🚀 go-pugleaf

    From: HOROWITZ at USC-ISIF
    To:   apollo at MIT-AI
    We are in the process of considering which personal workstations 
    to purchase.  Our current leader is the Xerox STAR as we believe
    Xerox will release Mesa and Smalltalk on the machine.  We find 
    the Lisp machines too expensive.

This message illustrates what I think is one of the really important
points in our discussions so far.  The above message indicates that
Horowitz's group has, for some reason, concluded that they want to buy
"personal workstations".  Having decided that, they are now considering
which ones to get.  Unfortunately, Horowitz does not tell us anything
else, like, WHY they want "personal workstations", and therefore it is
difficult to judge whether a Star or a Lisp Machine is better.

Since he mentions Mesa, presumably he doesn't just want to do the OI
that the Star can do; he wants to write programs.  Indeed, Lisp Machines
are more expensive, and maybe are too expensive for anything he wants to
do.  However, the Star has three disadvntages as a program development
system.  First of all, Xerox has not said for certain whether they will
ever release Mesa and the associated program development software; even
if they do release it, they probably will not be interested in
supporting it.  Their real market for Stars is the OI market, out in the
real world, and not a few random researchers here and there.  A
man-month put into improving the OI software is far more valuable to
Xerox then the same man-month put into maintaining the program
development system to the point where it can be released to remote
users.  As far as our discussions on this mailing list go, it should be
kept firmly in mind that Xerox's interest in the Star is as an OI
system, not a Mesa program-development system.  Secondly, even if you
did have Mesa and its program development system, the Lisp Machine's
language and program development system are far better anyway.  Thirdly,
the Star's processor doesn't have virtual memory, and so has very severe
address space limits.