Shell Programming and Scripting

Some sed versions do have the \< and \> boundary anchors; they do not consume a character. And not only allow a /g in one go but also make the substitution easier.

Code:

sed 's#\<[0-9]/#0&#g'

--
Just seeing that the Solaris /usr/xpg4/bin/sed counts the \< as a character that is truely a bug (/usr/bin/sed works).

Hi MIG,
How below highlighted part is working.

Code:

sed 's#\<[0-9]/#0&#g'

.Why only one sided boundary is being used. Please explain.

The right side is a /
If it were a \> then the following would be changed as well

Code:

4.2.2000
4-2-2001

I see no difference in behavior between the Solaris xpg4 and the regular svr4 sed implementations.

Your syntax only works with GNU sed, /usr/gnu/bin/sed on Solaris 11.

Why do you think this is a bug in /usr/xpg4/bin/sed? The standards say that the REs accepted by sed are Basic Regular Expressions (BREs) with three minor tweaks that don't affect this command. In a BRE the meaning of a backslash followed by a character that is not a BRE special character, a <left-parenthesis>, <right-parenthesis>, <left-curly-bracket>, <right-curly-bracket>, the current delimiter character (the character that delimits the BRE and replacement from the command and flags), and the decimal digits 1 through 9 is undefined (and all of these are valid after a <backslash> only in certain contexts). The BRE special characters are <period>, <left-square-bracket>, <backslash>, <asterisk>, <circumflex>, and <dollar-sign> and all of these are also only special in certain contexts.

According to the standards, using \< or \> in a BRE can be treated by the implementation as a request to match a < or >, respectively; match a word boundary; abort with a BRE syntax error; hunt down and kill the programmer who wrote that BRE, or anything else it chooses to do. (Legal and marketing implications make it unlikely that an implementation will make the next to the last choice in that list. )

The Solaris man page for both /usr/bin/sed and /usr/xpg4/bin/sed refers to the regexp man page, that states

Code:

...
     3.1             \< constrains a RE to match the beginning of
                     a  string  or  to follow a character that is
                     not a  digit,  underscore,  or  letter.  The
                     first  character  matching  the RE must be a
                     digit, underscore, or letter.

Even if it does not explicitly say that the \< should never count as a character, the implementations in /usr/bin/sed and /usr/xpg4/bin/sed should be consistent.

Hi MadeInGermany,
OK. Yes, the fact that /usr/xpg4/bin/sed does not behave as documented by the Solaris sed man page is definitely a bug. (It isn't a standards conformance issue, but it is a bug.)

Thanks,
Don