A suggestion for the Lisp Machine community to discuss

[Martin David Connor <MARTY at MIT-AI>]

I was not really sure who I should send this message to; It relates
to an issue that would seem important to most everyone, and especially
to those who consider Lisp Machines important to their research.  I
wish to beg the pardon of anyone who is inconvenienced by this
message.  I am leary of mailing lists for this very reason.

I thought now was a good time to raise certain questions that we will
very probably have to be deal with in the near future.  What follows
is my analysis of the current Symbolics Inc. agreement with the EE/CS
department, and how I feel we could benefit froma similar agreement.
     
     Date: Thursday, 18 March 1982, 14:29-EST
     From: The Software Staff of Symbolics at SCRC-TENEX
     Subject: Release of Symbolics software at MIT
     To: INFO-LISPM at MIT-AI
     
     Through an agreement between Symbolics, Inc. and MIT's EE/CS department,
     Symbolics' Lisp Machine software is now available for use by MIT users.
     
     Symbolics is making available all software that we maintain to MIT
     users.  This includes the Symbolics File System, Tape System, Canon
     Laser Printer Software, and Font Editor, as well as any other software
     that is being developed for distribution to Symbolics customers.  New
     releases will be made available to MIT at least as often as to Symbolics
     customers.  In practice, MIT will probably be given releases more
     often than this, to provide faster response to trouble reports.

I assume that this means that they are giving us the software at no
charge. What are giving them is possibly the best test market and
debugging user community in the world.  The tradeoff may be worthwhile.
     
     Symbolics systems can be distinguished because their system version
     numbers are greater than 200; the current system being released is
     System 204.  The master copies of the sources from which Symbolics
     systems are created are kept at Symbolics; copies of these sources are
     being kept on MIT-EECS (the TOPS-20 system of the EE/CS department,
     which is providing the disk space to hold these sources).  Please note
     that MIT-EECS is a very heavily loaded system, and its routine use
     should be avoided.  We expect lisp machine users to be considerate
     in this regard.

This begins to foreshadow my suggestion.  As an AI Lab researcher I
would not want to have to depend on a system outside the Lab for
accessing system sources.
     
     The sources are on the LIB: structure of the MIT-EECS machine.  The
     Symbolics systems are configured to know that all system sources live on
     this machine; if you access a system source file with the Meta-.
     command or otherwise, the system will connect to MIT-EECS to read in the
     file.  This way, Meta-. will find the correct version of the source,
     corresponding to the software in the Lisp environment.  Furthermore,
     if you do LOAD-PATCHES, the patch files will be searched for on MIT-EECS.
     
     When Symbolics bands for MIT are booted, they say "EE" instead of "MIT"
     in the last line of the herald message.  This is because these bands use
     a different set of site-configuration files, in order to make the SYS
     host be EE instead of AI.  This convention may change in the future.
     
     The MIT-EECS machine requires you to log in before you can access files.
     Anyone may log in by using the user name "Anonymous", without giving the
     file job a password; this will give you read-access to all of the
     sources.  Of course, if you have your own account on MIT-EECS, you may
     use that; all users have read-access to all the sources.

This situation is also far from ideal.  As an AI Lab person I feel a
bit restricted to have to ask for permission to access sources.
     
     We want to provide the same fast response to bugs that MIT Lisp Machine
     users are accustomed to.  Therefore, bug reports sent to Symbolics from
     MIT will be answered as soon as possible, and patch files will get moved
     to MIT-EECS promptly.
     
     In order to make it easy for MIT users to introduce bug fixes and
     changes into the Symbolics software, system maintainers at MIT may
     modify the copies of the sources that are stored on MIT-EECS.  Symbolics
     will normally merge such changes into the master source files at
     Symbolics; however, we reserve the right to reject any changes we feel

The right of Symbolics to refuse changes would be bothersome, but the
fact that there is a "site" file which I am told can contain patches
which are very specific to a particular site would seem to help
matters.  

     are incorrect or inappropriate.  When new releases are provided to
     MIT-EECS, sources will be copied from Symbolics to MIT-EECS, and any
     changes that were rejected will disappear from the sources.  (The source
     files on MIT-AI are merged with Symbolics sources only infrequently.)
     
     The administrators of the MIT-EECS machine have decided to control
     write-access to these sources, on a per-user basis.  If you feel you'd

op. cit. (already cited)

     like to have access to modify Symbolics sources stored on MIT-EECS, send
     mail to the list LISPM-ACCOUNTS@EE.  Please note that this control is
     being maintained by MIT-EECS personnel, and not Symbolics personnel.
     
     This software is provided to MIT under the terms of the Lisp System
     License Agreement between Symbolics and MIT.  MIT's license to use this
     software is non-transferable.  This means that the world loads,
     microloads, sources, etc. provided by Symbolics may not be distributed
     outside of MIT without the written permission of Symbolics.

This is a somewhat problematic point for some people.  I do not have
very big problems with it.  I assume that we can give application
programs freely to others, and as for the system, I think it a
distraction from most people's research to maintain a LISP machine
world.  The time it would take, and the hacker hours would not justify
it, not to mention that those best at it are not here.  If I am only going
to support a Lisp Machine world, I might as well get paid well for it.
As a hacker, I can say that I am here to do things that I could not do
at most companies.  I would suspect that many people have similar
motivations in their research.
     
     System 204 already exists in many forms on MIT machines; some of these
     bands have different systems loaded.  People who wish to receive advice
     on which software packages would best suit their needs, or who wish
     assistance in installing system 204 should send mail to
     BUG-LISPM-SCRC@SCRC.

     System 204, and all associated software, is now released to users.  It
     comes with Symbolics microcode 865, and ZMail 42.  There are a number of
     INCOMPATIBLE CHANGES in this software.  The system release announcement
     can be found in EE:LIB:<LISPM.DOCUMENTATION>SYSTEM-204.RELEASE and
     in AI: LMDOC; SYS204 MSG.

--------------

Now to my suggestion.

The new AI machine OZ, will have 1.4 gigabytes of disk space. 
I suggest that we store the Symbolics sources there.  
I think this idea has the following attractive features for the Lab:

- Each of the major groups of Lisp Machine users at the Lab may
  choose what software they wish to run.  The sources can be on OZ,
  and can be accessable to those who want them.  

- This does not negate the possibility that we could have other Lisp
  Machine system software on the same machine.

- We would not have to depend on the heavily loaded EE machine for
  storage of software.

- We would not have to ask permission of EE people to access the
  sources. I would not expect too much of a problem in this regard, but
  I see no reason for this indirection if we can just as well have
  our own sources.

- The AI Lab will quite likely be the major Lisp Machine user on
  Campus, and will have a very strong need to have the sources
  available.

What I would propose is that we seek an agreement with Symbolics
whereby they will store their sources at MIT-OZ, and for our
consideration of testing software and reporting bugs and allowing them
access to the software we write as well as using our computational
resources, they will incorporate our patches and at no charge will
supply any software that they would supply to their normal customers.
I would also like to see some words to the effect that we may give
away certain specific kinds of software that we develop to whoever
wants it.

That is my basic suggestion.  

I see the Lisp Machine as a tool, and thus I think we should use them
rather than build them. Use them to create new ideas.  

This represents my current thinking, and I wish it to be clear that
this is a suggestion.  I hope it will serve as a basis for clearer
more rational discussion between all the people who are affected by this
issue.

Hacking Tomorrow;

Marty