Subj : Re: Compiler and an interpreter
To   : comp.programming
From : Vesa Karvonen
Date : Fri Aug 05 2005 01:38 pm

CBFalconer <cbfalconer@yahoo.com> wrote:
> Vesa Karvonen wrote:
> > 
> ... snip ...
> > 
> > Psychological studies have indicated that short-term memory has a
> > capacity of 5-9 "items" (http://www.well.com/user/smalin/miller.html).
> > Code in first-order languages typically has lots of little details,
> > particularly details related to control flow and resource management,
> > that can not easily be abstracted away in those languages. My own
> > experience and intuition says that those details quickly eat all the
> > available short-term memory and ruin productivity.

> It's not the language, but the coding breakdown.  By doing a
> suitable top-down design you remain at the higher level well into
> the design, and the lower levels are exactly what the design needs.

I disagree. First of all, I have written hundreds of thousands of lines of
code in first-order languages and I know that building certain kinds of
abstractions in those languages is much more difficult than in
higher-order languages.

More to the point, my experience is that pure top-down design is not a
good approach to design and construct software. Some bottom-up design is
always necessary, but often much of the bottom-up work has already been
done by those providing languages and libraries. I consider languages and
libraries as artifacts of bottom-up design. Rather than starting with a
specific application, you start with a problem domain and gradually build
abstractions to allow concrete applications to be written easily. Here is
a link to an essay from an advocate of the bottom-up approach

  http://www.paulgraham.com/progbot.html .

I think that one way to gain intuition into the relative merits of
top-down and bottom-up programming is to think of the design space of a
program as a triangle (or pyramid):

           Top

           /\
          /  \
         /    \
        /      \
       /        \
      /          \
     /            \
    /              \
   /                \
   \/\/\/\/\/\/\/\/\/

         Bottom

The idea is that the area (or length of a horizontal section) corresponds
to the amount of "details" at a particular level of abstraction. There are
much more details at the bottom. High-level languages cut out a large
section of the bottom of the triangle. Domain specific combinator
libraries stake out large blocks of the bottom of the triangle.

-Vesa Karvonen

.