Question

I have this file:

100: pattern1
++++++++++++++++++++
1:pattern2
9:pattern2
+++++++++++++++++++
79: pattern1
61: pattern1
+++++++++++++++++++

and I want to sort it like this:

++++++++++++++++++++
1:pattern2
9:pattern2
+++++++++++++++++++
61:pattern1
79:pattern1
100:pattern1
+++++++++++++++++++

Is it possible using Linux sort command only ?

If I had :

4:pat1 
3:pat2
2:pat2
1:pat1

O/p should be:

1:pat1
++++++++++++ 
2:pat2
3:pat2
++++++++++++
4:pat1

So, want to sort on first group, but “group” on the pattern of second group.
Please note, the thing after : is a regex pattern not a literal.

Answer 1

I don’t believe sort alone can do what you need.

Create a new shell script and put this in its contents (ie mysort.sh):

#!/bin/sh
IFS=$'\n' # This makes the for loop below split on newline instead of whitespace.
delim=+++++++++++++++++++
for l in `grep -v ^+| sort -g`      # Ignore all + lines and sort by number
do
    current=`echo $l | sed s/^[0-9]*://g` # Get what comes after the number
    if [ ! -z "$prev" ] && [ "$prev" != "$current" ] # If it has changed...
    then                                  #  then output a ++++ delimiter line.
        echo $delim
    fi
    prev=$current
    echo $l                               # Output this line.
done

To use it, pipe in the contents of your file like so:

cat input | sh mysort.sh