I have a very large (~60MB) text file in which I want to replace specific block lines with a predefined text. The line number of every block (3 lines) start is known and they are stored in a file:
...
11
30
42
58
...
I know that I can use the following option in order to replace a block:
sed -i "Xs,(X+3)s/.*/REPLACEMENT/" filename.txt
However, executing this command in a for loop like:
for line in $(cat linenumbers.txt); do
eline=$((${line}+3))
sed -i "Xs,(X+3)s/.*/REPLACEMENT/" filename.txt
done
is very slow and takes a lot of time (> 10') and I have 100s of files in which I have to replace blocks.
Is there any other way to instruct sed to do that in one pass?
awk to the rescue!
$ awk 'NR==FNR {start[$1]; next}
FNR in start {c=3}
c&&c-- {print "replacement"; next}1' indices file
this is a one pass process, you can save the output into a new file and overwrite the original one if you want.
Similar to #karakfas answer but a different interpretation of your requirements (hint: an actual example with input and output would have cleared up the confusion):
awk '
NR==FNR { start[$1]; next }
FNR in start { print "replacement"; c=3 }
c&&c-- { next }
{ print }
' indices file
This might work for you (GNU sed):
sed 's/.*/&,+3cREPLACEMENT/' lineNumbersFile | sed -f - file
Convert the line numbers file into a sed script and run it against the data file.
Related
I am trying to replace every 4th occurrence of "_" with "#" in multiple files with bash.
E.g.
foo_foo_foo_foo_foo_foo_foo_foo_foo_foo..
would become
foo_foo_foo_foo#foo_foo_foo_foo#foo_foo...
#perl -pe 's{_}{++$n % 4 ? $& : "#"}ge' *.txt
I have tried perl but the problem is this replaces every 4th _ carrying on from the last file. So for example, some files the first _ is replaced because it is not starting each new file at a count of 0, it carries on from the previous file.
I have tried:
#awk '{for(i=1; i<=NF; i++) if($i=="_") if(++count%4==0) $i="#"}1' *.txt
but this also does not work.
Using sed I cannot find a way to keep replacing every 4th occurrence as there are different numbers of _ in each file. Some files have 20 _, some have 200 _. Therefore, I cant specify a range.
I am really lost what to do, can anybody help?
You just need to reset the counter in the perl one using eof to tell when it's done reading each file:
perl -pe 's{_}{++$n % 4 ? "_" : "#"}ge; $n = 0 if eof' *.txt
This MAY be what you want, using GNU awk for RT:
$ awk -v RS='_' '{ORS=(FNR%4 ? RT : "#")} 1' file
foo_foo_foo_foo#foo_foo_foo_foo#foo_foo..
It only reads each _-separated string into memory 1 at a time so should work no matter how large your input file, assuming there are _s in it.
It assumes you want to replace every 4th _ across the whole file as opposed to within individual lines.
A simple sed would handle this:
s='foo_foo_foo_foo_foo_foo_foo_foo_foo_foo'
sed -E 's/(([^_]+_){3}[^_]+)_/\1#/g' <<< "$s"
foo_foo_foo_foo#foo_foo_foo_foo#foo_foo
Explanation:
(: Start capture group #1
([^_]+_){3}: Match Match 1+ of non-_ characters followed by a _. Repeat this group 3 times to match 3 such words separated by _
[^_]+: Match 1+ of non-_ characters
): End capture group #1
_: Match a _
Replacement is \1# to replace 4th _ with a #
With GNU sed:
sed -nsE ':a;${s/(([^_]*_){3}[^_]*)_/\1#/g;p};N;ba' *.txt
-n suppresses the automatic printing, -s processes each file separately, -E uses extended regular expressions.
The script is a loop between label a (:a) and the branch-to-label-a command (ba). Each iteration appends the next line of input to the pattern space (N). This way, after the last line has been read, the pattern space contains the whole file(*). During the last iteration, when the last line has been read ($), a substitute command (s) replaces every 4th _ in the pattern space by a # (s/(([^_]*_){3}[^_]*)_/\1#/g) and prints (p) the result.
When you will be satisfied with the result you can change the options:
sed -i -nE ':a;${s/(([^_]*_){3}[^_]*)_/\1#/g;p};N;ba' *.txt
to modify the files in-place, or:
sed -i.bkp -nE ':a;${s/(([^_]*_){3}[^_]*)_/\1#/g;p};N;ba' *.txt
to modify the files in-place, but keep a *.txt.bkp backup of each file.
(*) Note that if you have very large files this could cause memory overflows.
With your shown samples, please try following awk program. Have created an awk variable named fieldNum where I have assigned 4 to it, since OP needs to enter # after every 4th _, you can keep it as per your need too.
awk -v fieldNum="4" '
BEGIN{ FS=OFS="_" }
{
val=""
for(i=1;i<=NF;i++){
val=(val?val:"") $i (i%fieldNum==0?"#":(i<NF?OFS:""))
}
print val
}
' Input_file
With GNU awk
$ cat ip.txt
foo_foo_foo_foo_foo_foo_foo_foo_foo_foo
123_45678_90
_
$ awk -v RS='(_[^_]+){3}_' -v ORS= '{sub(/_$/, "#", RT); print $0 RT}' ip.txt
foo_foo_foo_foo#foo_foo_foo_foo#foo_foo
123_45678_90
#
-v RS='(_[^_]+){3}_' set input record separator to cover sequence of four _ (text matched by this separator will be available via RT)
-v ORS= empty output record separator
sub(/_$/, "#", RT) change last _ to #
Use -i inplace for inplace editing.
If the count should reset for each line:
perl -pe's/(?:_[^_]*){3}\K_/\#/g'
$ cat a.txt
foo_foo_foo_foo_foo_foo_foo_foo_foo_foo
foo_foo_foo_foo_foo_foo_foo_foo_foo_foo
$ perl -pe's/(?:_[^_]*){3}\K_/\#/g' a.txt a.txt
foo_foo_foo_foo#foo_foo_foo_foo#foo_foo
foo_foo_foo_foo#foo_foo_foo_foo#foo_foo
foo_foo_foo_foo#foo_foo_foo_foo#foo_foo
foo_foo_foo_foo#foo_foo_foo_foo#foo_foo
If the count shouldn't reset for each line, but should reset for each file:
perl -0777pe's/(?:_[^_]*){3}\K_/\#/g'
The -0777 cause the whole file to be treated as one line. This causes the count to work properly across lines.
But since a new a match is used for each file, the count is reset between files.
$ cat a.txt
foo_foo_foo_foo_foo_foo_foo_foo_foo_foo
foo_foo_foo_foo_foo_foo_foo_foo_foo_foo
$ perl -0777pe's/(?:_[^_]*){3}\K_/\#/g' a.txt a.txt
foo_foo_foo_foo#foo_foo_foo_foo#foo_foo
foo_foo_foo#foo_foo_foo_foo#foo_foo_foo
foo_foo_foo_foo#foo_foo_foo_foo#foo_foo
foo_foo_foo#foo_foo_foo_foo#foo_foo_foo
To avoid that reading the entire file at once, you could continue using the same approach, but with the following added:
$n = 0 if eof;
Note that eof is not the same thing as eof()! See eof.
I have a file and I want to extract specific lines from that file like lines 2, 10, 15,21, .... and so on. There are around 200 thousand lines to be extracted from the file. How can I do it efficiently in bash
Maybe looking for:
sed -n -e 1p -e 4p afile
Put the linenumbers of the lines you want in a file called "wanted", like this:
2
10
15
21
Then run this script:
#!/bin/bash
while read w
do
sed -n ${w}p yourfile
done < wanted
TOTALLY ALTERNATIVE METHOD
Or you could let "awk" do it all for you, like this which is probably miles faster since you won't have to create 200,000 sed processes:
awk 'FNR==NR{a[$1]=1;next}{if(FNR in a){print;}}' wanted yourfile
The FNR==NR portion detects when awk is reading the file called "wanted" and if so, it sets element "$1" of array "a" to "1" so we know that this line number is wanted. The stuff in the second set of curly braces is active when processing your bigger file only and it prints the current line if its linenumber is in the array "a" we created when reading the "wanted" file.
$ gawk 'ARGIND==1 { L[$0]++ }; ARGIND==2 && FNR in L' lines file > file.lines
Wanted line numbers have to be stored in lines delimited by newline and they may safely be in random order. It almost exactly the same as #Mark Setchell’s second method, but uses a little more clear way to determine which file is current. Although this ARGIND is GNU extension, so gawk. If you are limited to original AWK or mawk, you can write it as:
$ awk 'FILENAME==ARGV[1] { L[$0]++ }; FILENAME==ARGV[2] && FNR in L' lines file > file.lines
Efficiency test:
$ awk 'BEGIN { for (i=1; i<=1000000; i++) print i }' > file
$ shuf -i 1-1000000 -n 200000 > lines
$ time gawk 'ARGIND==1 { L[$0]++ }; ARGIND==2 && FNR in L' lines file > file.lines
real 0m1.734s
user 0m1.460s
sys 0m0.052s
UPD:
As #Costi Ciudatu pointed out, there is room for impovement for the case when all wanted lines are in the head of a file.
#!/usr/bin/gawk -f
ARGIND==1 { L[$0]++ }
ENDFILE { L_COUNT = FNR }
ARGIND==2 && FNR in L { L_PRINTED++; print }
ARGIND==2 && L_PRINTED == L_COUNT { exit 0 }
Sript interrupts when last line is printed, so now it take few milliseconds to filter out 2000 random lines from first 1 % of a one million lines file.
$ time ./getlines.awk lines file > file.lines
real 0m0.016s
user 0m0.012s
sys 0m0.000s
While reading a whole file still takes about a second.
$ time gawk 'ARGIND==1 { L[$0]++ }; ARGIND==2 && FNR in L' lines file > file.lines
real 0m0.780s
user 0m0.756s
sys 0m0.016s
Provided your system supports sed -f - (i.e. for sed to read its script on standard input; it works on Linux, but not on some other platforms) you can turn the file of line numbers into a sed script, naturally using sed:
sed 's/$/p/' lines | sed -n -f - inputfile >output
If the lines you're interested in are close to the beginning of the file, you can make use of head and tail to efficiently extract specific lines.
For your example line numbers (assuming that list doesn't go on until close to 200,000), a dummy but still efficient approach to read those lines would be the following:
for n in 2 10 15 21; do
head -n $n /your/large/file | tail -1
done
sed Example
sed -n '2p' file
awk Example
awk 'NR==2' file
this will print 2nd line of file
use same logic in loop & try.
say a for loop
for VARIABLE in 2 10 15 21
do
awk "NR==$VARIABLE" file
done
Give your line numbers this way..
It's difficult to tell what is being asked here. This question is ambiguous, vague, incomplete, overly broad, or rhetorical and cannot be reasonably answered in its current form. For help clarifying this question so that it can be reopened, visit the help center.
Closed 9 years ago.
I have a sample file which has thousands of lines.
I want to print text between two line numbers in that file. I don't want to input line numbers manually, rather I have a file which contains list of line numbers between which text has to be printed.
Example : linenumbers.txt
345|789
999|1056
1522|1366
3523|3562
I need a shell script which will read line numbers from this file and print the text between each range of lines into a separate (new) file.
That is, it should print lines between 345 and 789 into a new file, say File1.txt, and print text between lines 999 and 1056 into a new file, say File2.txt, and so on.
considering your target file has only thousands of lines. here is a quick and dirty solution.
awk -F'|' '{system("sed -n \""$1","$2"p\" targetFile > file"NR)}' linenumbers.txt
the targetFile is your file containing thousands of lines.
the oneliner does not require your linenumbers.txt to be sorted.
the oneliner allows line range to be overlapped in your linenumbers.txt
after running the command above, you will have n filex files. n is the row counts of linenumbers.txt x is from 1-n you can change the filename pattern as you want.
Here's one way using GNU awk. Run like:
awk -f script.awk numbers.txt file.txt
Contents of script.awk:
BEGIN {
# set the field separator
FS="|"
}
# for the first file in the arguments list
FNR==NR {
# add the row number and field one as keys to a multidimensional array with
# a value of field two
a[NR][$1]=$2
# skip processing the rest of the code
next
}
# for the second file in the arguments list
{
# for every element in the array's first dimension
for (i in a) {
# for every element in the second dimension
for (j in a[i]) {
# ensure that the first field is treated numerically
j+=0
# if the line number is greater than the first field
# and smaller than the second field
if (FNR>=j && FNR<=a[i][j]) {
# print the line to a file with the suffix of the first file's
# line number (the first dimension)
print > "File" i
}
}
}
}
Alternatively, here's the one-liner:
awk -F "|" 'FNR==NR { a[NR][$1]=$2; next } { for (i in a) for (j in a[i]) { j+=0; if (FNR>=j && FNR<=a[i][j]) print > "File" i } }' numbers.txt file.txt
If you have an 'old' awk, here's the version with compatibility. Run like:
awk -f script.awk numbers.txt file.txt
Contents of script.awk:
BEGIN {
# set the field separator
FS="|"
}
# for the first file in the arguments list
FNR==NR {
# add the row number and field one as a key to a pseudo-multidimensional
# array with a value of field two
a[NR,$1]=$2
# skip processing the rest of the code
next
}
# for the second file in the arguments list
{
# for every element in the array
for (i in a) {
# split the element in to another array
# b[1] is the row number and b[2] is the first field
split(i,b,SUBSEP)
# if the line number is greater than the first field
# and smaller than the second field
if (FNR>=b[2] && FNR<=a[i]) {
# print the line to a file with the suffix of the first file's
# line number (the first pseudo-dimension)
print > "File" b[1]
}
}
}
Alternatively, here's the one-liner:
awk -F "|" 'FNR==NR { a[NR,$1]=$2; next } { for (i in a) { split(i,b,SUBSEP); if (FNR>=b[2] && FNR<=a[i]) print > "File" b[1] } }' numbers.txt file.txt
I would use sed to process the sample data file because it is simple and swift. This requires a mechanism for converting the line numbers file into the appropriate sed script. There are many ways to do this.
One way uses sed to convert the set of line numbers into a sed script. If everything was going to standard output, this would be trivial. With the output needing to go to different files, we need a line number for each line in the line numbers file. One way to give line numbers is the nl command. Another possibility would be to use pr -n -l1. The same sed command line works with both:
nl linenumbers.txt |
sed 's/ *\([0-9]*\)[^0-9]*\([0-9]*\)|\([0-9]*\)/\2,\3w file\1.txt/'
For the given data file, that generates:
345,789w > file1.txt
999,1056w > file2.txt
1522,1366w > file3.txt
3523,3562w > file4.txt
Another option would be to have awk generate the sed script:
awk -F'|' '{ printf "%d,%dw > file%d.txt\n", $1, $2, NR }' linenumbers.txt
If your version of sed will allow you to read its script from standard input with -f - (GNU sed does; BSD sed does not), then you can convert the line numbers file into a sed script on the fly, and use that to parse the sample data:
awk -F'|' '{ printf "%d,%dw > file%d.txt\n", $1, $2, NR }' linenumbers.txt |
sed -n -f - sample.data
If your system supports /dev/stdin, you can use one of:
awk -F'|' '{ printf "%d,%dw > file%d.txt\n", $1, $2, NR }' linenumbers.txt |
sed -n -f /dev/stdin sample.data
awk -F'|' '{ printf "%d,%dw > file%d.txt\n", $1, $2, NR }' linenumbers.txt |
sed -n -f /dev/fd/0 sample.data
Failing that, use an explicit script file:
awk -F'|' '{ printf "%d,%dw > file%d.txt\n", $1, $2, NR }' linenumbers.txt > sed.script
sed -n -f sed.script sample.data
rm -f sed.script
Strictly, you should deal with ensuring the temporary file name is unique (mktemp) and removed even if the script is interrupted (trap):
tmp=$(mktemp sed.script.XXXXXX)
trap "rm -f $tmp; exit 1" 0 1 2 3 13 15
awk -F'|' '{ printf "%d,%dw > file%d.txt\n", $1, $2, NR }' linenumbers.txt > $tmp
sed -n -f $tmp sample.data
rm -f $tmp
trap 0
The final trap 0 allows your script to exit successfully; omit it, and you script will always exit with status 1.
I've ignored Perl and Python; either could be used for this in a single command. The file management is just fiddly enough that using sed seems simpler. You could also use just awk, either with a first awk script writing an awk script to do the heavy duty work (trivial extension of the above), or having a single awk process read both files and produce the required output (harder, but far from impossible).
If nothing else, this shows that there are many possible ways of doing the job. If this is a one-off exercise, it really doesn't matter very much which you choose. If you will be doing this repeatedly, then choose the mechanism that you like. If you're worried about performance, measure. It is likely that converting the line numbers into a command script is a negligible cost; processing the sample data with the command script is where the time is taken. I would expect sed to excel at that point; I've not measured to confirm that it does.
You could do the following
# myscript.sh
linenumbers="linenumber.txt"
somefile="afile"
while IFS=\| read start end ; do
echo "sed -n '$start,${end}p;${end}q;' $somefile > $somefile-$start-$end"
done < $linenumbers
run it like so sh myscript.sh
sed -n '345,789p;789q;' afile > afile-345-789
sed -n '999,1056p;1056q;' afile > afile-999-1056
sed -n '1522,1366p;1366q;' afile > afile-1522-1366
sed -n '3523,3562p;3562q;' afile > afile-3523-3562
then when you're happy do sh myscript.sh | sh
EDIT Added William's excellent points on style and correctness.
EDIT Explanation
The basic idea is to get a script to generate a series of shell commands that can be checked for correctness first before being executed by "| sh".
sed -n '345,789p;789q; means use sed and don't echo each line (-n) ; there are two commands saying from line 345 to 789 p(rint) the lines and the second command is at line 789 q(uit) - by quitting on the last line you save having sed read all the input file.
The while loop reads from the $linenumbers file using read, read if given more than one variable name populates each with a field from the input, a field is usually separated by space and if there are too few variable names then read will put the remaining data into the last variable name.
You can put the following in at your shell prompt to understand that behaviour.
ls -l | while read first rest ; do
echo $first XXXX $rest
done
Try adding another variable second to the above to see what happens then, it should be obvious.
The problem is your data is delimited by |s and that's where using William's suggestion of IFS=\| works as now when reading from the input the IFS has changed and the input is now separated by |s and we get the desired result.
Others can feel free to edit,correct and expand.
To extract the first field from 345|789 you can e.g use awk
awk -F'|' '{print $1}'
Combine that with the answers received from your other question and you will have a solution.
This might work for you (GNU sed):
sed -r 's/(.*)\|(.*)/\1,\2w file-\1-\2.txt/' | sed -nf - file
I have a flat file as follows:
11|aaa
11|bbb|NO|xxx
11|ccc
11|ddd|NO|yyy
For lines that do not contain |NO|, I would like to add the string |YES| at the end. So my file should look like:
11|aaa|YES|
11|bbb|NO|xxx
11|ccc|YES|
11|ddd|NO|yyy
I am using AIX and sed -i option for inline replacements is not available. Hence, currently I'm using the following code to do this:
#Get the lines that do not contain |NO|
LINES=`grep -v "|NO|" file`
for i in LINES
do
sed "/$i/{s/$/|YES|/;}" file > temp
mv temp file
done
The above works, however, as my file contains over 40000 lines, it takes about 3 hours to run. I believe it is taking so much time because it has to search for each line and write to a temp file. Is there a faster way to achieve this ?
This will be quick:
sed '/NO/!s/$/|YES|/' filename
If temp.txt is your file, try:
awk '$0 !~ /NO/ {print $0 "|YES|"} $0 ~ /NO/ {print}' temp.txt
Simple with awk. Put the code below into a script and run it with awk -f script file > temp
/\|NO\|/ { print; next; } # just print anything which contains |NO| and read next line
{ print $0 "|YES|"; } # For any other line (no pattern), print the line + |YES|
I'm not sure about awk regexps; if it doesn't work, try to remove the two \ in the first pattern.
i have small file with around 50 lines and 2 fields like below
file1
-----
12345 8373
65236 7376
82738 2872
..
..
..
i have some around 100 files which are comma"," separated as below:
file2
-----
1,3,4,4,12345,,,23,3,,,2,8373,1,1
each file has many lines similar to the above line.
i want to extract from all these 100 files whose
5th field is eqaul to 1st field in the first file and
13th field is equal to 2nd field in the first file
I want to search all the 100 files using that single file?
i came up with the below in case of a single comma separated file.i am not even sure whether this is correct!
but i have multiple comma separated files.
awk -F"\t|," 'FNR==NR{a[$1$2]++;next}($5$13 in a)' file1 file2
can anyone help me pls?
EDIT:
the above command is working fine in case of a single file.
Here is another using an array, avoiding multiple work files:
#!/bin/awk -f
FILENAME == "file1" {
keys[$1] = ""
keys[$2] = ""
next
}
{
split($0, fields, "," )
if (fields[5] in keys && fields[13] in keys) print "*:",$0
}
I am using split because the field seperator in the two files are different. You could swap it around if necessary. You should call the script thus:
runit.awk file1 file2
An alternative is to open the first file explicitly (using "open") and reading it (readline) in a BEGIN block.
Here is a simple approach. Extract each line from the small file, split it into fields and then use awk to print lines from the other files which match those fields:
while read line
do
f1=$(echo $line | awk '{print $1}')
f2=$(echo $line | awk '{print $2}')
awk -v f1="$f1" -v f2="$f2" -F, '$5==f1 && $13==f2' file*
done < small_file