[SOLVED] sed - why '[.]' matches with the beginning and end of a line?

Issue

Why the the beginning and end of a line is matched by the [.]* and how to avoid it?

It seems the patten matches with the word boundary but not sure if it is as designed and which specification it implements.

$ echo " a " | sed -n 's/[.]*/X/pg'
X XaX X

$ echo " a " | sed -n 's/[\b]*/X/pg'
X XaX X

Solution

Note that \b inside brackets has no special meaning. It is just a list of characters.

[.]* and [\b]* match zero or more of the character(s) in brackets.

So they match any runs of one or more of the characters, where the empty strings immediately preceding and following a run of the character(s) are just part of that run.

They also match any empty string that is not immediately preceded or followed by a run of the character(s).

---

Replacing space with something more visible, your input is: sas

None of the characters ., \ nor b appear in the string. So there are no runs of length one or longer. With this input, both [.]* and [\b]* are equivalent to "match empty string".

• the empty string between start of line and first s matches
• the empty string between first s and a matches
• the empty string between a and second s matches
• the empty string between second s and end of line matches

These 4 matches explain the Xs added to your sample output.

---

Using \b to mean word boundary is not standard, although some versions of sed accept it (or the related \< and \>).

It is safer not to use this extension, and certainly not with *.

Even versions of sed that appear to support it produce non-intuitive and inconsistent results.

For example, with GNU sed 4.8:

$ echo ,aa, | sed 's/\b/x/g'
,xaax,
$ echo ,aa, | sed 's/\b*/x/g'
,aa,
$ echo ,aa, | sed 's/\b\{1,\}/x/g'
sed: -e expression #1, char 14: Invalid preceding regular expression
$ echo ,aa, | sed 's/\(\b\)\{1,\}/x/g'
,xaax,

With busybox sed 1.30.1:

$ echo ,aa, | busybox sed 's/\b/x/g'
,xaxa,
$ echo ,aa, | busybox sed 's/\b*/x/g'
,aa,
$ echo ,aa, | busybox sed 's/\b\{1,\}/x/g'
sed: bad regex '\b\{1,\}': Invalid preceding regular expression
$ echo ,aa, | busybox sed 's/\(\b\)\{1,\}/x/g'
,xaxa,

Even other programs like Perl require care:

$ echo ,aa, | perl -ple 's/\b/x/g'
,xaax,
$ echo ,aa, | perl -ple 's/\b*/x/g'
x,xaxax,x
$ echo ,aa, | perl -ple 's/\b{1,}/x/g'
'1,' is an unknown bound type in regex; marked by <-- HERE in m/\b{1, <-- HERE }/ at -e line 1.
$ echo ,aa, | perl -ple 's/(\b){1,}/x/g'
,xaax,
$ echo ,aa, | perl -ple 's/\b+/x/g'
,xaax,

Answered By - jhnc

Answer Checked By - Gilberto Lyons (WPSolving Admin)