[SOLVED] Why does sed not process all occurrences?

Issue

Or: How do I get sed to replace all occurrences?

$ echo ":::::::" |sed -e 's/::/:0:/g'
:0::0::0::

I would expect/want the output to be :0:0:0:0:0:0:

Solution

Because once :: seen, next :: pattern doesn't contain last : of 1st pair!

Your string is parsed as :: replaced by :0:, then ::, :: and :. So there are 3 :: followed by :. This could be replaced by :0::0::0::. The operation seem correct!

So operation have to be done two time to ensure all :: pair will be modified.

sed first

As requested, I answer by using sed first and because sed is the more efficient tool for this operation on big streamed files. See bash further, more efficient for variables and small files.

hardcoding two steps

Simply:

echo ":::::::" | sed -e 's/::/:0:/g;s/::/:0:/g'
:0:0:0:0:0:0:

Like many programmer (publishing here) this solution was not the first I was thinked about. Anyway, for this problem, hardcoding two step is the more efficient solution!

loop now

Because I hate to do hardcoding, the first solution I found was to do a loop until all :: pair will be changed:

So we have to add a branch condition:

echo ":::::::" |sed -e ':a;s/::/:0:/;ta'
:0:0:0:0:0:0:

t statment branch to :a label if last s/// operation does some modification.

So sed will loop over whole line while :: will be found.

Note I've supressed g switch, but this will make more loop over line. Compare (I use lower l sed command to dump current line. see info sed for more informations regarding this):

echo ":::::::" |sed -e ':a;s/::/:0:/;l;ta'
:0::::::$
:0:0:::::$
:0:0:0::::$
:0:0:0:0:::$
:0:0:0:0:0::$
:0:0:0:0:0:0:$
:0:0:0:0:0:0:$
:0:0:0:0:0:0:

with 6 loop (1 by 0 added), and

echo ":::::::" |sed -e ':a;s/::/:0:/g;l;ta'
:0::0::0::$
:0:0:0:0:0:0:$
:0:0:0:0:0:0:$
:0:0:0:0:0:0:

where 2 loop will do the whole job.

bash now.

Of course this could be done by using a lot of different tools like awk, python, perl, javascript, ruby...

But this could be done in a very efficient way, by using pure bash:

var=':::::::'
echo ${var//::/:0:}
:0::0::0::

With the same problem!

hardcoded

var=':::::::'
var=${var//::/:0:}
var=${var//::/:0:}
echo $var
:0:0:0:0:0:0:

In a read loop:

echo >testfile ':::::::'
while read -r var ;do
    var=${var//::/:0:}
    echo "${var//::/:0:}"
done <testfile
:0:0:0:0:0:0:

bash loop

So by doing a loop, this could look like:

var=':::::::'
while [ "$var" != "${var//::/:0:}" ] ;do
    var="${var//::/:0:}"
done
echo "$var"
:0:0:0:0:0:0:

As this don't have to fork to another binary, this could be quicker.

And if doing translation two time seem too much, you could write this:

var=':::::::'
while tmpvar="${var//::/:0:}" ; [ "$var" != "$tmpvar" ] ;do
    var="$tmpvar"
done
echo $var
:0:0:0:0:0:0:

Show steps:

var=':::::::'
while
  tmpvar="${var//::/:0:}"
  [ "$var" != "$tmpvar" ]
  do
    var="$tmpvar"
    printf 'Step: \047%s\047\n' "$tmpvar"
done
Step: ':0::0::0::'
Step: ':0:0:0:0:0:0:'

echo $var
:0:0:0:0:0:0:

Or if you drop 1st / to do only 1 modification by step:

var=':::::::'
while
  tmpvar="${var/::/:0:}"
  [ "$var" != "$tmpvar" ]
  do
    var="$tmpvar"
    printf 'Step: \047%s\047\n' "$tmpvar"
done
Step: ':0::::::'
Step: ':0:0:::::'
Step: ':0:0:0::::'
Step: ':0:0:0:0:::'
Step: ':0:0:0:0:0::'
Step: ':0:0:0:0:0:0:'

But this become slower!

Answered By - F. Hauri

Answer Checked By - David Marino (WPSolving Volunteer)