Review available....(OK, it's HERE!)

[Brad Miller <miller@UR-ACORN.ARPA>]

 From the immediate deluge of mail I got, I guess I should have just posted
this. I mistakenly thought there'd only be 2 or 3 of you all interested...
Anyway, here's what happened. We have 1 3670 and 2 3640s here at the
University of Rochester Computer Science department. For various reasons,
(NEAIC grant) we aquire 3 TI Explorers. Since I am the one locally responsible
for getting the machines integrated into our environment (several vaxen and
suns running UNIX 4.2) and we want to keep using the Symbolics 3670 as the
lisp machine fileserver, I end up with a 3640 and an explorer on my desk for 2
months (in fact, they're both still here - which is OK with me! :-)

So now I face the task of letting our users know which machine they should use
for their own particular purposes, and as a result of that, I wrote the
following short review. I have edited it slightly to make sense to the world
at large. I don't go into excrutiating detail, but I would be happy to supply
additional information to anyone who needs it. Quick background about me (as
you are possibly wondering if my opinion means anything -- probably not, but
here are my qualifications anyway): I have an MSCS, and have been supporting
the Symbolics here for about a year now. I am the full-time staff lisp hacker
for our department, and besides supporting the lisp machines in general, do NL
research support. I hope to get my PhD in CS here in about 3 years, part time.

*** Based on Symbolics release 6.1 vs. explorer release 2.1

Note that both machines are sufficiently similar that there is a lot of carry
over between them. So time spent in learning one machine isn't totally lost
when applied to the other. If you are writing code you want used by everyone
(or because you don't think you can count on getting the same machine for your
next session) you should pay some attention to the document distributed
earlier 'Common Lisp the Style' which gives some guidlines on differences
between the two vendors, and some local stylistic conventions. Unfortunately,
it was based on a TI document which was current for release 1.0, and we are
running 2.0+.

The main thing to remember, though, is that if you are writing software to be
general, just stay with CL constructs, and be sure to test your software on
BOTH machines. The make-system function is similar on the two machines, but
NOT IDENTICAL. 

Important little things to know:

The site directory for the symbolics is >sys>site.
The site directory for the explorers is >sys>explorer>site (if you don't know
what a site directory is for, don't worry about it - squirril this message
away so when it comes up, you'll know where to look).

Anything that wants to take advantage of fancy user interface features should
not count on being portable between the two machines - the user interface will
have to be REWRITTEN.

Symbolics flavors in 7.0 is not downward compatible with the explorer flavors,
but explorer flavors are almost upward compatible with Symbolics flavors, so
use the explorer version if you want less work to do. (Though Symbolics new
flavors look like they will be a substantial improvement for debugging)

Explorer lisp/flavors doc is MUCH better than the symbolics. Plus, the same
lisp will usually work on both machines.

OK, on to the review. For the following, an arrow indicates the vendor that
has the better item, and the rating is one + for minor advantage, not enough to
buy the machine on, ++ a major advantage and +++ a critical advantage. Note that
others will likely disagree with the following, but I tried to be objective
(e.g. I don't care how easy a machine is for a new user, but I still rated the
machines on that quality). Some of these items may only be of interest to the
labcomm.

EXPLORER					SYMBOLICS
		      Hardware, packaging,
	+       <     self-tests, ease of     (no self-tests, or smart boards)
		      isolating hardware
		      problems.

    (less pixels,	  Display			>    +
     output looks
     like dot-matrix
     printer)

    (non-generic	  NETWORKING			>    ++
     interface)

    (clumsy)		  NAMESPACE			>    +

    (none)		  EPHEMERAL GC			>    +++

    (4 char limit,
     non-generic	  DISK MAINTINANCE		>    +
     interface)

	+++     <	  Initial Ease of Use	  (no suggestions, less attention
			  and Overall User	   paid to make keyboard and menu   
			  Interface		   interface consistant, poorer
					           approach to readtables in
						   windows, etc.) 

    (can't create
     arbitrary file	  DIRED/FSE			>    + 
     properties, no
     FSE)

	++	    <     ZMACS			  (Explorer allows >1 editor PROCESS)

    (Almost worse	  MAIL				>    ++
     than UNIX!)

	+++	    <	  Full Sources		  (many unavailable at any price)
    (only toolkit
     sources un-
     available)

    (glossary		  Full Online Doc	        >    ++
     only)

	++	    <	  Completeness and	  (little internals doc, poorly
    (poor index)	  style of Doc		   written other doc)

	+	    <	  Performance metering tools

	+	    <     Debug tools	          (no MAR)

    (128M limit)	  Max Paging space		>    + (~1gig)

		      Speed (unrated - will
		      change in future. Right now,
		      Roughly speaking, an Explorer
		      is 50% of a 3640/3670.)

		      Lisp (unrated - virtually
		      identical)

		      My totals:
	+  3						5  +
       ++  2						3  ++
      +++  2						1  +++


With release 7.0 of the symbolics software, the above is subject to change
(new flavors system, improved window system including 2 way scrollable
windows, new doc, new performance metering tools, etc. will improve the
symbolics score) 

Release 3.0 of the explorer also promises improved mailer, improved
networking, and the possibility of better GC. At some point soon, they also
promise to support Sun NFS.

So overall, you can see they are close, and you need to look at what is
important to YOU to pick the right machine. My guess is that if you do not
plan to spend a LOT of time at the machine, the Explorer suggestions system is
worth the price of admission - you will be able to do useful work on the
machine a lot quicker. The doc is much better, though it is not as accessible
as the Symbolics, since it is not online and must be searched manually.

What I have found in my day-to-day use with both systems on my desk:

First, remember that we are using a symbolics file server / mail server for
all our lisp machines.

I tend to use the symbolics document examiner for looking up lisp forms, etc.
The index for the explorer manuals is only slightly above useless, and when I
need to find something that isn't compatible with the symbolics anyway, it can
take a while.

I use zmail on the symbolics. The explorer's mail system is so wretched that
I'll open a telnet window to UNIX and use that mail system before I'll use the
2.1 mailer. One problem is that the symbolics can be made to easily understand
the arpanet host layout (it can download a host table from NIC, and TI only
supplies unsupported software to do something similar (which is even
documented as something that will crash your machine if you use it on your
entire host table). Since the explorer will not talk to the symbolics
namespace, and the mailer thinks that addresses in mail should be in the
namespace, rather than letting the mail server figure it out, non local mail
will not work.

I use the editor on the explorer. Overall the symbolics editor has more nice
features, (like most variables can be set on a buffer, a major mode or
globally - on the explorer one usually can only set things globally, and for
non-prexisting editor processes as well) but the explorer has one thing that
is a major advantage. Each editor window has it's own process. So, if you are
in the middle of a search in one editor, you can switch to another and do
something else. On the symbolics, when you type to your new editor window,
chances are that the characters are being accepted by your original window if
it was in the middle of some command, or suspended, or whatever.

While the symbolics networking is better in general, I tend to use the vt100
simulator on the explorer. My only complaint here is that I wish it were small
enough to fit two on a screen, but it's a vt100 that really WORKS like a vt100
for all practical purposes.

 From a site maintainance point of view, I find the FEP filesystem on the
symbolics much easier to fiddle with than the disk labels on the explorer.
Symbolics lets you do incremental world saves as well, which allows more
students to save their own worlds on a particular machine without having to
connect all our machines to their own Eagle.

I find that due to ephemeral GC on the symbolics (I don't bother with normal
GC since it's such a performance drag - except overnight I do a
gc-immediately, if possible, on both machines) I can run my world about two
weeks before I have to reboot. The explorer would need rebooted about every
day, if it wasn't quite so stupid about what's undoable at logout, so I
usually have to reboot 2x/day.

Bottom line is that they both have good points and they both can learn from
the other's work. In my opinion.  The only real comment I've gotten from the
local user community so far is that the explorers seem friendlier. My
suspicion is that for every engineering hour both companies had, symbolics
spent it on adding new features, and TI spent it on making the CADR machine
easier to use....

Hope this is helpful. If I have any facts wrong, please correct me, but most
of this is opinion, and I told you it was opinion, so if we have religious
differences...sorry.

Brad Miller

------
miller@rochester.arpa
miller@ur-acorn.arpa