Hi,
sed was based on ed; ed came first. ed can do forwards and backwards searches; sed can't do backwards searches. The syntax for the ed g command is:

Code:

g/BRE/command

It tells ed to identify every line in the file that matches the basic regular expression BRE and for each line found, execute command on that line. And command in this case is:

Code:

?BRE1?,/BRE2/p

where p is the print command which takes zero, one, or two addresses to specify a range of lines to be printed. (No addresses prints the current line; one address prints the addressed line, and two addresses (separated by a comma) prints the lines from the 1st address up to and including the 2nd address.) The address specified by ?BRE1? searches for the line matching the basic regular expression BRE1 backwards from the current line and (as with sed) /BRE2/ searches forwards from the current line for a line matching the basic regular expression BRE2.

With your new sample input, the ed script I suggested should still print the lines you want. And, if you like to write less portable, 1-liners instead of code that will work with any POSIX-conforming shell, you can translate this to:

Code:

strear='myapp1-ear';ed -s xmlfile3.xml <<< "g/<application-name>$strear<\/application-name>/?<application>?,/<\/application>/p"

The above works with both 1993 or later versions of ksh and bash, but is a syntax error for many other POSIX-compliant shells.

And, if you want to strip two <tab> characters from the front of each of those lines, you could use:

Code:

#!/bin/ksh
strear='myapp1-ear'

ed -s xmlfile.xml <<EOF
g/<application-name>$strear<\/application-name>/?<application>?,/<\/application>/s/^..//
?<application-name>,.p
EOF

You could turn that into a 2-liner, but I much prefer readable and maintainable code to the minimal line approach.