The GNU Text Utilities

by David Mertz, Ph.D. <[email protected]>

Introduction: The Unix Philosophy

Small utilities combine to do large tasks

In Unix-inspired operating systems such as Linux, FreeBSD, MacOSX, Solaris, AIX, and so on, a common philosophy underlies the development environment, and even just the shells and working environment. The main jist of this philsophy is using small component utilities to do each small task well (and no other thing badly), then combining utilities to perform compound tasks. Most of what has been produced by the GNU project falls under this component philosophy--and indeed the specific GNU implementations have been ported to many platforms, even ones not traditionally thought of as Unix-like. The Linux kernel, however, is of necessity a more monolithic bit of software--though even there kernel modules, filesystems, video drivers, and so on, are largely componentized.

For this column, readers should be generally familiar with some Unix-like environment, and especially with a command-line shell. Readers need not be programmers per se; in fact, the techniques described will be most useful to system administrators and users who process ad hoc reports, log files, project documentation, and the like (and less so for formal programming code processing).

Files and Streams

If the Unix philosophy has a deontological aspect in advocating minimal modular components and cooperation, it also has an ontological aspect: "everything is a file." Abstractly, a file is simply an object that supports a few operations; firstly reading and writing bytes, but also some supporting operations like indicating its current position and knowing when it has reached its end. The Unix permission model is also oriented around its idea of file.

Concretely, a file might be an actual region on a recordable media (with appropriate tagging of its name, size, position on disk, and so on, supplied by the filesystem). But a file might also be a virtual device in the /dev/ hierarchy, or a remote stream coming over a TCP/IP socket or via a higher-level protocol like NFS. Importantly, the special files STDIN and STDOUT and STDERR can be used to read or write to the user console and/or to pass data between utilities. These special files can be indicated by virtual filenames, along with using special syntax: STDIN is /dev/stdin and/or /dev/fd/0 ; STDOUT is /dev/stdout and/or /dev/fd/1 ; STDERR is /dev/stderr and/or /dev/fd/2 .

The advantage and principle of Unix' file ontology is that most of the utilities discussed here will handle various data sources uniformly and neutrally, regardless of what storage or transmission mechanisms actually underly the delivery of bytes.

Redirection and Piping

The way that Unix/Linux utilities are typically combined is via piping and redirection. Many utilites either automatically or optionally take their input from STDIN, and send their output to STDOUT (with special messages sent to STDERR). A pipe sends the STDOUT of one utility to the STDIN of another utility (or to a new invocation of the same utility). A redirect either reads the content of a file as STDIN, or sends the STDOUT and/or STDERR output to a named file. Redirects are often used to save data for later or repeated processing (with the later utility runs using STDIN redirection).

In almost all shells, piping is performed with the vertical-bar | symbol, and redirection with the greater-than and less-than symbols: > and < . To redirect STDERR, use 2> , or &> to redirect both STDOUT and STDERR to the same place. You may also use a doubled greater-than (>>) to append to the end of an existing file. For example:

 $ foo fname | bar - > myout 2> myerr 

The utility foo probably processes the file named fname , and outputs to STDOUT. The utility bar uses a common convention of specifying a dash when output is to be taken from STDIN rather than a named file (some other utilities only take STDIN). The STDOUT from bar is saved in myout , and its STDERR in myerr .

What are the text utilities?

The GNU Text Utilities is a collection of some of the tools for processing and manipulating text files and streams that have proven most useful, and been refined, over the evolution of Unix-like operating systems. Most of them have been part of Unix from the earliest implementations, though many have grown additional options over time.

The suite of utilities collected in the archive textutils-2.1 includes twenty-seven tools; however, the GNU project maintainers have more recently decided to bundle these tools instead as part of the larger collection coreutils-5.0 (and presumably likewise for later versions). On systems derived from BSD rather than GNU tools, the same utilities might be bundled a bit differently, but most of the same utilities will still be present. This tutorial will focus on the twenty-seven utilities traditionally included in textutils , with some occassional mention and use of related tools that are generally available on Unix-like systems. However, I will skip the utility ptx (permuted indexes) which is both too narrow in purpose and too difficult to understand for inclusion here.

grep (Generalized Regular Experession Processor)

One tool that is not per se part of textutils still deserves special mention. The utility grep is one of the most widely used Unix utilities, and will very often be used in pipes to or from the text utilities.

What grep does is in one sense very simple, in another sense quite complex to understand. Basically, grep identifies lines in a file that match a regular expression. Some switches let you massage the output in various ways, such as printing surrounding context lines, numbering the matching lines, or identifying only the files in which the matches occur rather than individual lines. But at heart, grep is just a (very powerful) filter on the lines in a file. The complex part of grep is the regular expressions you can specify to describe matches of interest. But that's another tutorial (see Resources). A number of other utilities also support regular expression patterns, but grep is the most general such tool, and hence it is often easier to put grep in your pipeline than to use the weaker filters other tools provide. A quick grep example:

          $ grep -c [Ss]ystem$ * 2> /dev/null | grep :[^0]$

The example lists files that contain lines ending with word "system", perhaps with initial cap, at the end of lines; and also show the number of such occurrences (i.e. if non-zero). (Actually, the example does not not handle counts greater than 9 properly).

Shell Scripting

While the text utilities are designed to produce outputs in various useful formats--often modified by command-line switches--there are still times when being able to explicitly branch and loop is useful. Shells like bash let you combine utilities with flow control to perform more complex chores. Shell scripts are especially useful to encapsulate compound tasks that you will perform multiple times, especially those involving some parameterization of the task.

Explaining bash scripting is certainly outside the scope of this tutorial. See Resources for an introduction to bash . Once you understand the text utilities, it is fairly simple to combine them into saved shell scripts. Just for illustration, here is a quick (albeit somewhat contrived) example of flow control with bash :

          [~/bacchus/articles/scratch/tu]$ cat flow
          for fname in `ls $1`; do
            if (grep $2 $fname > /dev/null); then
              echo "Creating: $" ;
              tr "abc" "ABC" < $fname > $
          [~/bacchus/articles/scratch/tu]$ ./flow '*' bash
          [~/bacchus/articles/scratch/tu]$ cat
          for fnAme in `ls $1`; do
            if (grep $2 $fnAme > /dev/null); then
              eCho "CreAting: $" ;
              tr "ABC" "ABC" < $fnAme > $

Stream-Oriented Filtering

cat and tac

The simplest text utilities simply output the exact contents of a file or stream to STDOUT, or perhaps a portion or simple rearrangment of those contents.

The utility cat begins with the first line and ends with the last line. The utility tac outputs lines in reverse. Both utilites will read every file specified as an argument, but default to STDIN if none is specified. As with many utilities, you may explicitly specify STDIN using the special name - . Some examples:

          $ cat test2
          $ tac < test3
          $ cat test2 test3
          $ cat test2 | tac - test3

head and tail

The utilities head and tail output only an initial or final portion of a file or stream, respectively. The GNU version of both utilities support the switch -c to output a number of bytes; most often both utilities are used in their line-oriented mode which output a number of lines (whatever the actual line lengths). Both head and tail default to outputting ten lines. As with cat or tac , head and tail default to STDIN if files are not specified.

          $ head -c 8 test2 && echo # push prompt to new line
          $ /usr/local/bin/head -2 test2
          $ cat test3 | tail -n 2
          $ tail -r -n 2 test3 # reverse

By the way, the GNU versions of these utilities (and many others) have more flexible switches than do the BSD versions.

The tail utility has a special mode indicated with the switches -f and -F that continues to display new lines written to the end of a "followed" file. The capitalized switch watches for truncations and renaming of the file as well as the simple appends the lower case switch monitors. Follow mode is particularly useful for watching changes to a log file that another process might peform periodically.

od and hexdump

The utilities od and hexdump output octal, hex, or otherwise encoded bytes from a file or stream. These are useful for access to or visual examination of characters in a file that are not directly displayible on your terminal. For example, cat or tail do not directly disambiguate between tabs, spaces, or other whitespace--you can check which characters are used with hexdump . Depending on you system type, either or both of these two utilities will be available--BSD systems deprecate od for hexdump , GNU systems the reverse. The two utilities, however, have exactly the same purpose, just slightly different switches.

          $ od test3 # default output format
          0000000 054141 073151 062562 005131 067554 060556 062141 005132
          0000020 062553 062412
          $ od -w8 -x test3 # 8 hex digits per line
          0000000 5861 7669 6572 0a59
          0000010 6f6c 616e 6461 0a5a
          0000020 656b 650a
          $ od -c test3 # 5 escaped ASCII chars per line
          0000000   X   a   v   i   e   r  \n   Y   o   l   a   n   d   a  \n   Z
          0000020   e   k   e  \n

As with other utilities, od and hexdump accept input from STDIN or from one or more named files. As well, the od switches -j and -N let you skip initial bytes and limit the number read, respectively. You may customize output formats even further than with the standard switches using fprintf() -like formatting specifiers

HERE documents

There is a special kind of redirection that is worth noting in this tutorial. While HERE documents are, strictly speaking, a feature of shells like bash rather than anything to do with the text utilities, they provide a useful way of sending ad hoc data to the text utilities (or to other applications).

Direction with a double less-than can be used to take pseudo-file contents from the terminal. A HERE document must specifiy a terminating delimiter immediately after its << . For example:

          $ od -c <<END
          > Alice
          > Bob
          > END
          0000000   A   l   i   c   e  \n   B   o   b  \n

Any string may be used as a delimiter, input is terminated when the string occurs on a line by itself. This gives us a quick way to create a persistent file:

          $ cat > myfile <<EOF
          > Dave
          > Edna
          > EOF
          $ hexdump -C myfile
          00000000  44 61 76 65 0a 45 64 6e  61 0a            |Dave.Edna.|

Line-Oriented Filtering

Lines as records

Many Linux utilities view files as a line-oriented collection of records or data. This has proved a very convenient way of aggregating data collections in ways that is both readable to people, and easy to process with tools. The simple trick is to treat each newline as a delimiter between records, where each record has a similar format.

As a practical matter, line-oriented records usually should have a relatively limited length--perhaps up through a few hundred characters. While none of the text utilties have such a limit built in to them, human eyes have trouble working with extremely long lines, even if auto-wrapping or horizontal scrolling is used. Either a more complex structured data format might be used in such cases, or records might be broken into multiple lines (perhaps flagged for type in a way that grep can sort out). As a simple example, you might preserve a hierarchical multi-line data format using prefix characters:

          $ cat multiline
          A Alice Aaronson
          B System Administrator
          C 99 9th Street
          A Bob Babuk
          B Programmer
          C 7 77th Avenue
          $ grep '^A ' multiline # names only
          A Alice Aaronson
          A Bob Babuk
          $ grep '^C ' multiline # address only
          C 99 9th Street
          C 7 77th Avenue

The output from one of these grep filters is a usuable newline-delimited collection of partial records with the field(s) of interest.


The utility cut writes fields from a file to the standard output, where each line is treated as a delimited collection of fields. The default delimiting character is a tab, but this can be changed with the short form option -d <DELIM> or the long form option --delimiter=<DELIM> .

You may select one or more fields with the -f switch. The -c switch selects specific character positions from each line instead. Either switch will accept comma separated numbers or ranges as parameters (including open ranges). For example, we can see that the file employees is tab delimited:

          $ cat employees
          Alice Aaronson  System Administrator    99 9th Street
          Bob Babuk       Programmer      7 77th Avenue
          Carol Cavo      Manager 111 West 1st Blvd.
          $ hexdump -n 50 -c employees
          0000000   A   l   i   c   e       A   a   r   o   n   s   o   n  \t   S
          0000010   y   s   t   e   m       A   d   m   i   n   i   s   t   r   a
          0000020   t   o   r  \t   9   9       9   t   h       S   t   r   e   e
          0000030   t  \n
          $ cut -f 1,3 employees
          Alice Aaronson  99 9th Street
          Bob Babuk       7 77th Avenue
          Carol Cavo      111 West 1st Blvd.
          $ cut -c 1-3,20,25- employees
          Alieministrator 99 9th Street
          Bobr7th Avenue
          Car1est 1st Blvd.

Later examples will utilize custom delimiters, other than tabs.

expand and unexpand

The utilties expand and unexpand convert tabs to spaces and vice-versa. A tab is considered to align at specific columns, by default every eight columns, so the specific number of spaces that correspond to a tab depends on where those spaces or tab occur. Unless you specify the -a option, unexpand will only entab the initial whitespace (the default is useful for reformatting source code).

Continuing with the employees file of the last panel, we can peform some substitutions. Notice that after you run unexpand , tabs in the output may be follwed by some spaces in order to produce the needed overall alignment.

          $ cat -T employees  # show tabs explicitly
          Alice Aaronson^ISystem Administrator^I99 9th Street
          Bob Babuk^IProgrammer^I7 77th Avenue
          Carol Cavo^IManager^I111 West 1st Blvd.
          $ expand -25 employees
          Alice Aaronson           System Administrator     99 9th Street
          Bob Babuk                Programmer               7 77th Avenue
          Carol Cavo               Manager                  111 West 1st Blvd.
          $ expand -25 employees | unexpand -a | hexdump -n 50 -c
          0000000   A   l   i   c   e       A   a   r   o   n   s   o   n  \t  \t
          0000010       S   y   s   t   e   m       A   d   m   i   n   i   s   t
          0000020   r   a   t   o   r  \t           9   9       9   t   h       S
          0000030   t   r


The fold utility simply forces lines in a file to wrap. By default, wrapping is to 80 columns, but you may specify other widths. You get a limited sort of word-wrap formatting with fold , but it will not fully rewrap paragraphs. The option -s is useful for at least forcing new line breaks to occur on whitespace. Using a recent article of mine as a source (and clipping an example portion using tools we've seen earlier):

          $ tail -4 rexx.txt | cut -c 3-
          David Mertz' fondness for IBM dates back embarrassingly many decades.
          David may be reached at [email protected]; his life pored over at
 And buy his book: _Text Processing in
          Python_ (
          $ tail -4 rexx.txt | cut -c 3- | fold -w 50
          David Mertz' fondness for IBM dates back embarrass
          ingly many decades.
          David may be reached at [email protected]; his life
          pored over at
 And buy his book: _Text
           Processing in
          Python_ (
          $ tail -4 rexx.txt | cut -c 3- | fold -w 50 -s
          David Mertz' fondness for IBM dates back
          embarrassingly many decades.
          David may be reached at [email protected]; his life
          pored over at
 And buy his book:
          _Text Processing in
          Python_ (


For most purposes, fmt is a more useful tool for wrapping lines than is fold . The utility fmt will wrap lines, while both preserving initial indentation and aggregating lines for paragraph balance (as needed). fmt is useful for formatting documents such as email messages before transmission or final storage.

          $ tail -4 rexx.txt  | fmt -40 -w50 # goal 40, max 50
            David Mertz' fondness for IBM dates back
            embarrassingly many decades.  David may be
            reached at [email protected]; his life pored
            over at And
            buy his book: _Text Processing in Python_
          $ tail -4 rexx.txt  | fold -40
            David Mertz' fondness for IBM dates ba
          ck embarrassingly many decades.
            David may be reached at [email protected]
          x; his life pored over at
   And buy his
           book: _Text Processing in
            Python_ (

The GNU version of fmt provides several options regarding how indentation of first and subsequent lines is handled in determining indentation style. An option particularly likely to be useful is -u which normalizes word and sentence spaces.

nl (and cat)

The utility nl numbers the lines in a file, with a variety of options for how numbers appear. For the most part, cat contains the line numbering options you will need for most purposes--choose the more general tool, cat when it does what you need. Only in special cases such as controlling display of leading zeros is nl needed (historically, cat did not always include line numbering).

          $ nl -w4 -nrz -ba rexx.txt | head -6  # width 4, zero padded
          0001    LINUX ZONE FEATURE: Regina and NetRexx
          0002    Scripting with Free Software Rexx Implementations
          0004    David Mertz, Ph.D.
          0005    Text Processor, Gnosis Software, Inc.
          0006    January, 2004
          $ cat -b rexx.txt | head -6   # don't number bare lines
               1  LINUX ZONE FEATURE: Regina and NetRexx
               2  Scripting with Free Software Rexx Implementations

               3  David Mertz, Ph.D.
               4  Text Processor, Gnosis Software, Inc.
               5  January, 2004

Aside from making discussions of lines within files easier, line numbers potentially provide sort or filter criteria for downstream processes.

tr, Part 1

The utility tr is a powerful tool for tranforming the characters that occur within a file--or rather, within STDIN, since tr operates exclusively on STDIN and writes exclusively to STDOUT (redirection and piping is allowed, of course).

tr is somewhat more limited in capability than is its big sibling sed , which is not included in the text utilities (nor in this tutorial) but is still almost always available on Unix-like systems. Where sed can perform general replacements of regular expressions, tr is limited to replacing and deleting single characters (it has no real concept of context). At its most basic, tr replaces the characters of STDIN that are contained in a source string with those in a target string.

A simple example helps illustrate tr . We might have a file with variable numbers of tabs and spaces, and with to normalize these separators, and replace them with a new delimiter. The trick is to use the -s (squeeze) flag to eliminate runs of the same character:

          $ expand -26 employees | unexpand -a > empl.multitab
          $ cat -T empl.multitab
          Alice Aaronson^I^I  System Administrator^I    99 9th Street
          Bob Babuk^I^I  Programmer^I^I    7 77th Avenue
          Carol Cavo^I^I  Manager^I^I    111 West 1st Blvd.
          $ tr -s "\t " "| " < empl.multitab | /usr/local/bin/cat -T
          Alice Aaronson| System Administrator| 99 9th Street
          Bob Babuk| Programmer| 7 77th Avenue
          Carol Cavo| Manager| 111 West 1st Blvd.

tr, Part 2

As well as translating explicitly listed characters, tr supports ranges and several named character classes. For example, to translate lower-case characters to upper-case, you may use either of:

          $ tr "a-z" "A-Z" < employees
          BOB BABUK       PROGRAMMER      7 77TH AVENUE
          CAROL CAVO      MANAGER 111 WEST 1ST BLVD.
          $ tr [:lower:] [:upper:] < employees
          BOB BABUK       PROGRAMMER      7 77TH AVENUE
          CAROL CAVO      MANAGER 111 WEST 1ST BLVD.

If the second range is not as long as the first, the second is padded with occurrences of its last character:

          $ tr [:upper:] "a-l#" < employees
          alice aaronson  #ystem administrator    99 9th #treet
          bob babuk       #rogrammer      7 77th avenue
          carol cavo      #anager 111 #est 1st blvd.

You may also delete characters from the STDIN stream. Typically you might delete special characters like formfeeds or high-bit characters you want to filter. But for this, let us continue with the prior example:

          $ tr -d [:lower:] < employees
          A A     S A     99 9 S
          B B     P       7 77 A
          C C     M       111 W 1 B.

File-Oriented Filtering

Working with line collections

The tools we have seen so far operate on each line individually. Another subset of the text utilities treats files as collections of lines, and performs some kind of global manipulation on those lines.

Pipes under Unix-like operating systems can operate very efficiently in terms of memory and latency. When a process earlier in a pipe produces a line to STDOUT, that line is immediately available to the next stage. However, the below utilities will not produce output until they have (mostly) completed their processing. For large files, some of these utilities can take a while to complete (but they are nonetheless all well optimized for the tasks they perform).


The utility sort does just what the name suggests: it sorts the lines within a file or files. A variety of options exist to allow sorting on fields or character positions within the file, and to modify the comparison operation (numeric, date, case-insensitive, etc).

A common use of sort is in combining multiple files. Building on our earlier example:

          $ cat employees2
          Doug Dobrovsky  Accountant      333 Tri-State Road
          Adam Aman       Technician      4 Fourth Street
          $ sort employees employees2
          Adam Aman       Technician      4 Fourth Street
          Alice Aaronson  System Administrator    99 9th Street
          Bob Babuk       Programmer      7 77th Avenue
          Carol Cavo      Manager 111 West 1st Blvd.
          Doug Dobrovsky  Accountant      333 Tri-State Road

Field and character position within a field may be specified as sort criteria, as may use of numeric sorting:

          $ cat namenums
          Alice   123
          Bob     45
          Carol   6
          $ sort -k 2.1 -n namenums
          Carol   6
          Bob     45
          Alice   123


The utility uniq removes adjacent lines which are identical to each other--or if some switches are used, close enough to count as identical (you may skip fields, character postitions, or compare as case-insensitive). Most often, the input to uniq is the output from sort , though GNU sort itself contains a limited ability to eliminate duplicate lines with the -u switch.

The most typical use of uniq is in the expression sort list_of_things | uniq , producing a list with just one of each item (one per line). But some fancier uses let you analyze duplicates or use different duplication criteria:

          $ uniq -d test5 # identify duplicates
          $ uniq -c test5 # count occurrences
          1 Alice
          2 Bob
          1 Carol
          $ cat test4
          1       Alice
          2       Bob
          3       Bob
          4       Carol
          $ uniq -f 1 test4  # skip first field in comparisons
          1       Alice
          2       Bob
          4       Carol


The utility tsort is a bit of an oddity in the text utilities collection. The utility itself is quite useful in a limited context, but what it does is not something you would centrally think of as text processing-- tsort performs a topological sort on a directed graph. Don't panic just yet if this concept is not familiar to you: in simple terms, tsort is good for finding a suitable order among dependencies. For example, installing packages might need to occur with certain order constraints, or some system daemons might need to be initialized before others.

Using tsort is quite simple, really. Just create a file (or stream) that lists each known dependency (space separated). The utility will produce a suitable (not necessarily uniquely so) order for the whole collection. E.g.:

          $ cat dependencies # not necessarily exhaustive, but realistic
          libpng XFree86
          FreeType XFree86
          Fontconfig XFree86
          FreeType Fontconfig
          expat Fontconfig
          Zlib libpng
          Binutils Zlib
          Coreutils Zlib
          GCC Zlib
          Glibc Zlib
          Sed Zlibc
          $ tsort dependencies


The pr utility is a general page formatter for text files that provides facilities such as page headers, linefeeds, columnization of source texts, indentation margins, and configurable page and line width. However, pr does not itself rewrap paragraphs, and so might often be used in conjunction with fmt .

          $ tail -5 rexx.txt | pr -w 60 -f | head
          2004-01-31 03:22                                      Page 1

            {Picture of Author:}
            David Mertz' fondness for IBM dates back embarrassingly many decades.
            David may be reached at [email protected]; his life pored over at
   And buy his book: _Text Processing in
            Python_ (
          $ tail -5 rexx.txt | fmt -30 > blurb.txt
          $ pr blurb.txt -2 -w 65 -f | head
          2004-01-31 03:24                 blurb.txt                 Page 1

            {Picture of Author:              at [email protected]; his life
     pored over at
            David Mertz' fondness for IBM    And buy his book: _Text
            dates back embarrassingly many   Processing in Python_
            decades.  David may be reached   (

Combining and Splitting Multiple Files


The utility comm is used to compare the contents of already (alphabeticly) sorted files. This is useful when the lines of files are considered as unordered collections of items . The diff utility, though not included in the text utilities, is a more general way of comparing files that might have isolated modifications--but that are treated in an ordered manner (such as source code files or documents). On the other hand, files that are considered as fields of records do not have any inherent order, and sorting does not change the information content.

Let us look at the difference between two sorted lists of names; the columns displayed are those in first file only, those in the second only, and those in common:

          $ comm test2b test2c

Introducing an out-of-order name, we see that diff compares happily, while comm fails to identify overlaps anymore:

          $ cat test2d
          $ diff -U 2 test2d test2c
          --- test2d      Sun Feb  1 18:18:26 2004
          +++ test2c      Sun Feb  1 18:01:49 2004
          @@ -1,5 +1,4 @@
          $ comm test2d test2c


The utility join is quite interesting; it performs some basic relational calculus (as will be familiar to readers who know relational database theory). In short, join lets you find records that share fields between (sorted) record collections. For example, you might be interested in which IP addresses have vistited both your web site and your FTP site, along with information on these visits (resources requested, times, etc., which will be in your logs).

To present a simple example, suppose you issue color-coded access badges to various people: vendors, partners, employees. You'd like information on which badge types have been issued to employees. Notice that names are the first field in employees , but second in badges , all tab separated:

          $ cat employees
          Alice Aaronson  System Administrator    99 9th Street
          Bob Babuk       Programmer      7 77th Avenue
          Carol Cavo      Manager 111 West 1st Blvd.
          $ cat badges
          Red     Alice Aaronson
          Green   Alice Aaronson
          Red     David Decker
          Blue    Ernestine Eckel
          Green   Francis Fu
          $ join -1 2 -2 1 -t $'\t' badges employees
          Alice Aaronson  Red     System Administrator    99 9th Street
          Alice Aaronson  Green   System Administrator    99 9th Street


The utility paste is approximately the reverse operation of that performed by cut . That is, paste combines multiple files into columns, e.g. fields. By default, the corresponding lines between files are tab separated, but you may use a different delimiter by specifying a -d option.

While paste can combine unrelated files (leaving empty fields if one input is longer), it generally makes the most sense to paste synchronized data sources. One example of this is in reorganizing the fields of an existing data file, e.g.:

          $ cut -f 1 employees > names
          $ cut -f 2 employees > titles
          $ paste -d "," titles names
          System Administrator,Alice Aaronson
          Programmer,Bob Babuk
          Manager,Carol Cavo

The flag -s lets you reverse the use of rows and columns, which amounts to converting successive lines in a file into delimited fields:

          $ paste -s titles | cat -T
          System Administrator^IProgrammer^IManager


The utility split simply divides a file into multiple parts, each one of a specified number of lines or bytes (the last one perhaps smaller). The parts are written to files whose names are sequenced with two suffix letters (by default xaa , xab , ... xzz ) .

While split can be useful just in managing the size of large files or data sets, it is more intersting in processing more structured data. For example, in the panel "Lines as Records" we saw an example of splitting fields across lines--what if we want to assemble those back into employees style tab separated fields, one per line. Here is a way to do it:

          $ cut -b 3- multiline | split -l 3 - employee
          $ cat employeeab
          Bob Babuk
          7 77th Avenue
          $ paste -s employeea*
          Alice Aaronson  System Administrator    99 9th Street
          Bob Babuk       Programmer      7 77th Avenue


The utility csplit is similar to split , but it divides files based on context lines within them, rather than on simple line/byte counts. You may divide on one or more different criteria within a command, and may repeat each criterion however many times you wish. The most interesting criterion-type is regular expressions to match against lines. For example, as an odd cut-up of multiline :

          $ csplit multiline -zq 2 /99/ /Progr/ # line 2, find 99, find Progr
          $ cat xx00
          A Alice Aaronson
          $ cat xx01
          B System Administrator
          $ cat xx02
          C 99 9th Street
          A Bob Babuk
          $ cat xx03
          B Programmer
          C 7 77th Avenue

The above division is a bit perverse in that it does not correspond with the data structure. A more usual approach might be to arrange to have delimiter lines , and split on those throughout:

          $ head -5 multiline2
          Alice Aaronson
          System Administrator
          99 9th Street
          Bob Babuk
          $ csplit -zq multiline2 /-----/+1 {*} # incl dashes at end, per chunk
          $ cat xx01
          Bob Babuk
          7 77th Avenue

Summarizing and Identifying Files

The simplest summary: wc

Most of the tools we have seen before produce output that is largely reversible to create the original form--or at the least, each line of input contributes in some straightforward way to the output. A number of tools in the GNU Text Utilities can instead be best described as producing a summary a file. Specifically, the output of these utilities are generally much shorter than their inputs, and the utilities all discard most of the information in their input (technically, you could describe them as one-way functions .

About the simplest one-way function on an input file is to count its lines, words, and/or bytes, which is what wc does. These are interesting things to know about a file, but are clearly non-unique among distinct files. For example:

          $ wc rexx.txt # lines, words, chars, name
               402    2585   18231 rexx.txt
          $ wc -w < rexx.txt # bare word count
          $ wc -lc rexx.txt # lines, chars, name
               402   18231 rexx.txt

Put to a bit of use, suppose I wonder which developerWorks articles I have written are the wordiest, I might use (note inclusion of total, another pipe to tail could remove that):

          $ wc -w *.txt | sort -nr | head -4
              55190 total
               3905 quantum_computer.txt
               3785 filtering-spam.txt
               3098 linuxppc.txt

cksum and sum

The utilities cksum and sum produce checksums and block counts of files. The latter exists for historical reasons only, and implements a less robust method. Either utility produces a calculated value that is unlikely to be the same between randomly chosen files. In particular, a checksum lets you establish to a reasonable degree of certainty that a file has not become corrupted in transmission or accidentally modified. cksum implements four successively more robust techniques, where -o 1 is the behavior of sum , and the default (no switch) is best.

          $ cksum rexx.txt
          937454632 18231 rexx.txt
          $ cksum -o 3 < rexx.txt
          4101119105 18231
          $ cat rexx.txt | cksum -o 2
          47555 36
          $ cksum -o 1 rexx.txt
          10915 18 rexx.txt

md5sum and sha1sum

The utilities md5sum and sha1sum are similar in concept to that of cksum . Note, by the way, that in BSD-derived systems, the former command goes by the name md5 . However, md5sum and sha1sum produce 128-bit and 160-bit checksums, respectively, rather than the 16-bit or 32-bit outputs of cksum . Checksums are also called hashes.

The difference in checksum lengths gives a hint as to a difference in purpose. In truth, comparing a 32-bit hash value is quite unlikely to falsely indicate that a file was transmitted correctly and left unchanged. But protection against accidents is a much weaker standard than protection against malicious tamperers. And MD5 or SHA hash is a value that is computationally infeasable to spoof. The hash length of a cryptographic hash like MD5 or SHA is necessary for its strength, but a lot more than just the length went into their design.

The scenario to imagine is where you are sent a file on an insecure channel. In order to make sure that you receive the real data rather than some malicious substitute, the sender publishes (through a different channel) an MD5 or SHA hash for the file. An adversary cannot create a false file with the published MD5/SHA hash--the checksum, for practical purposes, uniquely identifies the desired file. While sha1sum is actually a bit better cryptographically, for historical reasons, md5sum is in more widespread use.

          $ md5sum rexx.txt
          2cbdbc5bc401b6eb70a0d127609d8772  rexx.txt
          $ cat md5s
          2cbdbc5bc401b6eb70a0d127609d8772  rexx.txt
          c8d19740349f9cd9776827a0135444d5  metaclass.txt
          $ md5sum -cv md5s
          rexx.txt       OK
          metaclass.txt  FAILED
          md5sum: 1 of 2 file(s) failed MD5 check

Working with Log Files

The structure of a weblog

A weblog file provides a good data source for demonstrating a variety of real world uses of the text utilities. Standard Apache log files contain a variety of space-separated fields per line, with each line describing one access to a web resource. Unfortunately for us, spaces also occur at times inside quoted fields, so processing is often not quite as simple as we might hope.(or as it might be if the delimiter was excluded from the fields). Oh well, we must work with what we are given.

Let us take a look at a line from one of my weblogs before we perform some tasks with it:

          $ wc access-log
             24422  448497 5075558 access-log
          $ head -1 access-log | fmt -25
 - -
          [28/Dec/2003:00:00:16 -0600]
          "GET /TPiP/cover-small.jpg
          HTTP/1.1" 200 10146
          "Mozilla/4.0 (compatible;
          MSIE 6.0; Windows NT 5.1)"

We can see that the original file is pretty large--24k records. Wrapping the fields with fmt does not always wrap on field boundaries, but the quotes let you see what the fields are.

Extracting the IP addresses of Website visitors

A very simple task to peform on a weblog file is to extract all the IP addresses of visitors to the site. This combines a few of our a common pipe pattern (let us only look at the first few):

          $ cut -f 1 -d " " access-log | sort | uniq | head -5


We might wonder, as well, just how many such distinct visitors have visited in total:

          $ cut -f 1 -d " " access-log | sort | uniq | wc -l

Counting occurences

In the last panel, we determined how many visitors our website got, but perhaps we are also interested in how much each of those 2820 visitors contribute to the overal 24422 hits. Or specifically, who are the most frequent visitors. In one line we can run:

          $ cut -f 1 -d " " access-log | sort | uniq -c | sort -nr | head -5 

While this approach works, it might be nice to pull out the histogram part into a reusable shell script:

          $ cat histogram 
          sort | uniq -c | sort -nr | head -n $1
          $ cut -f 1 -d " " access-log | ./histogram 3

Now we can pipe any line-oriented list of items to our histogram shell script. The number of most frequent items we want to display is a parameter passed to the script.

Generate a new ad hoc report

Sometimes existing data files contain information we need, but not necessarily in the arrangement needed by a downstream process. As a basic example, suppose you want to pull several fields out of the weblog shown above, and combine them in different order (and skipping unneeded fields):

          $ cut -f 6 -d \" access-log > browsers
          $ cut -f 1 -d " " access-log > ips
          $ cut -f 2 -d \" access-log | cut -f 2 -d " " 
                 | tr "/" ":" > resources
          $ paste resources browsers ips >
          $ head -2 | tr "\t" "\n"
          Mozilla/4.0 (compatible; MSIE 6.0; Windows NT 5.1)

          Mozilla/4.0 (compatible; MSIE 6.0; Windows NT 5.1)


The line the produces resources uses two passes through cut , with different delimiters. That is because what access-log thinks of as a REQUEST contains more information than we want as a RESOURCE, i.e.:

          $ cut -f 2 -d \" access-log | head -1
          GET /TPiP/cover-small.jpg HTTP/1.1

We also decide to massage the path delimiter in the Apache log to use a colon path separator (which is the old MacOS format, but we really just do it here to show a type of operation).


This next script combines much of what we have seen in this tutorial into a rather complex pipeline. Suppose that we know we want to cut a field from a data file, but we do not know its field position. Obviously, visual examination could provide the answer, but to automate processing different types of data files, we can cut whichever field matches a regular expression:

          $ cat cut_by_regex
          # USAGE: cut_by_regex <pattern> <file> <delim> 
          cut -d "$3" -f \
            `head -n 1 $2 | tr "$3" "\n" | nl | \
             egrep $1 | cut -f 1 | head -1` \

In practice, we might use this as, e.g.:

          $ ./cut_by_regex "([0-9]+\.){3}" access-log " " | ./histogram 3

Several parts of this could use further explanation. The backtick is a special syntax in bash to treat the result of a command as a an argument to another command. Specifically, the pipe in the backticks produces the first field number that matches the regular expression given as the first argument. How does it mangage this? First we pull off only the first line of the data file; then we transform the specified delimiter to a newline (one field per line now); then we number the resulting lines/fields; then we search for a line with the desired pattern pattern; then we cut just the field number from the line; and finally, we only take the first match, even if several fields match. It takes a bit of thought to put a good pipeline together, but a lot can be done this way

Summary and resources


This tutorial only directly presents a small portion of what you can achieve with creative use of the GNU Text Utilities. The final few examples start to give a good sense of just how powerful they can be with creative use of pipes and redirection. The key is to break an overall transformation down into useful intermediate data, either saving that intermediary to another file or piping it to a utility that deals with that data format.

I wish to thank my colleage Andrew Blais for assistance in preparation of this tutorial.


You can download the 27 GNU Text Utilities from their FTP site.

The most current utilities have been incorporated into the GNU Core Utilies (88 in all).

Peter Seebach's The art of writing Linux utilities: Developing small, useful command-line tools

David Mertz's Regular Expression Tutorial is a good starting point for understanding the tools like grep and csplit that utilize regular expressions.

David's book Text Processing in Python (Addison Wesley, 2003; ISBN: 0-321-11254-7) also contains an introduction to regular expressions, as well as extensive discussion of performing many of the techniques in this tutorial using Python.

The Linux Zone article Scripting with Free Software Rexx Implementations , written by David Mertz, might be useful for an alternative approach to simple text processing tasks. The scope of the text utilities is nearly identical to the core purpose of the Rexx programming language.


Please let us know whether this tutorial was helpful to you and how we could make it better. We'd also like to hear about other tutorial topics you'd like to see covered.

For questions about the content of this tutorial, contact the author, David Mertz, at [email protected] .