I'm trying to replace a character in a file at a random position. My file looks something like:


I'm trying to replace a random character for 'c'. So the output might look like:


I have tried removing all line breaks and saving in a file new_string.txt and then using sed but it isn't working.

This is the code I have tried:

rand1="$(shuf -i 0-$tot_len -n 1)"
sed "s/^\(.\{"${rand1}"\}\)./\1G/" new_string.txt

I keep getting the error:

sed: -e expression #1, char 25: Invalid content of \{\}
  • 1
    A random character? Or a random character other than newline? – Stéphane Chazelas Aug 11 '15 at 11:59
  • @StéphaneChazelas he meant a character at a random position in the string. – FelixJN Aug 11 '15 at 12:03
  • A random character apart from the newline. – user3891532 Aug 11 '15 at 12:05
  • A random character or a character at random position or a random character at random position? – jimmij Aug 11 '15 at 12:06
  • Sorry for not being clear. A specified character at a random position – user3891532 Aug 11 '15 at 12:07

No need for the curly brackets in your variable, and the variable should be quoted as well. Use:

sed "s/^\(.\{$rand1\}\)./\1G/" new_string.txt

UPDATE: as stated below in comments:

The original code is fine, however the integer for $rand1 is too large for sed. I found that the maximum value can be 32767 for GNU sed, i.e. sed still takes 16bit integers only.

You can obtain that limit for the system's regular expression library (though GNU sed generally uses a builtin version) with:

$ getconf RE_DUP_MAX

POSIX requires that limit to be at least _POSIX_RE_DUP_MAX (255), and that's the maximum you can expect portably (some systems like Solaris or OS/X have it as low as that).

  • I've tried that too. I still get the same error. – user3891532 Aug 11 '15 at 12:03
  • what is echo $rand1 giving you? because it worked in my test. – FelixJN Aug 11 '15 at 12:04
  • echo $rand1 works fine – user3891532 Aug 11 '15 at 12:05
  • what is the output of echo $rand1 | cat -A ? – FelixJN Aug 11 '15 at 12:09
  • 5255810$ . 5255810 is the value of $rand1 and my string is larger than that – user3891532 Aug 11 '15 at 12:11

On a GNU system, to substitute one character (other than newline) at random, you could do:

offset=$(grep -bo . < "$file" | cut -d: -f1 | shuf -n1)
[ -z "$offset" ] || # file doesn't have non-newline characters
  printf c | dd bs=1 seek="$offset" of="$file" conv=notrunc status=none

(with old versions of GNU dd (prior to 8.20), replace status=none with 2> /dev/null).

grep -bo . < "$file" would give you the offset in number of bytes in the file of each non-newline character. For instance, with a file encoded in UTF-8 that contains:


That gives us:

$ grep -bo . < "$file"

With cut -d: -f1, we retain the part before the first colon. Then, we pick one of those offsets at random with shuf -n1.

That assumes the replacement character has the same size as the replaced one. For instance, replacing that £ above (2 bytes) with c (1 byte) would leave the file with c followed by an invalid character.

To work around that, we can't overwrite the file in-place anymore as we'd need to shift data around.

We'd need something like:

perl -C -0777 -pi -e "substr \$_, $offset, 1, 'c'" -- "$file"

instead. With -C, perl honours the locale for what constitutes a character. -0777 -p turns on the slurp mode where the content of $file is slurped into $_ (see Security implications of running perl -ne '…' * though for security considerations with that construct). -pi gives you in-place editing, $_ is written back to the file after the code is run. Then we call substr to substitute the 1 character at the given offset with c.

  • I've just blindly copied your code as I don't quite understand what you've done but I have the error message: dd: invalid status flag: ‘none’ – user3891532 Aug 11 '15 at 12:25
  • @user3891532, see edit, you must have an old version of GNU dd. – Stéphane Chazelas Aug 11 '15 at 12:36

With new GNU sed you can do it even without \newline remove

sed -z "s/./@/$(($RANDOM%$(wc -m < file.txt)))" file.txt
  • Note that $RANDOM is ksh/zsh/bash specific and is limited to 0-32767. – Stéphane Chazelas Aug 11 '15 at 13:27
  • That assumes file.txt doesn't contain NUL characters (which should be a safe assumption for a text file, just as that it contains only valid characters in the current locale which is also a requirement (and of my answer as well)). – Stéphane Chazelas Aug 11 '15 at 13:28
  • Note that it may replace newline characters. – Stéphane Chazelas Aug 11 '15 at 13:29

Try this:

sed 's/^\(.\{'"${rand1}"'\}\)./\1G/'  new_string.txt
  • That doesn't work either – user3891532 Aug 11 '15 at 12:06
  • are you still geting same error?? – Shravan Yadav Aug 11 '15 at 12:07
  • It's still the same error. – user3891532 Aug 11 '15 at 12:08
  • What is value in tot_len?? Is it getting some value or it os blank?? – Shravan Yadav Aug 11 '15 at 12:33
  • may be rand1 value is getting some value larger than the string length. – Shravan Yadav Aug 11 '15 at 12:48

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.