home
***
CD-ROM
|
disk
|
FTP
|
other
***
search
/
ftp.cs.arizona.edu
/
ftp.cs.arizona.edu.tar
/
ftp.cs.arizona.edu
/
icon
/
newsgrp
/
group92c.txt
/
000116_icon-group-sender _Fri Dec 18 08:49:21 1992.msg
< prev
next >
Wrap
Internet Message Format
|
1993-01-04
|
2KB
Received: from owl.cs.arizona.edu by cheltenham.cs.arizona.edu; Sun, 20 Dec 1992 18:04:42 MST
Received: by owl.cs.arizona.edu; Sun, 20 Dec 1992 18:04:41 MST
Date: 18 Dec 92 08:49:21 GMT
From: enterpoop.mit.edu!eru.mt.luth.se!lunic!sunic!sics.se!eua.ericsson.se!euas27c34!konhks@ucbvax.Berkeley.EDU (Hakan Soderstrom)
Organization: Ellemtel Telecom Systems Labs, Stockholm, Sweden
Subject: Re: Bus Error
Message-Id: <1992Dec18.084921.2511@eua.ericsson.se>
References: <9212171730.AA26944@gremlin>
Sender: icon-group-request@cs.arizona.edu
To: icon-group@cs.arizona.edu
Status: R
Errors-To: icon-group-errors@cs.arizona.edu
Thanks to Bill Griswold for mentioning some of the mechanisms that may
cause a bus error.
My problem occurs in a program segment where coexpressions are used
heavily. (I don't have the Icon program library at hand, but it is the
procedures, by Ralph Griswold I seem to remember, that convert any data
structure to a string, and back again.) The problem appears when the
volume and 'knottiness' of the data structure increases, so I'd rather
not post a sample. In addition, I have a vague impression that in a
previous experiment some other memory related environment variable
made the bus error go away.
I have become accustomed to Icon as extremely reliable. Still, it is
known that stack overflow can go undetected. So I tend to believe that
in this case heavily recursive use of coexpressions cause the stack to
grow into some other memory region. This is the kind of situation
where there is no a priori upper bound on neither the number of active
coexpressions nor the depth of recursion.
I will re-implement this (small) part of the application.
-- Finally, a re-statement of the problem for those who wonder (and for
Richard Goerwitz :-)
I have an Icon program which I run under SunOS 4.1.1 on SPARC 1+
hardware. I run Icon 8.7 created by gcc 2.1 with -O turned on. This
program has been known to terminate with bus error for some input
data. It behaved the same under Icon 8.6 as under 8.7.
The problem went away when I increased STRSIZE. So my question was, is
it a known problem that too small a STRSIZE may go undetected and cause
a bus error under Unix?
------------------------------------------
Hakan Soderstrom (contractor)