Doug McIlroy recently sent me an email saying
"It occurred to me recently that 1959 is the 50th anniversary
of the most significant computer science talk I ever attended."
He went on to describe it:
Just fifty years ago, John McCarthy circulated a notice
that he would be giving an informal talk that he thought
would be of interest. I drove up to MIT from New Jersey to
attend this seminar of typical length in a typical classroom.
But the seminar was anything but typical. It was a revelation.
In one session at the blackboard, John introduced Lisp—all
you could do with car, cdr, cons, cond, lambda, and recursion.
Recursion had no place in mainstream programming at the time,
nor did lambda calculus. Only two years before, I had sat in a
coffee-room discussion of what it would mean for a subroutine
to call itself. Questions raised but unanswered were whether
recursive instances deserved to be deemed the "same" subroutine,
and, if you could do it, what good would it be? It turned out
you could do it: I programmed it for the IBM 704. Given the
challenge, the now standard stack solution arose inexorably.
But the question of what it was good for remained.
In the course of the lecture John introduced the usual basic
list functions like copy, append and reverse (quadratic
and linear), as well as tree manipulation. He went on to
higher-level functions, demonstrating maplis and lambda.
By the end of the hour he had put together a powerful little
toolkit of functions which he used in his finale: symbolic
differentiation of univariate expressions.
There it was—functional programming ex nihilo. McCarthy
acknowledged IPL V and recursive function theory, but the elegant
and practical face he put upon these antecedents was a work
of genius. Nobody would ever again wonder what good it was to
allow functions to call themselves. And it was all so clear one
could go home and build it oneself without any instruction book.