Subject: Re: Origin date of Wallman's COMPOSE
From: Tom Van Vleck 
Date: 22/10/2011 22:29
To: Kristaps Dzonsons


On Oct 22, 2011, at 1:11 PM, Kristaps Dzonsons wrote:
> > I'm researching the history of the UNIX manpage, which of course
> > extends to CTSS via Saltzer's RUNOFF.  After speaking with Saltzer
> > to fill some holes, I found mention of the COMPOSE utility in the
> > source archives.  Unfortunately, the author, Edwin Wallman, seems to
> > have passed away.
> > 
> > My question: would you happen to know when Wallman began working on
> > COMPOSE?  The PL/1 source and info manual are inconsistent in terms
> > of dates.

I took over as Ed's manager in Phoenix in mid-1978. He was already
working on compose (the correct name is lower case) by then.  I glanced
quickly at the source at MIT and saw no dates at all.. the archive dates
are all 1985, which is not helpful.  I believe that Ed started work on
compose in about 1976, as part of a project to produce photo-typeset
manuals and eliminate dependence on BCPL. 

Ed died in 1996 at the age on 68.

The Multics compose command, in the 1978 time frame, was very like CTSS
RUNOFF and Joe Ossanna's runoff for Multics written in BCPL: that is, it
read an input file and wrote a formatted document.  Saltzer's RUNOFF was
one-pass.  The Ossanna runoff, and subsequent descendants, could make
more than one pass over the input.

The compose command could drive different kinds of typesetting output
devices.  Users asked for new features all the time, and Wallman put
them in and fixed bugs. He had to fix a lot of bugs. Compose was our
second generation text processor, replacing the BCPL-coded runoff
command; we had hoped that it would have fewer bugs, but it seemed to
have more. Ed would fix a bug here and the fix would cause a bug
somewhere else. Adding a new feature in one place would cause some other
feature to stop working.

It occurred to me that the real problem was that compose was written in
PL/I, which was too low-level a language for the problem we were trying
to solve. If compose had been written in "compose-tran," a hypothetical
text formatting language, we could have tested it in two phases: first
test the compose-tran interpreter, and prove that its atoms worked
correctly no matter what order its operations were invoked in. Then,
test the compose program written in compose-tran. This strategy would
reduce the number of tests required from astronomical to merely huge.

Hope This Helps,
tom