we're looking at a likely
 > DOD-wide, or nationwide, or world-wide, benefit on the order of
 > $10**8 or 9, then it's well worth a *substantial* investment.

   We try.  MITRE, NASA and the University of Maryland have a lab
which has been doing this sort of data collection for years.
Unfortunately, the data is on langauge effects only, not programming
methods, and see story above.  The more you prove that (well
developed) Ada code doesn't have bugs--unless you change the
requirements--the more trouble you get Ada into.

   Another similar story.  There was a project a few years back, where
we (MITRE) wanted to insure that the code was VERY trustworthy.  (If
you have seen the movie War Games, the HAC/RMPE portion of REACT is
Whopper.)  It was written in Ada by a contractor selected after a
SEE--Software Engineering Exercise--which was designed to show
management and staff alike how important doing it right was.  The
bidders were required to write code to match a package specification
we provided, and to do error correction on EAMs.  One of the "test"
messages read "Let's do lunch." if you followed the rules.  If you
didn't follow the instructions carefully, you were likely to end up
with an extra 'a' in the last word.

   As far as I know, the Air Force still hasn't decided when to
transition to "organic" maintenance on the system.  They are supposed
to do that to reduce costs, but first they have to have some costs to
reduce.


                                        Robert I. Eachus

with Standard_Disclaimer;
use  Standard_Disclaimer;
function Message (Text: in Clever_Ideas) return Better_Ideas is...