Delivery-Date: 
Sender: info-hol-request@lal.cs.byu.edu
Errors-To: info-hol-request@lal.cs.byu.edu
Precedence: list
Date: Fri, 24 Mar 1995 16:57:12 +1000
From: Malcolm Newey <Malcolm.Newey@cs.anu.edu.au>
Message-Id: <199503240657.QAA29655@vega.anu.edu.au>
To: info-hol@leopard.cs.byu.edu
Subject: lisps for HOL88


Keywords:  HOL88, AKCL, GCL, HOL90, Database of Successes


All,

On arriving back in Australia, I found that the CS Dept. has shifted to
Solaris on all its machines.  In the process of getting HOL88 going in my
very problematic new environment, I sought advice from various sources
(including the info-hol archives) several times.  I formed the view that
we would collectively benefit if we individually logged problems and
successes when porting HOL to new environments.  I note that the people
at Austin get a mail message sent to them from every successful make of
akcl and gcl.  (see shell script notify in their directory xbin.)

I did not have any problems getting HOL90 going on a SPARCstation 5
running Solaris 2.3.

Getting HOL88 going (yes, it's still alive!) made me realise we need an update
to the advice we give about porting it since the old recipes don`t work
on many new systems.  The `tried and true' version of akcl that has been
(and is being) distributed from ftp.cl.cam.ac.uk is version 605 (circa 1991)
and has a list of possible hardware/OS options which doesn't include ncr,
NeXT, mac2, 386-linux, 386-bsd, sun5/solaris which newer versions do.

The following is wisdom that should be more widely known, but which
doesn't appear in the three most likely places (WWW site at BYU, the ftp
site for HOL at Cambridge, or even the AKCL ftp directory at Austin).  AKCL,
as an evolving product, is DEAD and has been replaced by GCL, Gnu Common
Lisp.  It is distributed from the same site (ftp.cli.com) and attempts
to implement Common Lisp on the machines I mentioned above, as well as the
ones that AKCL-605 supported.  (There have been a couple of messages that
have mentioned gcl but they did not catch my eye as being of general import.)

My experience:

   Under solaris, gcl-1.1 works better than the last akcl (vers 624), but
still gives a significant problem.   basic-hol and hol appear to be made
correctly but when hol is fired up it fails with a message about dlopen
not being implemented by Sun.  A temporary workaround is to remove `~/`
from the list of paths given to set_search_path.  This same problem has
been mentioned by another subscriber to the gcl mail list (not in relation
to HOL) so maybe its a gcl problem and more info will surface.

    By the way, the only other change I made was to treat GCL like IBCL
in lisp/f-cl (apart from the prescribed changes in the Makefile).

Suggestion:

    If you use HOL88, you might be interested in rebuilding it on your machine
using gcl-1.1 and sharing your experience.  As a group we don't yet have
a database of who uses which HOL and what machines have been successfully used.

Maybe we should adopt the Austin school's trick of sending mail to a central
archive - in our case,  whenever HOL is built successfully.  (The size of the
mail file would not be a problem.)

Malcolm