[PREV - BLADES_OF_PERL]    [TOP]

MODEST_PROPOSAL

                                               April 11, 2007
                                              
Computer Geeks argue endlessly with each other         
about the virtues of different computer languages,     
programming methodologies and so on.                   
                                                       
These arguments revolve around anecdotes              
and hand-waving to the point where many        CASE_STUDY_1
people dismiss them as "religious arguments".
                                                     
And yet, these are about subjects of much practical           
significance.  Someone planning a large software project      
is about to invest a large sum of money on a scheme whose     
success is far from certain.  Any technique of minimizing     
that risk would be extremely valuable.                        
                                                              
I hereby propose that experiments should                     
be conducted to settle these questions.     
                                                      
What is needed for this is a number of teams of       
volunteer programmers all willing to work on an       
assigned problem in parallel with each other, with    
each team using a different technique of solution.    
                                                         
You take the results of these teams efforts, and then    
repeat the experiment with new teams, until you get      
enough data for statistical significance.             
                                                         
Coming up with controlled experiments to               (March 6, 2004)
measure qualities such as "maintability" would
certainly not be trivial, but it's not beyond
the resources of a large university.

To get things started, it would be enough to
have a single interdisciplinary partnership
between two professors, one in computer science
and one in psychology.  My presumption is that
they would have access to a pack of student
volunteers willing to play guinea pigs in
return for senior project credit.

Many claims are made about software languages,
it would seem that if you were a "computer
scientist" you would be interested in trying to
experimentally verify some of the claims.

      We've been hearing about how Java
      code is more maintainable for about
      a decade now, and massive amounts of
      industry investment has been made in       How many Java projects
      that language.  Wouldn't you think         tend to die on the vine,
      that someone would be interested in        versus Perl projects?
      investigating how maintainable it
      really is?                                    Wouldn't that be
                                                    worth knowing?

                                                    Among perl programmers
                                                    it's a truism that you
                                                    need more Java programmers
                                                    than perl to accomplish
                                                    the same task.  You hear
                                                    factors ranging from 3-10.
                                                    Wouldn't it be worth
                                                    knowing what the number
                                                    really is?

                                                           (March 14, 2007)
   "But it's exceptionally difficult
   to measure software developer
   productivity, for all sorts of
   famous reasons."  -- Steve Yegge

           I think the main reason is
           that no one tries to do it.
                                            Except for maybe:

                                              CASE_STUDY_3



--------
[NEXT - SOFT_ART]