A friend was recently looking for help with a problem, and that gave me the chance to freshen up my sed a bit.

His problem was the not uncommon one of wanting to replace a two-word string which might go over line-breaks. He wanted to keep the newlines in the replacement.

Most people are unaware that sed can do multiline edition with the commands N, D, P, G and H. Here's the script which did the job in the end (assuming I want to replace "I hope" with "we expect"):

sed -e '
N
s/\<I\([ \n]\{1,\}\)hope\>/we\1expect/g
P
D' input.txt > output.txt

N adds a new input line to the pattern buffer, the s replaces I hope in that pattern buffer, saving the thing which separates I and hope (spaces or a newline), and inserts it between the we and think in the output, P prints up to the first newline, and finally D deletes up to the first newline, and forks back to the N if there's anything left in the patter buffer. The last line, which is the active pattern buffer, gets dumped when the N command fails (that is, when there is no more input).

I'd forgotten how much sed rocks.