In general, using the shell for text-parsing is very slow and cumbersome. Here are some other options:
Perl in "paragraph mode"
perl -00pe 's/^/$./' file
Explanation
The -00 turns on paragraph mode where "lines" are defined by consecutive \n\n, paragraphs in other words. The s/^/$./ will replace the start of the line (^) with the current "line" (paragraph) number $.. The -p tells perl to print each line of the input file after running the script given by -e on it.
Awk
awk -vRS='\n\n' -vORS='\n\n' '{print NR$0}' file
Explanation
-vRS='\n\n' sets awk's record separator to consecutive newline characters. Like perl's paragraph mode, this makes it treat paragraphs as "lines". We then tell it to print the current line number (NR) and the current "line" $0. The -vORS= sets the output record separator to consecutive newlines so that paragraphs are separated by blank lines in the output as well. Note that this will add 2 empty lines at the end of the output. To avoid that, you can use head:
awk -v RS='\n\n' -vORS='\n\n' '{print NR$0}' file | head -n -2
By way of comparison, here are the times that the various solutions took on my system when run on a 10M test file:
$ time a.sh > /dev/null ## a.sh is Cyrus's solution real 0m1.419s user 0m1.308s sys 0m0.104s $ time perl -00pe 's/^/$./' file > /dev/null real 0m0.087s user 0m0.084s sys 0m0.000s $ time awk -v RS='\n\n' -vORS='\n\n' '{print NR$0}' file | head -n -2 >/dev/null real 0m0.074s user 0m0.056s sys 0m0.020s
As you can see above, both the perl and awk solutions are an order of magnitude faster than the shell approach.
