From chemistry-request $#at#$ ccl.net Tue Dec 10 12:09:49 1991
Date: Tue, 10 Dec 91 17:00:41 GMT
From: jnh- at -iris41.biosym.com (Jon Hurley)
To: chemistry%!at!%ccl.net
Subject: FORTRAN vs. C
Status: RO

OK,
  I normally steer away from these political issues, but I have had
enough of these comments favoring FORTRAN.

  In all these messages about structured programming, people seem
to be missing the point.

Sure it is important that individual subroutines be written clearly,
with indentation, meaningful variable names, straighforward control
structures and suitable comments.

But the crux of writting code that is extensible and easily
maintable is modularity (i.e. using subroutines (or methods on
objects or any other object oriented analog) that do a well defined
thing and work on a known set of variables).

FORTRAN does not have structures (although a few extensions have
records). Since most reasonably complex (as scientific codes usually are)
programs have dozens or hundreds of variables, the only way of passing
the data into routines in FORTRAN is:

  1) Use common blocks to keep global data.

  2) Have subroutine calls with literally dozens or hundreds of arguments.

The second approach is clearly infeasible and the former approach
guarantees lack of modularity. For instance, let's say you want to
modify a program to support two separate molecules instead of one. With
FORTRAN global common block variables, you have no recourse but to
rewrite the whole program; with C where you pass a pointer to a molecule
structure into a routine, all you have to do is rewrite the calling
routine.


Yes FORTRAN is great for computational kernels. It has highly optimized
compilers, and the form of the code strongly resembles the equations that
it represents. But for data manipulation and extension of code based on
pre-existing modules, and as I can attest the VAST majority of commercial
computational code is just that, FORTRAN, without structures, is totally
inadequate.

	Jon