Re: Lisp history: IF, etc.

On 4/16/2024 1:22 AM, Lawrence D'Oliveiro wrote:

On Mon, 15 Apr 2024 22:58:21 -0600, Jeff Barnett wrote:

I met him in the early/middle 1960s when DARPA gave us a contract to do
a thing called Lisp 2 - a Lisp system with extended language facilities
and borrows from Algol.

That project was considered a “failure”, but I wonder why? Did it turn out
that getting rid of the (ahem) quirky Lisp syntax in fact got rid of some
of its expressive power, too?

Paul Rubin discusses this in another reply to your question. But there
where many issues. In the first place, there was a full normal prefix
Lisp interface to the system - compiler, assembler, debugger, reader,
etc. In the second place, there was a full meta-compiler setting in the
Lisp2 system. This meta-compiler was good enough to generate some of the
early Jovial compilers developed and used at SDC and could do
sophisticate things, e.g., remove left recursion from a grammar. In
other words, you could use prefix, a complimentary infix, or roll most
(but) not all interface languages that made you giddy.

The problem in my mind was simply lack of computer resources. The
implementation on the ANFS-Q32 had something like 48K 36+ bit words.
That simply wasn't enough so some of the fellows devised a swapping
scheme: the function linkage cliche - the common code invoked by the
function call sequence -- noted what was called and frequency. It also
noted whether the routine was in memory; if not it tried to move it to
memory from external storage; if there wasn't enough room, current
memory resident routines were moved or excised (and written out if
necessary) being selected using the frequency information; pieces of the
GC may have been used to do such tasks as compacting program storage to
make a hole big enough to do the swap in. In other words, we were doing
swapping a binary program at a time. This, as you can imagine, was a
Royal Pain In The Ass (RPITA).

So DARPA (or was it still ARPA?) says they wanted SDC to do something or
other with massive databases on IBM 360 machines so they are going to
get us a better machine. The machine turns out to be about the same size
as the Q32. Further, the next goodie a 360/370 with virtual memory could
be significantly better. However user virtual address space was to be
limited to about a megabyte - not enough to consider a copy-collect GC.
I, and several others had enough of this sort of development in limited
resources. We were motivated by experimenting with language forms and
ways to implement them, not research in how to scrounge memory while
losing sleep and forsaking a social life. Not every one agreed with the
decision to shut it down but most did. So RPITA->RIP.
--
Jeff Barnett