grab last few lines until meet the first pattern

cristalp · 10-24-2011, 09:52 AM

Dear Experts,

I want to read a file and save last few lines into a new file.

The number of the lines I need was defined by the first "PATTERN" counted from the bottom line of the file. For example, suppose my file looks like:

Code:

First line of the file
line 2
line 3
...
...
line 98
"PATTERN"
line 100
line 101
line 102
...
...
line 187
"PATTERN"
line 189
line 190
...
...
line 308
"PATTERN"
line 310
line 311
...
...
line 359
line 360
bottom line(the last line of the file)

What I want to grab from the file and save to the new file is:

Code:

line 310
line 311
...
...
line 359
line 360
bottom line(the last line of the file)

I do NOT know in advanced where the "PATTERN" locate and how many lines between each of them.

How could I do this easily in bash script?

I would appreciate any help! Thanks for your time.

sycamorex · 10-24-2011, 10:06 AM

How about:

Code:

sed '1,/pattern/d' file > newfile

Would it be what you're after?

cristalp · 10-24-2011, 10:19 AM

Quote:

Originally Posted by sycamorex

How about:

Code:

sed '1,/pattern/d' file > newfile

Would it be what you're after?

Thanks, I tried your code, but it doesn't work for my purpose. It print out a lot more lines which I do not need. I only need the lines bewteen the last line and the last "PATTERN".

sycamorex · 10-24-2011, 10:25 AM

Quote:

Originally Posted by cristalp

Thanks, I tried your code, but it doesn't work for my purpose. It print out a lot more lines which I do not need. I only need the lines bewteen the last line and the last "PATTERN".

I'm not at my linux computer right now. Once I get home I'll double check it. In the meantime, someone else will probably correct it.

cristalp · 10-24-2011, 10:33 AM

Quote:

Originally Posted by sycamorex

I'm not at my linux computer right now. Once I get home I'll double check it. In the meantime, someone else will probably correct it.

Thanks a lot! Take your time, don't need to be so hurry. I will also try something more and let you know if I get some results.

grail · 10-24-2011, 10:58 AM

Well a not so clean way:

Code:

tac file | sed '/PATTERN/,$ d' | tac

crts · 10-24-2011, 11:25 AM

You could also try:

Code:

sed -n '/PATTERN/! {H;${g;s/\n//;p};b};z;h' file

sycamorex · 10-24-2011, 01:10 PM

hmmm, indeed. My solution doesn't work. I thought you could mix line numbers and patterns.

David the H. · 10-24-2011, 05:18 PM

@sycamorex

sed does allow you to mix&match. It's just that your command matches from line 1 to the first instance of the pattern in the file, and he needs it to match to the last instance.

Notice how grail's solution works, by reversing the file first.

Edit: While they don't appear to address this problem specifically, these links can may you to work out such complex situations.

http://sed.sourceforge.net/sedfaq.html
http://sed.sourceforge.net/sed1line.txt

And the ever-popular grymoire tutorial:

http://www.grymoire.com/Unix/Sed.html

sycamorex · 10-24-2011, 06:13 PM

Quote:

Originally Posted by David the H.

@sycamorex

sed does allow you to mix&match. It's just that your command matches from line 1 to the first instance of the pattern in the file, and he needs it to match to the last instance.

Notice how grail's solution works, by reversing the file first.

Edit: While they don't appear to address this problem specifically, these links can may you to work out such complex situations.

http://sed.sourceforge.net/sedfaq.html
http://sed.sourceforge.net/sed1line.txt

And the ever-popular grymoire tutorial:

http://www.grymoire.com/Unix/Sed.html

Thanks David. I know the links. Mind you I haven't visited them in months.

Initially, I thought that sed, being greedy as always, will match as many instances of PATTERN as possible.

My bad.

David the H. · 10-24-2011, 06:22 PM

I actually posted the links mostly for the OP, but I guess I didn't make that clear.

And I wouldn't call sed itself greedy, just the regular expressions that you often use in it. When it comes to multi-line operations, sed can often be a real pain to use.

grail · 10-24-2011, 09:58 PM

Maybe a little awk:

Code:

awk '{l = l (l?"\n":"") $0}/PATTERN/{l=""}END{print l}' file