How to use sed to parse and modify the result of a find?

Question

I stumbled upon following issue: on a single line, I would like to alter the result of my find command to do a specific cp depending of the files I found, with a destination name that depends of the files returned by the find.

Here are my files,

$  find jcho -name *.data
jcho/category1/001.data
jcho/category2/002.data

and they must be copied to

jcho2/category1_001.data
jcho2/category2_002.data

I tried this,

$ find . -name *.data \
-exec cp  {} ` echo {} | sed -re 's/(jcho\/)(category[0-9]*)(\/)(.*data)/jcho2\/\2_\4/' ` \;

but it says it want to copy to same file -- my substitution is not made; got below error:

cp: './jcho/category1/001.data' and './jcho/category1/001.data' are the same file

So I tried something with a subshell to which I give the result of the find. This worked (a little).

find jcho -name *.data -exec sh -c \ 
'f="${0}"; d=$(echo ${f} | sed -re 's/0/2/' ); cp ${f} ${d} ' {} \;

==>

find jcho -name 2*.data
jcho/category1/201.data
jcho/category2/202.data

If I could include `/` in the `sed`pattern, my solution would be at hand...

But I get:

find jcho -name *.data -exec sh -c \
'f="${0}"; d=$(echo ${f} | sed -re 's/(jcho\/)(category[0-9]*)(\/)(.*data)/jcho2\/\2_\4/' ); cp ${f} ${d} ' {} \;

-ksh: syntax error: `(' unexpected

... I tried escaping the / with \, with \\... not better.

idem with

find jcho -name *.data -exec sh -c \
'f="${0}"; d=$(echo ${f} | sed -re 's~\([^)]*\)/\([^()]*\)$~\1_\2~' ); cp ${f} ${d} ' {} \;

Thank you for your help!

This is the same issue than `find -name` with regex pattern and filename replacement using `cp` [link] (http://stackoverflow.com/questions/36396401/find-name-with-regex-pattern-and-filename-replacement-using-cp) Check the answer there. — Jay jargot, Apr 04 '16 at 14:22
No, it is more specific. I tried answering it differently, but could not. My question ask a solution with **sed** — J. Chomel, Apr 04 '16 at 14:26

anubhava · Answer 1 · 2016-04-04T14:56:01.363

1

You can use this sed with an alternative delimiter ~:

find jcho -name '*.data' | 
while read -r f; do cp "$f" "$(echo "$f" | sed 's~\([^)]*\)/\([^()]*\)$~\1_\2~')"; done

With your find output it will give:

jcho/category1_001.data
jcho/category2_002.data

edited Apr 04 '16 at 14:56

answered Apr 04 '16 at 14:14

anubhava

761,203
64
569
643

Thank you, but I still get `-ksh: syntax error: `)' unexpected` – J. Chomel Apr 04 '16 at 14:28
Why `ksh`? You tagged question as `bash` and should use `bash` only. – anubhava Apr 04 '16 at 14:30
Sorry about this. I retagged it. – J. Chomel Apr 04 '16 at 14:33
`sed` remains same across `bash` or `ksh` so this is something else. Can you tell me what is full command you're trying? – anubhava Apr 04 '16 at 14:36

Jay jargot · Accepted Answer · 2016-04-04T23:25:34.400

The ( and ) need to be escaped to be referenced later.

The / must be escaped with [/] in regex pattern, or with \/ in the replacement part.

You could try the command-line below:

$ find jcho -name \*.data | sed -n '
{
h
s/^\(.*\)[/]\(category[^/]*\)[/]\(.*[.]data\)$/\12\/\2_\3/
H
x
s/\n/ /
p
}' | xargs -L 1 cp

This is working with the example provided but there will be issues if pathnames or filenames contain special chars: \n " ' space.

There are famous pages to read on these topics:

Let see how sed will handle each filename read from the pipe.

0- -n option instructs sed to not print lines except when p is used.

1- jcho/category1/001.data is read from the pipe and it is stored in the pattern buffer.

pattern buffer = jcho/category1/001.data

2- h overwrites the content of the hold buffer with a copy of pattern buffer.

hold buffer = jcho/category1/001.data

3- the first s changes the content of the pattern buffer

pattern buffer = jcho2/category1_001.data

4- H adds pattern buffer to hold buffer

hold buffer = jcho/category1/001.data
jcho2/category1_001.data

5- x exchanges buffers content

pattern buffer = jcho/category1/001.data
jcho2/category1_001.data

6- the last s command replaces '\n' with ' ' in the pattern buffer

pattern buffer = jcho/category1/001.data jcho2/category1_001.data

7- p prints the current pattern buffer.

8- End of the block: return to 1 and work with next filename found.

Thanks Jay, but I am trying to do it the `sed` way with a subscript - without a loop. — J. Chomel, Apr 04 '16 at 14:41
@J.Chomel the answer had been updated. Let me know your remarks. — Jay jargot, Apr 04 '16 at 14:51
This worked for me, thanks. I guess I will have hard time with sed and understanding the `h`, `H`, `x`, `p`you used there. — J. Chomel, Apr 04 '16 at 14:55
@J.Chomel the answer had been updated, see the new sectino at the bottom. The letters are described. — Jay jargot, Apr 04 '16 at 23:12

score 1 · Answer 3 · answered Apr 04 '16 at 14:41

1

probably there is an easier way than this, but with bash basename/dirname combination you can achieve the same

for f in $(find ...); do cp $f $(echo $(dirname $f)"_"$(basename $f)); done

without getting into string manipulation with regex matching.

answered Apr 04 '16 at 14:41

karakfa

66,216
7
41
56

I tried this way but `dirname` is not the right destination. It is a different path I want to copy to: `jcho2`. – J. Chomel Apr 04 '16 at 14:52

How to use sed to parse and modify the result of a find?

If I could include / in the sedpattern, my solution would be at hand...

3 Answers3

If I could include `/` in the `sed`pattern, my solution would be at hand...