Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!utgpu!water!watmath!clyde!rutgers!ames!amdcad!sun!rdh
From: rdh@sun.UUCP
Newsgroups: comp.unix.wizards
Subject: shell and sed conflicts (was embedded NL's in shell vars)
Message-ID: <25712@sun.uucp>
Date: Sat, 15-Aug-87 02:33:22 EDT
Article-I.D.: sun.25712
Posted: Sat Aug 15 02:33:22 1987
Date-Received: Sun, 16-Aug-87 09:08:20 EDT
References: <228@cjsa.UUCP>
Reply-To: rdh@sun.UUCP (Robert Hartman)
Organization: Sun Microsystems, Mountain View
Lines: 53
Keywords: sed sh shell

In article <228@cjsa.UUCP> jeff@cjsa.UUCP (C. Jeffery Small) writes:
>I have been attempting to construct a 'sed' command line script and load it
>into a shell variable (called ACTION) for later use within a bourne shell
>script. ...

[ describes problem passing a NEWLINE to sed via shell variable ]

While it may be inelegant in principle, a hack I often use is simply to
build a temporary sed script in a separate file, and then run sed -f 
with that script.  This avoids many such conflicts between sed and the shell.

	sed ... > /tmp/$file

	# Build a temporary sed script
	echo "/$address/a\\" > $sedscript
	echo "$appendln"     >> $sedscript

	# Apply temporary sed script
	sed -f $sedscript /tmp/$file > $outfile
	rm -f /tmp/$file $sedscript

It's gross, but it works.  If you can't stand the thought of having TWO
temporary files, a somewhat more elegant way is to use a subshell:

	sed ... | \
	(	echo "/$address/a\\" > $sedscript ; \
		echo "$appendln"     >> $sedscript \
	) | sed -f $sedscript > $outfile ; rm -f $sedscript

But, I've had only mixed success with this.  When a shell script gets
this complicated, I tend to opt for temp files.  I'd rather waste a little
temp space than have to debug my way thru 2 shells and 2 complex seds.

I wholeheartedly agree with anyone who is appalled by this kludge, but after
all, we're talking about shell scripts, hopefully for personal or limited use.
I would never suggest that anyone do this in production scripts.

You'll undoubtedly get a million suggestions to "just use awk," but I
don't know.  I don't see why one should have to learn yet another
programming language (with less-flexible RE handling), to implement
knock-off tasks that are already easy to think about in terms of the
regular editing commands common to ed, ex, sed, and vi (and the
already-familiar shell control constructs).

A "real" solution to this problem would be to teach sed to recognize SOME 
PRINTABLE ESCAPE SEQUENCE as an embedded NEWLINE.

By the way, has anyone come up with a better way to get
sed to give you "just the line after `address'" than:

	... | sed -n -e "/addr/N" -e "/addr.*/p" | sed "/addr/d"

-bob.