Shell Programming and Scripting

Hi guys, I am new here. I have done many scripts in VisualBasic and batch.
Now going to *inux and learning bash and perl, quite amazing commands I find in linux, like sed, grep, awk and so on...

the task is simple, find a string in folders and remove the line where that string exists.
I can find all lines with that string doing:
grep -R "STRING" /home/comm/nagios/confs/*
witch outputs:
filename.ext line

But now I actually need to remove these lines. I tried something like:
sed '/^.* {/ {:a /}/ !{N;ba} }; s/.*STRING.*//' test.cfg
That just ouputs without the lines with string X but i need to write into that same file, and need to do in many files

thanks in advance,

grep "sample" <infile
will show all lines from infile that have the text "sample"

grep -v "sample" <infile
will show all lines from infile except those that have the text sample

thanks but that was not what i needed.
I need to remove an entire line where I find certain STRING.
But I shuold not have to specify the filename, just the folder. I have to do this
in multiple files.

The right way is to combine two tools you are already using and glue them together via shell. This is how most of this class of problems are solved in Unix. (Welcome to the fun side of the isle, btw. ;-)) )

Your first task is to find all the files containing a specific string, the second task is to feed the list of filenames found that way one by one to sed which deletes the respective lines.

First, how would sed do that with one file:

Code:

sed '/some_expression/d' /path/to/file

Not quite. The mechanism is correct (the lines containing "some_expression" would be deleted) but sed would put its results to <stdout> because habitually it does not change the input file. So we need to save the results to a file and then replace the original file with the result-file:

Code:

sed '/some_expression/d' /path/to/file > /path/to/new_file
mv /path/to/new_file /path/to/file

This does exactly what you want to do, it is the solution to the second task. Now we want this to work on one file after the other: You already noticed that using "grep" you can produce a list of filenames on which the mechanism above should work. Lets feed this list to the command sequence above. This is done with a "while"-loop. Since we nee only the filenames we trim the output of grep to display only the filenames by using the "-l" switch:

Code:

grep -Rl "EXPRESSION" /path/to/search/* | while read filename ; do
     sed '/EXPRESSION/d' $filename > tmpfile
     mv tmpfile $filename
done

Consider the grep-command as a sort-of file. You can display its contents by simply execute the command. You can feed it to other commands by using the pipeline ("|") operator. The while-loop takes the content of this "file", one line by one, and assigns this to the variable "$filename" which is used throughout the loop.

So, that was it, yes? Hold on: the sed-expression doesn't even touch lines which do not contain expressions. What would happen, if the file consists only of such lines? Correct! Nothing. So we could skip the whole grep-part and directly modify all the files found:

Code:

ls -R /path/to/search/* | while read filename ; do
     sed '/EXPRESSION/d' $filename > tmpfile
     mv tmpfile $filename
done

This list is now longer but looks similar: filenames, one on a line each. But as files not containing "EXPRESSION" do pass through the mechanism unchanged this is as good as the first solution.

Still, as every file has to be copied it may or may not put considerably more load on the machine. That depends on the ratio of number of files to be changed to files to go unchanged, the number of files as a whole, etc. You will have to decide. You might decide to fine-tune the second program by using seds exit-codes to determine if it has changed anything at all and skip the "mv"-step accordingly. But this is another story....

I hope this helps.

bakunin

I hope this helps.