I Sed, she sed

[Goose]

Does anyone here use sed much? I've got a bit stuck, could do with some help! :(

[Garp]

Does anyone here use sed much? I've got a bit stuck, could do with some help! :(

Whatcha trying to achieve? I use it on and off here.

[Goose]

Right, this is a simple example. In real times, my data is a bit hairy but the idea should work.

I have a file populated a little like this:

abcdefghijklmn<<o>>pqrstuvwxyz
abcdefghijklmnopqrstuvwxyz
abcdefg<h>ijklmnopqrstuvwxyz
abcdefghijkl<<m>>nopqr<<s>>tuvwxyz
abcdef<g>hijklmnopqrs<<t>>uvwxyz
abcdefghijklmno<<p>>qrstuvwxyz

I need to use sed to extract out anything surrounded by << >>. So ideally I'd end up with:

o
m
s
t
p


It's frustrating me somewhat.

[Mark]

The fact you can have more than one pattern per line caused me some major grief, but if it's OK to use sed twice, then I managed to get around that one:

sed '/\(<<[^<>]*>>\)/s//\n\1\n/g' | sed '/^<<\([^<>]*\)>>$/s//\1/p;d'

The first sed breaks out the <<...>> so that each is split out onto a line by itself, then the second one picks out just those lines.

[Goose]

Thanks Mark. I've not used sed much, takes some getting used to.

[Mark]

Make sure you use the latest edit - I spotted a few bits of redundant code in the original and removed them.

[Fayshun]

sed '/\(<<[^<>]*>>\)/s//\n\1\n/g' | sed '/^<<\([^<>]*\)>>$/s//\1/p;d'

Such language on a family forum!

[Mark]

I do wonder what that would say when translated into semaphore and back. ;D ;D

[Goose]

OK, second challange. This should be easy, I've nearly got it working but seem to be running in circles.

Say you have this:

{.One$="One",.Number=12,.Two$="Two",.Number=13,.Three$="Three",/

I need to grab everything between .Two$=" and the following "

A pint to whoever digs me out of sed hell #2.

I really should buy a book on this...

[Garp]

http://www.amazon.co.uk/sed-awk-Pocket-Reference-OReilly/dp/0596003528/ref=sr_1_2?ie=UTF8&s=books&qid=1214233854&sr=8-2

[Goose]

*click* bought.

[LeperousDust]

Jesus, that actually looks foreign to me... :huh:

[Goose]

Book shmook, got there in the end.

Still, it's a handy thing to learn.

[Goose]

The fact you can have more than one pattern per line caused me some major grief, but if it's OK to use sed twice, then I managed to get around that one:

sed '/\(<<[^<>]*>>\)/s//\n\1\n/g' | sed '/^<<\([^<>]*\)>>$/s//\1/p;d'

The first sed breaks out the <<...>> so that each is split out onto a line by itself, then the second one picks out just those lines.
Just realised this screws up when it comes across something like:

abcdefgh<<ijklmno>pqrst>>uvwxyz
abcdefgh<<ijklmno<pqrst>>uvwxyz

:'(

[Mark]

You may have reached the limits of what sed can do for you there. It'll also break with this one:
abcdefgh<<ijklmno<>pqrst>>uvwxyz

[Goose]

Any suggestions, Mr Mark? I think I need to buy you a pub... ;D

[Mark]

Windows or Linux? There's a way to cheat it that I hadn't thought of before, but I'll have to put it into a script as it uses unprintable characters.

[Goose]

Linux. Any suggestions would be more than welcome.

[Mark]

OK - one more.

If I gave it:

abcdefg<<<<hijkl>>mno>>pqrstuvwxyz
Would the answer 'hijkl' be OK, or do you want '<<hijkl>>mno', or is that just a case of 'never going to happen'?

I've solved all the other cases, but this one might resist.

[Goose]

It'd never exist. The '<<' and '>>' will always match, there may be more than 1 per line, they would never overlap, and there may be single '<' and '>' in each.

[Mark]

http://www.ouroboros.me.uk/misc/for-goose.sh

Converts:

abcdefghijklmn<<o>>pqrstuvwxyz
abcdefghijklmnopqrstuvwxyz
abcdefg<h>ijklmnopqrstuvwxyz
abcdefghijkl<<m>>nopqr<<s>>tuvwxyz
abcdef<g>hijklmnopqrs<<t>>uvwxyz
abcdefghijklmno<<p>>qrstuvwxyz
abcdefgh<<ijklmno>pqrst>>uvwxyz
abcdefgh<<ijklmno<pqrst>>uvwxyz
abcdefgh<<ijklmno<>pqrst>>uvwxyz
abcdefg<<<<hijkl>>mno>>pqrstuvwxyz

To:

o
m
s
t
p
ijklmno>pqrst
ijklmno<pqrst
ijklmno<>pqrst
hijkl

I think (hope) that's everything. ;D

[Goose]

I want your babies.