Shell Programming and Scripting

Input: a/b/c
Output required: a\/b\/c

This does not work:
sed s/'\/'/'\//'/g

Try:

sed 's!/!\\/!g'

Thanks
How does this work? Can you please explain.

Here ! is the seperater, it will look for the pattern "/" and replace it with \/. \ is escaped.

OK a question:



Why this doesnt work:
node=`echo $node | sed 's!/!\\/!g'`
echo $node
output: a/b/c

But this works
echo $node | sed 's!/!\\/!g'
output: a\/b\/c

This is two times the same expression - either of the same variant works. ;-)

I suppose you want to know why this works and yours (from the first post) doesn't:

Normally (that is: out of habit) regular expressions are written with "/" as delimiter:

Code:

sed 's/<pattern>/<replacement>/'

but this is just a tradition, not enforced by the rules of regex language. You can use every character as delimiter, sed will take the character following the initial "s" and look for exactly the same character to determine where the parts of the regex end. The following example are all equivalent and will change an "x" to a "y".

Code:

s/x/y/
s@x@y@
s x y

You see that even a space character can act as delimiter (i do NOT recommend using a space, its just possible). Btw.: if the last example fails it is probably because you forgot the last character, a trailing space! Otherwise the regex would miss its delimiter. Use 's x y ' instead of 's x y'.

Now, regardless of which character we actually use as a delimiter, we always have a problem when we need to use the character literally (that is: NOT in its function as delimiter). To phrase it more generally (because this is really a very general principle): whenever we use a character to not adress the character itself but for some "other meaning" (like the asterisk, which means not an asterisk but "zero or more of the preceeding character") we have a problem when we want to use the character literally.

In the case of the delimiter character we could simply change the delimiter character - this is ultimately, what dennis.jacob did. But it is also possible to "flag" the character to tell the regex program it is not used in its function but merely as the character itself - in other words: to escape the character. This is usually done by preceeding the character in question with a backslash. The backslash tells sed that the following character should not be interpreted with its special meaning but simply as the character it is. The following example changes the "/" to "\/" without changing the delimiter character:

Code:

echo "a/b/c" | sed 's/\//\\\//g'

Lets talk this through, part by part:

s/ - so far, normal sed: the command (substitute) and the first delimiter
\/ - this is the slash, but escaped to tell sed it is not used as delimiter
/ - this is the delimiter, though
\\ - Because we use the backslash with a special meaning - to escape other characters - we need to escape itself too, to declare we want to use it literally. This is why a literal backslash is written "\\".
\/ - Again our slash, escaped to set it apart from the delimiter
/g - the delimiter and an eventual command to process all occurrences instead of just the first

I hope this helps.

bakunin

Thanks for detailed explanation, bakunin!

No there was a slight problem, I figured out myself why this didn't work:
node=`echo $node | sed 's!/!\\/!g'`
echo $node
a/b/c
I needed to put another "\" because the expression is enclosed in single quotes. When it expands it it escapes the \ and net effect is i am left with the same thing. So this works:
node=`echo $node | sed 's!/!\\\/!g'`
echo $node
a\/b\/c