Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!tut.cis.ohio-state.edu!cs.utexas.edu!csd4.milw.wisc.edu!uakari.primate.wisc.edu!indri!ames!vsi1!altos86!dtynan
From: ehkinmonth@ucdavis.edu (Earl H. Kinmonth)
Newsgroups: comp.unix
Subject: set -e breaks some shell scripts
Message-ID: <12382@altos86.Altos.COM>
Date: 31 Jul 89 21:46:32 GMT
Sender: dtynan@altos86.Altos.COM
Reply-To: uunet!ucdavis.edu!ehkinmonth (Earl H. Kinmonth)
Organization: University of California, Davis
Lines: 31
Approved: dtynan@zorba.Tynan.COM


Presumably what I'm going to describe is common knowledge for adepts,
but for the benefit of a few other amateurs out there, I'm sharing
my information (and frustrations).

Situation: a shell script that runs under a half dozen varieties of
the Bourne shell (and even the Korn shell under MSDOS) dies without
explanation on a particular set of machines.  In each case the death
without explanation occurs in a construct of the form

if test -r $F
then
echo found $F
fi

Using the set -x option shows that test is being executed AND that
the shell is instantly exiting.

After hours of frustration, switching back to other machines to verify
that the code actually works, I see a "set -e" in the shell script.
I comment it out, run the shell script, and it works!

Analysis: for archaic versions of the Bourne shell, test is a separate
program.  If the -e option is on, a non-zero status is considered an
execution failure, and the shell exits (unfortunately with no
message).  With more recent versions of /bin/sh and ksh, test is a
built in function, and a non-zero status is recognized as a normal
state of affairs and execution continues.

Save your RTFM comments.  I did RTFM, and if this problem is
explained, I missed it.