Why is sed giving me an error about an unterminated `s'?

Question

I have a set of sed replacements in a bash script and am getting an error about an unterminated `s' command. Here's what the sed line looks like:

sed -n -e "s/TMPFOO1/$FOO1/" -e "s/TMPFOO2/$FOO2/" -e "s/TMPFOO3/$FOO3/" -e "s/TMPFOO4/$FOO4/" -e "s/TMPFOO5/$FOO5/" /home/foo/template > /home/foo/template/finishedresult

For some reason though, bash doesn't like this and I get an error about

sed: -e expression #4, char 69: unterminated `s' command

What am I missing here? How can I have SED input the variable? It looks to me like they are all terminated.

Answer 1

You cannot safely interpolate a variable in a sed command, because the substitution is performed by the shell, not by sed. The value of the variable becomes sed syntax. For example, in "s/TMPFOO1/$FOO1/", if $FOO1 contains a newline, this will cause a syntax error like the one you observed. If $FOO1 contains a /, this terminates the s command and may cause an error or may cause other commands to be executed (if what's after the / happens to be valid sed syntax).

While you can do a first pass of substitution of FOO1 to quote its special characters for inclusion in that sed command, it's a lot simpler to use awk. Awk has a notion of variable, and a command line syntax to set the initial value of a variable.

awk -v FOO1="$FOO1" -v FOO2="$FOO2" -v FOO3="$FOO3" -v FOO4="$FOO4" -v FOO5="$FOO5" '{
    sub(/TMPFOO1/, FOO1);
    sub(/TMPFOO2/, FOO2);
    sub(/TMPFOO3/, FOO3);
    sub(/TMPFOO4/, FOO4);
    sub(/TMPFOO5/, FOO5);
    print;
}' /home/foo/template > /home/foo/template/finishedresult

Answer 2

Most likely one of your $FOO variables contains special characters that are interpreted by sed.

I have another version of sed which generates other error messages but here is an example of a similar problem:

$ VAR=a
$ echo i | sed -e "s/i/"$VAR"/"
a
$ tmp> VAR=/
$ echo i | sed -e "s/i/"$VAR"/"
sed: 1: "s/i///
": bad flag in substitute command: '/'

In this case $VAR contains a character which is interpreted by sed as the trailing slash.

Answer 3

As others have mentioned here, it depends on the content of your FOO* variables. In your case, using sed is the wrong choice because it probably contain some special characters.

Take a look at this link and see the function gsub_literal

Since gsub_literal reads from stdin and writes to stdout, the usage will be:

gsub_literal "$search" "$replace" < /home/foo/template > /home/foo/template/finishedresult

Sample output:

rany$ cat > foo.txt
a'
a'

rany$ gsub_literal a\' b < foo.txt 
b
b

Answer 4

The problem has been resolved for me by trying additional backslashes within the sed 's/.../.../g' command. This command

sed 's/\"a/ä/g' input_file

gave me this error message

sed: -e expression #1, char 9: unterminated `s' command

when it was executed as part of a c-shell script, and it did so only with a recent version of sed, not with an older version. (Strangely, the same command works fine when it is given on the command line (in a tc-shell)).

The problem with the c-shell script has been remedied by preceding the ä character by a backslash:

sed 's/\"a/\ä/g' input_file