home
***
CD-ROM
|
disk
|
FTP
|
other
***
search
/
ftp.cs.arizona.edu
/
ftp.cs.arizona.edu.tar
/
ftp.cs.arizona.edu
/
icon
/
newsgrp
/
group93b.txt
/
000020_icon-group-sender _Sun Apr 25 00:41:30 1993.msg
< prev
next >
Wrap
Internet Message Format
|
1993-06-16
|
5KB
Received: by cheltenham.cs.arizona.edu; Sun, 25 Apr 1993 18:22:55 MST
Date: 25 Apr 93 00:41:30 GMT
From: agate!howland.reston.ans.net!usc!cs.utexas.edu!news.uta.edu!utafll.uta.edu!bruce@ucbvax.Berkeley.EDU (Bruce Samuelson)
Organization: UTexas at Arlington, Linguistics
Subject: Lack of robustness limits Icon's popularity
Message-Id: <BRUCE.93Apr24184130@utafll.utafll.uta.edu>
References: <9304230936.AA23022@univ-caen.fr>
Sender: icon-group-request@cs.arizona.edu
To: icon-group@cs.arizona.edu
Status: R
Errors-To: icon-group-errors@cs.arizona.edu
Another possible reason Icon is not more popular is lack of
robustness in some circumstances.
A friend who programs in C needed to write some software to index and
compress several megabytes of text data for use in a commercial
product. Each multimegabyte text is indexed and compressed once and
placed on floppy disks. I suggested Icon as the ideal language.
Although he had never used it before, he followed my suggestion. He
was impressed by how effective it was at implementing his parsing and
compression algorithms, but he was plagued by its unreliability.
Compression runs would take several hours on his 486/33 with 8MB of
memory running DOS 5.0. That wasn't a particular problem, since he ran
it overnight. The real problem was that when he checked the results
the next morning, he could never be sure whether the run would
complete or crash. His and my best guess was that he was bumping up
against some memory constraint.
He did try adjusting the various parameters you can set when compiling
Icon. I think string space was one he modified to be able to process
larger texts. However, he was never able to find settings which would
guarantee consistently reliable runs. He insisted the program was so
fragile that even the presence or absence of a comment in the soure
code could determine whether it would crash. I don't remember the Icon
version number, other than that when I ftp-ed it a year or so ago it
was the most current one for 32-bit Intel--8.something I think. It was
interpreted, not compled.
I told him that my experience with Icon on a Sun workstation running
Unix had been more positive, and that I only encountered one serious
bug for which there was a workaround. (With DOS, using an older
version of Icon that was restricted to 640K, my experiences were even
worse than my friend's, but that was to be exptected.) My positive Sun
experience failed to console him.
In the end, he did manage to get all the texts compressed, writing C
modules in place of Icon ones when necessary. The company had
considered turning their compression software into a commercial
product. It started out being written for internal use only. My friend
is a good programmer and documents his systems well. Unfortunately,
his Icon programs are too fragile to sell commercially. Another
problem, of course, is the niche nature of the language. Source code
written in C would seem to offer more commercial potential.
In a postmortem analysis, my friend said that if he could do it all
over again, he would have simply used C.
He does not have internet or usenet access. I tried to get him to find
a benefactor at the local university who could give him an account,
but he didn't pursue this vigorously. I told him that if he had posted
his symptoms in detail to this newsgroup, and perhaps reported them to
the Icon project, he might have gotten some practical help.
It is not fair for me to complain about Icon's robustness to this
group without giving sufficient documentation so that one of you could
track down the problem and fix it. It could be as trivial as a
compiler flag or setting. And perhaps the 32-bit DOS version of Icon
is one of the poorer ones. I have no idea. But I think this does
illustrate that for at least one version of Icon, intermittant,
unreproduceable runtime failures in building indexes for and
compressed representations of fairly large texts limits its
popularity. There were so many ways my friend caused intermittant
crashes by legitimately modifying his source code that he will
probably never again use Icon again for serious work. On a more
positive note, I would, under the right circumstances.
--
**********************************************************
* Bruce Samuelson Department of Linguistics *
* bruce@ling.uta.edu University of Texas at Arlington *
**********************************************************