sed between two patterns where result contains a third pattern

I'm attempting to filter log files for xml responses, and using sed as below, it is fairly easy to find all the xml messages.

sed -n '/<element/,/</element/p' file


Returns:

<element>
<id>12345</id>
...
</element>
<element>
<id>54321</id>
...
</element>


However I have been unable to figure out how to apply a second filter which means only xml responses containing a certain pattern, such as an ID, are returned.

In the above example, how would I filter on ID to only return the first one?

3 Answers
3

sed is for doing s/old/new/ THAT IS ALL. All of it's wacky single character runic language constructs became obsolete in the mid 1980s when awk was invented.

$ cat tst.awk
/<element>/ { inElt = 1 }
inElt {
elt = (elt == "" ? "" : elt ORS) $0
if ( /</element>/ ) {
if ( elt ~ /<id>12345</id>/ ) {
print elt
}
elt = ""
inElt = 0
}
next
}
{ print }

$ awk -f tst.awk file
<element>
<id>12345</id>
...
</element>


The main benefits of the above over the currently accepted sed solution are:

</element

For example, lets say you wanted to print the first element in the file regardless of it's ID rather than the one containing a specific ID. That'd be a trivial tweak of the above to:

$ cat tst.awk
/<element>/ { inElt = 1 }
inElt {
elt = (elt == "" ? "" : elt ORS) $0
if ( /</element>/ ) {
if ( ++cnt == 1 ) {
print elt
}
elt = ""
inElt = 0
}
next
}
{ print }

$ awk -f tst.awk file
<element>
<id>12345</id>
...
</element>


If you want to print the 27th instead of the 1st element, just change ++cnt == 1 to ++cnt == 27. Try modifying the sed script for such a trivial requirements change and you can look forward to a complete rewrite and having to invoke additional tools. Want to print multiple elements and/or other parts of the file not within element tags? Also absolutely trivial with awk. Hopefully you get the point.

++cnt == 1

++cnt == 27

You can group commands for your ranges:

sed -n '/<element/,/</element/{ /id/p }'


But, you should really consider using XML tools when dealing with XML, such as xmlstarlet.

In order to print the complete entry when searching for a specific ID, you need to accumulate the lines inside the <element> node using the hold space, if you reached the end tag of an <element> node, you can replace the hold and pattern spaces, match for your ID and print it:

<element>

<element>

sed -n -e '
/<element/,/</element/H # append to the hold space
/</element/{
g # replace pattern space with hold space
/<id>12345</id>/p # print if matching ID
s/.*// # clear pattern space
x # clear hold space
b # start next cycle without further output
}' input-file


You see, this gets messy really fast.

This might work for you (GNU sed):

sed -n '/<element>/{:a;/</element>/!{N;ba};/<id>12345</id>/p}' file


Use seds grep-like nature by using the -n option which turn off automatic printing of every line. On encountering a line that contains <element> gather up a collection of lines until the end tag </element> is reached. Now check the collection for <id>12345</id> and print the collection if true otherwise the collection is passed over.

-n

<element>

</element>

<id>12345</id>

If instead you want a specific element e.g. the second, use:

sed -n '/<element>/{:a;/</element>/!{N;ba};x;s/^/x/;/^x{2}$/{x;p;b};x}' file


This uses a counter held in the hold space which is incremented on each complete collection and checked for a specific number.

N.B. The range operator , can be used as a flip-flop type command but in general the start address{:a;N;end address!ba; commands on collection} is more useful.

,

start address{:a;N;end address!ba; commands on collection}

By clicking "Post Your Answer", you acknowledge that you have read our updated terms of service, privacy policy and cookie policy, and that your continued use of the website is subject to these policies.