Scheme compilers

[reti@ai.mit.edu]

    Date: Thu, 10 Sep 1998 12:31 EDT
    From: Mark Dulcey <mark@ziplink.net>

    On Thu, 10 Sep 1998, Kragen wrote:

    > On Wed, 9 Sep 1998, Mark Dulcey wrote:
    > > Most LMI CADRs and Lambdas were sold with complete source code (all the
    > > way down to the microcode!). If you can find one that is still in working
    > > condition, it will probably have the source.
    > 
    > If you can find one in working condition, I'm sure there are computer
    > museums that would love to have it.  In fact, I'm sure there are
    > computer museums that would love to have a broken one.
    > 
    > How reparable are they?  I had a (much cheaper) machine of the same
    > vintage that had only a couple of common failure modes (the keyboard
    > encoder getting zapped with static, and a power-supply transistor
    > shorting out and smoking some power-supply resistors), both of which
    > could be fixed with a simple replacement of a couple of parts costing
    > less than $10.
    > 
    > I know the LispMs all used lots of custom chips.  Did the custom chips
    > die often?

    The LispMs had a lot of PLAs and the like in them (not true custom chips).
    They aren't all that prone to failure, but it only takes one blown one to
    make the machine fail, and there isn't anybody left around with the
    information needed to burn new ones.

Not true. Symbolics is still alive (though just barely) and still has all the
engineering information about the systems from the 3600 on (as well as spare
parts, etc.).  But a quick check for the LM-2 information revealed that it was
no longer on line.

    The other problem is that the LMI systems all used a huge wirewrapped
    backplane. Since the newest Lambda would be over 10 years old at this
    point, backplane reliability is likely to be a problem. Some of the plugin
    cards were wirewrapped as well - same problems apply there, too.

Only a few of the early 3600's from Symbolics were wire-wrapped, all the
later machines had printed circuit boards (as did Symbolics' version of the
CADR, called the LM-2).

    Finally, there is the issue of the hard disks. The Fujitsu Eagles were
    pretty good drives, but they ARE old at this point, and may have died as
    well. On the other hand, if you could find a non-working LispM with a
    working hard drive, you might be able to connect the drive to some other
    sort of computer and get the data off it. You'd still be left with the
    task of writing code that could understand an LMFS file system, which
    probably isn't documented anywhere other than the unavailable source code,
    but I don't think it was all that complicated a file system, so it might
    be possible to figure it out, or find someone who knows how it worked.

It wasn't all that complicated, and is still in active use and maintenance.

    There is one more possible way to get source code - I don't know whether
    anyone has ever pursued this one. At the MIT AI Lab, the sources weren't
    stored on the LispMs; they were kept in a central repository on the lab's
    DEC-20 system. That computer is long gone, of course, but they made
    regular backups of it on 9-track magnetic tape. Whether the tapes from 10+
    years ago still exist, let alone can be read, is another question...
    Unlike the LispM disk drives, though, this won't get you a built binary
    image; those were only stored on the LispMs. (A LispM booted by loading
    memory from a "band" file on the disk; it contained a binary image of the
    contents of memory at the time it was dumped.)