31

I'm looking for the simplest method to print the longest line in a file. I did some googling and surprisingly couldn't seem to find an answer. I frequently print the length of the longest line in a file, but I don't know how to actually print the longest line. Can anyone provide a solution to print the longest line in a file? Thanks in advance.

  • 1
    What about when there are multiple "longest" lines?. Because you want more than a simple maximum length, do you want to see all instances of lines which are equal longest? – Peter.O Nov 13 '11 at 14:20
34
cat ./text | awk ' { if ( length > x ) { x = length; y = $0 } }END{ print y }'

UPD: summarizing all the advices in the comments

awk 'length > max_length { max_length = length; longest_line = $0 } END { print longest_line }' ./text 
  • 3
    It is, both calling another command (cat), and using a pipe are expensive operations, not to mention that it's more efficient for awk to just read the file. The performance implications are definitely noticeable if this is done frequently, and even so, you are completely misusing cat. – Chris Down Nov 13 '11 at 2:37
  • 6
    @laebshade There absolutely is a reason -- it's so you don't need to remember which commands take filenames and which don't, or care about which command is going to execute first in the pipeline. If you're going to be writing a script that gets run frequently, by all means worry about something like this. If you're writing a one-off thing to find the longest line in a file, the extra process and fractional amount of time consumed is completely irrelevant. It's silly that people are so obsessed with it here, it's incredibly minor – Michael Mrozek Nov 13 '11 at 17:41
  • 4
    @Keith Thompson: cat is not useless here. It might be useless to a computer but for a human reader it could provide value. The first variant clearly shows the input. The flow is more natural (from left to right). In the second case you don't know what the input is unless you scroll the window. – jfs Nov 14 '11 at 3:02
  • 1
    @J.F.Sebastian Even if you want it at the left, you don't need cat. < file command works just fine. – Chris Down Nov 14 '11 at 3:03
  • 2
    @J.F.Sebastian: The fact that a redirection can be written at the beginning of a command is somewhat obscure; < filename command is equivalent to filename < command in every shell I've tried. But once you're aware of it, you can take advantage of it when writing long pipes that clearly show the direction of the data flow (without invoking an extra command): < input-file command1 | command2 | command3 > output-file – Keith Thompson Nov 14 '11 at 3:16
5
sed -rn "/.{$(<file expand -t1 |wc -L)}/{p;q}" file

This first reads the file inside the command substitution and outputs the length of the longest line, (previously, expand converts tabs to spaces, to overcome the semantics of wc -L -- each tab in the line will add 8 instead of 1 to line length). This length is then used in a sed expression meaning "find a line this number of characters long, print it, then quit". So this actually can be as optimal as the longest line is near to the top of the file, heheh (thanks fered for the awesome and constructive comments).

Another, I had thought earlier than the sed one (in bash):

#!/bin/bash
while read -r line; do
    (( ${#line} > max )) && max=${#line} && longest="$line"
done
echo "$longest"
  • 2
    This method is very expensive and slow. – Chris Down Nov 13 '11 at 2:47
  • 2
    @Chris Down: Oh yes it is. But the question was about the sortest method, not the most efficient. WOrks finely for small to medium files or non critical tasks, though. – ata Nov 13 '11 at 11:49
  • 3
    WARNING: wc's option -L, --max-line-length prints the length of the longest line, according to the man page, but if you dig deeper (as in when you get wrong / unexpected results), you find that this option increments the length by 8 for each 1 tab char \x09 see this Unix & Linux Q/A – Peter.O Nov 13 '11 at 13:05
  • PS. Your answer will print all the "equally longest" lines, which is probably a good thing... To force wc to count only 1 char per tab, this works. sed -rn "/.{$(<file expand -t1 |wc -L)}/p" file – Peter.O Nov 13 '11 at 14:13
  • 1
    read line will interpret backslash-escaped chars as the literal char, eg \A resloves to A, which of course effectively reports a shorter than actual byte-usage... To prevent this escaped interpretation, use: read -r line . . . . Also, to make the sed+wc version quit after the first "longest line", change p to {p;q} .. sed -rn "/.{$(<file expand -t1 |wc -L)}/{p;q}" file – Peter.O Nov 14 '11 at 3:15
4

Here's a Perl solution:

perl -e 'while(<>){
           $l=length;  
           $l>$m && do {$c=$_; $m=$l}  
         } print $c' file.txt 

Or, if you want to print all the longest lines

perl -e 'while(<>){
           $l=length;
           push @{$k{$l}},$_;
           $m=$l if $l>$m;
         } print @{$k{$m}}' file.txt 

Since I had nothing better to do, I ran some benchmarks on a 625M text file. Surprisingly, my Perl solution was consistently faster than the others. Granted, the difference with the accepted awk solution is tiny, but it is there. Obviously, solutions that print multiple lines are slower so I have sorted by type, fastest to slowest.

Print only one of the longest lines:

$ time perl -e 'while(<>){
           $l=length;  
           $l>$m && do {$c=$_; $m=$l}  
         } print $c' file.txt 
real    0m3.837s
user    0m3.724s
sys     0m0.096s



$ time awk 'length > max_length { max_length = length; longest_line = $0 }
 END { print longest_line }' file.txt
real    0m5.835s
user    0m5.604s
sys     0m0.204s



$ time sed -rn "/.{$(<file.txt expand -t1 |wc -L)}/{p;q}" file.txt 
real    2m37.348s
user    2m39.990s
sys     0m1.868s

Print all longest lines :

$ time perl -e 'while(<>){
           $l=length;
           push @{$k{$l}},$_;
           $m=$l if $l>$m;
         } print @{$k{$m}}' file.txt 
real    0m9.263s
user    0m8.417s
sys     0m0.760s


$ time awk 'length >x { delete y; x=length }
     length==x { y[NR]=$0 } END{ for (z in y) print y[z] }' file.txt
real    0m10.220s
user    0m9.925s
sys     0m0.252s


## This is Chris Down's bash solution
$ time ./a.sh < file.txt 
Max line length: 254
Lines matched with that length: 2
real    8m36.975s
user    8m17.495s
sys     0m17.153s
4
cat filename | awk '{ print length }' | sort -n | tail -1
  • +1 There were lots of interesting solutions to this but this was the simplest. (It would be simpler without the cat by letting awk read the file but why quibble?) – user1683793 Apr 19 at 15:14
3

Grep the first longest line

grep -Em1 "^.{$(wc -L <file.txt)}\$" file.txt 

The command is unusually hard to read without practise because it mixes shell- and regexp syntax.
For explanation, I will use simplified pseudocode first. The lines starting with ## do not run in the shell.
This simplified code uses the file name F, and leaves out quoting and parts of regexps for readability.

How it works

The command has two parts, a grep- and a wc invocation:

## grep "^.{$( wc -L F )}$" F

The wc is used in a process expansion, $( ... ), so it is run before grep. It calculates the length of the longest line. The shell expansion syntax is mixed with the regular expression pattern syntax in a confusing way, so I will decompose the process expansion:

## wc -L F
42
## grep "^.{42}$" F

Here, the process expansion was replaced with the value it would return, creating the grep commandline that is used. We can now read the regular expression more easily: It matches exactly from start (^) to end ($) of the line. The expression between them matches any character except newline, repeated by 42 times. Combined, that is lines that consist of exactly 42 characters.


Now, back to real shell commands: The grep option -E (--extended-regexp) allows to not escape the {} for readability. Option -m 1 (--max-count=1) makes it stop after the first line is found. The < in the wc command writes the file to its stdin, to prevent wc from printing the file name together with the length.

Which longest lines?

To make the examples more readable with the filename occurring twice, I will use a variable f for the filename; Each $f in the example could be replaced by the file name.

f="file.txt"

Show the first longest line - the first line that is as long as the longest line:

grep -E -m1 "^.{$(wc -L <"$f")}\$" "$f"

Show all longest lines - all lines that are as long as the longest line:

grep -E "^.{$(wc -L <"$f")}\$" "$f" 

Show the last longest line - the last line that is as long as the longest line:

tac "$f" | grep -E -m1 "^.{$(wc -L <"$f")}\$"

Show the single longest line - the longest line longer than all other lines, or fail:

[ $(grep -E "^.{$(wc -L <"$f")}\$" "$f" | wc -l) = 1 ] && grep -E "^.{$(wc -L <"$f")}\$" "$f" 

(The last command is even more inefficient than the others, as it repeats the complete grep command. It should obviously be decomposed so that the output of wc and the lines written by grep are saved to variables.
Note that all longest lines may actually be all lines. For saving in a variable, only the first two lines need to be kept.)

  • Wow great answer, learned a lot from it. thanks – somethingSomething Oct 3 '18 at 18:45
2

The following example was going to be, and should have been, a comment to dmitry.malikov's answer, but because of the Useless Use of Visible Comment Space there, I've chosen to present it here, where it will at least be seen...

This is a simple variation of the dmitry's single-pass awk method.
It prints all "equal longest" lines. (Note. delete array is a gawk extension).

awk 'length >x { delete y; x=length }
     length==x { y[NR]=$0 } END{ for (z in y) print y[z] }' file
1

In pure bash:

#!/bin/bash

_max_length=0
while IFS= read -r _line; do
    _length="${#_line}"
    if (( _length > _max_length )); then
        _max_length=${_length}
        _max_line=( "${_line}" )
    elif (( _length == _max_length )); then
        _max_line+=( "${_line}" )
    fi
done

printf 'Max line length: %d\n' "${_max_length}"
printf 'Lines matched with that length: %d\n' "${#_max_line[@]}"
(( ${#_max_line[@]} )) && printf '%s\n' '----------------' "${_max_line[@]}"
  • As-is, the code can return invalid results. Setting _max_line[0]=${_line} does not remove the rest of any previously accumulated shorter "longest lines"... unset _max_line will clear the entire array... – Peter.O Nov 13 '11 at 21:00
  • @fered Thanks for that, was written pretty quickly. Fixed. – Chris Down Nov 13 '11 at 21:03
0

I have developed a small shell script for this. It displays length, line # and line itself by length that exceeds a particular size like 80 characters:

#!/bin/sh

# Author: Surinder

if test $# -lt 2
then
   echo "usage: $0 length file1 file2 ..."
   echo "usage: $0 80 hello.c"
   exit 1
fi

length=$1

shift

LONGLINE=/tmp/longest-line-$$.awk

cat << EOF > $LONGLINE
  BEGIN {
  }

  /.*/ {
    current_length=length(\$0);
    if (current_length >= expected_length) {
       printf("%d at line # %d %s\n", current_length, NR, \$0);
    }
  }

  END {
  }
EOF

for file in $*
do
  echo "$file"
  cat $file | awk -v expected_length=$length -f $LONGLINE |sort -nr
done

rm $LONGLINE

https://github.com/lordofrain/tools/blob/master/longest-line/longest-line.sh

  • 1
    There are a few improvements you could make. Quote your variables. This will break on any file names that contain whitespace or other strange characters. Using $* is rarely a good idea, you want "$@". The /.*/ in your awk doesn't do anything since that matches empty lines as well. You could avoid escaping the \$0 if you single quote the 'EOF'. Why use an empty BEGIN{} block? Finally, you don't need cat, just awk . . . "$file" | . . . – terdon Oct 19 '16 at 9:24
  • 1
    You couuld also just do the whole thing in awk directly: awk -vmax=15 '{len=length($0); if(len>=max){printf("%s, %d at line # %d %s\n", FILENAME, len, NR, $0);}}' file* – terdon Oct 19 '16 at 9:25
-3

You can use wc:

wc -L fileName
  • 3
    Please read the question again. The required output is the longest line itself, not the length of the longest line. Also see Peter.O's comment regarding wc -L's drawback. – manatwork May 1 '13 at 11:47

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.