Shell Programming and Scripting

Thanks anurag.singh!

Your awk example definitively gives me the output I was after. Think there was a typo between the a and the $1, had to change the '{' to a '['. Would you know why my script does not work? As per DGPickett's remark, it appears that sed is not able to see $SEARCH and $REPLACE. sed just sees the whole output of each line as $SEARCH. Almost like its not seeing the format of "SEARCH[space]REPLACE". I can't seem to figure out how to get sed to see the expanded/variablized values of $SEARCH and $REPLACE. Thanks again.

---------- Post updated at 05:18 PM ---------- Previous update was at 05:02 PM ----------

Thanks again DGPickett but I get the same output with the usage of sed to clean up the control file. Here is a copy of the updated script:

Code:

#!/bin/bash

sed '
s/#.*//
s/[ \t]*$//
/^$/d
' controlfile | while read SEARCH REPLACE
    do
        sed  's/$"SEARCH"/$"REPLACE"/' sourcefile
    done

Here is its output:

Code:

search1


search2
search3



search4
search1


search2
search3



search4
search1


search2
search3



search4
search1


search2
search3



search4

I think you are correct with your statement that sed does not see the expanded values for $SEARCH and $REPLACE. If I do echo $SEACH, I get the whole line echoed (not what I would expect). If I do echo $"SEARCH" I only get the first column (what I would expect). Doing echo $"REPLACE" gives the second column. Doing just a echo $REPLACE gives me blank lines. I think the scripts logic is correct - figuring I am just missing some semantic in passing variables to sed.

The side effect of putting sed inside of the while loop is that the sourcefile is processed for each search/replace pair in the control file. May I suggest:

Code:

# Args: Control-File Source-File

NAME=$(basename ${0})
FILE=/tmp/${NAME}.${$}
trap "rm -f '${FILE}'" EXIT

awk '/^#/ { next; } NF == 2 { print "s/" $1 "/" $2 "/"; }' "${1}" > "${FILE}"
sed -f "${FILE}" "${2}"

In post #8, sed under while loop should be (-e not really needed here):

Code:

sed "s/$SEARCH/$REPLACE/" sourcefile

Also as mentioed in post #9, sed under while, will run for each search/replace pair in the control file (source file being processed that many times, that is why output is repeated that many times), so you need to use -i option if supported.

Code:

sed -i "s/$SEARCH/$REPLACE/" sourcefile

This will modify source file itself.
I believe in such scenarios (where fields in one file are being modified based on values of fields in another file), awk provides better solutions.

One side effect of using awk as shown in post #6: is that whitespace would not be preserved. This may, or may not be, important. But the request was to replace string "pattern" with "replacement", not replace a field with a given value with a new value. Or was it implied?

IMHO, if the request was for word/field replacements, or with more complexity, I would probably have gone with PERL with \b in the regex. I also have to admit I am not a fan of in-place updates. Yes, it is very useful at times, but it has bit me in the posterior more times than I care to count.

Well, just take it in pieces, printing variables to stderr with delimiters so you can see what is up "'$xxxx'" ! Often an extra space or a tab not a space ruins the match, and then there is no translation.

---------- Post updated at 09:06 PM ---------- Previous update was at 09:04 PM ----------

Code:

 
sed "s/$SEARCH/$REPLACE/"
 
or 
 
sed 's/'"$SEARCH"'/'"$REPLACE"'/'

Thanks everyone for your help. m.d.ludwig's combination of awk and sed in post #9 produces just what I was looking for.