Sed match string and copy it to another line (with added lines) in specific place

To store things, sed has a hold space and h/H commands to copy/append the pattern space to the host space (hold), g/G to get what's held and replace/append to the pattern space (and x to exchange hold and pattern space).

sed -E ' /^ {4}rtr-/h /snmp_location/ { s/$/\ \ routers:\ hosts:/ G }' 

Would hold any line that starts with 4 spaces followed by rtr-, and for any line that contains snmp_location, would append those 3 extra literal lines (by substituting the end of it ($) with them), plus the contents of the hold space.

In a less antiquated language, the equivalent would look like:

perl -pe ' $hold = $_ if /^ {4}rtr-/; $_ .= "\nrouters:\n hosts:\n$hold" if /snmp_location/' 

Or to avoid adding that routers section if rtr- was not found:

perl -pe ' $hold = $_ if /^ {4}rtr-/; $_ .= "\nrouters:\n hosts:\n$hold" if /snmp_location/ && defined $hold' 

In sed:

sed -E ' /^ {4}rtr-/h /snmp_location/ { G s/(\n)(.)/\1\1routers:\1 hosts:&/; t s/\n// }' 

Where we first append the hold space to the pattern space for lines that contain snmp_location, and only insert the routers section before what was appended if non-empty (matches .). Also note the then command that branches to the end if the substitution succeeded.

Note that if there are several occurrences of rtr- before the snmp_location line, all those will consider the last, not first. In perl you can change $host = to $host //= to skip the assignment if $hold is already defined. In sed, you could change it to:

sed -E ' /^ {4}rtr-/ { x /./ { # already found x b after_g } g : after_g } /snmp_location/ { G s/(\n)(.)/\1\1routers:\1 hosts:&/; t s/\n// }' 

Or, assuming the first occurrence of rtr- won't be on the first line (with GNU sed, you could replace 1 with 0 to cover for that case):

sed -E ' 1, /^ {4}rtr-/ { /^ {4}rtr-/ h } /snmp_location/ { G s/(\n)(.)/\1\1routers:\1 hosts:&/; t s/\n// }' 

sed processes the input line by line. You can use the hold space and pattern space to hold additional data, but it's far from readable.

Perl seems to be a better tool for the job:

perl -pe ' $r = $1 if ! defined $r && /\s(rtr-[^:]+)/; $vars = 1 if /^\s*vars:/; $vars = 2, print "\nrouters:\n hosts:\n $r:\n" if defined $r && 1 == $vars && /^$/; undef $r, undef $vars if eof; ' -- file1 file2 ... 

• -p processes the input line by line, printing each line after processing;
• $r is used to store the first instance of rtr-* (we only populate it if it hasn't been set yet);
• $vars is set to 1 once we start processing the vars section;
• The additional lines are printed when $vars is 1 (i.e. the vars section has started) and we process an empty line;
• Once the section is printed, $vars is set to 2 which means no further sections will be printed;
• The variables are reset when we reach the end of a file.

But even better would be to use a library that can read YAML files.

perl -MYAML::XS=LoadFile,Dump -e ' $y = LoadFile(shift); undef $y->{routers}{hosts}{ (grep /^rtr-/, keys %{ $y->{abc_123}{hosts} })[0] }; print Dump($y); ' "$file" 

It's not exactly correct as it doesn't find the first occurrence of rtr-*: they are keys, so they have no defined order.

For anything more than a simple s/old/new/ on a single line, an awk script will almost always be some combination of clearer, simpler, more robust, more maintainable, more portable, etc. than a sed script. For example, using any awk with the sample input from the question:

$ awk ' { print } /rtr-/ { rtr="\nrouters:\n hosts:\n" $0 ORS } /snmp_location/ { printf "%s", rtr } ' file abc_123: hosts: rtr-abc-123.us.dom: sw-abc-123-01.us.dom: sw-abc-123-02.us.dom: vars: snmp_location: "123 Main Street, Nowhere, IL 12345" routers: hosts: rtr-abc-123.us.dom: terminals: children: abc_123: 

Note that that won't insert anything if rtr isn't present in the input:

$ cat file2 abc_123: hosts: sw-abc-123-01.us.dom: sw-abc-123-02.us.dom: vars: snmp_location: "123 Main Street, Nowhere, IL 12345" terminals: children: abc_123: 

$ awk '{print} /rtr-/{rtr="\nrouters:\n hosts:\n" $0 ORS} /snmp_location/{printf "%s", rtr}' file2 abc_123: hosts: sw-abc-123-01.us.dom: sw-abc-123-02.us.dom: vars: snmp_location: "123 Main Street, Nowhere, IL 12345" terminals: children: abc_123: 

If you need to update the input file with the command output then either use GNU awk for -i inplace:

awk -i inplace '...' file 

or with any awk just redirect the output to a temp file then overwrite the input file with the temp file:

awk '...' file > tmp && mv tmp file 

Change the regexps /rtr-/ and /snmp_location/ however you like if those aren't robust enough, I just copied what you had in the question assuming it's good enough.

If you really care about only selecting the first occurrence of rtr- from the input, just change /rtr-/ to !rtr && /rtr-/. Any other such changes would be similarly trivial.