Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!mnetor!seismo!lll-lcc!mordor!ehj
From: ehj@mordor.s1.gov (Eric H Jensen)
Newsgroups: comp.lang.lisp
Subject: Re: frustration with LUCID on Sun 3
Message-ID: <8465@mordor.s1.gov>
Date: Tue, 5-May-87 17:02:14 EDT
Article-I.D.: mordor.8465
Posted: Tue May  5 17:02:14 1987
Date-Received: Thu, 7-May-87 05:21:49 EDT
References: <611@savax.UUCP>
Reply-To: ehj@mordor.UUCP (Eric H Jensen)
Organization: S-1 Project, LLNL
Lines: 38


I share your frustrations.  Paging on an 8 meg 3/75 is outrageous.
Things get satisfactory with at least 16 meg of real memory. I suspect
that Sun's paging algorithm may be partially to blame here - I have
seen brain dead paging behaviour on other large programs.

The lack of condition handling is also a real lose - I hope they
supply something soon.

As for an environment, I wrote an extensive sub-process interface for
GNU emacs some time ago and that allows command history, installing a 
new defun, runing the lisp process on a variety of servers and other
small features.  I haven't been spoiled by a lisp machine environment
so I guess I don't know what I am missing.

Why did they have to invent yet another window interface ?? How I wish
they provided a well integrated X interface.

Among the bugs I've come across the following is one I find most
intriguing.

Occasionally I have problems with the garbage collector. Once in
a while it gets wedged in a loop that seems to cons garbage as fast as
it can, gc and then cons again.  When I send an interupt and then
backtrace I get something like

    SOMETHING <- WRITE-* <- unnamed-function

(where * means I forget what was there) when I know my stack should be
much bigger and I haven't been doing any writing to streams.  A simple
abort makes everybody happy again.  This behaviour has happened in two
cases. The first case is in reading a form during a load that is ~
80,000 characters.  This problem went away with the inclusion of a
reader macro that read the subforms printing out a little progress
blurb.  The second case was applying nconc to a large list of lists.
This bug also went away with the insertion of diagnostic code.  Any
body else seen this behaviour or have some insight as to why it
occurs?