[SOLVED] Insert line of text prior to pattern only once

grail · 06-29-2010, 07:02 AM

So I have obviously been looking at this problem for too long now as
I am unable to come up with what I know should be a simple solution,
so here goes:

I have a file like so:

Code:

one
two
three
four

I would like to insert prior to the word "three" all items from this
second file with the following contents:

Code:

four
three
two
one

Now my issue is, and I have been using both sed and awk currently,
that after the second line of the new file is read there will of course now be 2 copies of the word "three" but I would like to only insert the final 2 words, ie "two" and "one" prior to only the first occurrence of the word "three" so final file will look like:

Code:

one
two
four
two
one
three
three
four

So here there is now only one of each word from the second file joined to make the new file.
For simple code I have tried something like the following:

Code:

while read line
do
    awk -v n=$line '!f && /three/{print n;f++}1' file1 > tmp_file
    mv tmp_file file1
done < file2

Now this works but seems very clumsy to me. There is obviously a better sed and / or awk out there.

Look forward to replies

David the H. · 06-29-2010, 08:08 AM

Does it have to be done in a single operation? You could split file two into two parts, saving lines "four and "three" into one variable or temp file, and "two" and "one" in another. Then you can use two separate sed commands to insert them into file one in the sequence you want. Or some variation of the above. I think that would be much cleaner than trying to finagle everything into a single command.

grail · 06-29-2010, 09:03 AM

As you probably guessed this is a trivial example to accomplish a larger task. The issue would be that I will eventually be looking at
potentially 50 - 100 files each of these will be coalesced with the original and be searching for potentially different values each time.
The one rule that must remain the same no matter how many files are used is that the insertion should only come prior to the line being searched
for and if there are other copies within file1 then it should stop after the first occurrence.

I hope this is a little clearer.

ntubski · 06-29-2010, 10:04 AM

Quote:

Originally Posted by grail

Now this works but seems very clumsy to me. There is obviously a better sed and / or awk out there.

The problem seems awkward, so I think any solution will be too. Here is a sh solution which is a bit longer than yours, but might be considered "better" in that it only uses 1 temp file:

Code:

#!/bin/sh

while read line1 ; do
    while [ "$line1" = three ] && read line2 ; do
        [ "$line2" = three ] && cat
        echo "$line2"
    done < file2
    echo "$line1"
done < file1 > tmp_file
mv tmp_file file1

crts · 06-29-2010, 10:51 AM

Hi,

at least with sed I do not see a solution without a temporary file. However, the follwing code produced the same result for your sample data.

Code:

#!/bin/bash

sed -e '/three/ d' reverse > tmp
sed -e '/three/ {
	r tmp
	a \
three\
three
	d
}' forward

It is important that there is exactly one space between the r command and tmp, and nothing after tmp.

grail · 06-29-2010, 11:10 AM

Hi crts

Thanks for your input

The ultimate goal here is this is my attempt to produce dependencies for package management we have spoken of before.
So I am happy to try any given solution

My initial thought was recursion but was concerned that if the path grew too deep that bash would
throw a hissy fit.

So I will build on yours and ntubski's solutions to see if they work with a deeper set