[SOLVED] sed escape characters issue

bmxakias · 12-05-2019, 03:18 PM

Hello

I am trying to to edit a file and change this line:

Code:

[ ${NETWORKING} = "no" ] && exit 0

to

Code:

[ "${NETWORKING}" = "no" ] && exit 0

i tried to use \ for some special characters and my command is:

Code:

sed -i "s|[ \${NETWORKING} = \"no\" ] && exit 0|[ \"\${NETWORKING}\" = \"no\" ] && exit 0|" /folder/file

But it doesn't work...

astrogeek · 12-05-2019, 05:52 PM

"Doesn't work" is not a very helpful description of what you see as a result.

Is all that context necessary and always identical?

If not, then assuming the test brackets to be sufficient context, this would safely work:

Code:

sed 's/[[] \${NETWORKING} = "no" []]/[ "${NETWORKING}" = "no" ]/' src.txt

If you do not actually need context beyond the immediate space characters this should work well enough too (GNU sed):

Code:

sed 's/ \(\${NETWORKING}\) / "\1" /' src.txt

It would be best to state the actual requirements and avoid the complications if not really necessary.

syg00 · 12-05-2019, 06:01 PM

I hate escapes - this should be equivalent with a bit of luck

Code:

sed 's/${NETWORKING}/"&"/' src.txt

(and yes I know the back-reference isn't an escape ...)

Totally agree re the necessity for all the context - if needed use it as an address selection.

astrogeek · 12-05-2019, 06:11 PM

Quote:

Originally Posted by syg00

I hate escapes - this should be equivalent with a bit of luck

Code:

sed 's/${NETWORKING}/"&"/' src.txt

(and yes I know the back-reference isn't an escape ...)

Totally agree re the necessity for all the context - if needed use it as an address selection.

I included the surrounding spaces to avoid ending up with things like ""${NETWORKING}"", but agree that using an addressed context would be in order if needed rather than trying to squeeze it all into the regex.

bmxakias · 12-05-2019, 06:49 PM

My target is to add to ${NETWORKING} the "" like "${NETWORKING}" without changing anything else in the line.

Also using the full line i think is better because if the file has a second ${NETWORKING} then it will changed it also and i don't want that.

Is that the best solution for that scenario?

Code:

sed 's/[[] \${NETWORKING} = "no" []]/[ "${NETWORKING}" = "no" ]/' src.txt

All solutions above working great !

Thank you

syg00 · 12-05-2019, 07:24 PM

Quote:

Originally Posted by bmxakias

Is that the best solution for that scenario?

Somebody recently wanted to ban all "best of " questions ... this is Linux, hopefully everybody can find an option that suits. I would do it as

Code:

sed '/[[] \${NETWORKING} = "no" []]/ s/${NETWORKING}/"&"/' src.txt

Untested, but I reckon you shouldn't need to escape that $ either since you are using single quotes.

NevemTeve · 12-05-2019, 08:07 PM

A (hopefully) generic method:

Code:

From='[]^$\/*.'
To='Replaced'
SedMagic='s:[]\[\\\.\&*^$/]:\\&:g'
From1=$(echo "$From" | sed "$SedMagic")
To1=$(echo "$To" | sed "$SedMagic")
echo "AAA${From}AAA" | sed "s/$From1/$To1/"

grail · 12-06-2019, 04:32 AM

As the code being changed appears to be bash, then the spaces around the changed item are required, so you should be able to just look for the item to change.
Here is another alternative:

Code:

sed -r 's/ (\$\{NETWORKING\}) / "\1" /' file

Here you do need the escapes shown as -r is extended regex so will use $ and {} for other things

Or without -r:

Code:

sed  's/ \(${NETWORKING}\) / "\1" /' file

One less escape here

bmxakias · 12-06-2019, 06:51 AM

Working great thank you !

pan64 · 12-06-2019, 07:13 AM

you can also use . (dot) for special chars:

Code:

sed 's/^. ..NETWORKING. =/[ "${NETWORKING}" =/'

or something similar may work too.

David the H. · 01-01-2020, 09:50 AM

I would re-think this a bit. The proper use of sed is to use the /address/ section to match the lines you want, then the command(s) following it (especially, but not limited to, 's///') to modify them.

So first figure out what makes the line unique and easy to target. It doesn't have to be the string you want to change. If, for example, it's the "exit 0", then use that for the address. sed would thus only modify lines containing it. You can use regex patterns, line numbers, line ranges, and more. Whatever works.

The commands following the address then target and modify the part of the matched lines that you want to change. Multiple commands can be grouped and/or nested with '{ ; }'. Here, I use a second, nested /address/ string to further filter the line:

Code:

#The input file. We want to change only the first line.#
$ cat infile.txt
[ ${NETWORKING} = "no" ] && exit 0
[ "${NETWORKING}" = "no" ] && exit 0
[ ${NETWORKING} = "no" ] && exit 1
[ Another unimportant ${NETWORKING} string ]

Code:

$ sed '/exit 0/ { /[^"]${N.*G}[^"]/ s/${N.*G}/"&"/ }' infile.txt
[ "${NETWORKING}" = "no" ] && exit 0
[ "${NETWORKING}" = "no" ] && exit 0
[ ${NETWORKING} = "no" ] && exit 1
[ Another unimportant ${NETWORKING} string ]

It first finds any line containing 'exit 0', then checks those lines for an unquoted '${N.*G}' string. If it finds one, it adds quotes around it. Notice how I simplified the regex down to (close to) the minimal phrase needed to get the match. It helps keep the code short and readable.

Actually, though, the following is slightly simpler. No nested commands are needed:

Code:

$ sed '/exit 0/ s|"\?\(${N.*G}\)"\?|"\1"|' infile.txt
[ "${NETWORKING}" = "no" ] && exit 0
[ "${NETWORKING}" = "no" ] && exit 0
[ ${NETWORKING} = "no" ] && exit 1
[ Another unimportant ${NETWORKING} string ]

or

$ sed -r '/exit 0/ s|"?(\$\{N.*G\})"?|"\1"|' infile.txt
[ "${NETWORKING}" = "no" ] && exit 0
[ "${NETWORKING}" = "no" ] && exit 0
[ ${NETWORKING} = "no" ] && exit 1
[ Another unimportant ${NETWORKING} string ]

We don't actually have to check for quotes, when we can simply replace them with themselves if they already exist.

Unfortunately, with this particular input, we either have to use sed in basic regex mode and escape the extended regex characters, or use extended regex mode (-r) and escape the special characters in the string. So to avoid the "picket fence" effect and help readability, I also changed 's///' to 's|||'. (Any ascii character not found in the pattern or the input string can be used as the delimiter.)

And finally, I found the sed FAQ to be very useful for learning how to use the full power of the command... although it is a bit out of date now:

http://sed.sourceforge.net/sedfaq.html

David the H. · 01-01-2020, 11:33 AM

By the way, here's one more option that's just dawned on me. If the input allows it, you could use two 's' patterns and modify the start and end quotes separately, matching either quoting possibility. Just be sure to use strings long enough to match uniquely.

Code:

$ sed -r '/exit 0/ { s/"?\$\{N/"${N/ ; s/G}"?/G}"/ }' infile.txt
[ "${NETWORKING}" = "no" ] && exit 0
[ "${NETWORKING}" = "no" ] && exit 0
[ ${NETWORKING} = "no" ] && exit 1
Another unimportant ${NETWORKING} string

Be careful, though, when it comes to chaining commands. Later commands can potentially match and modify the output of earlier ones.