Escape characters in sed stream
 

I'm trying to use sed on the linux command line, and am running into an issue with special characters.

I'd like to replace "~" with "E;" in the target file (without quotations). So far, using this command:

sed {s/\~/\&#7E\;/g}

on this data:

blah~blah

produces:

blah~#7E;blah

Close, but no cigar.

Thanks in advance for any insight you can provide.

I would first ask how you arrived at the expression you are using? Why does it include "&#7" in the replacement string if you do not want them in the replacement text? And why the curly braces?

And by extension, what happens if you remove those characters, have you tried to do that?

I don't believe '~' is a special character. At least not when using regular expressions with sed. It is in the Bash shell. It expands the user's home directory. So it doesn't need to be escaped with sed.

FWIW sed -i 's/~/E;/g' file.txt worked for me

The forum software "edited" my desired target string. The curly braces are standard for a sed command-line script, by my understanding.

The ampersand is the issue, I believe. It is a sed special character.

The curly braces are the problem (and I've never seen them used with sed before). This works:And so does this:And this:Note the double escaping in the last example, which is necessary due to both \ and & being treated as special characters in that context.

Edit: The forum messed up the characters again.

Thanks all. My ignorance is showing.

Just to close this out, curly brackets are used to group commands - typically when a selection criteria is met; say particular line number or regex match. It is a standard, if not everyday, usage in sed.

Ah! I tripped over the forum-mangled post myself...

You don't need to escape the tilde, '~' for the regular expression, but you do need to escape the ampersand, '&' because it is special, and the semi-colon to hide it from the shell (i.e. for a different reason than the &).

The curly braces are unnecessary, but I would suggest single quotes around the whole expression in which case there is no need to additionally escape the tilde and semi-colon as the shell does not see them:

To clarify further, in Sed replacement strings, ampersand represents "the matched text" - i.e. it's the same as \0 (back-reference zero), and thus needs to be escaped to provide a literal ampersand character.

This doesn't apply in the match pattern - ampersands do not need to be escaped there.

(Aside: Some regex engines have a concept of a union operator for combining multiple character classes, and use a double ampersand for this; sed is not one of those.)