Parity with conventional programs

[boyle@antares.mcs.anl.gov]

It seems to me that one barrier to the wider acceptance of functional
programming methodology, especially by scientists and programmers
interested in applications, is the perception that functional programs 
*necessarily* sacrifice efficiency for simplicity, clarity,
adaptability, etc.

I also believe that the time is ripe to attempt to demonstrate that
this is NOT the case--that one can write in the functional style and
still obtain programs that execute as fast (or faster) and that use as
little storage as do well-written procedural programs on the same
hardware.  

I would call such a demonstration a "demonstration of parity" between
functional and conventional programming.  I believe that several
demonstrations of parity on programs for solving real problems would
remove a large barrier to the use of functional programming, and
encourage applications scientists and programmers to consider fp on
its (obvious) merits.

How about some discussion on this topic?  Does anyone out there agree
or disagree with these sentiments?  Is anyone out there working on
demonstrations of parity, or better yet, has anyone achieved some?

Terence Harmer (now at Queens University, Belfast) and I have been
working toward this goal at Argonne over the past year using
automated program transformation techniques.  We are within sight of
the goal for several specifications for numerical and non-numerical
problems of practical interest.  We express these specifications in
pure Lisp (lambda calculus); some use higher order functions.  We
transform them into Fortran or C programs for sequential, vector,
and/or parallel computers.  We'd like to hear from others working
along similar lines.

Jim Boyle
boyle@mcs.anl.gov