General Introduction

This file documents awk, a program that you can use to select particular records in a file and perform operations upon them.

Copyright © 1989, 1991, 1992, 1993, 1996–2005, 2007, 2009–2023
Free Software Foundation, Inc.



This is Edition 5.2 of GAWK: Effective AWK Programming: A User’s Guide for GNU Awk, for the 5.2.2 (or later) version of the GNU implementation of AWK.

Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.3 or any later version published by the Free Software Foundation; with the Invariant Sections being “GNU General Public License”, with the Front-Cover Texts being “A GNU Manual”, and with the Back-Cover Texts as in (a) below. A copy of the license is included in the section entitled “GNU Free Documentation License”.

  1. The FSF’s Back-Cover Text is: “You have the freedom to copy and modify this GNU manual.”

Table of Contents

Short Table of Contents


Foreword to the Third Edition

Arnold Robbins and I are good friends. We were introduced in 1990 by circumstances—and our favorite programming language, AWK. The circumstances started a couple of years earlier. I was working at a new job and noticed an unplugged Unix computer sitting in the corner. No one knew how to use it, and neither did I. However, a couple of days later, it was running, and I was root and the one-and-only user. That day, I began the transition from statistician to Unix programmer.

On one of many trips to the library or bookstore in search of books on Unix, I found the gray AWK book, a.k.a. Alfred V. Aho, Brian W. Kernighan, and Peter J. Weinberger’s The AWK Programming Language (Addison-Wesley, 1988). awk’s simple programming paradigm—find a pattern in the input and then perform an action—often reduced complex or tedious data manipulations to a few lines of code. I was excited to try my hand at programming in AWK.

Alas, the awk on my computer was a limited version of the language described in the gray book. I discovered that my computer had “old awk” and the book described “new awk.” I learned that this was typical; the old version refused to step aside or relinquish its name. If a system had a new awk, it was invariably called nawk, and few systems had it. The best way to get a new awk was to ftp the source code for gawk from prep.ai.mit.edu. gawk was a version of new awk written by David Trueman and Arnold, and available under the GNU General Public License.

(Incidentally, it’s no longer difficult to find a new awk. gawk ships with GNU/Linux, and you can download binaries or source code for almost any system; my wife uses gawk on her VMS box.)

My Unix system started out unplugged from the wall; it certainly was not plugged into a network. So, oblivious to the existence of gawk and the Unix community in general, and desiring a new awk, I wrote my own, called mawk. Before I was finished, I knew about gawk, but it was too late to stop, so I eventually posted to a comp.sources newsgroup.

A few days after my posting, I got a friendly email from Arnold introducing himself. He suggested we share design and algorithms and attached a draft of the POSIX standard so that I could update mawk to support language extensions added after publication of The AWK Programming Language.

Frankly, if our roles had been reversed, I would not have been so open and we probably would have never met. I’m glad we did meet. He is an AWK expert’s AWK expert and a genuinely nice person. Arnold contributes significant amounts of his expertise and time to the Free Software Foundation.

This book is the gawk reference manual, but at its core it is a book about AWK programming that will appeal to a wide audience. It is a definitive reference to the AWK language as defined by the 1987 Bell Laboratories release and codified in the 1992 POSIX Utilities standard.

On the other hand, the novice AWK programmer can study a wealth of practical programs that emphasize the power of AWK’s basic idioms: data-driven control flow, pattern matching with regular expressions, and associative arrays. Those looking for something new can try out gawk’s interface to network protocols via special /inet files.

The programs in this book make clear that an AWK program is typically much smaller and faster to develop than a counterpart written in C. Consequently, there is often a payoff to prototyping an algorithm or design in AWK to get it running quickly and expose problems early. Often, the interpreted performance is adequate and the AWK prototype becomes the product.

The new pgawk (profiling gawk), produces program execution counts. I recently experimented with an algorithm that for n lines of input, exhibited ~ C n^2 performance, while theory predicted ~ C n log n behavior. A few minutes poring over the awkprof.out profile pinpointed the problem to a single line of code. pgawk is a welcome addition to my programmer’s toolbox.

Arnold has distilled over a decade of experience writing and using AWK programs, and developing gawk, into this book. If you use AWK or want to learn how, then read this book.

Michael Brennan
Author of mawk
March 2001

Foreword to the Fourth Edition

Some things don’t change. Thirteen years ago I wrote: “If you use AWK or want to learn how, then read this book.” True then, and still true today.

Learning to use a programming language is about more than mastering the syntax. One needs to acquire an understanding of how to use the features of the language to solve practical programming problems. A focus of this book is many examples that show how to use AWK.

Some things do change. Our computers are much faster and have more memory. Consequently, speed and storage inefficiencies of a high-level language matter less. Prototyping in AWK and then rewriting in C for performance reasons happens less, because more often the prototype is fast enough.

Of course, there are computing operations that are best done in C or C++. With gawk 4.1 and later, you do not have to choose between writing your program in AWK or in C/C++. You can write most of your program in AWK and the aspects that require C/C++ capabilities can be written in C/C++, and then the pieces glued together when the gawk module loads the C/C++ module as a dynamic plug-in. Writing Extensions for gawk, has all the details, and, as expected, many examples to help you learn the ins and outs.

I enjoy programming in AWK and had fun (re)reading this book. I think you will too.

Michael Brennan
Author of mawk
October 2014

Preface

Several kinds of tasks occur repeatedly when working with text files. You might want to extract certain lines and discard the rest. Or you may need to make changes wherever certain patterns appear, but leave the rest of the file alone. Such jobs are often easy with awk. The awk utility interprets a special-purpose programming language that makes it easy to handle simple data-reformatting jobs.

The GNU implementation of awk is called gawk; if you invoke it with the proper options or environment variables, it is fully compatible with the POSIX1 specification of the awk language and with the Unix version of awk maintained by Brian Kernighan. This means that all properly written awk programs should work with gawk. So most of the time, we don’t distinguish between gawk and other awk implementations.

Using awk you can:

In addition, gawk provides facilities that make it easy to:

This Web page teaches you about the awk language and how you can use it effectively. You should already be familiar with basic system commands, such as cat and ls,2 as well as basic shell facilities, such as input/output (I/O) redirection and pipes.

Implementations of the awk language are available for many different computing environments. This Web page, while describing the awk language in general, also describes the particular implementation of awk called gawk (which stands for “GNU awk”). gawk runs on a broad range of Unix systems, ranging from Intel-architecture PC-based computers up through large-scale systems. gawk has also been ported to macOS, z/OS, Microsoft Windows (all versions), and OpenVMS.3


History of awk and gawk

Recipe for a Programming Language
1 part egrep1 part snobol
2 parts ed3 parts C

Blend all parts well using lex and yacc. Document minimally and release.

After eight years, add another part egrep and two more parts C. Document very well and release.

The name awk comes from the initials of its designers: Alfred V. Aho, Peter J. Weinberger, and Brian W. Kernighan. The original version of awk was written in 1977 at AT&T Bell Laboratories. In 1985, a new version made the programming language more powerful, introducing user-defined functions, multiple input streams, and computed regular expressions. This new version became widely available with Unix System V Release 3.1 (1987). The version in System V Release 4 (1989) added some new features and cleaned up the behavior in some of the “dark corners” of the language. The specification for awk in the POSIX Command Language and Utilities standard further clarified the language. Both the gawk designers and the original awk designers at Bell Laboratories provided feedback for the POSIX specification.

Paul Rubin wrote gawk in 1986. Jay Fenlason completed it, with advice from Richard Stallman. John Woods contributed parts of the code as well. In 1988 and 1989, David Trueman, with help from me, thoroughly reworked gawk for compatibility with the newer awk. Circa 1994, I became the primary maintainer. Current development focuses on bug fixes, performance improvements, standards compliance, and, occasionally, new features.

In May 1997, Jürgen Kahrs felt the need for network access from awk, and with a little help from me, set about adding features to do this for gawk. At that time, he also wrote the bulk of TCP/IP Internetworking with gawk (a separate document, available as part of the gawk distribution). His code finally became part of the main gawk distribution with gawk version 3.1.

John Haque rewrote the gawk internals, in the process providing an awk-level debugger. This version became available as gawk version 4.0 in 2011.

See Major Contributors to gawk for a full list of those who have made important contributions to gawk.


A Rose by Any Other Name

The awk language has evolved over the years. Full details are provided in The Evolution of the awk Language. The language described in this Web page is often referred to as “new awk.” By analogy, the original version of awk is referred to as “old awk.”

On most current systems, when you run the awk utility you get some version of new awk.4 If your system’s standard awk is the old one, you will see something like this if you try the following test program:

$ awk 1 /dev/null
error→ awk: syntax error near line 1
error→ awk: bailing out near line 1

In this case, you should find a version of new awk, or just install gawk!

Throughout this Web page, whenever we refer to a language feature that should be available in any complete implementation of POSIX awk, we simply use the term awk. When referring to a feature that is specific to the GNU implementation, we use the term gawk.


Using This Book

The term awk refers to a particular program as well as to the language you use to tell this program what to do. When we need to be careful, we call the language “the awk language,” and the program “the awk utility.” This Web page explains both how to write programs in the awk language and how to run the awk utility. The term “awk program” refers to a program written by you in the awk programming language.

Primarily, this Web page explains the features of awk as defined in the POSIX standard. It does so in the context of the gawk implementation. While doing so, it also attempts to describe important differences between gawk and other awk implementations.5 Finally, it notes any gawk features that are not in the POSIX standard for awk.

This Web page has the difficult task of being both a tutorial and a reference. If you are a novice, feel free to skip over details that seem too complex. You should also ignore the many cross-references; they are for the expert user and for the Info and HTML versions of the Web page.

There are sidebars scattered throughout the Web page. They add a more complete explanation of points that are relevant, but not likely to be of interest on first reading. All appear in the index, under the heading “sidebar.”

Most of the time, the examples use complete awk programs. Some of the more advanced sections show only the part of the awk program that illustrates the concept being described.

Although this Web page is aimed principally at people who have not been exposed to awk, there is a lot of information here that even the awk expert should find useful. In particular, the description of POSIX awk and the example programs in A Library of awk Functions, and in Practical awk Programs, should be of interest.

This Web page is split into several parts, as follows:

  • Part I describes the awk language and the gawk program in detail. It starts with the basics, and continues through all of the features of awk. It contains the following chapters:
    • Getting Started with awk, provides the essentials you need to know to begin using awk.
    • Running awk and gawk, describes how to run gawk, the meaning of its command-line options, and how it finds awk program source files.
    • Regular Expressions, introduces regular expressions in general, and in particular the flavors supported by POSIX awk and gawk.
    • Reading Input Files, describes how awk reads your data. It introduces the concepts of records and fields, as well as the getline command. I/O redirection is first described here. Network I/O is also briefly introduced here.
    • Printing Output, describes how awk programs can produce output with print and printf.
    • Expressions, describes expressions, which are the basic building blocks for getting most things done in a program.
    • Patterns, Actions, and Variables, describes how to write patterns for matching records, actions for doing something when a record is matched, and the predefined variables awk and gawk use.
    • Arrays in awk, covers awk’s one-and-only data structure: the associative array. Deleting array elements and whole arrays is described, as well as sorting arrays in gawk. The chapter also describes how gawk provides arrays of arrays.
    • Functions, describes the built-in functions awk and gawk provide, as well as how to define your own functions. It also discusses how gawk lets you call functions indirectly.
  • Part II shows how to use awk and gawk for problem solving. There is lots of code here for you to read and learn from. This part contains the following chapters:

    Reading these two chapters allows you to see awk solving real problems.

  • Part III focuses on features specific to gawk. It contains the following chapters:
  • Part IV provides the appendices, the Glossary, and two licenses that cover the gawk source code and this Web page, respectively. It contains the following appendices:
    • The Evolution of the awk Language, describes how the awk language has evolved since its first release to the present. It also describes how gawk has acquired features over time.
    • Installing gawk, describes how to get gawk, how to compile it on POSIX-compatible systems, and how to compile and use it on different non-POSIX systems. It also describes how to report bugs in gawk and where to get other freely available awk implementations.
    • Implementation Notes, describes how to disable gawk’s extensions, as well as how to contribute new code to gawk, and some possible future directions for gawk development.
    • Basic Programming Concepts, provides some very cursory background material for those who are completely unfamiliar with computer programming.
    • The Glossary, defines most, if not all, of the significant terms used throughout the Web page. If you find terms that you aren’t familiar with, try looking them up here.
    • GNU General Public License, and GNU Free Documentation License, present the licenses that cover the gawk source code and this Web page, respectively.

Typographical Conventions

This Web page is written in Texinfo, the GNU documentation formatting language. A single Texinfo source file is used to produce both the printed and online versions of the documentation. Because of this, the typographical conventions are slightly different than in other books you may have read.

Examples you would type at the command line are preceded by the common shell primary and secondary prompts, ‘$’ and ‘>’, respectively. Input that you type is shown like this. Output from the command is preceded by the glyph “-|”. This typically represents the command’s standard output. Error messages and other output on the command’s standard error are preceded by the glyph “error→”. For example:

$ echo hi on stdout
-| hi on stdout
$ echo hello on stderr 1>&2
error→ hello on stderr

In the text, almost anything related to programming, such as command names, variable and function names, and string, numeric and regexp constants appear in this font. Code fragments appear in the same font and quoted, ‘like this’. Things that are replaced by the user or programmer appear in this font. Options look like this: -f. File names are indicated like this: /path/to/ourfile. Some things are emphasized like this, and if a point needs to be made strongly, it is done like this. The first occurrence of a new term is usually its definition and appears in the same font as the previous occurrence of “definition” in this sentence.

Characters that you type at the keyboard look like this. In particular, there are special characters called “control characters.” These are characters that you type by holding down both the CONTROL key and another key, at the same time. For example, a Ctrl-d is typed by first pressing and holding the CONTROL key, next pressing the d key, and finally releasing both keys.

For the sake of brevity, throughout this Web page, we refer to Brian Kernighan’s version of awk as “BWK awk.” (See Other Freely Available awk Implementations for information on his and other versions.)

Dark Corners

Dark corners are basically fractal—no matter how much you illuminate, there’s always a smaller but darker one.

Brian Kernighan

Until the POSIX standard (and GAWK: Effective AWK Programming), many features of awk were either poorly documented or not documented at all. Descriptions of such features (often called “dark corners”) are noted in this Web page with “(d.c.).” They also appear in the index under the heading “dark corner.”

But, as noted by the opening quote, any coverage of dark corners is by definition incomplete.

Extensions to the standard awk language that are supported by more than one awk implementation are marked “(c.e.),” and listed in the index under “common extensions” and “extensions, common.”


The GNU Project and This Book

The Free Software Foundation (FSF) is a nonprofit organization dedicated to the production and distribution of freely distributable software. It was founded by Richard M. Stallman, the author of the original Emacs editor. GNU Emacs is the most widely used version of Emacs today.

The GNU6 Project is an ongoing effort on the part of the Free Software Foundation to create a complete, freely distributable, POSIX-compliant computing environment. The FSF uses the GNU General Public License (GPL) to ensure that its software’s source code is always available to the end user. A copy of the GPL is included in this Web page for your reference (see GNU General Public License). The GPL applies to the C language source code for gawk. To find out more about the FSF and the GNU Project online, see the GNU Project’s home page. This Web page may also be read from GNU’s website.

A shell, an editor (Emacs), highly portable optimizing C, C++, and Objective-C compilers, a symbolic debugger and dozens of large and small utilities (such as gawk), have all been completed and are freely available. The GNU operating system kernel (the HURD), has been released but remains in an early stage of development.

Until the GNU operating system is more fully developed, you should consider using GNU/Linux, a freely distributable, Unix-like operating system for Intel, Power Architecture, Sun SPARC, IBM S/390, and other systems.7 Many GNU/Linux distributions are available for download from the Internet.

The Web page you are reading is actually free—at least, the information in it is free to anyone. The machine-readable source code for the Web page comes with gawk. (Take a moment to check the Free Documentation License in GNU Free Documentation License.)

The Web page itself has gone through multiple previous editions. Paul Rubin wrote the very first draft of The GAWK Manual; it was around 40 pages long. Diane Close and Richard Stallman improved it, yielding a version that was around 90 pages and barely described the original, “old” version of awk.

I started working with that version in the fall of 1988. As work on it progressed, the FSF published several preliminary versions (numbered 0.x). In 1996, edition 1.0 was released with gawk 3.0.0. The FSF published the first two editions under the title The GNU Awk User’s Guide.

This edition maintains the basic structure of the previous editions. For FSF edition 4.0, the content was thoroughly reviewed and updated. All references to gawk versions prior to 4.0 were removed. Of significant note for that edition was the addition of Debugging awk Programs.

For FSF edition 5.0, the content has been reorganized into parts, and the major new additions are Arithmetic and Arbitrary-Precision Arithmetic with gawk, and Writing Extensions for gawk.

This Web page will undoubtedly continue to evolve. If you find an error in the Web page, please report it! See Reporting Problems and Bugs for information on submitting problem reports electronically.


How to Contribute

As the maintainer of GNU awk, I once thought that I would be able to manage a collection of publicly available awk programs and I even solicited contributions. Making things available on the Internet helps keep the gawk distribution down to manageable size.

The initial collection of material, such as it is, is still available at ftp://ftp.freefriends.org/arnold/Awkstuff.

In the hopes of doing something broader, I acquired the awklang.org domain. Late in 2017, a volunteer took on the task of managing it.

If you have written an interesting awk program that you would like to share with the rest of the world, please see http://www.awklang.org and use the “Contact” link.

If you have written a gawk extension, please see The gawkextlib Project.


Acknowledgments

The initial draft of The GAWK Manual had the following acknowledgments:

Many people need to be thanked for their assistance in producing this manual. Jay Fenlason contributed many ideas and sample programs. Richard Mlynarik and Robert Chassell gave helpful comments on drafts of this manual. The paper A Supplemental Document for AWK by John W. Pierce of the Chemistry Department at UC San Diego, pinpointed several issues relevant both to awk implementation and to this manual, that would otherwise have escaped us.

I would like to acknowledge Richard M. Stallman, for his vision of a better world and for his courage in founding the FSF and starting the GNU Project.

Earlier editions of this Web page had the following acknowledgements:

The following people (in alphabetical order) provided helpful comments on various versions of this book: Rick Adams, Dr. Nelson H.F. Beebe, Karl Berry, Dr. Michael Brennan, Rich Burridge, Claire Cloutier, Diane Close, Scott Deifik, Christopher (“Topher”) Eliot, Jeffrey Friedl, Dr. Darrel Hankerson, Michal Jaegermann, Dr. Richard J. LeBlanc, Michael Lijewski, Pat Rankin, Miriam Robbins, Mary Sheehan, and Chuck Toporek.

Robert J. Chassell provided much valuable advice on the use of Texinfo. He also deserves special thanks for convincing me not to title this Web page How to Gawk Politely. Karl Berry helped significantly with the TeX part of Texinfo.

I would like to thank Marshall and Elaine Hartholz of Seattle and Dr. Bert and Rita Schreiber of Detroit for large amounts of quiet vacation time in their homes, which allowed me to make significant progress on this Web page and on gawk itself.

Phil Hughes of SSC contributed in a very important way by loaning me his laptop GNU/Linux system, not once, but twice, which allowed me to do a lot of work while away from home.

David Trueman deserves special credit; he has done a yeoman job of evolving gawk so that it performs well and without bugs. Although he is no longer involved with gawk, working with him on this project was a significant pleasure.

The intrepid members of the GNITS mailing list, and most notably Ulrich Drepper, provided invaluable help and feedback for the design of the internationalization features.

Chuck Toporek, Mary Sheehan, and Claire Cloutier of O’Reilly & Associates contributed significant editorial help for this Web page for the 3.1 release of gawk.

Dr. Nelson Beebe, Andreas Buening, Dr. Manuel Collado, Antonio Colombo, Stephen Davies, Scott Deifik, Akim Demaille, Daniel Richard G., Juan Manuel Guerrero, Darrel Hankerson, Michal Jaegermann, Jürgen Kahrs, Stepan Kasal, John Malmberg, Chet Ramey, Pat Rankin, Andrew Schorr, Corinna Vinschen, and Eli Zaretskii (in alphabetical order) make up the current gawk “crack portability team.” Without their hard work and help, gawk would not be nearly the robust, portable program it is today. It has been and continues to be a pleasure working with this team of fine people.

Notable code and documentation contributions were made by a number of people. See Major Contributors to gawk for the full list.

Thanks to Michael Brennan for the Forewords.

Thanks to Patrice Dumas for the new makeinfo program. Thanks to Karl Berry for his past work on Texinfo, and to Gavin Smith, who continues to work to improve the Texinfo markup language.

Robert P.J. Day, Michael Brennan, and Brian Kernighan kindly acted as reviewers for the 2015 edition of this Web page. Their feedback helped improve the final work.

I would also like to thank Brian Kernighan for his invaluable assistance during the testing and debugging of gawk, and for his ongoing help and advice in clarifying numerous points about the language. We could not have done nearly as good a job on either gawk or its documentation without his help.

Brian is in a class by himself as a programmer and technical author. I have to thank him (yet again) for his ongoing friendship and for being a role model to me for over 30 years! Having him as a reviewer is an exciting privilege. It has also been extremely humbling...

I must thank my wonderful wife, Miriam, for her patience through the many versions of this project, for her proofreading, and for sharing me with the computer. I would like to thank my parents for their love, and for the grace with which they raised and educated me. Finally, I also must acknowledge my gratitude to G-d, for the many opportunities He has sent my way, as well as for the gifts He has given me with which to take advantage of those opportunities.



Arnold Robbins
Nof Ayalon
Israel
March, 2020


Part I:
The awk Language


1 Getting Started with awk

The basic function of awk is to search files for lines (or other units of text) that contain certain patterns. When a line matches one of the patterns, awk performs specified actions on that line. awk continues to process input lines in this way until it reaches the end of the input files.

Programs in awk are different from programs in most other languages, because awk programs are data driven (i.e., you describe the data you want to work with and then what to do when you find it). Most other languages are procedural; you have to describe, in great detail, every step the program should take. When working with procedural languages, it is usually much harder to clearly describe the data your program will process. For this reason, awk programs are often refreshingly easy to read and write.

When you run awk, you specify an awk program that tells awk what to do. The program consists of a series of rules (it may also contain function definitions, an advanced feature that we will ignore for now; see User-Defined Functions). Each rule specifies one pattern to search for and one action to perform upon finding the pattern.

Syntactically, a rule consists of a pattern followed by an action. The action is enclosed in braces to separate it from the pattern. Newlines usually separate rules. Therefore, an awk program looks like this:

pattern { action }
pattern { action }
…

1.1 How to Run awk Programs

There are several ways to run an awk program. If the program is short, it is easiest to include it in the command that runs awk, like this:

awk 'program' input-file1 input-file2

When the program is long, it is usually more convenient to put it in a file and run it with a command like this:

awk -f program-file input-file1 input-file2

This section discusses both mechanisms, along with several variations of each.


1.1.1 One-Shot Throwaway awk Programs

Once you are familiar with awk, you will often type in simple programs the moment you want to use them. Then you can write the program as the first argument of the awk command, like this:

awk 'program' input-file1 input-file2

where program consists of a series of patterns and actions, as described earlier.

This command format instructs the shell, or command interpreter, to start awk and use the program to process records in the input file(s). There are single quotes around program so the shell won’t interpret any awk characters as special shell characters. The quotes also cause the shell to treat all of program as a single argument for awk, and allow program to be more than one line long.

This format is also useful for running short or medium-sized awk programs from shell scripts, because it avoids the need for a separate file for the awk program. A self-contained shell script is more reliable because there are no other files to misplace.

Later in this chapter, in Some Simple Examples, we’ll see examples of several short, self-contained programs.


1.1.2 Running awk Without Input Files

You can also run awk without any input files. If you type the following command line:

awk 'program'

awk applies the program to the standard input, which usually means whatever you type on the keyboard. This continues until you indicate end-of-file by typing Ctrl-d. (On non-POSIX operating systems, the end-of-file character may be different.)

As an example, the following program prints a friendly piece of advice (from Douglas Adams’s The Hitchhiker’s Guide to the Galaxy), to keep you from worrying about the complexities of computer programming:

$ awk 'BEGIN { print "Don\47t Panic!" }'
-| Don't Panic!

awk executes statements associated with BEGIN before reading any input. If there are no other statements in your program, as is the case here, awk just stops, instead of trying to read input it doesn’t know how to process. The ‘\47’ is a magic way (explained later) of getting a single quote into the program, without having to engage in ugly shell quoting tricks.

NOTE: If you use Bash as your shell, you should execute the command ‘set +H’ before running this program interactively, to disable the C shell-style command history, which treats ‘!’ as a special character. We recommend putting this command into your personal startup file.

This next simple awk program emulates the cat utility; it copies whatever you type on the keyboard to its standard output (why this works is explained shortly):

$ awk '{ print }'
Now is the time for all good men
-| Now is the time for all good men
to come to the aid of their country.
-| to come to the aid of their country.
Four score and seven years ago, ...
-| Four score and seven years ago, ...
What, me worry?
-| What, me worry?
Ctrl-d

1.1.3 Running Long Programs

Sometimes awk programs are very long. In these cases, it is more convenient to put the program into a separate file. In order to tell awk to use that file for its program, you type:

awk -f source-file input-file1 input-file2

The -f instructs the awk utility to get the awk program from the file source-file (see Command-Line Options). Any file name can be used for source-file. For example, you could put the program:

BEGIN { print "Don't Panic!" }

into the file advice. Then this command:

awk -f advice

does the same thing as this one:

awk 'BEGIN { print "Don\47t Panic!" }'

This was explained earlier (see Running awk Without Input Files). Note that you don’t usually need single quotes around the file name that you specify with -f, because most file names don’t contain any of the shell’s special characters. Notice that in advice, the awk program did not have single quotes around it. The quotes are only needed for programs that are provided on the awk command line. (Also, placing the program in a file allows us to use a literal single quote in the program text, instead of the magic ‘\47’.)

If you want to clearly identify an awk program file as such, you can add the extension .awk to the file name. This doesn’t affect the execution of the awk program but it does make “housekeeping” easier.


1.1.4 Executable awk Programs

Once you have learned awk, you may want to write self-contained awk scripts, using the ‘#!’ script mechanism. You can do this on many systems.8 For example, you could update the file advice to look like this:

#! /bin/awk -f
BEGIN { print "Don't Panic!" }

After making this file executable (with the chmod utility), simply type ‘advice’ at the shell and the system arranges to run awk as if you had typed ‘awk -f advice’:

$ chmod +x advice
$ ./advice
-| Don't Panic!

Self-contained awk scripts are useful when you want to write a program that users can invoke without their having to know that the program is written in awk.

Understanding ‘#!

awk is an interpreted language. This means that the awk utility reads your program and then processes your data according to the instructions in your program. (This is different from a compiled language such as C, where your program is first compiled into machine code that is executed directly by your system’s processor.) The awk utility is thus termed an interpreter. Many modern languages are interpreted.

The line beginning with ‘#!’ lists the full file name of an interpreter to run and a single optional initial command-line argument to pass to that interpreter. The operating system then runs the interpreter with the given argument and the full argument list of the executed program. The first argument in the list is the full file name of the awk program. The rest of the argument list contains either options to awk, or data files, or both. (Note that on many systems awk is found in /usr/bin instead of in /bin.)

Some systems limit the length of the interpreter name to 32 characters. Often, this can be dealt with by using a symbolic link.

You should not put more than one argument on the ‘#!’ line after the path to awk. It does not work. The operating system treats the rest of the line as a single argument and passes it to awk. Doing this leads to confusing behavior—most likely a usage diagnostic of some sort from awk.

Finally, the value of ARGV[0] (see Predefined Variables) varies depending upon your operating system. Some systems put ‘awk’ there, some put the full pathname of awk (such as /bin/awk), and some put the name of your script (‘advice’). (d.c.) Don’t rely on the value of ARGV[0] to provide your script name.


1.1.5 Comments in awk Programs

A comment is some text that is included in a program for the sake of human readers; it is not really an executable part of the program. Comments can explain what the program does and how it works. Nearly all programming languages have provisions for comments, as programs are typically hard to understand without them.

In the awk language, a comment starts with the number sign character (‘#’) and continues to the end of the line. The ‘#’ does not have to be the first character on the line. The awk language ignores the rest of a line following a number sign. For example, we could have put the following into advice:

# This program prints a nice, friendly message.  It helps
# keep novice users from being afraid of the computer.
BEGIN    { print "Don't Panic!" }

You can put comment lines into keyboard-composed throwaway awk programs, but this usually isn’t very useful; the purpose of a comment is to help you or another person understand the program when reading it at a later time.

CAUTION: As mentioned in One-Shot Throwaway awk Programs, you can enclose short to medium-sized programs in single quotes, in order to keep your shell scripts self-contained. When doing so, don’t put an apostrophe (i.e., a single quote) into a comment (or anywhere else in your program). The shell interprets the quote as the closing quote for the entire program. As a result, usually the shell prints a message about mismatched quotes, and if awk actually runs, it will probably print strange messages about syntax errors. For example, look at the following:

$ awk 'BEGIN { print "hello" } # let's be cute'
>

The shell sees that the first two quotes match, and that a new quoted object begins at the end of the command line. It therefore prompts with the secondary prompt, waiting for more input. With Unix awk, closing the quoted string produces this result:

$ awk '{ print "hello" } # let's be cute'
> '
error→ awk: can't open file be
error→  source line number 1

Putting a backslash before the single quote in ‘let's’ wouldn’t help, because backslashes are not special inside single quotes. The next subsection describes the shell’s quoting rules.


1.1.6 Shell Quoting Issues

For short to medium-length awk programs, it is most convenient to enter the program on the awk command line. This is best done by enclosing the entire program in single quotes. This is true whether you are entering the program interactively at the shell prompt, or writing it as part of a larger shell script:

awk 'program text' input-file1 input-file2

Once you are working with the shell, it is helpful to have a basic knowledge of shell quoting rules. The following rules apply only to POSIX-compliant, Bourne-style shells (such as Bash, the GNU Bourne-Again Shell). If you use the C shell, you’re on your own.

Before diving into the rules, we introduce a concept that appears throughout this Web page, which is that of the null, or empty, string.

The null string is character data that has no value. In other words, it is empty. It is written in awk programs like this: "". In the shell, it can be written using single or double quotes: "" or ''. Although the null string has no characters in it, it does exist. For example, consider this command:

$ echo ""

Here, the echo utility receives a single argument, even though that argument has no characters in it. In the rest of this Web page, we use the terms null string and empty string interchangeably. Now, on to the quoting rules:

  • Quoted items can be concatenated with nonquoted items as well as with other quoted items. The shell turns everything into one argument for the command.
  • Preceding any single character with a backslash (‘\’) quotes that character. The shell removes the backslash and passes the quoted character on to the command.
  • Single quotes protect everything between the opening and closing quotes. The shell does no interpretation of the quoted text, passing it on verbatim to the command. It is impossible to embed a single quote inside single-quoted text. Refer back to Comments in awk Programs for an example of what happens if you try.
  • Double quotes protect most things between the opening and closing quotes. The shell does at least variable and command substitution on the quoted text. Different shells may do additional kinds of processing on double-quoted text.

    Because certain characters within double-quoted text are processed by the shell, they must be escaped within the text. Of note are the characters ‘$’, ‘`’, ‘\’, and ‘"’, all of which must be preceded by a backslash within double-quoted text if they are to be passed on literally to the program. (The leading backslash is stripped first.) Thus, the example seen previously in Running awk Without Input Files:

    awk 'BEGIN { print "Don\47t Panic!" }'
    

    could instead be written this way:

    $ awk "BEGIN { print \"Don't Panic!\" }"
    -| Don't Panic!
    

    Note that the single quote is not special within double quotes.

  • Null strings are removed when they occur as part of a non-null command-line argument, while explicit null objects are kept. For example, to specify that the field separator FS should be set to the null string, use:
    awk -F "" 'program' files # correct
    

    Don’t use this:

    awk -F"" 'program' files  # wrong!
    

    In the second case, awk attempts to use the text of the program as the value of FS, and the first file name as the text of the program! This results in syntax errors at best, and confusing behavior at worst.

Mixing single and double quotes is difficult. You have to resort to shell quoting tricks, like this:

$ awk 'BEGIN { print "Here is a single quote <'"'"'>" }'
-| Here is a single quote <'>

This program consists of three concatenated quoted strings. The first and the third are single-quoted, and the second is double-quoted.

This can be “simplified” to:

$ awk 'BEGIN { print "Here is a single quote <'\''>" }'
-| Here is a single quote <'>

Judge for yourself which of these two is the more readable.

Another option is to use double quotes, escaping the embedded, awk-level double quotes:

$ awk "BEGIN { print \"Here is a single quote <'>\" }"
-| Here is a single quote <'>

This option is also painful, because double quotes, backslashes, and dollar signs are very common in more advanced awk programs.

A third option is to use the octal escape sequence equivalents (see Escape Sequences) for the single- and double-quote characters, like so:

$ awk 'BEGIN { print "Here is a single quote <\47>" }'
-| Here is a single quote <'>
$ awk 'BEGIN { print "Here is a double quote <\42>" }'
-| Here is a double quote <">

This works nicely, but you should comment clearly what the escape sequences mean.

A fourth option is to use command-line variable assignment, like this:

$ awk -v sq="'" 'BEGIN { print "Here is a single quote <" sq ">" }'
-| Here is a single quote <'>

(Here, the two string constants and the value of sq are concatenated into a single string that is printed by print.)

If you really need both single and double quotes in your awk program, it is probably best to move it into a separate file, where the shell won’t be part of the picture and you can say what you mean.


1.1.6.1 Quoting in MS-Windows Batch Files

Although this Web page generally only worries about POSIX systems and the POSIX shell, the following issue arises often enough for many users that it is worth addressing.

The “shells” on Microsoft Windows systems use the double-quote character for quoting, and make it difficult or impossible to include an escaped double-quote character in a command-line script. The following example, courtesy of Jeroen Brink, shows how to escape the double quotes from this one liner script that prints all lines in a file surrounded by double quotes:

{ print "\"" $0 "\"" }

In an MS-Windows command-line the one-liner script above may be passed as follows:

gawk "{ print \"\042\" $0 \"\042\" }" file

In this example the ‘\042’ is the octal code for a double-quote; gawk converts it into a real double-quote for output by the print statement.

In MS-Windows escaping double-quotes is a little tricky because you use backslashes to escape double-quotes, but backslashes themselves are not escaped in the usual way; indeed they are either duplicated or not, depending upon whether there is a subsequent double-quote. The MS-Windows rule for double-quoting a string is the following:

  1. For each double quote in the original string, let N be the number of backslash(es) before it, N might be zero. Replace these N backslash(es) by 2*N+1 backslash(es)
  2. Let N be the number of backslash(es) tailing the original string, N might be zero. Replace these N backslash(es) by 2*N backslash(es)
  3. Surround the resulting string by double-quotes.

So to double-quote the one-liner script ‘{ print "\"" $0 "\"" }’ from the previous example you would do it this way:

gawk "{ print \"\\\"\" $0 \"\\\"\" }" file

However, the use of ‘\042’ instead of ‘\\\"’ is also possible and easier to read, because backslashes that are not followed by a double-quote don’t need duplication.


1.2 Data files for the Examples

Many of the examples in this Web page take their input from two sample data files. The first, mail-list, represents a list of peoples’ names together with their email addresses and information about those people. The second data file, called inventory-shipped, contains information about monthly shipments. In both files, each line is considered to be one record.

In mail-list, each record contains the name of a person, his/her phone number, his/her email address, and a code for his/her relationship with the author of the list. The columns are aligned using spaces. An ‘A’ in the last column means that the person is an acquaintance. An ‘F’ in the last column means that the person is a friend. An ‘R’ means that the person is a relative:

Amelia       555-5553     [email protected]    F
Anthony      555-3412     [email protected]   A
Becky        555-7685     [email protected]      A
Bill         555-1675     [email protected]       A
Broderick    555-0542     [email protected] R
Camilla      555-2912     [email protected]     R
Fabius       555-1234     [email protected]    F
Julie        555-6699     [email protected]   F
Martin       555-6480     [email protected]    A
Samuel       555-3430     [email protected]        A
Jean-Paul    555-2127     [email protected]     R

The data file inventory-shipped represents information about shipments during the year. Each record contains the month, the number of green crates shipped, the number of red boxes shipped, the number of orange bags shipped, and the number of blue packages shipped, respectively. There are 16 entries, covering the 12 months of last year and the first four months of the current year. An empty line separates the data for the two years:

Jan  13  25  15 115
Feb  15  32  24 226
Mar  15  24  34 228
Apr  31  52  63 420
May  16  34  29 208
Jun  31  42  75 492
Jul  24  34  67 436
Aug  15  34  47 316
Sep  13  55  37 277
Oct  29  54  68 525
Nov  20  87  82 577
Dec  17  35  61 401
Jan  21  36  64 620
Feb  26  58  80 652
Mar  24  75  70 495
Apr  21  70  74 514

The sample files are included in the gawk distribution, in the directory awklib/eg/data.


1.3 Some Simple Examples

The following command runs a simple awk program that searches the input file mail-list for the character string ‘li’ (a grouping of characters is usually called a string; the term string is based on similar usage in English, such as “a string of pearls” or “a string of cars in a train”):

awk '/li/ { print $0 }' mail-list

When lines containing ‘li’ are found, they are printed because ‘print $0 means print the current line. (Just ‘print’ by itself means the same thing, so we could have written that instead.)

You will notice that slashes (‘/’) surround the string ‘li’ in the awk program. The slashes indicate that ‘li’ is the pattern to search for. This type of pattern is called a regular expression, which is covered in more detail later (see Regular Expressions). The pattern is allowed to match parts of words. There are single quotes around the awk program so that the shell won’t interpret any of it as special shell characters.

Here is what this program prints:

$ awk '/li/ { print $0 }' mail-list
-| Amelia       555-5553     [email protected]    F
-| Broderick    555-0542     [email protected] R
-| Julie        555-6699     [email protected]   F
-| Samuel       555-3430     [email protected]        A

In an awk rule, either the pattern or the action can be omitted, but not both. If the pattern is omitted, then the action is performed for every input line. If the action is omitted, the default action is to print all lines that match the pattern.

Thus, we could leave out the action (the print statement and the braces) in the previous example and the result would be the same: awk prints all lines matching the pattern ‘li’. By comparison, omitting the print statement but retaining the braces makes an empty action that does nothing (i.e., no lines are printed).

Many practical awk programs are just a line or two long. Following is a collection of useful, short programs to get you started. Some of these programs contain constructs that haven’t been covered yet. (The description of the program will give you a good idea of what is going on, but you’ll need to read the rest of the Web page to become an awk expert!) Most of the examples use a data file named data. This is just a placeholder; if you use these programs yourself, substitute your own file names for data.

Some of the following examples use the output of ‘ls -l as input. ls is a system command that gives you a listing of the files in a directory. With the -l option, this listing includes each file’s size and the date the file was last modified. Its output looks like this:

-rw-r--r--  1 arnold   user   1933 Nov  7 13:05 Makefile
-rw-r--r--  1 arnold   user  10809 Nov  7 13:03 awk.h
-rw-r--r--  1 arnold   user    983 Apr 13 12:14 awk.tab.h
-rw-r--r--  1 arnold   user  31869 Jun 15 12:20 awkgram.y
-rw-r--r--  1 arnold   user  22414 Nov  7 13:03 awk1.c
-rw-r--r--  1 arnold   user  37455 Nov  7 13:03 awk2.c
-rw-r--r--  1 arnold   user  27511 Dec  9 13:07 awk3.c
-rw-r--r--  1 arnold   user   7989 Nov  7 13:03 awk4.c

The first field contains read-write permissions, the second field contains the number of links to the file, and the third field identifies the file’s owner. The fourth field identifies the file’s group. The fifth field contains the file’s size in bytes. The sixth, seventh, and eighth fields contain the month, day, and time, respectively, that the file was last modified. Finally, the ninth field contains the file name.

For future reference, note that there is often more than one way to do things in awk. At some point, you may want to look back at these examples and see if you can come up with different ways to do the same things shown here:

  • Print every line that is longer than 80 characters:
    awk 'length($0) > 80' data
    

    The sole rule has a relational expression as its pattern and has no action—so it uses the default action, printing the record.

  • Print the length of the longest input line:
    awk '{ if (length($0) > max) max = length($0) }
         END { print max }' data
    

    The code associated with END executes after all input has been read; it’s the other side of the coin to BEGIN.

  • Print the length of the longest line in data:
    expand data | awk '{ if (x < length($0)) x = length($0) }
                       END { print "maximum line length is " x }'
    

    This example differs slightly from the previous one: the input is processed by the expand utility to change TABs into spaces, so the widths compared are actually the right-margin columns, as opposed to the number of input characters on each line.

  • Print every line that has at least one field:
    awk 'NF > 0' data
    

    This is an easy way to delete blank lines from a file (or rather, to create a new file similar to the old file but from which the blank lines have been removed).

  • Print seven random numbers from 0 to 100, inclusive:
    awk 'BEGIN { for (i = 1; i <= 7; i++)
                     print int(101 * rand()) }'
    
  • Print the total number of bytes used by files:
    ls -l files | awk '{ x += $5 }
                       END { print "total bytes: " x }'
    
  • Print the total number of kilobytes used by files:
    ls -l files | awk '{ x += $5 }
       END { print "total K-bytes:", x / 1024 }'
    
  • Print a sorted list of the login names of all users:
    awk -F: '{ print $1 }' /etc/passwd | sort
    
  • Count the lines in a file:
    awk 'END { print NR }' data
    
  • Print the even-numbered lines in the data file:
    awk 'NR % 2 == 0' data
    

    If you used the expression ‘NR % 2 == 1’ instead, the program would print the odd-numbered lines.


1.4 An Example with Two Rules

The awk utility reads the input files one line at a time. For each line, awk tries the patterns of each rule. If several patterns match, then several actions execute in the order in which they appear in the awk program. If no patterns match, then no actions run.

After processing all the rules that match the line (and perhaps there are none), awk reads the next line. (However, see The next Statement and also see The nextfile Statement.) This continues until the program reaches the end of the file. For example, the following awk program contains two rules:

/12/  { print $0 }
/21/  { print $0 }

The first rule has the string ‘12’ as the pattern and ‘print $0’ as the action. The second rule has the string ‘21’ as the pattern and also has ‘print $0’ as the action. Each rule’s action is enclosed in its own pair of braces.

This program prints every line that contains the string ‘12or the string ‘21’. If a line contains both strings, it is printed twice, once by each rule.

This is what happens if we run this program on our two sample data files, mail-list and inventory-shipped:

$ awk '/12/ { print $0 }
>      /21/ { print $0 }' mail-list inventory-shipped
-| Anthony      555-3412     [email protected]   A
-| Camilla      555-2912     [email protected]     R
-| Fabius       555-1234     [email protected]    F
-| Jean-Paul    555-2127     [email protected]     R
-| Jean-Paul    555-2127     [email protected]     R
-| Jan  21  36  64 620
-| Apr  21  70  74 514

Note how the line beginning with ‘Jean-Paul’ in mail-list was printed twice, once for each rule.


1.5 A More Complex Example

Now that we’ve mastered some simple tasks, let’s look at what typical awk programs do. This example shows how awk can be used to summarize, select, and rearrange the output of another utility. It uses features that haven’t been covered yet, so don’t worry if you don’t understand all the details:

ls -l | awk '$6 == "Nov" { sum += $5 }
             END { print sum }'

This command prints the total number of bytes in all the files in the current directory that were last modified in November (of any year).

As a reminder, the output of ‘ls -l gives you a listing of the files in a directory, including each file’s size and the date the file was last modified. The first field contains read-write permissions, the second field contains the number of links to the file, and the third field identifies the file’s owner. The fourth field identifies the file’s group. The fifth field contains the file’s size in bytes. The sixth, seventh, and eighth fields contain the month, day, and time, respectively, that the file was last modified. Finally, the ninth field contains the file name.

The ‘$6 == "Nov"’ in our awk program is an expression that tests whether the sixth field of the output from ‘ls -l matches the string ‘Nov’. Each time a line has the string ‘Nov’ for its sixth field, awk performs the action ‘sum += $5’. This adds the fifth field (the file’s size) to the variable sum. As a result, when awk has finished reading all the input lines, sum is the total of the sizes of the files whose lines matched the pattern. (This works because awk variables are automatically initialized to zero.)

After the last line of output from ls has been processed, the END rule executes and prints the value of sum. In this example, the value of sum is 80600.

These more advanced awk techniques are covered in later sections (see Actions). Before you can move on to more advanced awk programming, you have to know how awk interprets your input and displays your output. By manipulating fields and using print statements, you can produce some very useful and impressive-looking reports.


1.6 awk Statements Versus Lines

Most often, each line in an awk program is a separate statement or separate rule, like this:

awk '/12/  { print $0 }
     /21/  { print $0 }' mail-list inventory-shipped

However, gawk ignores newlines after any of the following symbols and keywords:

,    {    ?    :    ||    &&    do    else

A newline at any other point is considered the end of the statement.9

If you would like to split a single statement into two lines at a point where a newline would terminate it, you can continue it by ending the first line with a backslash character (‘\’). The backslash must be the final character on the line in order to be recognized as a continuation character. A backslash followed by a newline is allowed anywhere in the statement, even in the middle of a string or regular expression. For example:

awk '/This regular expression is too long, so continue it\
 on the next line/ { print $1 }'

We have generally not used backslash continuation in our sample programs. gawk places no limit on the length of a line, so backslash continuation is never strictly necessary; it just makes programs more readable. For this same reason, as well as for clarity, we have kept most statements short in the programs presented throughout the Web page.

Backslash continuation is most useful when your awk program is in a separate source file instead of entered from the command line. You should also note that many awk implementations are more particular about where you may use backslash continuation. For example, they may not allow you to split a string constant using backslash continuation. Thus, for maximum portability of your awk programs, it is best not to split your lines in the middle of a regular expression or a string.

CAUTION: Backslash continuation does not work as described with the C shell. It works for awk programs in files and for one-shot programs, provided you are using a POSIX-compliant shell, such as the Unix Bourne shell or Bash. But the C shell behaves differently! There you must use two backslashes in a row, followed by a newline. Note also that when using the C shell, every newline in your awk program must be escaped with a backslash. To illustrate:

% awk 'BEGIN { \
?   print \\
?       "hello, world" \
? }'
-| hello, world

Here, the ‘%’ and ‘?’ are the C shell’s primary and secondary prompts, analogous to the standard shell’s ‘$’ and ‘>’.

Compare the previous example to how it is done with a POSIX-compliant shell:

$ awk 'BEGIN {
>   print \
>       "hello, world"
> }'
-| hello, world

awk is a line-oriented language. Each rule’s action has to begin on the same line as the pattern. To have the pattern and action on separate lines, you must use backslash continuation; there is no other option.

Another thing to keep in mind is that backslash continuation and comments do not mix. As soon as awk sees the ‘#’ that starts a comment, it ignores everything on the rest of the line. For example:

$ gawk 'BEGIN { print "dont panic" # a friendly \
>                                    BEGIN rule
> }'
error→ gawk: cmd. line:2:                BEGIN rule
error→ gawk: cmd. line:2:                ^ syntax error

In this case, it looks like the backslash would continue the comment onto the next line. However, the backslash-newline combination is never even noticed because it is “hidden” inside the comment. Thus, the BEGIN is noted as a syntax error.

If you’re interested, see https://lists.gnu.org/archive/html/bug-gawk/2022-10/msg00025.html for a source code patch that allows lines to be continued when inside parentheses. This patch was not added to gawk since it would quietly decrease the portability of awk programs.

When awk statements within one rule are short, you might want to put more than one of them on a line. This is accomplished by separating the statements with a semicolon (‘;’). This also applies to the rules themselves. Thus, the program shown at the start of this section could also be written this way:

/12/ { print $0 } ; /21/ { print $0 }

NOTE: The requirement that states that rules on the same line must be separated with a semicolon was not in the original awk language; it was added for consistency with the treatment of statements within an action.


1.7 Other Features of awk

The awk language provides a number of predefined, or built-in, variables that your programs can use to get information from awk. There are other variables your program can set as well to control how awk processes your data.

In addition, awk provides a number of built-in functions for doing common computational and string-related operations. gawk provides built-in functions for working with timestamps, performing bit manipulation, for runtime string translation (internationalization), determining the type of a variable, and array sorting.

As we develop our presentation of the awk language, we will introduce most of the variables and many of the functions. They are described systematically in Predefined Variables and in Built-in Functions.


1.8 When to Use awk

Now that you’ve seen some of what awk can do, you might wonder how awk could be useful for you. By using utility programs, advanced patterns, field separators, arithmetic statements, and other selection criteria, you can produce much more complex output. The awk language is very useful for producing reports from large amounts of raw data, such as summarizing information from the output of other utility programs like ls. (See A More Complex Example.)

Programs written with awk are usually much smaller than they would be in other languages. This makes awk programs easy to compose and use. Often, awk programs can be quickly composed at your keyboard, used once, and thrown away. Because awk programs are interpreted, you can avoid the (usually lengthy) compilation part of the typical edit-compile-test-debug cycle of software development.

Complex programs have been written in awk, including a complete retargetable assembler for eight-bit microprocessors (see Glossary, for more information), and a microcode assembler for a special-purpose Prolog computer. The original awk’s capabilities were strained by tasks of such complexity, but modern versions are more capable.

If you find yourself writing awk scripts of more than, say, a few hundred lines, you might consider using a different programming language. The shell is good at string and pattern matching; in addition, it allows powerful use of the system utilities. Python offers a nice balance between high-level ease of programming and access to system facilities.10


1.9 Summary

  • Programs in awk consist of patternaction pairs.
  • An action without a pattern always runs. The default action for a pattern without one is ‘{ print $0 }’.
  • Use either ‘awk 'program' files’ or ‘awk -f program-file files’ to run awk.
  • You may use the special ‘#!’ header line to create awk programs that are directly executable.
  • Comments in awk programs start with ‘#’ and continue to the end of the same line.
  • Be aware of quoting issues when writing awk programs as part of a larger shell script (or MS-Windows batch file).
  • You may use backslash continuation to continue a source line. Lines are automatically continued after a comma, open brace, question mark, colon, ‘||’, ‘&&’, do, and else.

2 Running awk and gawk

This chapter covers how to run awk, both POSIX-standard and gawk-specific command-line options, and what awk and gawk do with nonoption arguments. It then proceeds to cover how gawk searches for source files, reading standard input along with other files, gawk’s environment variables, gawk’s exit status, using include files, and obsolete and undocumented options and/or features.

Many of the options and features described here are discussed in more detail later in the Web page; feel free to skip over things in this chapter that don’t interest you right now.


2.1 Invoking awk

There are two ways to run awk—with an explicit program or with one or more program files. Here are templates for both of them; items enclosed in […] in these templates are optional:

awk [options] -f progfile [--] fileawk [options] [--] 'program' file

In addition to traditional one-letter POSIX-style options, gawk also supports GNU long options.

It is possible to invoke awk with an empty program:

awk '' datafile1 datafile2

Doing so makes little sense, though; awk exits silently when given an empty program. (d.c.) If --lint has been specified on the command line, gawk issues a warning that the program is empty.


2.2 Command-Line Options

Options begin with a dash and consist of a single character. GNU-style long options consist of two dashes and a keyword. The keyword can be abbreviated, as long as the abbreviation allows the option to be uniquely identified. If the option takes an argument, either the keyword is immediately followed by an equals sign (‘=’) and the argument’s value, or the keyword and the argument’s value are separated by whitespace (spaces or TABs). If a particular option with a value is given more than once, it is (usually) the last value that counts.

Each long option for gawk has a corresponding POSIX-style short option. The long and short options are interchangeable in all contexts. The following list describes options mandated by the POSIX standard:

-F fs
--field-separator fs

Set the FS variable to fs (see Specifying How Fields Are Separated).

-f source-file
--file source-file

Read the awk program source from source-file instead of in the first nonoption argument. This option may be given multiple times; the awk program consists of the concatenation of the contents of each specified source-file.

Files named with -f are treated as if they had ‘@namespace "awk"’ at their beginning. See Changing The Namespace, for more information on this advanced feature.

-v var=val
--assign var=val

Set the variable var to the value val before execution of the program begins. Such variable values are available inside the BEGIN rule (see Other Command-Line Arguments).

The -v option can only set one variable, but it can be used more than once, setting another variable each time, like this: ‘awk -v foo=1 -v bar=2’.

CAUTION: Using -v to set the values of the built-in variables may lead to surprising results. awk will reset the values of those variables as it needs to, possibly ignoring any initial value you may have given.

-W gawk-opt

Provide an implementation-specific option. This is the POSIX convention for providing implementation-specific options. These options also have corresponding GNU-style long options. Note that the long options may be abbreviated, as long as the abbreviations remain unique. The full list of gawk-specific options is provided next.

--

Signal the end of the command-line options. The following arguments are not treated as options even if they begin with ‘-’. This interpretation of -- follows the POSIX argument parsing conventions.

This is useful if you have file names that start with ‘-’, or in shell scripts, if you have file names that will be specified by the user that could start with ‘-’. It is also useful for passing options on to the awk program; see Processing Command-Line Options.

The following list describes gawk-specific options:

-b
--characters-as-bytes

Cause gawk to treat all input data as single-byte characters. In addition, all output written with print or printf is treated as single-byte characters.

Normally, gawk follows the POSIX standard and attempts to process its input data according to the current locale (see Where You Are Makes a Difference). This can often involve converting multibyte characters into wide characters (internally), and can lead to problems or confusion if the input data does not contain valid multibyte characters. This option is an easy way to tell gawk, “Hands off my data!”

-c
--traditional

Specify compatibility mode, in which the GNU extensions to the awk language are disabled, so that gawk behaves just like BWK awk. See Extensions in gawk Not in POSIX awk, which summarizes the extensions. Also see Downward Compatibility and Debugging.

-C
--copyright

Print the short version of the General Public License and then exit.

-d[file]
--dump-variables[=file]

Print a sorted list of global variables, their types, and final values to file. If no file is provided, print this list to a file named awkvars.out in the current directory. No space is allowed between the -d and file, if file is supplied.

Having a list of all global variables is a good way to look for typographical errors in your programs. You would also use this option if you have a large program with a lot of functions, and you want to be sure that your functions don’t inadvertently use global variables that you meant to be local. (This is a particularly easy mistake to make with simple variable names like i, j, etc.)

-D[file]
--debug[=file]

Enable debugging of awk programs (see Introduction to the gawk Debugger). By default, the debugger reads commands interactively from the keyboard (standard input). The optional file argument allows you to specify a file with a list of commands for the debugger to execute noninteractively. No space is allowed between the -D and file, if file is supplied.

-e program-text
--source program-text

Provide program source code in the program-text. This option allows you to mix source code in files with source code that you enter on the command line. This is particularly useful when you have library functions that you want to use from your command-line programs (see The AWKPATH Environment Variable).

Note that gawk treats each string as if it ended with a newline character (even if it doesn’t). This makes building the total program easier.

CAUTION: Prior to version 5.0, there was no requirement that each program-text be a full syntactic unit. I.e., the following worked:

$ gawk -e 'BEGIN { a = 5 ;' -e 'print a }'
-| 5

However, this is no longer true. If you have any scripts that rely upon this feature, you should revise them.

This is because each program-text is treated as if it had ‘@namespace "awk"’ at its beginning. See Changing The Namespace, for more information.

-E file
--exec file

Similar to -f, read awk program text from file. There are two differences from -f:

  • This option terminates option processing; anything else on the command line is passed on directly to the awk program.
  • Command-line variable assignments of the form ‘var=value’ are disallowed.

This option is particularly necessary for World Wide Web CGI applications that pass arguments through the URL; using this option prevents a malicious (or other) user from passing in options, assignments, or awk source code (via -e) to the CGI application.11 This option should be used with ‘#!’ scripts (see Executable awk Programs), like so:

#! /usr/local/bin/gawk -E
awk program here …
-g
--gen-pot

Analyze the source program and generate a GNU gettext portable object template file on standard output for all string constants that have been marked for translation. See Internationalization with gawk, for information about this option.

-h
--help

Print a “usage” message summarizing the short- and long-style options that gawk accepts and then exit.

-i source-file
--include source-file

Read an awk source library from source-file. This option is completely equivalent to using the @include directive inside your program. It is very similar to the -f option, but there are two important differences. First, when -i is used, the program source is not loaded if it has been previously loaded, whereas with -f, gawk always loads the file. Second, because this option is intended to be used with code libraries, gawk does not recognize such files as constituting main program input. Thus, after processing an -i argument, gawk still expects to find the main source code via the -f option or on the command line.

Files named with -i are treated as if they had ‘@namespace "awk"’ at their beginning. See Changing The Namespace, for more information.

-I
--trace

Print the internal byte code names as they are executed when running the program. The trace is printed to standard error. Each “op code” is preceded by a + sign in the output.

-l ext
--load ext

Load a dynamic extension named ext. Extensions are stored as system shared libraries. This option searches for the library using the AWKLIBPATH environment variable. The correct library suffix for your platform will be supplied by default, so it need not be specified in the extension name. The extension initialization routine should be named dl_load(). An alternative is to use the @load keyword inside the program to load a shared library. This advanced feature is described in detail in Writing Extensions for gawk.

-L[value]
--lint[=value]

Warn about constructs that are dubious or nonportable to other awk implementations. No space is allowed between the -L and value, if value is supplied. Some warnings are issued when gawk first reads your program. Others are issued at runtime, as your program executes. The optional argument may be one of the following:

fatal

Cause lint warnings become fatal errors. This may be drastic, but its use will certainly encourage the development of cleaner awk programs.

invalid

Only issue warnings about things that are actually invalid are issued. (This is not fully implemented yet.)

no-ext

Disable warnings about gawk extensions.

Some warnings are only printed once, even if the dubious constructs they warn about occur multiple times in your awk program. Thus, when eliminating problems pointed out by --lint, you should take care to search for all occurrences of each inappropriate construct. As awk programs are usually short, doing so is not burdensome.

-M
--bignum

Select arbitrary-precision arithmetic on numbers. This option has no effect if gawk is not compiled to use the GNU MPFR and MP libraries (see Arithmetic and Arbitrary-Precision Arithmetic with gawk).

As of version 5.2, the arbitrary precision arithmetic features in gawk are “on parole.” The primary maintainer is no longer willing to support this feature, but another member of the development team has stepped up to take it over. As long as this situation remains stable, MPFR will be supported. If it changes, the MPFR support will be removed from gawk.

-n
--non-decimal-data

Enable automatic interpretation of octal and hexadecimal values in input data (see Allowing Nondecimal Input Data).

CAUTION: This option can severely break old programs. Use with care. Also note that this option may disappear in a future version of gawk.

-N
--use-lc-numeric

Force the use of the locale’s decimal point character when parsing numeric input data (see Where You Are Makes a Difference).

-o[file]
--pretty-print[=file]

Enable pretty-printing of awk programs. Implies --no-optimize. By default, the output program is created in a file named awkprof.out (see Profiling Your awk Programs). The optional file argument allows you to specify a different file name for the output. No space is allowed between the -o and file, if file is supplied.

NOTE: In the past, this option would also execute your program. This is no longer the case.

-O
--optimize

Enable gawk’s default optimizations on the internal representation of the program. At the moment, this includes just simple constant folding.

Optimization is enabled by default. This option remains primarily for backwards compatibility. However, it may be used to cancel the effect of an earlier -s option (see later in this list).

-p[file]
--profile[=file]

Enable profiling of awk programs (see Profiling Your awk Programs). Implies --no-optimize. By default, profiles are created in a file named awkprof.out. The optional file argument allows you to specify a different file name for the profile file. No space is allowed between the -p and file, if file is supplied.

The profile contains execution counts for each statement in the program in the left margin, and function call counts for each function.

-P
--posix

Operate in strict POSIX mode. This disables all gawk extensions (just like --traditional) and disables all extensions not allowed by POSIX. See Common Extensions Summary for a summary of the extensions in gawk that are disabled by this option. Also, the following additional restrictions apply:

If you supply both --traditional and --posix on the command line, --posix takes precedence. gawk issues a warning if both options are supplied.

-r
--re-interval

Allow interval expressions (see Regular Expression Operators) in regexps. This is now gawk’s default behavior. Nevertheless, this option remains for backward compatibility.

-s
--no-optimize

Disable gawk’s default optimizations on the internal representation of the program.

-S
--sandbox

Disable the system() function, input redirections with getline, output redirections with print and printf, and dynamic extensions. Also, disallow adding file names to ARGV that were not there when gawk started running. This is particularly useful when you want to run awk scripts from questionable sources and need to make sure the scripts can’t access your system (other than the specified input data files).

-t
--lint-old

Warn about constructs that are not available in the original version of awk from Version 7 Unix (see Major Changes Between V7 and SVR3.1).

-V
--version

Print version information for this particular copy of gawk. This allows you to determine if your copy of gawk is up to date with respect to whatever the Free Software Foundation is currently distributing. It is also useful for bug reports (see Reporting Problems and Bugs).

--

Mark the end of all options. Any command-line arguments following -- are placed in ARGV, even if they start with a minus sign.

In compatibility mode, as long as program text has been supplied, any other options are flagged as invalid with a warning message but are otherwise ignored.

In compatibility mode, as a special case, if the value of fs supplied to the -F option is ‘t’, then FS is set to the TAB character ("\t"). This is true only for --traditional and not for --posix (see Specifying How Fields Are Separated).

The -f option may be used more than once on the command line. If it is, awk reads its program source from all of the named files, as if they had been concatenated together into one big file. This is useful for creating libraries of awk functions. These functions can be written once and then retrieved from a standard place, instead of having to be included in each individual program. The -i option is similar in this regard. (As mentioned in Function Definition Syntax, function names must be unique.)

With standard awk, library functions can still be used, even if the program is entered at the keyboard, by specifying ‘-f /dev/tty’. After typing your program, type Ctrl-d (the end-of-file character) to terminate it. (You may also use ‘-f -’ to read program source from the standard input, but then you will not be able to also use the standard input as a source of data.)

Because it is clumsy using the standard awk mechanisms to mix source file and command-line awk programs, gawk provides the -e option. This does not require you to preempt the standard input for your source code, and it allows you to easily mix command-line and library source code (see The AWKPATH Environment Variable). As with -f, the -e and -i options may also be used multiple times on the command line.

If no -f option (or -e option for gawk) is specified, then awk uses the first nonoption command-line argument as the text of the program source code. Arguments on the command line that follow the program text are entered into the ARGV array; awk does not continue to parse the command line looking for options.

If the environment variable POSIXLY_CORRECT exists, then gawk behaves in strict POSIX mode, exactly as if you had supplied --posix. Many GNU programs look for this environment variable to suppress extensions that conflict with POSIX, but gawk behaves differently: it suppresses all extensions, even those that do not conflict with POSIX, and behaves in strict POSIX mode. If --lint is supplied on the command line and gawk turns on POSIX mode because of POSIXLY_CORRECT, then it issues a warning message indicating that POSIX mode is in effect. You would typically set this variable in your shell’s startup file. For a Bourne-compatible shell (such as Bash), you would add these lines to the .profile file in your home directory:

POSIXLY_CORRECT=true
export POSIXLY_CORRECT

For a C shell-compatible shell,12 you would add this line to the .login file in your home directory:

setenv POSIXLY_CORRECT true

Having POSIXLY_CORRECT set is not recommended for daily use, but it is good for testing the portability of your programs to other environments.


2.3 Other Command-Line Arguments

Any additional arguments on the command line are normally treated as input files to be processed in the order specified. However, an argument that has the form var=value, assigns the value value to the variable var—it does not specify a file at all. (See Assigning Variables on the Command Line.) In the following example, ‘count=1’ is a variable assignment, not a file name:

awk -f program.awk file1 count=1 file2

As a side point, should you really need to have awk process a file named count=1 (or any file whose name looks like a variable assignment), precede the file name with ‘./’, like so:

awk -f program.awk file1 ./count=1 file2

All the command-line arguments are made available to your awk program in the ARGV array (see Predefined Variables). Command-line options and the program text (if present) are omitted from ARGV. All other arguments, including variable assignments, are included. As each element of ARGV is processed, gawk sets ARGIND to the index in ARGV of the current element. (gawk makes the full command line, including program text and options, available in PROCINFO["argv"]; see Built-in Variables That Convey Information.)

Changing ARGC and ARGV in your awk program lets you control how awk processes the input files; this is described in more detail in Using ARGC and ARGV.

The distinction between file name arguments and variable-assignment arguments is made when awk is about to open the next input file. At that point in execution, it checks the file name to see whether it is really a variable assignment; if so, awk sets the variable instead of reading a file.

Therefore, the variables actually receive the given values after all previously specified files have been read. In particular, the values of variables assigned in this fashion are not available inside a BEGIN rule (see The BEGIN and END Special Patterns), because such rules are run before awk begins scanning the argument list.

The variable values given on the command line are processed for escape sequences (see Escape Sequences). (d.c.)

In some very early implementations of awk, when a variable assignment occurred before any file names, the assignment would happen before the BEGIN rule was executed. awk’s behavior was thus inconsistent; some command-line assignments were available inside the BEGIN rule, while others were not. Unfortunately, some applications came to depend upon this “feature.” When awk was changed to be more consistent, the -v option was added to accommodate applications that depended upon the old behavior.

The variable assignment feature is most useful for assigning to variables such as RS, OFS, and ORS, which control input and output formats, before scanning the data files. It is also useful for controlling state if multiple passes are needed over a data file. For example:

awk 'pass == 1  { pass 1 stuff }
     pass == 2  { pass 2 stuff }' pass=1 mydata pass=2 mydata

Given the variable assignment feature, the -F option for setting the value of FS is not strictly necessary. It remains for historical compatibility.

Quoting Shell Variables On The awk Command Line

Small awk programs are often embedded in larger shell scripts, so it’s worthwhile to understand some shell basics. Consider the following:

f=""
awk '{ print("hi") }' $f

In this case, awk reads from standard input instead of trying to open any command line files. To the unwary, this looks like awk is hanging.

However awk doesn’t see an explicit empty string. When a variable expansion is the null string, and it’s not quoted, the shell simply removes it from the command line. To demonstrate:

$ f=""
$ awk 'BEGIN { print ARGC }' $f
-| 1
$ awk 'BEGIN { print ARGC }' "$f"
-| 2

2.4 Naming Standard Input

Often, you may wish to read standard input together with other files. For example, you may wish to read one file, read standard input coming from a pipe, and then read another file.

The way to name the standard input, with all versions of awk, is to use a single, standalone minus sign or dash, ‘-’. For example:

some_command | awk -f myprog.awk file1 - file2

Here, awk first reads file1, then it reads the output of some_command, and finally it reads file2.

You may also use "-" to name standard input when reading files with getline (see Using getline from a File). And, you can even use "-" with the -f option to read program source code from standard input (see Command-Line Options).

In addition, gawk allows you to specify the special file name /dev/stdin, both on the command line and with getline. Some other versions of awk also support this, but it is not standard. (Some operating systems provide a /dev/stdin file in the filesystem; however, gawk always processes this file name itself.)


2.5 The Environment Variables gawk Uses

A number of environment variables influence how gawk behaves.


2.5.1 The AWKPATH Environment Variable

In most awk implementations, you must supply a precise pathname for each program file, unless the file is in the current directory. But with gawk, if the file name supplied to the -f or -i options does not contain a directory separator ‘/’, then gawk searches a list of directories (called the search path) one by one, looking for a file with the specified name.

The search path is a string consisting of directory names separated by colons.13 gawk gets its search path from the AWKPATH environment variable. If that variable does not exist, or if it has an empty value, gawk uses a default path (described shortly).

The search path feature is particularly helpful for building libraries of useful awk functions. The library files can be placed in a standard directory in the default path and then specified on the command line with a short file name. Otherwise, you would have to type the full file name for each file.

By using the -i or -f options, your command-line awk programs can use facilities in awk library files (see A Library of awk Functions). Path searching is not done if gawk is in compatibility mode. This is true for both --traditional and --posix. See Command-Line Options.

If the source code file is not found after the initial search, the path is searched again after adding the suffix ‘.awk’ to the file name.

gawk’s path search mechanism is similar to the shell’s. (See The Bourne-Again SHell manual.) It treats a null entry in the path as indicating the current directory. (A null entry is indicated by starting or ending the path with a colon or by placing two colons next to each other [‘::’].)

NOTE: To include the current directory in the path, either place . as an entry in the path or write a null entry in the path.

Different past versions of gawk would also look explicitly in the current directory, either before or after the path search. As of version 4.1.2, this no longer happens; if you wish to look in the current directory, you must include . either as a separate entry or as a null entry in the search path.

The default value for AWKPATH is ‘.:/usr/local/share/awk’.14 Since . is included at the beginning, gawk searches first in the current directory and then in /usr/local/share/awk. In practice, this means that you will rarely need to change the value of AWKPATH.

See Shell Startup Files, for information on functions that help to manipulate the AWKPATH variable.

gawk places the value of the search path that it used into ENVIRON["AWKPATH"]. This provides access to the actual search path value from within an awk program.

Although you can change ENVIRON["AWKPATH"] within your awk program, this has no effect on the running program’s behavior. This makes sense: the AWKPATH environment variable is used to find the program source files. Once your program is running, all the files have been found, and gawk no longer needs to use AWKPATH.


2.5.2 The AWKLIBPATH Environment Variable

The AWKLIBPATH environment variable is similar to the AWKPATH variable, but it is used to search for loadable extensions (stored as system shared libraries) specified with the -l option rather than for source files. If the extension is not found, the path is searched again after adding the appropriate shared library suffix for the platform. For example, on GNU/Linux systems, the suffix ‘.so’ is used. The search path specified is also used for extensions loaded via the @load keyword (see Loading Dynamic Extensions into Your Program).

If AWKLIBPATH does not exist in the environment, or if it has an empty value, gawk uses a default path; this is typically ‘/usr/local/lib/gawk’, although it can vary depending upon how gawk was built.15

See Shell Startup Files, for information on functions that help to manipulate the AWKLIBPATH variable.

gawk places the value of the search path that it used into ENVIRON["AWKLIBPATH"]. This provides access to the actual search path value from within an awk program.

Although you can change ENVIRON["AWKLIBPATH"] within your awk program, this has no effect on the running program’s behavior. This makes sense: the AWKLIBPATH environment variable is used to find any requested extensions, and they are loaded before the program starts to run. Once your program is running, all the extensions have been found, and gawk no longer needs to use AWKLIBPATH.


2.5.3 Other Environment Variables

A number of other environment variables affect gawk’s behavior, but they are more specialized. Those in the following list are meant to be used by regular users:

GAWK_MSEC_SLEEP

Specifies the interval between connection retries, in milliseconds. On systems that do not support the usleep() system call, the value is rounded up to an integral number of seconds.

GAWK_PERSIST_FILE

Specifies the backing file to use for persistent storage of gawk’s variables and arrays. See Preserving Data Between Runs.

GAWK_READ_TIMEOUT

Specifies the time, in milliseconds, for gawk to wait for input before returning with an error. See Reading Input with a Timeout.

GAWK_SOCK_RETRIES

Controls the number of times gawk attempts to retry a two-way TCP/IP (socket) connection before giving up. See Using gawk for Network Programming. Note that when nonfatal I/O is enabled (see Enabling Nonfatal Output), gawk only tries to open a TCP/IP socket once.

PMA_VERBOSITY

Controls the verbosity of the persistent memory allocator. See Preserving Data Between Runs.

POSIXLY_CORRECT

Causes gawk to switch to POSIX-compatibility mode, disabling all traditional and GNU extensions. See Command-Line Options.

The environment variables in the following list are meant for use by the gawk developers for testing and tuning. They are subject to change. The variables are:

AWKBUFSIZE

This variable only affects gawk on POSIX-compliant systems. With a value of ‘exact’, gawk uses the size of each input file as the size of the memory buffer to allocate for I/O. Otherwise, the value should be a number, and gawk uses that number as the size of the buffer to allocate. (When this variable is not set, gawk uses the smaller of the file’s size and the “default” blocksize, which is usually the filesystem’s I/O blocksize.)

AWK_HASH

If this variable exists with a value of ‘gst’, gawk switches to using the hash function from GNU Smalltalk for managing arrays. With a value of ‘fnv1a’, gawk uses the FNV1-A hash function. These functions may be marginally faster than the standard function.

AWKREADFUNC

If this variable exists, gawk switches to reading source files one line at a time, instead of reading in blocks. This exists for debugging problems on filesystems on non-POSIX operating systems where I/O is performed in records, not in blocks.

GAWK_MSG_SRC

If this variable exists, gawk includes the file name and line number within the gawk source code from which warning and/or fatal messages are generated. Its purpose is to help isolate the source of a message, as there are multiple places that produce the same warning or error message.

GAWK_LOCALE_DIR

Specifies the location of compiled message object files for gawk itself. This is passed to the bindtextdomain() function when gawk starts up.

GAWK_NO_DFA

If this variable exists, gawk does not use the DFA regexp matcher for “does it match” kinds of tests. This can cause gawk to be slower. Its purpose is to help isolate differences between the two regexp matchers that gawk uses internally. (There aren’t supposed to be differences, but occasionally theory and practice don’t coordinate with each other.)

GAWK_STACKSIZE

This specifies the amount by which gawk should grow its internal evaluation stack, when needed.

INT_CHAIN_MAX

This specifies intended maximum number of items gawk will maintain on a hash chain for managing arrays indexed by integers.

STR_CHAIN_MAX

This specifies intended maximum number of items gawk will maintain on a hash chain for managing arrays indexed by strings.

TIDYMEM

If this variable exists, gawk uses the mtrace() library calls from the GNU C library to help track down possible memory leaks. This cannot be used together with the persistent memory allocator.


2.6 gawk’s Exit Status

If the exit statement is used with a value (see The exit Statement), then gawk exits with the numeric value given to it.

Otherwise, if there were no problems during execution, gawk exits with the value of the C constant EXIT_SUCCESS. This is usually zero.

If an error occurs, gawk exits with the value of the C constant EXIT_FAILURE. This is usually one.

If gawk exits because of a fatal error, the exit status is two. On non-POSIX systems, this value may be mapped to EXIT_FAILURE.


2.7 Including Other Files into Your Program

This section describes a feature that is specific to gawk.

The @include keyword can be used to read external awk source files. This gives you the ability to split large awk source files into smaller, more manageable pieces, and also lets you reuse common awk code from various awk scripts. In other words, you can group together awk functions used to carry out specific tasks into external files. These files can be used just like function libraries, using the @include keyword in conjunction with the AWKPATH environment variable. Note that source files may also be included using the -i option.

Let’s see an example. We’ll start with two (trivial) awk scripts, namely test1 and test2. Here is the test1 script:

BEGIN {
    print "This is script test1."
}

and here is test2:

@include "test1"
BEGIN {
    print "This is script test2."
}

Running gawk with test2 produces the following result:

$ gawk -f test2
-| This is script test1.
-| This is script test2.

gawk runs the test2 script, which includes test1 using the @include keyword. So, to include external awk source files, you just use @include followed by the name of the file to be included, enclosed in double quotes.

NOTE: Keep in mind that this is a language construct and the file name cannot be a string variable, but rather just a literal string constant in double quotes.

The files to be included may be nested; e.g., given a third script, namely test3:

@include "test2"
BEGIN {
    print "This is script test3."
}

Running gawk with the test3 script produces the following results:

$ gawk -f test3
-| This is script test1.
-| This is script test2.
-| This is script test3.

The file name can, of course, be a pathname. For example:

@include "../io_funcs"

and:

@include "/usr/awklib/network"

are both valid. The AWKPATH environment variable can be of great value when using @include. The same rules for the use of the AWKPATH variable in command-line file searches (see The AWKPATH Environment Variable) apply to @include also.

This is very helpful in constructing gawk function libraries. If you have a large script with useful, general-purpose awk functions, you can break it down into library files and put those files in a special directory. You can then include those “libraries,” either by using the full pathnames of the files, or by setting the AWKPATH environment variable accordingly and then using @include with just the file part of the full pathname. Of course, you can keep library files in more than one directory; the more complex the working environment is, the more directories you may need to organize the files to be included.

Given the ability to specify multiple -f options, the @include mechanism is not strictly necessary. However, the @include keyword can help you in constructing self-contained gawk programs, thus reducing the need for writing complex and tedious command lines. In particular, @include is very useful for writing CGI scripts to be run from web pages.

The @include directive and the -i/--include command line option are completely equivalent. An included program source is not loaded if it has been previously loaded.

The rules for finding a source file described in The AWKPATH Environment Variable also apply to files loaded with @include.

Finally, files included with @include are treated as if they had ‘@namespace "awk"’ at their beginning. See Changing The Namespace, for more information.


2.8 Loading Dynamic Extensions into Your Program

This section describes a feature that is specific to gawk.

The @load keyword can be used to read external awk extensions (stored as system shared libraries). This allows you to link in compiled code that may offer superior performance and/or give you access to extended capabilities not supported by the awk language. The AWKLIBPATH variable is used to search for the extension. Using @load is completely equivalent to using the -l command-line option.

If the extension is not initially found in AWKLIBPATH, another search is conducted after appending the platform’s default shared library suffix to the file name. For example, on GNU/Linux systems, the suffix ‘.so’ is used:

$ gawk '@load "ordchr"; BEGIN {print chr(65)}'
-| A

This is equivalent to the following example:

$ gawk -lordchr 'BEGIN {print chr(65)}'
-| A

For command-line usage, the -l option is more convenient, but @load is useful for embedding inside an awk source file that requires access to an extension.

Writing Extensions for gawk, describes how to write extensions (in C or C++) that can be loaded with either @load or the -l option. It also describes the ordchr extension.


2.9 Obsolete Options and/or Features

This section describes features and/or command-line options from previous releases of gawk that either are not available in the current version or are still supported but deprecated (meaning that they will not be in a future release).

The arbitrary precision arithmetic feature is deprecated as of gawk version 5.2. Use of -M/--bignum produces a warning message. The feature will be removed in the release of 2024.


2.10 Undocumented Options and Features

Use the Source, Luke!

Obi-Wan

This section intentionally left blank.


2.11 Summary

  • gawk parses arguments on the command line, left to right, to determine if they should be treated as options or as non-option arguments.
  • gawk recognizes several options which control its operation, as described in Command-Line Options. All options begin with ‘-’.
  • Any argument that is not recognized as an option is treated as a non-option argument, even if it begins with ‘-’.
    • However, when an option itself requires an argument, and the option is separated from that argument on the command line by at least one space, the space is ignored, and the argument is considered to be related to the option. Thus, in the invocation, ‘gawk -F x’, the ‘x’ is treated as belonging to the -F option, not as a separate non-option argument.
  • Once gawk finds a non-option argument, it stops looking for options. Therefore, all following arguments are also non-option arguments, even if they resemble recognized options.
  • If no -e or -f options are present, gawk expects the program text to be in the first non-option argument.
  • All non-option arguments, except program text provided in the first non-option argument, are placed in ARGV as explained in Using ARGC and ARGV, and are processed as described in Other Command-Line Arguments. Adjusting ARGC and ARGV affects how awk processes input.
  • The three standard options for all versions of awk are -f, -F, and -v. gawk supplies these and many others, as well as corresponding GNU-style long options.
  • Nonoption command-line arguments are usually treated as file names, unless they have the form ‘var=value’, in which case they are taken as variable assignments to be performed at that point in processing the input.
  • You can use a single minus sign (‘-’) to refer to standard input on the command line. gawk also lets you use the special file name /dev/stdin.
  • gawk pays attention to a number of environment variables. AWKPATH, AWKLIBPATH, and POSIXLY_CORRECT are the most important ones.
  • gawk’s exit status conveys information to the program that invoked it. Use the exit statement from within an awk program to set the exit status.
  • gawk allows you to include other awk source files into your program using the @include statement and/or the -i and -f command-line options.
  • gawk allows you to load additional functions written in C or C++ using the @load statement and/or the -l option. (This advanced feature is described later, in Writing Extensions for gawk.)

3 Regular Expressions

A regular expression, or regexp, is a way of describing a set of strings. Because regular expressions are such a fundamental part of awk programming, their format and use deserve a separate chapter.

A regular expression enclosed in slashes (‘/’) is an awk pattern that matches every input record whose text belongs to that set. The simplest regular expression is a sequence of letters, numbers, or both. Such a regexp matches any string that contains that sequence. Thus, the regexp ‘foo’ matches any string containing ‘foo’. Thus, the pattern /foo/ matches any input record containing the three adjacent characters ‘fooanywhere in the record. Other kinds of regexps let you specify more complicated classes of strings.

Initially, the examples in this chapter are simple. As we explain more about how regular expressions work, we present more complicated instances.


3.1 How to Use Regular Expressions

A regular expression can be used as a pattern by enclosing it in slashes. Then the regular expression is tested against the entire text of each record. (Normally, it only needs to match some part of the text in order to succeed.) For example, the following prints the second field of each record where the string ‘li’ appears anywhere in the record:

$ awk '/li/ { print $2 }' mail-list
-| 555-5553
-| 555-0542
-| 555-6699
-| 555-3430

Regular expressions can also be used in matching expressions. These expressions allow you to specify the string to match against; it need not be the entire current input record. The two operators ‘~’ and ‘!~’ perform regular expression comparisons. Expressions using these operators can be used as patterns, or in if, while, for, and do statements. (See Control Statements in Actions.) For example, the following is true if the expression exp (taken as a string) matches regexp:

exp ~ /regexp/

This example matches, or selects, all input records with the uppercase letter ‘J’ somewhere in the first field:

$ awk '$1 ~ /J/' inventory-shipped
-| Jan  13  25  15 115
-| Jun  31  42  75 492
-| Jul  24  34  67 436
-| Jan  21  36  64 620

So does this:

awk '{ if ($1 ~ /J/) print }' inventory-shipped

This next example is true if the expression exp (taken as a character string) does not match regexp:

exp !~ /regexp/

The following example matches, or selects, all input records whose first field does not contain the uppercase letter ‘J’:

$ awk '$1 !~ /J/' inventory-shipped
-| Feb  15  32  24 226
-| Mar  15  24  34 228
-| Apr  31  52  63 420
-| May  16  34  29 208
…

When a regexp is enclosed in slashes, such as /foo/, we call it a regexp constant, much like 5.27 is a numeric constant and "foo" is a string constant.


3.2 Escape Sequences

Some characters cannot be included literally in string constants ("foo") or regexp constants (/foo/). Instead, they should be represented with escape sequences, which are character sequences beginning with a backslash (‘\’). One use of an escape sequence is to include a double-quote character in a string constant. Because a plain double quote ends the string, you must use ‘\"’ to represent an actual double-quote character as a part of the string. For example:

$ awk 'BEGIN { print "He said \"hi!\" to her." }'
-| He said "hi!" to her.

The backslash character itself is another character that cannot be included normally; you must write ‘\\’ to put one backslash in the string or regexp. Thus, the string whose contents are the two characters ‘"’ and ‘\’ must be written "\"\\".

Other escape sequences represent unprintable characters such as TAB or newline. There is nothing to stop you from entering most unprintable characters directly in a string constant or regexp constant, but they may look ugly.

The following list presents all the escape sequences used in awk and what they represent. Unless noted otherwise, all these escape sequences apply to both string constants and regexp constants:

\\

A literal backslash, ‘\’.

\a

The “alert” character, Ctrl-g, ASCII code 7 (BEL). (This often makes some sort of audible noise.)

\b

Backspace, Ctrl-h, ASCII code 8 (BS).

\f

Formfeed, Ctrl-l, ASCII code 12 (FF).

\n

Newline, Ctrl-j, ASCII code 10 (LF).

\r

Carriage return, Ctrl-m, ASCII code 13 (CR).

\t

Horizontal TAB, Ctrl-i, ASCII code 9 (HT).

\v

Vertical TAB, Ctrl-k, ASCII code 11 (VT).

\nnn

The octal value nnn, where nnn stands for 1 to 3 digits between ‘0’ and ‘7’. For example, the code for the ASCII ESC (escape) character is ‘\033’.

\xhh

The hexadecimal value hh, where hh stands for a sequence of hexadecimal digits (‘0’–‘9’, and either ‘A’–‘F’ or ‘a’–‘f’). A maximum of two digits are allowed after the ‘\x’. Any further hexadecimal digits are treated as simple letters or numbers. (c.e.) (The ‘\x’ escape sequence is not allowed in POSIX awk.)

CAUTION: In ISO C, the escape sequence continues until the first nonhexadecimal digit is seen. For many years, gawk would continue incorporating hexadecimal digits into the value until a non-hexadecimal digit or the end of the string was encountered. However, using more than two hexadecimal digits produced undefined results. As of version 4.2, only two digits are processed.

\/

A literal slash (should be used for regexp constants only). This sequence is used when you want to write a regexp constant that contains a slash (such as /.*:\/home\/[[:alnum:]]+:.*/; the ‘[[:alnum:]]’ notation is discussed in Using Bracket Expressions). Because the regexp is delimited by slashes, you need to escape any slash that is part of the pattern, in order to tell awk to keep processing the rest of the regexp.

\"

A literal double quote (should be used for string constants only). This sequence is used when you want to write a string constant that contains a double quote (such as "He said \"hi!\" to her."). Because the string is delimited by double quotes, you need to escape any quote that is part of the string, in order to tell awk to keep processing the rest of the string.

In gawk, a number of additional two-character sequences that begin with a backslash have special meaning in regexps. See gawk-Specific Regexp Operators.

In a regexp, a backslash before any character that is not in the previous list and not listed in gawk-Specific Regexp Operators means that the next character should be taken literally, even if it would normally be a regexp operator. For example, /a\+b/ matches the three characters ‘a+b’.

For complete portability, do not use a backslash before any character not shown in the previous list or that is not an operator.

Backslash Before Regular Characters

If you place a backslash in a string constant before something that is not one of the characters previously listed, POSIX awk purposely leaves what happens as undefined. There are two choices:

Strip the backslash out

This is what BWK awk and gawk both do. For example, "a\qc" is the same as "aqc". (Because this is such an easy bug both to introduce and to miss, gawk warns you about it.) Consider ‘FS = "[ \t]+\|[ \t]+"’ to use vertical bars surrounded by whitespace as the field separator. There should be two backslashes in the string: ‘FS = "[ \t]+\\|[ \t]+"’.)

Leave the backslash alone

Some other awk implementations do this. In such implementations, typing "a\qc" is the same as typing "a\\qc".

To summarize:

  • The escape sequences in the preceding list are always processed first, for both string constants and regexp constants. This happens very early, as soon as awk reads your program.
  • gawk processes both regexp constants and dynamic regexps (see Using Dynamic Regexps), for the special operators listed in gawk-Specific Regexp Operators.
  • A backslash before any other character means to treat that character literally.
Escape Sequences for Metacharacters

Suppose you use an octal or hexadecimal escape to represent a regexp metacharacter. (See Regular Expression Operators.) Does awk treat the character as a literal character or as a regexp operator?

Historically, such characters were taken literally. (d.c.) However, the POSIX standard indicates that they should be treated as real metacharacters, which is what gawk does. In compatibility mode (see Command-Line Options), gawk treats the characters represented by octal and hexadecimal escape sequences literally when used in regexp constants. Thus, /a\52b/ is equivalent to /a\*b/.


3.3 Regular Expression Operators

You can combine regular expressions with special characters, called regular expression operators or metacharacters, to increase the power and versatility of regular expressions.


3.3.1 Regexp Operators in awk

The escape sequences described earlier in Escape Sequences are valid inside a regexp. They are introduced by a ‘\’ and are recognized and converted into corresponding real characters as the very first step in processing regexps.

Here is a list of metacharacters. All characters that are not escape sequences and that are not listed here stand for themselves:

\

This suppresses the special meaning of a character when matching. For example, ‘\$’ matches the character ‘$’.

^

This matches the beginning of a string. ‘^@chapter’ matches ‘@chapter’ at the beginning of a string, for example, and can be used to identify chapter beginnings in Texinfo source files. The ‘^’ is known as an anchor, because it anchors the pattern to match only at the beginning of the string.

It is important to realize that ‘^’ does not match the beginning of a line (the point right after a ‘\n’ newline character) embedded in a string. The condition is not true in the following example:

if ("line1\nLINE 2" ~ /^L/) …
$

This is similar to ‘^’, but it matches only at the end of a string. For example, ‘p$’ matches a record that ends with a ‘p’. The ‘$’ is an anchor and does not match the end of a line (the point right before a ‘\n’ newline character) embedded in a string. The condition in the following example is not true:

if ("line1\nLINE 2" ~ /1$/) …
. (period)

This matches any single character, including the newline character. For example, ‘.P’ matches any single character followed by a ‘P’ in a string. Using concatenation, we can make a regular expression such as ‘U.A’, which matches any three-character sequence that begins with ‘U’ and ends with ‘A’.

In strict POSIX mode (see Command-Line Options), ‘.’ does not match the NUL character, which is a character with all bits equal to zero. Otherwise, NUL is just another character. Other versions of awk may not be able to match the NUL character.

[]

This is called a bracket expression.16 It matches any one of the characters that are enclosed in the square brackets. For example, ‘[MVX]’ matches any one of the characters ‘M’, ‘V’, or ‘X’ in a string. A full discussion of what can be inside the square brackets of a bracket expression is given in Using Bracket Expressions.

[^]

This is a complemented bracket expression. The first character after the ‘[must be a ‘^’. It matches any characters except those in the square brackets. For example, ‘[^awk]’ matches any character that is not an ‘a’, ‘w’, or ‘k’.

|

This is the alternation operator and it is used to specify alternatives. The ‘|’ has the lowest precedence of all the regular expression operators. For example, ‘^P|[aeiouy]’ matches any string that matches either ‘^P’ or ‘[aeiouy]’. This means it matches any string that starts with ‘P’ or contains (anywhere within it) a lowercase English vowel.

The alternation applies to the largest possible regexps on either side.

()

Parentheses are used for grouping in regular expressions, as in arithmetic. They can be used to concatenate regular expressions containing the alternation operator, ‘|’. For example, ‘@(samp|code)\{[^}]+\}’ matches both ‘@code{foo}’ and ‘@samp{bar}’. (These are Texinfo formatting control sequences. The ‘+’ is explained further on in this list.)

The left or opening parenthesis is always a metacharacter; to match one literally, precede it with a backslash. However, the right or closing parenthesis is only special when paired with a left parenthesis; an unpaired right parenthesis is (silently) treated as a regular character.

*

This symbol means that the preceding regular expression should be repeated as many times as necessary to find a match. For example, ‘ph*’ applies the ‘*’ symbol to the preceding ‘h’ and looks for matches of one ‘p’ followed by any number of ‘h’s. This also matches just ‘p’ if no ‘h’s are present.

There are two subtle points to understand about how ‘*’ works. First, the ‘*’ applies only to the single preceding regular expression component (e.g., in ‘ph*’, it applies just to the ‘h’). To cause ‘*’ to apply to a larger subexpression, use parentheses: ‘(ph)*’ matches ‘ph’, ‘phph’, ‘phphph’, and so on.

Second, ‘*’ finds as many repetitions as possible. If the text to be matched is ‘phhhhhhhhhhhhhhooey’, ‘ph*’ matches all of the ‘h’s.

+

This symbol is similar to ‘*’, except that the preceding expression must be matched at least once. This means that ‘wh+y’ would match ‘why’ and ‘whhy’, but not ‘wy’, whereas ‘wh*y’ would match all three.

?

This symbol is similar to ‘*’, except that the preceding expression can be matched either once or not at all. For example, ‘fe?d’ matches ‘fed’ and ‘fd’, but nothing else.

{n}
{n,}
{n,m}

One or two numbers inside braces denote an interval expression. If there is one number in the braces, the preceding regexp is repeated n times. If there are two numbers separated by a comma, the preceding regexp is repeated n to m times. If there is one number followed by a comma, then the preceding regexp is repeated at least n times:

wh{3}y

Matches ‘whhhy’, but not ‘why’ or ‘whhhhy’.

wh{3,5}y

Matches ‘whhhy’, ‘whhhhy’, or ‘whhhhhy’ only.

wh{2,}y

Matches ‘whhy’, ‘whhhy’, and so on.

In regular expressions, the ‘*’, ‘+’, and ‘?’ operators, as well as the braces ‘{’ and ‘}’, have the highest precedence, followed by concatenation, and finally by ‘|’. As in arithmetic, parentheses can change how operators are grouped.

In POSIX awk and gawk, the ‘*’, ‘+’, and ‘?’ operators stand for themselves when there is nothing in the regexp that precedes them. For example, /+/ matches a literal plus sign. However, many other versions of awk treat such a usage as a syntax error.

What About The Empty Regexp?

We describe here an advanced regexp usage. Feel free to skip it upon first reading.

You can supply an empty regexp constant (‘//’) in all places where a regexp is expected. Is this useful? What does it match?

It is useful. It matches the (invisible) empty string at the start and end of a string of characters, as well as the empty string between characters. This is best illustrated with the gsub() function, which makes global substitutions in a string (see String-Manipulation Functions). Normal usage of gsub() is like so:

$ awk '
> BEGIN {
>     x = "ABC_CBA"
>     gsub(/B/, "bb", x)
>     print x
> }'
-| AbbC_CbbA

We can use gsub() to see where the empty strings are that match the empty regexp:

$ awk '
> BEGIN {
>     x = "ABC"
>     gsub(//, "x", x)
>     print x
> }'
-| xAxBxCx

3.3.2 Some Notes On Interval Expressions

Interval expressions were not traditionally available in awk. They were added as part of the POSIX standard to make awk and egrep consistent with each other.

Initially, because old programs may use ‘{’ and ‘}’ in regexp constants, gawk did not match interval expressions in regexps.

However, beginning with version 4.0, gawk does match interval expressions by default. This is because compatibility with POSIX has become more important to most gawk users than compatibility with old programs.

For programs that use ‘{’ and ‘}’ in regexp constants, it is good practice to always escape them with a backslash. Then the regexp constants are valid and work the way you want them to, using any version of awk.17

When ‘{’ and ‘}’ appear in regexp constants in a way that cannot be interpreted as an interval expression (such as /q{a}/), then they stand for themselves.

As mentioned, interval expressions were not traditionally available in awk. In March of 2019, BWK awk (finally) acquired them. Starting with version 5.2, gawk’s --traditional option no longer disables interval expressions in regular expressions.

POSIX says that interval expressions containing repetition counts greater than 255 produce unspecified results.

In the manual for GNU grep, Paul Eggert notes the following:

Interval expressions may be implemented internally via repetition. For example, ‘^(a|bc){2,4}$’ might be implemented as ‘^(a|bc)(a|bc)((a|bc)(a|bc)?)?$’. A large repetition count may exhaust memory or greatly slow matching. Even small counts can cause problems if cascaded; for example, ‘grep -E ".*{10,}{10,}{10,}{10,}{10,}"’ is likely to overflow a stack. Fortunately, regular expressions like these are typically artificial, and cascaded repetitions do not conform to POSIX so cannot be used in portable programs anyway.

This same caveat applies to gawk.


3.4 Using Bracket Expressions

As mentioned earlier, a bracket expression matches any character among those listed between the opening and closing square brackets.

Within a bracket expression, a range expression consists of two characters separated by a hyphen. It matches any single character that sorts between the two characters, based upon the system’s native character set. For example, ‘[0-9]’ is equivalent to ‘[0123456789]’. (See Regexp Ranges and Locales: A Long Sad Story for an explanation of how the POSIX standard and gawk have changed over time. This is mainly of historical interest.)

With the increasing popularity of the Unicode character standard, there is an additional wrinkle to consider. Octal and hexadecimal escape sequences inside bracket expressions are taken to represent only single-byte characters (characters whose values fit within the range 0–256). To match a range of characters where the endpoints of the range are larger than 256, enter the multibyte encodings of the characters directly.

To include one of the characters ‘\’, ‘]’, ‘-’, or ‘^’ in a bracket expression, put a ‘\’ in front of it. For example:

[d\]]

matches either ‘d’ or ‘]’. Additionally, if you place ‘]’ right after the opening ‘[’, the closing bracket is treated as one of the characters to be matched.

The treatment of ‘\’ in bracket expressions is compatible with other awk implementations and is also mandated by POSIX. The regular expressions in awk are a superset of the POSIX specification for Extended Regular Expressions (EREs). POSIX EREs are based on the regular expressions accepted by the traditional egrep utility.

Character classes are a feature introduced in the POSIX standard. A character class is a special notation for describing lists of characters that have a specific attribute, but the actual characters can vary from country to country and/or from character set to character set. For example, the notion of what is an alphabetic character differs between the United States and France.

A character class is only valid in a regexp inside the brackets of a bracket expression. Character classes consist of ‘[:’, a keyword denoting the class, and ‘:]’. Table 3.1 lists the character classes defined by the POSIX standard.

ClassMeaning
[:alnum:]Alphanumeric characters
[:alpha:]Alphabetic characters
[:blank:]Space and TAB characters
[:cntrl:]Control characters
[:digit:]Numeric characters
[:graph:]Characters that are both printable and visible (a space is printable but not visible, whereas an ‘a’ is both)
[:lower:]Lowercase alphabetic characters
[:print:]Printable characters (characters that are not control characters)
[:punct:]Punctuation characters (characters that are not letters, digits, control characters, or space characters)
[:space:]Space characters (these are: space, TAB, newline, carriage return, formfeed and vertical tab)
[:upper:]Uppercase alphabetic characters
[:xdigit:]Characters that are hexadecimal digits

Table 3.1: POSIX character classes

For example, before the POSIX standard, you had to write /[A-Za-z0-9]/ to match alphanumeric characters. If your character set had other alphabetic characters in it, this would not match them. With the POSIX character classes, you can write /[[:alnum:]]/ to match the alphabetic and numeric characters in your character set.

Some utilities that match regular expressions provide a nonstandard ‘[:ascii:]’ character class; awk does not. However, you can simulate such a construct using ‘[\x00-\x7F]’. This matches all values numerically between zero and 127, which is the defined range of the ASCII character set. Use a complemented character list (‘[^\x00-\x7F]’) to match any single-byte characters that are not in the ASCII range.

NOTE: Some older versions of Unix awk treat [:blank:] like [:space:], incorrectly matching more characters than they should. Caveat Emptor.

Two additional special sequences can appear in bracket expressions. These apply to non-ASCII character sets, which can have single symbols (called collating elements) that are represented with more than one character. They can also have several characters that are equivalent for collating, or sorting, purposes. (For example, in French, a plain “e” and a grave-accented “è” are equivalent.) These sequences are:

Collating symbols

Multicharacter collating elements enclosed between ‘[.’ and ‘.]’. For example, if ‘ch’ is a collating element, then ‘[[.ch.]]’ is a regexp that matches this collating element, whereas ‘[ch]’ is a regexp that matches either ‘c’ or ‘h’.

Equivalence classes

Locale-specific names for a list of characters that are equal. The name is enclosed between ‘[=’ and ‘=]’. For example, the name ‘e’ might be used to represent all of “e,” “ê,” “è,” and “é.” In this case, ‘[[=e=]]’ is a regexp that matches any of ‘e’, ‘ê’, ‘é’, or ‘è’.

These features are very valuable in non-English-speaking locales.

CAUTION: The library functions that gawk uses for regular expression matching currently recognize only POSIX character classes; they do not recognize collating symbols or equivalence classes.

Inside a bracket expression, an opening bracket (‘[’) that does not start a character class, collating element or equivalence class is taken literally. This is also true of ‘.’ and ‘*’.


3.5 How Much Text Matches?

Consider the following:

echo aaaabcd | awk '{ sub(/a+/, "<A>"); print }'

This example uses the sub() function to make a change to the input record. (sub() replaces the first instance of any text matched by the first argument with the string provided as the second argument; see String-Manipulation Functions.) Here, the regexp /a+/ indicates “one or more ‘a’ characters,” and the replacement text is ‘<A>’.

The input contains four ‘a’ characters. awk (and POSIX) regular expressions always match the leftmost, longest sequence of input characters that can match. Thus, all four ‘a’ characters are replaced with ‘<A>’ in this example:

$ echo aaaabcd | awk '{ sub(/a+/, "<A>"); print }'
-| <A>bcd

For simple match/no-match tests, this is not so important. But when doing text matching and substitutions with the match(), sub(), gsub(), and gensub() functions, it is very important. Understanding this principle is also important for regexp-based record and field splitting (see How Input Is Split into Records, and also see Specifying How Fields Are Separated).


3.6 Using Dynamic Regexps

The righthand side of a ‘~’ or ‘!~’ operator need not be a regexp constant (i.e., a string of characters between slashes). It may be any expression. The expression is evaluated and converted to a string if necessary; the contents of the string are then used as the regexp. A regexp computed in this way is called a dynamic regexp or a computed regexp:

BEGIN { digits_regexp = "[[:digit:]]+" }
$0 ~ digits_regexp    { print }

This sets digits_regexp to a regexp that describes one or more digits, and tests whether the input record matches this regexp.

NOTE: When using the ‘~’ and ‘!~’ operators, be aware that there is a difference between a regexp constant enclosed in slashes and a string constant enclosed in double quotes. If you are going to use a string constant, you have to understand that the string is, in essence, scanned twice: the first time when awk reads your program, and the second time when it goes to match the string on the lefthand side of the operator with the pattern on the right. This is true of any string-valued expression (such as digits_regexp, shown in the previous example), not just string constants.

What difference does it make if the string is scanned twice? The answer has to do with escape sequences, and particularly with backslashes. To get a backslash into a regular expression inside a string, you have to type two backslashes.

For example, /\*/ is a regexp constant for a literal ‘*’. Only one backslash is needed. To do the same thing with a string, you have to type "\\*". The first backslash escapes the second one so that the string actually contains the two characters ‘\’ and ‘*’.

Given that you can use both regexp and string constants to describe regular expressions, which should you use? The answer is “regexp constants,” for several reasons:

  • String constants are more complicated to write and more difficult to read. Using regexp constants makes your programs less error-prone. Not understanding the difference between the two kinds of constants is a common source of errors.
  • It is more efficient to use regexp constants. awk can note that you have supplied a regexp and store it internally in a form that makes pattern matching more efficient. When using a string constant, awk must first convert the string into this internal form and then perform the pattern matching.
  • Using regexp constants is better form; it shows clearly that you intend a regexp match.
Using \n in Bracket Expressions of Dynamic Regexps

Some older versions of awk do not allow the newline character to be used inside a bracket expression for a dynamic regexp:

$ awk '$0 ~ "[ \t\n]"'
error→ awk: newline in character class [
error→ ]...
error→  source line number 1
error→  context is
error→        $0 ~ "[ >>>  \t\n]" <<<

But a newline in a regexp constant works with no problem:

$ awk '$0 ~ /[ \t\n]/'
here is a sample line
-| here is a sample line
Ctrl-d

gawk does not have this problem, and it isn’t likely to occur often in practice, but it’s worth noting for future reference.


3.7 gawk-Specific Regexp Operators

GNU software that deals with regular expressions provides a number of additional regexp operators. These operators are described in this section and are specific to gawk; they are not available in other awk implementations. Most of the additional operators deal with word matching. For our purposes, a word is a sequence of one or more letters, digits, or underscores (‘_’):

\s

Matches any space character as defined by the current locale. Think of it as shorthand for ‘[[:space:]].

\S

Matches any character that is not a space, as defined by the current locale. Think of it as shorthand for ‘[^[:space:]].

\w

Matches any word-constituent character—that is, it matches any letter, digit, or underscore. Think of it as shorthand for ‘[[:alnum:]_].

\W

Matches any character that is not word-constituent. Think of it as shorthand for ‘[^[:alnum:]_].

\<

Matches the empty string at the beginning of a word. For example, /\<away/ matches ‘away’ but not ‘stowaway’.

\>

Matches the empty string at the end of a word. For example, /stow\>/ matches ‘stow’ but not ‘stowaway’.

\y

Matches the empty string at either the beginning or the end of a word (i.e., the word boundary). For example, ‘\yballs?\y’ matches either ‘ball’ or ‘balls’, as a separate word.

\B

Matches the empty string that occurs between two word-constituent characters. For example, /\Brat\B/ matches ‘crate’, but it does not match ‘dirty rat’. ‘\B’ is essentially the opposite of ‘\y’. Another way to think of this is that ‘\B’ matches the empty string provided it’s not at the edge of a word.

There are two other operators that work on buffers. In Emacs, a buffer is, naturally, an Emacs buffer. Other GNU programs, including gawk, consider the entire string to match as the buffer. The operators are:

\`

Matches the empty string at the beginning of a buffer (string)

\'

Matches the empty string at the end of a buffer (string)

Because ‘^’ and ‘$’ always work in terms of the beginning and end of strings, these operators don’t add any new capabilities for awk. They are provided for compatibility with other GNU software.

In other GNU software, the word-boundary operator is ‘\b’. However, that conflicts with the awk language’s definition of ‘\b’ as backspace, so gawk uses a different letter. An alternative method would have been to require two backslashes in the GNU operators, but this was deemed too confusing. The current method of using ‘\y’ for the GNU ‘\b’ appears to be the lesser of two evils.

The various command-line options (see Command-Line Options) control how gawk interprets characters in regexps:

No options

In the default case, gawk provides all the facilities of POSIX regexps and the previously described GNU regexp operators. GNU regexp operators described in Regular Expression Operators.

--posix

Match only POSIX regexps; the GNU operators are not special (e.g., ‘\w’ matches a literal ‘w’). Interval expressions are allowed.

--traditional

Match traditional Unix awk regexps. The GNU operators are not special. Because BWK awk supports them, the POSIX character classes (‘[[:alnum:]]’, etc.) are available. So too, interval expressions are allowed. Characters described by octal and hexadecimal escape sequences are treated literally, even if they represent regexp metacharacters.

--re-interval

This option remains for backwards compatibility but no longer has any real effect.


3.8 Case Sensitivity in Matching

Case is normally significant in regular expressions, both when matching ordinary characters (i.e., not metacharacters) and inside bracket expressions. Thus, a ‘w’ in a regular expression matches only a lowercase ‘w’ and not an uppercase ‘W’.

The simplest way to do a case-independent match is to use a bracket expression—for example, ‘[Ww]’. However, this can be cumbersome if you need to use it often, and it can make the regular expressions harder to read. There are two alternatives that you might prefer.

One way to perform a case-insensitive match at a particular point in the program is to convert the data to a single case, using the tolower() or toupper() built-in string functions (which we haven’t discussed yet; see String-Manipulation Functions). For example:

tolower($1) ~ /foo/  { … }

converts the first field to lowercase before matching against it. This works in any POSIX-compliant awk.

Another method, specific to gawk, is to set the variable IGNORECASE to a nonzero value (see Predefined Variables). When IGNORECASE is not zero, all regexp and string operations ignore case.

Changing the value of IGNORECASE dynamically controls the case sensitivity of the program as it runs. Case is significant by default because IGNORECASE (like most variables) is initialized to zero:

x = "aB"
if (x ~ /ab/) …   # this test will fail
IGNORECASE = 1
if (x ~ /ab/) …   # now it will succeed

In general, you cannot use IGNORECASE to make certain rules case insensitive and other rules case sensitive, as there is no straightforward way to set IGNORECASE just for the pattern of a particular rule.18 To do this, use either bracket expressions or tolower(). However, one thing you can do with IGNORECASE only is dynamically turn case sensitivity on or off for all the rules at once.

IGNORECASE can be set on the command line or in a BEGIN rule (see Other Command-Line Arguments; also see Startup and Cleanup Actions). Setting IGNORECASE from the command line is a way to make a program case insensitive without having to edit it.

In multibyte locales, the equivalences between upper- and lowercase characters are tested based on the wide-character values of the locale’s character set. Prior to version 5.0, single-byte characters were tested based on the ISO-8859-1 (ISO Latin-1) character set. However, as of version 5.0, single-byte characters are also tested based on the values of the locale’s character set.19

The value of IGNORECASE has no effect if gawk is in compatibility mode (see Command-Line Options). Case is always significant in compatibility mode.


3.9 Summary

  • Regular expressions describe sets of strings to be matched. In awk, regular expression constants are written enclosed between slashes: //.
  • Regexp constants may be used standalone in patterns and in conditional expressions, or as part of matching expressions using the ‘~’ and ‘!~’ operators.
  • Escape sequences let you represent nonprintable characters and also let you represent regexp metacharacters as literal characters to be matched.
  • Regexp operators provide grouping, alternation, and repetition.
  • Bracket expressions give you a shorthand for specifying sets of characters that can match at a particular point in a regexp. Within bracket expressions, POSIX character classes let you specify certain groups of characters in a locale-independent fashion.
  • Regular expressions match the leftmost longest text in the string being matched. This matters for cases where you need to know the extent of the match, such as for text substitution and when the record separator is a regexp.
  • Matching expressions may use dynamic regexps (i.e., string values treated as regular expressions).
  • gawk’s IGNORECASE variable lets you control the case sensitivity of regexp matching. In other awk versions, use tolower() or toupper().

4 Reading Input Files

In the typical awk program, awk reads all input either from the standard input (by default, this is the keyboard, but often it is a pipe from another command) or from files whose names you specify on the awk command line. If you specify input files, awk reads them in order, processing all the data from one before going on to the next. The name of the current input file can be found in the predefined variable FILENAME (see Predefined Variables).

The input is read in units called records, and is processed by the rules of your program one record at a time. By default, each record is one line. Each record is automatically split into chunks called fields. This makes it more convenient for programs to work on the parts of a record.

On rare occasions, you may need to use the getline command. The getline command is valuable both because it can do explicit input from any number of files, and because the files used with it do not have to be named on the awk command line (see Explicit Input with getline).


4.1 How Input Is Split into Records

awk divides the input for your program into records and fields. It keeps track of the number of records that have been read so far from the current input file. This value is stored in a predefined variable called FNR, which is reset to zero every time a new file is started. Another predefined variable, NR, records the total number of input records read so far from all data files. It starts at zero, but is never automatically reset to zero.

Normally, records are separated by newline characters. You can control how records are separated by assigning values to the built-in variable RS. If RS is any single character, that character separates records. Otherwise (in gawk), RS is treated as a regular expression. This mechanism is explained in greater detail shortly.


4.1.1 Record Splitting with Standard awk

Records are separated by a character called the record separator. By default, the record separator is the newline character. This is why records are, by default, single lines. To use a different character for the record separator, simply assign that character to the predefined variable RS.

Like any other variable, the value of RS can be changed in the awk program with the assignment operator, ‘=’ (see Assignment Expressions). The new record-separator character should be enclosed in quotation marks, which indicate a string constant. Often, the right time to do this is at the beginning of execution, before any input is processed, so that the very first record is read with the proper separator. To do this, use the special BEGIN pattern (see The BEGIN and END Special Patterns). For example:

awk 'BEGIN { RS = "u" }
     { print $0 }' mail-list

changes the value of RS to ‘u’, before reading any input. The new value is a string whose first character is the letter “u”; as a result, records are separated by the letter “u”. Then the input file is read, and the second rule in the awk program (the action with no pattern) prints each record. Because each print statement adds a newline at the end of its output, this awk program copies the input with each ‘u’ changed to a newline. Here are the results of running the program on mail-list:

$ awk 'BEGIN { RS = "u" }
>      { print $0 }' mail-list
-| Amelia       555-5553     amelia.zodiac
-| sq
-| [email protected]    F
-| Anthony      555-3412     anthony.assert
-| [email protected]   A
-| Becky        555-7685     becky.algebrar
-| [email protected]      A
-| Bill         555-1675     [email protected]       A
-| Broderick    555-0542     broderick.aliq
-| [email protected] R
-| Camilla      555-2912     camilla.inf
-| sar
-| [email protected]     R
-| Fabi
-| s       555-1234     fabi
-| s.
-| ndevicesim
-| s@
-| cb.ed
-|     F
-| J
-| lie        555-6699     j
-| lie.perscr
-| [email protected]   F
-| Martin       555-6480     martin.codicib
-| [email protected]    A
-| Sam
-| el       555-3430     sam
-| el.lanceolis@sh
-| .ed
-|         A
-| Jean-Pa
-| l    555-2127     jeanpa
-| l.campanor
-| m@ny
-| .ed
-|      R
-|

Note that the entry for the name ‘Bill’ is not split. In the original data file (see Data files for the Examples), the line looks like this:

Bill         555-1675     [email protected]       A

It contains no ‘u’, so there is no reason to split the record, unlike the others, which each have one or more occurrences of the ‘u’. In fact, this record is treated as part of the previous record; the newline separating them in the output is the original newline in the data file, not the one added by awk when it printed the record!

Another way to change the record separator is on the command line, using the variable-assignment feature (see Other Command-Line Arguments):

awk '{ print $0 }' RS="u" mail-list

This sets RS to ‘u’ before processing mail-list.

Using an alphabetic character such as ‘u’ for the record separator is highly likely to produce strange results. Using an unusual character such as ‘/’ is more likely to produce correct behavior in the majority of cases, but there are no guarantees. The moral is: Know Your Data.

gawk allows RS to be a full regular expression (discussed shortly; see Record Splitting with gawk). Even so, using a regular expression metacharacter, such as ‘.’ as the single character in the value of RS has no special effect: it is treated literally. This is required for backwards compatibility with both Unix awk and with POSIX.

Reaching the end of an input file terminates the current input record, even if the last character in the file is not the character in RS. (d.c.)

The empty string "" (a string without any characters) has a special meaning as the value of RS. It means that records are separated by one or more blank lines and nothing else. See Multiple-Line Records for more details.

If you change the value of RS in the middle of an awk run, the new value is used to delimit subsequent records, but the record currently being processed, as well as records already processed, are not affected.

After the end of the record has been determined, gawk sets the variable RT to the text in the input that matched RS.


4.1.2 Record Splitting with gawk

When using gawk, the value of RS is not limited to a one-character string. If it contains more than one character, it is treated as a regular expression (see Regular Expressions). (c.e.) In general, each record ends at the next string that matches the regular expression; the next record starts at the end of the matching string. This general rule is actually at work in the usual case, where RS contains just a newline: a record ends at the beginning of the next matching string (the next newline in the input), and the following record starts just after the end of this string (at the first character of the following line). The newline, because it matches RS, is not part of either record.

When RS is a single character, RT contains the same single character. However, when RS is a regular expression, RT contains the actual input text that matched the regular expression.

If the input file ends without any text matching RS, gawk sets RT to the null string.

The following example illustrates both of these features. It sets RS equal to a regular expression that matches either a newline or a series of one or more uppercase letters with optional leading and/or trailing whitespace:

$ echo record 1 AAAA record 2 BBBB record 3 |
> gawk 'BEGIN { RS = "\n|( *[[:upper:]]+ *)" }
>             { print "Record =", $0,"and RT = [" RT "]" }'
-| Record = record 1 and RT = [ AAAA ]
-| Record = record 2 and RT = [ BBBB ]
-| Record = record 3 and RT = [
-| ]

The square brackets delineate the contents of RT, letting you see the leading and trailing whitespace. The final value of RT is a newline. See A Simple Stream Editor for a more useful example of RS as a regexp and RT.

If you set RS to a regular expression that allows optional trailing text, such as ‘RS = "abc(XYZ)?"’, it is possible, due to implementation constraints, that gawk may match the leading part of the regular expression, but not the trailing part, particularly if the input text that could match the trailing part is fairly long. gawk attempts to avoid this problem, but currently, there’s no guarantee that this will never happen.

Caveats When Using Regular Expressions for RS

Remember that in awk, the ‘^’ and ‘$’ anchor metacharacters match the beginning and end of a string, and not the beginning and end of a line. As a result, something like ‘RS = "^[[:upper:]]"’ can only match at the beginning of a file. This is because gawk views the input file as one long string that happens to contain newline characters. It is thus best to avoid anchor metacharacters in the value of RS.

Record splitting with regular expressions works differently than regexp matching with the sub(), gsub(), and gensub() (see String-Manipulation Functions). Those functions allow a regexp to match the empty string; record splitting does not. Thus, for example ‘RS = "()"’ does not split records between characters.

The use of RS as a regular expression and the RT variable are gawk extensions; they are not available in compatibility mode (see Command-Line Options). In compatibility mode, only the first character of the value of RS determines the end of the record.

mawk has allowed RS to be a regexp for decades. As of October, 2019, BWK awk also supports it. Neither version supplies RT, however.

RS = "\0" Is Not Portable

There are times when you might want to treat an entire data file as a single record. The only way to make this happen is to give RS a value that you know doesn’t occur in the input file. This is hard to do in a general way, such that a program always works for arbitrary input files.

You might think that for text files, the NUL character, which consists of a character with all bits equal to zero, is a good value to use for RS in this case:

BEGIN { RS = "\0" }  # whole file becomes one record?

gawk in fact accepts this, and uses the NUL character for the record separator. This works for certain special files, such as /proc/environ on GNU/Linux systems, where the NUL character is in fact the record separator. However, this usage is not portable to most other awk implementations.

Almost all other awk implementations20 store strings internally as C-style strings. C strings use the NUL character as the string terminator. In effect, this means that ‘RS = "\0"’ is the same as ‘RS = ""’. (d.c.)

It happens that recent versions of mawk can use the NUL character as a record separator. However, this is a special case: mawk does not allow embedded NUL characters in strings. (This may change in a future version of mawk.)

See Reading a Whole File at Once for an interesting way to read whole files. If you are using gawk, see Reading an Entire File for another option.


4.2 Examining Fields

When awk reads an input record, the record is automatically parsed or separated by the awk utility into chunks called fields. By default, fields are separated by whitespace, like words in a line. Whitespace in awk means any string of one or more spaces, TABs, or newlines; other characters that are considered whitespace by other languages (such as formfeed, vertical tab, etc.) are not considered whitespace by awk.

The purpose of fields is to make it more convenient for you to refer to these pieces of the record. You don’t have to use them—you can operate on the whole record if you want—but fields are what make simple awk programs so powerful.

You use a dollar sign (‘$’) to refer to a field in an awk program, followed by the number of the field you want. Thus, $1 refers to the first field, $2 to the second, and so on. (Unlike in the Unix shells, the field numbers are not limited to single digits. $127 is the 127th field in the record.) For example, suppose the following is a line of input:

This seems like a pretty nice example.

Here the first field, or $1, is ‘This’, the second field, or $2, is ‘seems’, and so on. Note that the last field, $7, is ‘example.’. Because there is no space between the ‘e’ and the ‘.’, the period is considered part of the seventh field.

NF is a predefined variable whose value is the number of fields in the current record. awk automatically updates the value of NF each time it reads a record. No matter how many fields there are, the last field in a record can be represented by $NF. So, $NF is the same as $7, which is ‘example.’. If you try to reference a field beyond the last one (such as $8 when the record has only seven fields), you get the empty string. (If used in a numeric operation, you get zero.)

The use of $0, which looks like a reference to the “zeroth” field, is a special case: it represents the whole input record. Use it when you are not interested in specific fields. Here are some more examples:

$ awk '$1 ~ /li/ { print $0 }' mail-list
-| Amelia       555-5553     [email protected]    F
-| Julie        555-6699     [email protected]   F

This example prints each record in the file mail-list whose first field contains the string ‘li’.

By contrast, the following example looks for ‘li’ in the entire record and prints the first and last fields for each matching input record:

$ awk '/li/ { print $1, $NF }' mail-list
-| Amelia F
-| Broderick R
-| Julie F
-| Samuel A

4.3 Nonconstant Field Numbers

A field number need not be a constant. Any expression in the awk language can be used after a ‘$’ to refer to a field. The value of the expression specifies the field number. If the value is a string, rather than a number, it is converted to a number. Consider this example:

awk '{ print $NR }'

Recall that NR is the number of records read so far: one in the first record, two in the second, and so on. So this example prints the first field of the first record, the second field of the second record, and so on. For the twentieth record, field number 20 is printed; most likely, the record has fewer than 20 fields, so this prints a blank line. Here is another example of using expressions as field numbers:

awk '{ print $(2*2) }' mail-list

awk evaluates the expression ‘(2*2)’ and uses its value as the number of the field to print. The ‘*’ represents multiplication, so the expression ‘2*2’ evaluates to four. The parentheses are used so that the multiplication is done before the ‘$’ operation; they are necessary whenever there is a binary operator21 in the field-number expression. This example, then, prints the type of relationship (the fourth field) for every line of the file mail-list. (All of the awk operators are listed, in order of decreasing precedence, in Operator Precedence (How Operators Nest).)

If the field number you compute is zero, you get the entire record. Thus, ‘$(2-2)’ has the same value as $0. Negative field numbers are not allowed; trying to reference one usually terminates the program. (The POSIX standard does not define what happens when you reference a negative field number. gawk notices this and terminates your program. Other awk implementations may behave differently.)

As mentioned in Examining Fields, awk stores the current record’s number of fields in the built-in variable NF (also see Predefined Variables). Thus, the expression $NF is not a special feature—it is the direct consequence of evaluating NF and using its value as a field number.


4.4 Changing the Contents of a Field

The contents of a field, as seen by awk, can be changed within an awk program; this changes what awk perceives as the current input record. (The actual input is untouched; awk never modifies the input file.) Consider the following example and its output:

$ awk '{ nboxes = $3 ; $3 = $3 - 10
>        print nboxes, $3 }' inventory-shipped
-| 25 15
-| 32 22
-| 24 14
…

The program first saves the original value of field three in the variable nboxes. The ‘-’ sign represents subtraction, so this program reassigns field three, $3, as the original value of field three minus ten: ‘$3 - 10’. (See Arithmetic Operators.) Then it prints the original and new values for field three. (Someone in the warehouse made a consistent mistake while inventorying the red boxes.)

For this to work, the text in $3 must make sense as a number; the string of characters must be converted to a number for the computer to do arithmetic on it. The number resulting from the subtraction is converted back to a string of characters that then becomes field three. See Conversion of Strings and Numbers.

When the value of a field is changed (as perceived by awk), the text of the input record is recalculated to contain the new field where the old one was. In other words, $0 changes to reflect the altered field. Thus, this program prints a copy of the input file, with 10 subtracted from the second field of each line:

$ awk '{ $2 = $2 - 10; print $0 }' inventory-shipped
-| Jan 3 25 15 115
-| Feb 5 32 24 226
-| Mar 5 24 34 228
…

It is also possible to assign contents to fields that are out of range. For example:

$ awk '{ $6 = ($5 + $4 + $3 + $2)
>        print $6 }' inventory-shipped
-| 168
-| 297
-| 301
…

We’ve just created $6, whose value is the sum of fields $2, $3, $4, and $5. The ‘+’ sign represents addition. For the file inventory-shipped, $6 represents the total number of parcels shipped for a particular month.

Creating a new field changes awk’s internal copy of the current input record, which is the value of $0. Thus, if you do ‘print $0’ after adding a field, the record printed includes the new field, with the appropriate number of field separators between it and the previously existing fields.

This recomputation affects and is affected by NF (the number of fields; see Examining Fields). For example, the value of NF is set to the number of the highest field you create. The exact format of $0 is also affected by a feature that has not been discussed yet: the output field separator, OFS, used to separate the fields (see Output Separators).

Note, however, that merely referencing an out-of-range field does not change the value of either $0 or NF. Referencing an out-of-range field only produces an empty string. For example:

if ($(NF+1) != "")
    print "can't happen"
else
    print "everything is normal"

should print ‘everything is normal’, because NF+1 is certain to be out of range. (See The if-else Statement for more information about awk’s if-else statements. See Variable Typing and Comparison Expressions for more information about the ‘!=’ operator.)

It is important to note that making an assignment to an existing field changes the value of $0 but does not change the value of NF, even when you assign the empty string to a field. For example:

$ echo a b c d | awk '{ OFS = ":"; $2 = ""
>                       print $0; print NF }'
-| a::c:d
-| 4

The field is still there; it just has an empty value, delimited by the two colons between ‘a’ and ‘c’. This example shows what happens if you create a new field:

$ echo a b c d | awk '{ OFS = ":"; $2 = ""; $6 = "new"
>                       print $0; print NF }'
-| a::c:d::new
-| 6

The intervening field, $5, is created with an empty value (indicated by the second pair of adjacent colons), and NF is updated with the value six.

Decrementing NF throws away the values of the fields after the new value of NF and recomputes $0. (d.c.) Here is an example:

$ echo a b c d e f | awk '{ print "NF =", NF;
>                           NF = 3; print $0 }'
-| NF = 6
-| a b c

CAUTION: Some versions of awk don’t rebuild $0 when NF is decremented. Until August, 2018, this included BWK awk; fortunately his version now handles this correctly.

Finally, there are times when it is convenient to force awk to rebuild the entire record, using the current values of the fields and OFS. To do this, use the seemingly innocuous assignment:

$1 = $1   # force record to be reconstituted
print $0  # or whatever else with $0

This forces awk to rebuild the record. It does help to add a comment, as we’ve shown here.

There is a flip side to the relationship between $0 and the fields. Any assignment to $0 causes the record to be reparsed into fields using the current value of FS. This also applies to any built-in function that updates $0, such as sub() and gsub() (see String-Manipulation Functions).

Understanding $0

It is important to remember that $0 is the full record, exactly as it was read from the input. This includes any leading or trailing whitespace, and the exact whitespace (or other characters) that separates the fields.

It is a common error to try to change the field separators in a record simply by setting FS and OFS, and then expecting a plain ‘print’ or ‘print $0’ to print the modified record.

But this does not work, because nothing was done to change the record itself. Instead, you must force the record to be rebuilt, typically with a statement such as ‘$1 = $1’, as described earlier.


4.5 Specifying How Fields Are Separated

The field separator, which is either a single character or a regular expression, controls the way awk splits an input record into fields. awk scans the input record for character sequences that match the separator; the fields themselves are the text between the matches.

In the examples that follow, we use the bullet symbol (•) to represent spaces in the output. If the field separator is ‘oo’, then the following line:

moo goo gai pan

is split into three fields: ‘m’, ‘•g’, and ‘•gai•pan’. Note the leading spaces in the values of the second and third fields.

The field separator is represented by the predefined variable FS. Shell programmers take note: awk does not use the name IFS that is used by the POSIX-compliant shells (such as the Unix Bourne shell, sh, or Bash).

The value of FS can be changed in the awk program with the assignment operator, ‘=’ (see Assignment Expressions). Often, the right time to do this is at the beginning of execution before any input has been processed, so that the very first record is read with the proper separator. To do this, use the special BEGIN pattern (see The BEGIN and END Special Patterns). For example, here we set the value of FS to the string ",":

awk 'BEGIN { FS = "," } ; { print $2 }'

Given the input line:

John Q. Smith, 29 Oak St., Walamazoo, MI 42139

this awk program extracts and prints the string ‘•29•Oak•St.’.

Sometimes the input data contains separator characters that don’t separate fields the way you thought they would. For instance, the person’s name in the example we just used might have a title or suffix attached, such as:

John Q. Smith, LXIX, 29 Oak St., Walamazoo, MI 42139

The same program would extract ‘•LXIX’ instead of ‘•29•Oak•St.’. If you were expecting the program to print the address, you would be surprised. The moral is to choose your data layout and separator characters carefully to prevent such problems. (If the data is not in a form that is easy to process, perhaps you can massage it first with a separate awk program.)


4.5.1 Whitespace Normally Separates Fields

Fields are normally separated by whitespace sequences (spaces, TABs, and newlines), not by single spaces. Two spaces in a row do not delimit an empty field. The default value of the field separator FS is a string containing a single space, " ". If awk interpreted this value in the usual way, each space character would separate fields, so two spaces in a row would make an empty field between them. The reason this does not happen is that a single space as the value of FS is a special case—it is taken to specify the default manner of delimiting fields.

If FS is any other single character, such as ",", then each occurrence of that character separates two fields. Two consecutive occurrences delimit an empty field. If the character occurs at the beginning or the end of the line, that too delimits an empty field. The space character is the only single character that does not follow these rules.


4.5.2 Using Regular Expressions to Separate Fields

The previous subsection discussed the use of single characters or simple strings as the value of FS. More generally, the value of FS may be a string containing any regular expression. In this case, each match in the record for the regular expression separates fields. For example, the assignment:

FS = ", \t"

makes every area of an input line that consists of a comma followed by a space and a TAB into a field separator.

For a less trivial example of a regular expression, try using single spaces to separate fields the way single commas are used. FS can be set to "[ ]" (left bracket, space, right bracket). This regular expression matches a single space and nothing else (see Regular Expressions).

There is an important difference between the two cases of ‘FS = " "’ (a single space) and ‘FS = "[ \t\n]+"’ (a regular expression matching one or more spaces, TABs, or newlines). For both values of FS, fields are separated by runs (multiple adjacent occurrences) of spaces, TABs, and/or newlines. However, when the value of FS is " ", awk first strips leading and trailing whitespace from the record and then decides where the fields are. For example, the following pipeline prints ‘b’:

$ echo ' a b c d ' | awk '{ print $2 }'
-| b

However, this pipeline prints ‘a’ (note the extra spaces around each letter):

$ echo ' a  b  c  d ' | awk 'BEGIN { FS = "[ \t\n]+" }
>                                  { print $2 }'
-| a

In this case, the first field is null, or empty.

The stripping of leading and trailing whitespace also comes into play whenever $0 is recomputed. For instance, study this pipeline:

$ echo '   a b c d' | awk '{ print; $2 = $2; print }'
-|    a b c d
-| a b c d

The first print statement prints the record as it was read, with leading whitespace intact. The assignment to $2 rebuilds $0 by concatenating $1 through $NF together, separated by the value of OFS (which is a space by default). Because the leading whitespace was ignored when finding $1, it is not part of the new $0. Finally, the last print statement prints the new $0.

There is an additional subtlety to be aware of when using regular expressions for field splitting. It is not well specified in the POSIX standard, or anywhere else, what ‘^’ means when splitting fields. Does the ‘^’ match only at the beginning of the entire record? Or is each field separator a new string? It turns out that different awk versions answer this question differently, and you should not rely on any specific behavior in your programs. (d.c.)

As a point of information, BWK awk allows ‘^’ to match only at the beginning of the record. gawk also works this way. For example:

$ echo 'xxAA  xxBxx  C' |
> gawk -F '(^x+)|( +)' '{ for (i = 1; i <= NF; i++)
>                             printf "-->%s<--\n", $i }'
-| --><--
-| -->AA<--
-| -->xxBxx<--
-| -->C<--

Finally, field splitting with regular expressions works differently than regexp matching with the sub(), gsub(), and gensub() (see String-Manipulation Functions). Those functions allow a regexp to match the empty string; field splitting does not. Thus, for example ‘FS = "()"’ does not split fields between characters.


4.5.3 Making Each Character a Separate Field

There are times when you may want to examine each character of a record separately. This can be done in gawk by simply assigning the null string ("") to FS. (c.e.) In this case, each individual character in the record becomes a separate field. For example:

$ echo a b | gawk 'BEGIN { FS = "" }
>                  {
>                      for (i = 1; i <= NF; i = i + 1)
>                          print "Field", i, "is", $i
>                  }'
-| Field 1 is a
-| Field 2 is
-| Field 3 is b

Traditionally, the behavior of FS equal to "" was not defined. In this case, most versions of Unix awk simply treat the entire record as only having one field. (d.c.) In compatibility mode (see Command-Line Options), if FS is the null string, then gawk also behaves this way.


4.5.4 Setting FS from the Command Line

FS can be set on the command line. Use the -F option to do so. For example:

awk -F, 'program' input-files

sets FS to the ‘,’ character. Notice that the option uses an uppercase ‘F’ instead of a lowercase ‘f’. The latter option (-f) specifies a file containing an awk program.

The value used for the argument to -F is processed in exactly the same way as assignments to the predefined variable FS. Any special characters in the field separator must be escaped appropriately. For example, to use a ‘\’ as the field separator on the command line, you would have to type:

# same as FS = "\\"
awk -F\\\\ '…' files …

Because ‘\’ is used for quoting in the shell, awk sees ‘-F\\’. Then awk processes the ‘\\’ for escape characters (see Escape Sequences), finally yielding a single ‘\’ to use for the field separator.

As a special case, in compatibility mode (see Command-Line Options), if the argument to -F is ‘t’, then FS is set to the TAB character. If you type ‘-F\t’ at the shell, without any quotes, the ‘\’ gets deleted, so awk figures that you really want your fields to be separated with TABs and not ‘t’s. Use ‘-v FS="t"’ or ‘-F"[t]"’ on the command line if you really do want to separate your fields with ‘t’s. Use ‘-F '\t'’ when not in compatibility mode to specify that TABs separate fields.

As an example, let’s use an awk program file called edu.awk that contains the pattern /edu/ and the action ‘print $1’:

/edu/   { print $1 }

Let’s also set FS to be the ‘-’ character and run the program on the file mail-list. The following command prints a list of the names of the people that work at or attend a university, and the first three digits of their phone numbers:

$ awk -F- -f edu.awk mail-list
-| Fabius       555
-| Samuel       555
-| Jean

Note the third line of output. The third line in the original file looked like this:

Jean-Paul    555-2127     [email protected]     R

The ‘-’ as part of the person’s name was used as the field separator, instead of the ‘-’ in the phone number that was originally intended. This demonstrates why you have to be careful in choosing your field and record separators.

Perhaps the most common use of a single character as the field separator occurs when processing the Unix system password file. On many Unix systems, each user has a separate entry in the system password file, with one line per user. The information in these lines is separated by colons. The first field is the user’s login name and the second is the user’s encrypted or shadow password. (A shadow password is indicated by the presence of a single ‘x’ in the second field.) A password file entry might look like this:

arnold:x:2076:10:Arnold Robbins:/home/arnold:/bin/bash

The following program searches the system password file and prints the entries for users whose full name is not indicated:

awk -F: '$5 == ""' /etc/passwd

4.5.5 Making the Full Line Be a Single Field

Occasionally, it’s useful to treat the whole input line as a single field. This can be done easily and portably simply by setting FS to "\n" (a newline):22

awk -F'\n' 'program' files …

When you do this, $1 is the same as $0.

Changing FS Does Not Affect the Fields

According to the POSIX standard, awk is supposed to behave as if each record is split into fields at the time it is read. In particular, this means that if you change the value of FS after a record is read, the values of the fields (i.e., how they were split) should reflect the old value of FS, not the new one.

However, many older implementations of awk do not work this way. Instead, they defer splitting the fields until a field is actually referenced. The fields are split using the current value of FS! (d.c.) This behavior can be difficult to diagnose. The following example illustrates the difference between the two methods:

sed 1q /etc/passwd | awk '{ FS = ":" ; print $1 }'

which usually prints:

root

on an incorrect implementation of awk, while gawk prints the full first line of the file, something like:

root:x:0:0:Root:/:

(The sed23 command prints just the first line of /etc/passwd.)


4.5.6 Field-Splitting Summary

It is important to remember that when you assign a string constant as the value of FS, it undergoes normal awk string processing. For example, with Unix awk and gawk, the assignment ‘FS = "\.."’ assigns the character string ".." to FS (the backslash is stripped). This creates a regexp meaning “fields are separated by occurrences of any two characters.” If instead you want fields to be separated by a literal period followed by any single character, use ‘FS = "\\.."’.

The following list summarizes how fields are split, based on the value of FS (‘==’ means “is equal to”):

FS == " "

Fields are separated by runs of whitespace. Leading and trailing whitespace are ignored. This is the default.

FS == any other single character

Fields are separated by each occurrence of the character. Multiple successive occurrences delimit empty fields, as do leading and trailing occurrences. The character can even be a regexp metacharacter; it does not need to be escaped.

FS == regexp

Fields are separated by occurrences of characters that match regexp. Leading and trailing matches of regexp delimit empty fields.

FS == ""

Each individual character in the record becomes a separate field. (This is a common extension; it is not specified by the POSIX standard.)

FS and IGNORECASE

The IGNORECASE variable (see Built-in Variables That Control awk) affects field splitting only when the value of FS is a regexp. It has no effect when FS is a single character, even if that character is a letter. Thus, in the following code:

FS = "c"
IGNORECASE = 1
$0 = "aCa"
print $1

The output is ‘aCa’. If you really want to split fields on an alphabetic character while ignoring case, use a regexp that will do it for you (e.g., ‘FS = "[c]"’). In this case, IGNORECASE will take effect.


4.6 Reading Fixed-Width Data

This section discusses an advanced feature of gawk. If you are a novice awk user, you might want to skip it on the first reading.

gawk provides a facility for dealing with fixed-width fields with no distinctive field separator. We discuss this feature in the following subsections.


4.6.1 Processing Fixed-Width Data

An example of fixed-width data would be the input for old Fortran programs where numbers are run together, or the output of programs that did not anticipate the use of their output as input for other programs.

An example of the latter is a table where all the columns are lined up by the use of a variable number of spaces and empty fields are just spaces. Clearly, awk’s normal field splitting based on FS does not work well in this case. Although a portable awk program can use a series of substr() calls on $0 (see String-Manipulation Functions), this is awkward and inefficient for a large number of fields.

The splitting of an input record into fixed-width fields is specified by assigning a string containing space-separated numbers to the built-in variable FIELDWIDTHS. Each number specifies the width of the field, including columns between fields. If you want to ignore the columns between fields, you can specify the width as a separate field that is subsequently ignored. It is a fatal error to supply a field width that has a negative value.

The following data is the output of the Unix w utility. It is useful to illustrate the use of FIELDWIDTHS:

 10:06pm  up 21 days, 14:04,  23 users
User     tty       login  idle   JCPU   PCPU  what
hzuo     ttyV0     8:58pm            9      5  vi p24.tex
hzang    ttyV3     6:37pm    50                -csh
eklye    ttyV5     9:53pm            7      1  em thes.tex
dportein ttyV6     8:17pm  1:47                -csh
gierd    ttyD3    10:00pm     1                elm
dave     ttyD4     9:47pm            4      4  w
brent    ttyp0    26Jun91  4:46  26:46   4:41  bash
dave     ttyq4    26Jun9115days     46     46  wnewmail

The following program takes this input, converts the idle time to number of seconds, and prints out the first two fields and the calculated idle time:

BEGIN  { FIELDWIDTHS = "9 6 10 6 7 7 35" }
NR > 2 {
    idle = $4
    sub(/^ +/, "", idle)   # strip leading spaces
    if (idle == "")
        idle = 0
    if (idle ~ /:/) {      # hh:mm
        split(idle, t, ":")
        idle = t[1] * 60 + t[2]
    }
    if (idle ~ /days/)
        idle *= 24 * 60 * 60
    print $1, $2, idle
}

NOTE: The preceding program uses a number of awk features that haven’t been introduced yet.

Running the program on the data produces the following results:

hzuo      ttyV0  0
hzang     ttyV3  50
eklye     ttyV5  0
dportein  ttyV6  107
gierd     ttyD3  1
dave      ttyD4  0
brent     ttyp0  286
dave      ttyq4  1296000

Another (possibly more practical) example of fixed-width input data is the input from a deck of balloting cards. In some parts of the United States, voters mark their choices by punching holes in computer cards. These cards are then processed to count the votes for any particular candidate or on any particular issue. Because a voter may choose not to vote on some issue, any column on the card may be empty. An awk program for processing such data could use the FIELDWIDTHS feature to simplify reading the data. (Of course, getting gawk to run on a system with card readers is another story!)


4.6.2 Skipping Intervening Fields

Starting in version 4.2, each field width may optionally be preceded by a colon-separated value specifying the number of characters to skip before the field starts. Thus, the preceding program could be rewritten to specify FIELDWIDTHS like so:

BEGIN  { FIELDWIDTHS = "8 1:5 4:7 6 1:6 1:6 2:33" }

This strips away some of the white space separating the fields. With such a change, the program produces the following results:

hzang    ttyV3 50
eklye    ttyV5 0
dportein ttyV6 107
gierd    ttyD3 1
dave     ttyD4 0
brent    ttyp0 286
dave     ttyq4 1296000

4.6.3 Capturing Optional Trailing Data

There are times when fixed-width data may be followed by additional data that has no fixed length. Such data may or may not be present, but if it is, it should be possible to get at it from an awk program.

Starting with version 4.2, in order to provide a way to say “anything else in the record after the defined fields,” gawk allows you to add a final ‘*’ character to the value of FIELDWIDTHS. There can only be one such character, and it must be the final non-whitespace character in FIELDWIDTHS. For example:

$ cat fw.awk                         Show the program
-| BEGIN { FIELDWIDTHS = "2 2 *" }
-| { print NF, $1, $2, $3 }
$ cat fw.in                          Show sample input
-| 1234abcdefghi
$ gawk -f fw.awk fw.in               Run the program
-| 3 12 34 abcdefghi

4.6.4 Field Values With Fixed-Width Data

So far, so good. But what happens if there isn’t as much data as there should be based on the contents of FIELDWIDTHS? Or, what happens if there is more data than expected?

For many years, what happens in these cases was not well defined. Starting with version 4.2, the rules are as follows:

Enough data for some fields

For example, if FIELDWIDTHS is set to "2 3 4" and the input record is ‘aabbb’. In this case, NF is set to two.

Not enough data for a field

For example, if FIELDWIDTHS is set to "2 3 4" and the input record is ‘aab’. In this case, NF is set to two and $2 has the value "b". The idea is that even though there aren’t as many characters as were expected, there are some, so the data should be made available to the program.

Too much data

For example, if FIELDWIDTHS is set to "2 3 4" and the input record is ‘aabbbccccddd’. In this case, NF is set to three and the extra characters (‘ddd’) are ignored. If you want gawk to capture the extra characters, supply a final ‘*’ in the value of FIELDWIDTHS.

Too much data, but with ‘*’ supplied

For example, if FIELDWIDTHS is set to "2 3 4 *" and the input record is ‘aabbbccccddd’. In this case, NF is set to four, and $4 has the value "ddd".


4.7 Defining Fields by Content

This section discusses an advanced feature of gawk. If you are a novice awk user, you might want to skip it on the first reading.

Normally, when using FS, gawk defines the fields as the parts of the record that occur in between each field separator. In other words, FS defines what a field is not, instead of what a field is. However, there are times when you really want to define the fields by what they are, and not by what they are not.

The most notorious such case is so-called comma-separated values (CSV) data. Many spreadsheet programs, for example, can export their data into text files, where each record is terminated with a newline, and fields are separated by commas. If commas only separated the data, there wouldn’t be an issue. The problem comes when one of the fields contains an embedded comma. In such cases, most programs embed the field in double quotes.24 So, we might have data like this:

Robbins,Arnold,"1234 A Pretty Street, NE",MyTown,MyState,12345-6789,USA

The FPAT variable offers a solution for cases like this. The value of FPAT should be a string that provides a regular expression. This regular expression describes the contents of each field.

In the case of CSV data as presented here, each field is either “anything that is not a comma,” or “a double quote, anything that is not a double quote, and a closing double quote.” (There are more complicated definitions of CSV data, treated shortly.) If written as a regular expression constant (see Regular Expressions), we would have /([^,]+)|("[^"]+")/. Writing this as a string requires us to escape the double quotes, leading to:

FPAT = "([^,]+)|(\"[^\"]+\")"

Putting this to use, here is a simple program to parse the data:

BEGIN {
    FPAT = "([^,]+)|(\"[^\"]+\")"
}
{
    print "NF = ", NF
    for (i = 1; i <= NF; i++) {
        printf("$%d = <%s>\n", i, $i)
    }
}

When run, we get the following:

$ gawk -f simple-csv.awk addresses.csv
NF =  7
$1 = <Robbins>
$2 = <Arnold>
$3 = <"1234 A Pretty Street, NE">
$4 = <MyTown>
$5 = <MyState>
$6 = <12345-6789>
$7 = <USA>

Note the embedded comma in the value of $3.

A straightforward improvement when processing CSV data of this sort would be to remove the quotes when they occur, with something like this:

if (substr($i, 1, 1) == "\"") {
    len = length($i)
    $i = substr($i, 2, len - 2)    # Get text within the two quotes
}

NOTE: Some programs export CSV data that contains embedded newlines between the double quotes. gawk provides no way to deal with this. Even though a formal specification for CSV data exists, there isn’t much more to be done; the FPAT mechanism provides an elegant solution for the majority of cases, and the gawk developers are satisfied with that.

As written, the regexp used for FPAT requires that each field contain at least one character. A straightforward modification (changing the first ‘+’ to ‘*’) allows fields to be empty:

FPAT = "([^,]*)|(\"[^\"]+\")"

As with FS, the IGNORECASE variable (see Built-in Variables That Control awk) affects field splitting with FPAT.

Assigning a value to FPAT overrides field splitting with FS and with FIELDWIDTHS.

Finally, the patsplit() function makes the same functionality available for splitting regular strings (see String-Manipulation Functions).


4.7.1 More on CSV Files

Manuel Collado notes that in addition to commas, a CSV field can also contains quotes, that have to be escaped by doubling them. The previously described regexps fail to accept quoted fields with both commas and quotes inside. He suggests that the simplest FPAT expression that recognizes this kind of fields is /([^,]*)|("([^"]|"")+")/. He provides the following input data to test these variants:

p,"q,r",s
p,"q""r",s
p,"q,""r",s
p,"",s
p,,s

And here is his test program:

BEGIN {
     fp[0] = "([^,]+)|(\"[^\"]+\")"
     fp[1] = "([^,]*)|(\"[^\"]+\")"
     fp[2] = "([^,]*)|(\"([^\"]|\"\")+\")"
     FPAT = fp[fpat+0]
}
{
     print "<" $0 ">"
     printf("NF = %s ", NF)
     for (i = 1; i <= NF; i++) {
         printf("<%s>", $i)
     }
     print ""
}

When run on the third variant, it produces:

$ gawk -v fpat=2 -f test-csv.awk sample.csv
-| <p,"q,r",s>
-| NF = 3 <p><"q,r"><s>
-| <p,"q""r",s>
-| NF = 3 <p><"q""r"><s>
-| <p,"q,""r",s>
-| NF = 3 <p><"q,""r"><s>
-| <p,"",s>
-| NF = 3 <p><""><s>
-| <p,,s>
-| NF = 3 <p><><s>

In general, using FPAT to do your own CSV parsing is like having a bed with a blanket that’s not quite big enough. There’s always a corner that isn’t covered. We recommend, instead, that you use Manuel Collado’s CSVMODE library for gawk.


4.7.2 FS Versus FPAT: A Subtle Difference

As we discussed earlier, FS describes the data between fields (“what fields are not”) and FPAT describes the fields themselves (“what fields are”). This leads to a subtle difference in how fields are found when using regexps as the value for FS or FPAT.

In order to distinguish one field from another, there must be a non-empty separator between each field. This makes intuitive sense—otherwise one could not distinguish fields from separators.

Thus, regular expression matching as done when splitting fields with FS is not allowed to match the null string; it must always match at least one character, in order to be able to proceed through the entire record.

On the other hand, regular expression matching with FPAT can match the null string, and the non-matching intervening characters function as the separators.

This same difference is reflected in how matching is done with the split() and patsplit() functions (see String-Manipulation Functions).


4.8 Checking How gawk Is Splitting Records

As we’ve seen, gawk provides three independent methods to split input records into fields. The mechanism used is based on which of the three variables—FS, FIELDWIDTHS, or FPAT—was last assigned to. In addition, an API input parser may choose to override the record parsing mechanism; please refer to Customized Input Parsers for further information about this feature.

To restore normal field splitting after using FIELDWIDTHS and/or FPAT, simply assign a value to FS. You can use ‘FS = FS’ to do this, without having to know the current value of FS.

In order to tell which kind of field splitting is in effect, use PROCINFO["FS"] (see Built-in Variables That Convey Information). The value is "FS" if regular field splitting is being used, "FIELDWIDTHS" if fixed-width field splitting is being used, or "FPAT" if content-based field splitting is being used:

if (PROCINFO["FS"] == "FS")
    regular field splitting …
else if (PROCINFO["FS"] == "FIELDWIDTHS")
    fixed-width field splitting …
else if (PROCINFO["FS"] == "FPAT")
    content-based field splitting …
else
    API input parser field splitting(advanced feature)

This information is useful when writing a function that needs to temporarily change FS or FIELDWIDTHS, read some records, and then restore the original settings (see Reading the User Database for an example of such a function).


4.9 Multiple-Line Records

In some databases, a single line cannot conveniently hold all the information in one entry. In such cases, you can use multiline records. The first step in doing this is to choose your data format.

One technique is to use an unusual character or string to separate records. For example, you could use the formfeed character (written ‘\f’ in awk, as in C) to separate them, making each record a page of the file. To do this, just set the variable RS to "\f" (a string containing the formfeed character). Any other character could equally well be used, as long as it won’t be part of the data in a record.

Another technique is to have blank lines separate records. By a special dispensation, an empty string as the value of RS indicates that records are separated by one or more blank lines. When RS is set to the empty string, each record always ends at the first blank line encountered. The next record doesn’t start until the first nonblank line that follows. No matter how many blank lines appear in a row, they all act as one record separator. (Blank lines must be completely empty; lines that contain only whitespace do not count.)

You can achieve the same effect as ‘RS = ""’ by assigning the string "\n\n+" to RS. This regexp matches the newline at the end of the record and one or more blank lines after the record. In addition, a regular expression always matches the longest possible sequence when there is a choice (see How Much Text Matches?). So, the next record doesn’t start until the first nonblank line that follows—no matter how many blank lines appear in a row, they are considered one record separator.

However, there is an important difference between ‘RS = ""’ and ‘RS = "\n\n+"’. In the first case, leading newlines in the input data file are ignored, and if a file ends without extra blank lines after the last record, the final newline is removed from the record. In the second case, this special processing is not done. (d.c.)

Now that the input is separated into records, the second step is to separate the fields in the records. One way to do this is to divide each of the lines into fields in the normal manner. This happens by default as the result of a special feature. When RS is set to the empty string and FS is set to a single character, the newline character always acts as a field separator. This is in addition to whatever field separations result from FS.

NOTE: When FS is the null string ("") or a regexp, this special feature of RS does not apply. It does apply to the default field separator of a single space: ‘FS = " "’.

Note that language in the POSIX specification implies that this special feature should apply when FS is a regexp. However, Unix awk has never behaved that way, nor has gawk. This is essentially a bug in POSIX.

The original motivation for this special exception was probably to provide useful behavior in the default case (i.e., FS is equal to " "). This feature can be a problem if you really don’t want the newline character to separate fields, because there is no way to prevent it. However, you can work around this by using the split() function to break up the record manually (see String-Manipulation Functions). If you have a single-character field separator, you can work around the special feature in a different way, by making FS into a regexp for that single character. For example, if the field separator is a percent character, instead of ‘FS = "%"’, use ‘FS = "[%]"’.

Another way to separate fields is to put each field on a separate line: to do this, just set the variable FS to the string "\n". (This single-character separator matches a single newline.) A practical example of a data file organized this way might be a mailing list, where blank lines separate the entries. Consider a mailing list in a file named addresses, which looks like this:

Jane Doe
123 Main Street
Anywhere, SE 12345-6789
John Smith
456 Tree-lined Avenue
Smallville, MW 98765-4321
…

A simple program to process this file is as follows:

# addrs.awk --- simple mailing list program
# Records are separated by blank lines.
# Each line is one field.
BEGIN { RS = "" ; FS = "\n" }
{
      print "Name is:", $1
      print "Address is:", $2
      print "City and State are:", $3
      print ""
}

Running the program produces the following output:

$ awk -f addrs.awk addresses
-| Name is: Jane Doe
-| Address is: 123 Main Street
-| City and State are: Anywhere, SE 12345-6789
-|
-| Name is: John Smith
-| Address is: 456 Tree-lined Avenue
-| City and State are: Smallville, MW 98765-4321
-|
…

See Printing Mailing Labels for a more realistic program dealing with address lists. The following list summarizes how records are split, based on the value of RS:

RS == "\n"

Records are separated by the newline character (‘\n’). In effect, every line in the data file is a separate record, including blank lines. This is the default.

RS == any single character

Records are separated by each occurrence of the character. Multiple successive occurrences delimit empty records.

RS == ""

Records are separated by runs of blank lines. When FS is a single character, then the newline character always serves as a field separator, in addition to whatever value FS may have. Leading and trailing newlines in a file are ignored.

RS == regexp

Records are separated by occurrences of characters that match regexp. Leading and trailing matches of regexp delimit empty records. (This is a gawk extension; it is not specified by the POSIX standard.)

If not in compatibility mode (see Command-Line Options), gawk sets RT to the input text that matched the value specified by RS. But if the input file ended without any text that matches RS, then gawk sets RT to the null string.


4.10 Explicit Input with getline

So far we have been getting our input data from awk’s main input stream—either the standard input (usually your keyboard, sometimes the output from another program) or the files specified on the command line. The awk language has a special built-in command called getline that can be used to read input under your explicit control.

The getline command is used in several different ways and should not be used by beginners. The examples that follow the explanation of the getline command include material that has not been covered yet. Therefore, come back and study the getline command after you have reviewed the rest of this Web page and have a good knowledge of how awk works.

The getline command returns 1 if it finds a record and 0 if it encounters the end of the file. If there is some error in getting a record, such as a file that cannot be opened, then getline returns −1. In this case, gawk sets the variable ERRNO to a string describing the error that occurred.

If ERRNO indicates that the I/O operation may be retried, and PROCINFO["input", "RETRY"] is set, then getline returns −2 instead of −1, and further calls to getline may be attempted. See Retrying Reads After Certain Input Errors for further information about this feature.

In the following examples, command stands for a string value that represents a shell command.

NOTE: When --sandbox is specified (see Command-Line Options), reading lines from files, pipes, and coprocesses is disabled.


4.10.1 Using getline with No Arguments

The getline command can be used without arguments to read input from the current input file. All it does in this case is read the next input record and split it up into fields. This is useful if you’ve finished processing the current record, but want to do some special processing on the next record right now. For example:

# Remove text between /* and */, inclusive
{
    while ((start = index($0, "/*")) != 0) {
        out = substr($0, 1, start - 1)  # leading part of the string
        rest = substr($0, start + 2)    # ... */ ...
        while ((end = index(rest, "*/")) == 0) {  # is */ in trailing part?
            # get more text
            if (getline <= 0) {
                print("unexpected EOF or error:", ERRNO) > "/dev/stderr"
                exit
            }
            # build up the line using string concatenation
            rest = rest $0
        }
        rest = substr(rest, end + 2)  # remove comment
        # build up the output line using string concatenation
        $0 = out rest
    }
    print $0
}

This awk program deletes C-style comments (‘/* … */’) from the input. It uses a number of features we haven’t covered yet, including string concatenation (see String Concatenation) and the index() and substr() built-in functions (see String-Manipulation Functions). By replacing the ‘print $0’ with other statements, you could perform more complicated processing on the decommented input, such as searching for matches of a regular expression.

Here is some sample input:

mon/*comment*/key
rab/*commen
t*/bit
horse /*comment*/more text
part 1 /*comment*/part 2 /*comment*/part 3
no comment

When run, the output is:

$ awk -f strip_comments.awk example_text
-| monkey
-| rabbit
-| horse more text
-| part 1 part 2 part 3
-| no comment

This form of the getline command sets NF, NR, FNR, RT, and the value of $0.

NOTE: The new value of $0 is used to test the patterns of any subsequent rules. The original value of $0 that triggered the rule that executed getline is lost. By contrast, the next statement reads a new record but immediately begins processing it normally, starting with the first rule in the program. See The next Statement.


4.10.2 Using getline into a Variable

You can use ‘getline var’ to read the next record from awk’s input into the variable var. No other processing is done. For example, suppose the next line is a comment or a special string, and you want to read it without triggering any rules. This form of getline allows you to read that line and store it in a variable so that the main read-a-line-and-check-each-rule loop of awk never sees it. The following example swaps every two lines of input:

{
     if ((getline tmp) > 0) {
          print tmp
          print $0
     } else
          print $0
}

It takes the following list:

wan
tew
free
phore

and produces these results:

tew
wan
phore
free

The getline command used in this way sets only the variables NR, FNR, and RT (and, of course, var). The record is not split into fields, so the values of the fields (including $0) and the value of NF do not change.


4.10.3 Using getline from a File

Use ‘getline < file’ to read the next record from file. Here, file is a string-valued expression that specifies the file name. ‘< file’ is called a redirection because it directs input to come from a different place. For example, the following program reads its input record from the file secondary.input when it encounters a first field with a value equal to 10 in the current input file:

{
    if ($1 == 10) {
         getline < "secondary.input"
         print
    } else
         print
}

Because the main input stream is not used, the values of NR and FNR are not changed. However, the record it reads is split into fields in the normal manner, so the values of $0 and the other fields are changed, resulting in a new value of NF. RT is also set.

According to POSIX, ‘getline < expression’ is ambiguous if expression contains unparenthesized operators other than ‘$’; for example, ‘getline < dir "/" file’ is ambiguous because the concatenation operator (not discussed yet; see String Concatenation) is not parenthesized. You should write it as ‘getline < (dir "/" file)’ if you want your program to be portable to all awk implementations.


4.10.4 Using getline into a Variable from a File

Use ‘getline var < file’ to read input from the file file, and put it in the variable var. As earlier, file is a string-valued expression that specifies the file from which to read.

In this version of getline, none of the predefined variables are changed and the record is not split into fields. The only variable changed is var.25 For example, the following program copies all the input files to the output, except for records that say ‘@include filename. Such a record is replaced by the contents of the file filename:

{
     if (NF == 2 && $1 == "@include") {
          while ((getline line < $2) > 0)
               print line
          close($2)
     } else
          print
}

Note here how the name of the extra input file is not built into the program; it is taken directly from the data, specifically from the second field on the @include line.

The close() function is called to ensure that if two identical @include lines appear in the input, the entire specified file is included twice. See Closing Input and Output Redirections.

One deficiency of this program is that it does not process nested @include statements (i.e., @include statements in included files) the way a true macro preprocessor would. See An Easy Way to Use Library Functions for a program that does handle nested @include statements.


4.10.5 Using getline from a Pipe

Omniscience has much to recommend it. Failing that, attention to details would be useful.

Brian Kernighan

The output of a command can also be piped into getline, using ‘command | getline’. In this case, the string command is run as a shell command and its output is piped into awk to be used as input. This form of getline reads one record at a time from the pipe. For example, the following program copies its input to its output, except for lines that begin with ‘@execute’, which are replaced by the output produced by running the rest of the line as a shell command:

{
     if ($1 == "@execute") {
          tmp = substr($0, 10)        # Remove "@execute"
          while ((tmp | getline) > 0)
               print
          close(tmp)
     } else
          print
}

The close() function is called to ensure that if two identical ‘@execute’ lines appear in the input, the command is run for each one. See Closing Input and Output Redirections. Given the input:

foo
bar
baz
@execute who
bletch

the program might produce:

foo
bar
baz
arnold     ttyv0   Jul 13 14:22
miriam     ttyp0   Jul 13 14:23     (murphy:0)
bill       ttyp1   Jul 13 14:23     (murphy:0)
bletch

Notice that this program ran the command who and printed the result. (If you try this program yourself, you will of course get different results, depending upon who is logged in on your system.)

This variation of getline splits the record into fields, sets the value of NF, and recomputes the value of $0. The values of NR and FNR are not changed. RT is set.

According to POSIX, ‘expression | getline’ is ambiguous if expression contains unparenthesized operators other than ‘$’—for example, ‘"echo " "date" | getline’ is ambiguous because the concatenation operator is not parenthesized. You should write it as ‘("echo " "date") | getline’ if you want your program to be portable to all awk implementations.

NOTE: Unfortunately, gawk has not been consistent in its treatment of a construct like ‘"echo " "date" | getline’. Most versions, including the current version, treat it as ‘("echo " "date") | getline’. (This is also how BWK awk behaves.) Some versions instead treat it as ‘"echo " ("date" | getline)’. (This is how mawk behaves.) In short, always use explicit parentheses, and then you won’t have to worry.


4.10.6 Using getline into a Variable from a Pipe

When you use ‘command | getline var’, the output of command is sent through a pipe to getline and into the variable var. For example, the following program reads the current date and time into the variable current_time, using the date utility, and then prints it:

BEGIN {
     "date" | getline current_time
     close("date")
     print "Report printed on " current_time
}

In this version of getline, none of the predefined variables are changed and the record is not split into fields. However, RT is set.


4.10.7 Using getline from a Coprocess

Reading input into getline from a pipe is a one-way operation. The command that is started with ‘command | getline’ only sends data to your awk program.

On occasion, you might want to send data to another program for processing and then read the results back. gawk allows you to start a coprocess, with which two-way communications are possible. This is done with the ‘|&’ operator. Typically, you write data to the coprocess first and then read the results back, as shown in the following:

print "some query" |& "db_server"
"db_server" |& getline

which sends a query to db_server and then reads the results.

The values of NR and FNR are not changed, because the main input stream is not used. However, the record is split into fields in the normal manner, thus changing the values of $0, of the other fields, and of NF and RT.

Coprocesses are an advanced feature. They are discussed here only because this is the section on getline. See Two-Way Communications with Another Process, where coprocesses are discussed in more detail.


4.10.8 Using getline into a Variable from a Coprocess

When you use ‘command |& getline var’, the output from the coprocess command is sent through a two-way pipe to getline and into the variable var.

In this version of getline, none of the predefined variables are changed and the record is not split into fields. The only variable changed is var. However, RT is set.


4.10.9 Points to Remember About getline

Here are some miscellaneous points about getline that you should bear in mind:

  • When getline changes the value of $0 and NF, awk does not automatically jump to the start of the program and start testing the new record against every pattern. However, the new record is tested against any subsequent rules.
  • Some very old awk implementations limit the number of pipelines that an awk program may have open to just one. In gawk, there is no such limit. You can open as many pipelines (and coprocesses) as the underlying operating system permits.
  • An interesting side effect occurs if you use getline without a redirection inside a BEGIN rule. Because an unredirected getline reads from the command-line data files, the first getline command causes awk to set the value of FILENAME. Normally, FILENAME does not have a value inside BEGIN rules, because you have not yet started to process the command-line data files. (d.c.) (See The BEGIN and END Special Patterns; also see Built-in Variables That Convey Information.)
  • Using FILENAME with getline (‘getline < FILENAME’) is likely to be a source of confusion. awk opens a separate input stream from the current input file. However, by not using a variable, $0 and NF are still updated. If you’re doing this, it’s probably by accident, and you should reconsider what it is you’re trying to accomplish.
  • Summary of getline Variants, presents a table summarizing the getline variants and which variables they can affect. It is worth noting that those variants that do not use redirection can cause FILENAME to be updated if they cause awk to start reading a new input file.
  • If the variable being assigned is an expression with side effects, different versions of awk behave differently upon encountering end-of-file. Some versions don’t evaluate the expression; many versions (including gawk) do. Here is an example, courtesy of Duncan Moore:
    BEGIN {
        system("echo 1 > f")
        while ((getline a[++c] < "f") > 0) { }
        print c
    }
    

    Here, the side effect is the ‘++c’. Is c incremented if end-of-file is encountered before the element in a is assigned?

    gawk treats getline like a function call, and evaluates the expression ‘a[++c]’ before attempting to read from f. However, some versions of awk only evaluate the expression once they know that there is a string value to be assigned.


4.10.10 Summary of getline Variants

Table 4.1 summarizes the eight variants of getline, listing which predefined variables are set by each one, and whether the variant is standard or a gawk extension. Note: for each variant, gawk sets the RT predefined variable.

VariantEffectawk / gawk
getlineSets $0, NF, FNR, NR, and RTawk
getline varSets var, FNR, NR, and RTawk
getline < fileSets $0, NF, and RTawk
getline var < fileSets var and RTawk
command | getlineSets $0, NF, and RTawk
command | getline varSets var and RTawk
command |& getlineSets $0, NF, and RTgawk
command |& getline varSets var and RTgawk

Table 4.1: getline variants and what they set


4.11 Reading Input with a Timeout

This section describes a feature that is specific to gawk.

You may specify a timeout in milliseconds for reading input from the keyboard, a pipe, or two-way communication, including TCP/IP sockets. This can be done on a per-input, per-command, or per-connection basis, by setting a special element in the PROCINFO array (see Built-in Variables That Convey Information):

PROCINFO["input_name", "READ_TIMEOUT"] = timeout in milliseconds

When set, this causes gawk to time out and return failure if no data is available to read within the specified timeout period. For example, a TCP client can decide to give up on receiving any response from the server after a certain amount of time:

Service = "/inet/tcp/0/localhost/daytime"
PROCINFO[Service, "READ_TIMEOUT"] = 100
if ((Service |& getline) > 0)
    print $0
else if (ERRNO != "")
    print ERRNO

Here is how to read interactively from the user26 without waiting for more than five seconds:

PROCINFO["/dev/stdin", "READ_TIMEOUT"] = 5000
while ((getline < "/dev/stdin") > 0)
    print $0

gawk terminates the read operation if input does not arrive after waiting for the timeout period, returns failure, and sets ERRNO to an appropriate string value. A negative or zero value for the timeout is the same as specifying no timeout at all.

A timeout can also be set for reading from the keyboard in the implicit loop that reads input records and matches them against patterns, like so:

$ gawk 'BEGIN { PROCINFO["-", "READ_TIMEOUT"] = 5000 }
> { print "You entered: " $0 }'
gawk
-| You entered: gawk

In this case, failure to respond within five seconds results in the following error message:

error→ gawk: cmd. line:2: (FILENAME=- FNR=1) fatal: error reading input file `-': Connection timed out

The timeout can be set or changed at any time, and will take effect on the next attempt to read from the input device. In the following example, we start with a timeout value of one second, and progressively reduce it by one-tenth of a second until we wait indefinitely for the input to arrive:

PROCINFO[Service, "READ_TIMEOUT"] = 1000
while ((Service |& getline) > 0) {
    print $0
    PROCINFO[Service, "READ_TIMEOUT"] -= 100
}

NOTE: You should not assume that the read operation will block exactly after the tenth record has been printed. It is possible that gawk will read and buffer more than one record’s worth of data the first time. Because of this, changing the value of timeout like in the preceding example is not very useful.

If the PROCINFO element is not present and the GAWK_READ_TIMEOUT environment variable exists, gawk uses its value to initialize the timeout value. The exclusive use of the environment variable to specify timeout has the disadvantage of not being able to control it on a per-command or per-connection basis.

gawk considers a timeout event to be an error even though the attempt to read from the underlying device may succeed in a later attempt. This is a limitation, and it also means that you cannot use this to multiplex input from two or more sources. See Retrying Reads After Certain Input Errors for a way to enable later I/O attempts to succeed.

Assigning a timeout value prevents read operations from being blocked indefinitely. But bear in mind that there are other ways gawk can stall waiting for an input device to be ready. A network client can sometimes take a long time to establish a connection before it can start reading any data, or the attempt to open a FIFO special file for reading can be blocked indefinitely until some other process opens it for writing.


4.12 Retrying Reads After Certain Input Errors

This section describes a feature that is specific to gawk.

When gawk encounters an error while reading input, by default getline returns −1, and subsequent attempts to read from that file result in an end-of-file indication. However, you may optionally instruct gawk to allow I/O to be retried when certain errors are encountered by setting a special element in the PROCINFO array (see Built-in Variables That Convey Information):

PROCINFO["input_name", "RETRY"] = 1

When this element exists, gawk checks the value of the system (C language) errno variable when an I/O error occurs. If errno indicates a subsequent I/O attempt may succeed, getline instead returns −2 and further calls to getline may succeed. This applies to the errno values EAGAIN, EWOULDBLOCK, EINTR, or ETIMEDOUT.

This feature is useful in conjunction with PROCINFO["input_name", "READ_TIMEOUT"] or situations where a file descriptor has been configured to behave in a non-blocking fashion.


4.13 Directories on the Command Line

According to the POSIX standard, files named on the awk command line must be text files; it is a fatal error if they are not. Most versions of awk treat a directory on the command line as a fatal error.

By default, gawk produces a warning for a directory on the command line, but otherwise ignores it. This makes it easier to use shell wildcards with your awk program:

$ gawk -f whizprog.awk *        Directories could kill this program

If either of the --posix or --traditional options is given, then gawk reverts to treating a directory on the command line as a fatal error.

See Reading Directories for a way to treat directories as usable data from an awk program.


4.14 Summary

  • Input is split into records based on the value of RS. The possibilities are as follows:
    Value of RSRecords are split on …awk / gawk
    Any single characterThat characterawk
    The empty string ("")Runs of two or more newlinesawk
    A regexpText that matches the regexpgawk
  • FNR indicates how many records have been read from the current input file; NR indicates how many records have been read in total.
  • gawk sets RT to the text matched by RS.
  • After splitting the input into records, awk further splits the records into individual fields, named $1, $2, and so on. $0 is the whole record, and NF indicates how many fields there are. The default way to split fields is between whitespace characters.
  • Fields may be referenced using a variable, as in $NF. Fields may also be assigned values, which causes the value of $0 to be recomputed when it is later referenced. Assigning to a field with a number greater than NF creates the field and rebuilds the record, using OFS to separate the fields. Incrementing NF does the same thing. Decrementing NF throws away fields and rebuilds the record.
  • Field splitting is more complicated than record splitting:
    Field separator valueFields are split …awk / gawk
    FS == " "On runs of whitespaceawk
    FS == any single characterOn that characterawk
    FS == regexpOn text matching the regexpawk
    FS == ""Such that each individual character is a separate fieldgawk
    FIELDWIDTHS == list of columnsBased on character positiongawk
    FPAT == regexpOn the text surrounding text matching the regexpgawk
  • Using ‘FS = "\n"’ causes the entire record to be a single field (assuming that newlines separate records).
  • FS may be set from the command line using the -F option. This can also be done using command-line variable assignment.
  • Use PROCINFO["FS"] to see how fields are being split.
  • Use getline in its various forms to read additional records from the default input stream, from a file, or from a pipe or coprocess.
  • Use PROCINFO[file, "READ_TIMEOUT"] to cause reads to time out for file.
  • Directories on the command line are fatal for standard awk; gawk ignores them if not in POSIX mode.

4.15 Exercises

  1. Using the FIELDWIDTHS variable (see Reading Fixed-Width Data), write a program to read election data, where each record represents one voter’s votes. Come up with a way to define which columns are associated with each ballot item, and print the total votes, including abstentions, for each item.

5 Printing Output

One of the most common programming actions is to print, or output, some or all of the input. Use the print statement for simple output, and the printf statement for fancier formatting. The print statement is not limited when computing which values to print. However, with two exceptions, you cannot specify how to print them—how many columns, whether to use exponential notation or not, and so on. (For the exceptions, see Output Separators and Controlling Numeric Output with print.) For printing with specifications, you need the printf statement (see Using printf Statements for Fancier Printing).

Besides basic and formatted printing, this chapter also covers I/O redirections to files and pipes, introduces the special file names that gawk processes internally, and discusses the close() built-in function.


5.1 The print Statement

Use the print statement to produce output with simple, standardized formatting. You specify only the strings or numbers to print, in a list separated by commas. They are output, separated by single spaces, followed by a newline. The statement looks like this:

print item1, item2, …

The entire list of items may be optionally enclosed in parentheses. The parentheses are necessary if any of the item expressions uses the ‘>’ relational operator; otherwise it could be confused with an output redirection (see Redirecting Output of print and printf).

The items to print can be constant strings or numbers, fields of the current record (such as $1), variables, or any awk expression. Numeric values are converted to strings and then printed.

The simple statement ‘print’ with no items is equivalent to ‘print $0’: it prints the entire current record. To print a blank line, use ‘print ""’. To print a fixed piece of text, use a string constant, such as "Don't Panic", as one item. If you forget to use the double-quote characters, your text is taken as an awk expression, and you will probably get an error. Keep in mind that a space is printed between any two items.

Note that the print statement is a statement and not an expression—you can’t use it in the pattern part of a pattern–action statement, for example.


5.3 Output Separators

As mentioned previously, a print statement contains a list of items separated by commas. In the output, the items are normally separated by single spaces. However, this doesn’t need to be the case; a single space is simply the default. Any string of characters may be used as the output field separator by setting the predefined variable OFS. The initial value of this variable is the string " " (i.e., a single space).

The output from an entire print statement is called an output record. Each print statement outputs one output record, and then outputs a string called the output record separator (or ORS). The initial value of ORS is the string "\n" (i.e., a newline character). Thus, each print statement normally makes a separate line.

In order to change how output fields and records are separated, assign new values to the variables OFS and ORS. The usual place to do this is in the BEGIN rule (see The BEGIN and END Special Patterns), so that it happens before any input is processed. It can also be done with assignments on the command line, before the names of the input files, or using the -v command-line option (see Command-Line Options). The following example prints the first and second fields of each input record, separated by a semicolon, with a blank line added after each newline:

$ awk 'BEGIN { OFS = ";"; ORS = "\n\n" }
>            { print $1, $2 }' mail-list
-| Amelia;555-5553
-|
-| Anthony;555-3412
-|
-| Becky;555-7685
-|
-| Bill;555-1675
-|
-| Broderick;555-0542
-|
-| Camilla;555-2912
-|
-| Fabius;555-1234
-|
-| Julie;555-6699
-|
-| Martin;555-6480
-|
-| Samuel;555-3430
-|
-| Jean-Paul;555-2127
-|

If the value of ORS does not contain a newline, the program’s output runs together on a single line.


5.4 Controlling Numeric Output with print

When printing numeric values with the print statement, awk internally converts each number to a string of characters and prints that string. awk uses the sprintf() function to do this conversion (see String-Manipulation Functions). For now, it suffices to say that the sprintf() function accepts a format specification that tells it how to format numbers (or strings), and that there are a number of different ways in which numbers can be formatted. The different format specifications are discussed more fully in Format-Control Letters.

The predefined variable OFMT contains the format specification that print uses with sprintf() when it wants to convert a number to a string for printing. The default value of OFMT is "%.6g". The way print prints numbers can be changed by supplying a different format specification for the value of OFMT, as shown in the following example:

$ awk 'BEGIN {
>   OFMT = "%.0f"  # print numbers as integers (rounds)
>   print 17.23, 17.54 }'
-| 17 18

According to the POSIX standard, awk’s behavior is undefined if OFMT contains anything but a floating-point conversion specification. (d.c.)


5.5 Using printf Statements for Fancier Printing

For more precise control over the output format than what is provided by print, use printf. With printf you can specify the width to use for each item, as well as various formatting choices for numbers (such as what output base to use, whether to print an exponent, whether to print a sign, and how many digits to print after the decimal point).


5.5.1 Introduction to the printf Statement

A simple printf statement looks like this:

printf format, item1, item2, …

As for print, the entire list of arguments may optionally be enclosed in parentheses. Here too, the parentheses are necessary if any of the item expressions uses the ‘>’ relational operator; otherwise, it can be confused with an output redirection (see Redirecting Output of print and printf).

The difference between printf and print is the format argument. This is an expression whose value is taken as a string; it specifies how to output each of the other arguments. It is called the format string.

The format string is very similar to that in the ISO C library function printf(). Most of format is text to output verbatim. Scattered among this text are format specifiers—one per item. Each format specifier says to output the next item in the argument list at that place in the format.

The printf statement does not automatically append a newline to its output. It outputs only what the format string specifies. So if a newline is needed, you must include one in the format string. The output separator variables OFS and ORS have no effect on printf statements. For example:

$ awk 'BEGIN {
>    ORS = "\nOUCH!\n"; OFS = "+"
>    msg = "Don\47t Panic!"
>    printf "%s\n", msg
> }'
-| Don't Panic!

Here, neither the ‘+’ nor the ‘OUCH!’ appears in the output message.


5.5.2 Format-Control Letters

A format specifier starts with the character ‘%’ and ends with a format-control letter—it tells the printf statement how to output one item. The format-control letter specifies what kind of value to print. The rest of the format specifier is made up of optional modifiers that control how to print the value, such as the field width. Here is a list of the format-control letters:

%a, %A

A floating point number of the form [-]0xh.hhhhp+-dd (C99 hexadecimal floating point format). For %A, uppercase letters are used instead of lowercase ones.

NOTE: The current POSIX standard requires support for %a and %A in awk. As far as we know, besides gawk, the only other version of awk that actually implements it is BWK awk. It’s use is thus highly nonportable!

Furthermore, these formats are not available on any system where the underlying C library printf() function does not support them. As of this writing, among current systems, only OpenVMS is known to not support them.

%c

Print a number as a character; thus, ‘printf "%c", 65’ outputs the letter ‘A’. The output for a string value is the first character of the string.

NOTE: The POSIX standard says the first character of a string is printed. In locales with multibyte characters, gawk attempts to convert the leading bytes of the string into a valid wide character and then to print the multibyte encoding of that character. Similarly, when printing a numeric value, gawk allows the value to be within the numeric range of values that can be held in a wide character. If the conversion to multibyte encoding fails, gawk uses the low eight bits of the value as the character to print.

Other awk versions generally restrict themselves to printing the first byte of a string or to numeric values within the range of a single byte (0–255). (d.c.)

%d, %i

Print a decimal integer. The two control letters are equivalent. (The ‘%i’ specification is for compatibility with ISO C.)

%e, %E

Print a number in scientific (exponential) notation. For example:

printf "%4.3e\n", 1950

prints ‘1.950e+03’, with a total of four significant figures, three of which follow the decimal point. (The ‘4.3’ represents two modifiers, discussed in the next subsection.) ‘%E’ uses ‘E’ instead of ‘e’ in the output.

%f

Print a number in floating-point notation. For example:

printf "%4.3f", 1950

prints ‘1950.000’, with a minimum of four significant figures, three of which follow the decimal point. (The ‘4.3’ represents two modifiers, discussed in the next subsection.)

On systems supporting IEEE 754 floating-point format, values representing negative infinity are formatted as ‘-inf’ or ‘-infinity’, and positive infinity as ‘inf’ or ‘infinity’. The special “not a number” value formats as ‘-nan’ or ‘nan’ (see Floating Point Values They Didn’t Talk About In School).

%F

Like ‘%f’, but the infinity and “not a number” values are spelled using uppercase letters.

The ‘%F’ format is a POSIX extension to ISO C; not all systems support it. On those that don’t, gawk uses ‘%f’ instead.

%g, %G

Print a number in either scientific notation or in floating-point notation, whichever uses fewer characters; if the result is printed in scientific notation, ‘%G’ uses ‘E’ instead of ‘e’.

%o

Print an unsigned octal integer (see Octal and Hexadecimal Numbers).

%s

Print a string.

%u

Print an unsigned decimal integer. (This format is of marginal use, because all numbers in awk are floating point; it is provided primarily for compatibility with C.)

%x, %X

Print an unsigned hexadecimal integer; ‘%X’ uses the letters ‘A’ through ‘F’ instead of ‘a’ through ‘f’ (see Octal and Hexadecimal Numbers).

%%

Print a single ‘%’. This does not consume an argument and it ignores any modifiers.

NOTE: When using the integer format-control letters for values that are outside the range of the widest C integer type, gawk switches to the ‘%g’ format specifier. If --lint is provided on the command line (see Command-Line Options), gawk warns about this. Other versions of awk may print invalid values or do something else entirely. (d.c.)

NOTE: The IEEE 754 standard for floating-point arithmetic allows for special values that represent “infinity” (positive and negative) and values that are “not a number” (NaN).

Input and output of these values occurs as text strings. This is somewhat problematic for the awk language, which predates the IEEE standard. Further details are provided in Standards Versus Existing Practice; please see there.


5.5.3 Modifiers for printf Formats

A format specification can also include modifiers that can control how much of the item’s value is printed, as well as how much space it gets. The modifiers come between the ‘%’ and the format-control letter. We use the bullet symbol “•” in the following examples to represent spaces in the output. Here are the possible modifiers, in the order in which they may appear:

N$

An integer constant followed by a ‘$’ is a positional specifier. Normally, format specifications are applied to arguments in the order given in the format string. With a positional specifier, the format specification is applied to a specific argument, instead of what would be the next argument in the list. Positional specifiers begin counting with one. Thus:

printf "%s %s\n", "don't", "panic"
printf "%2$s %1$s\n", "panic", "don't"

prints the famous friendly message twice.

At first glance, this feature doesn’t seem to be of much use. It is in fact a gawk extension, intended for use in translating messages at runtime. See Rearranging printf Arguments, which describes how and why to use positional specifiers. For now, we ignore them.

- (Minus)

The minus sign, used before the width modifier (see later on in this list), says to left-justify the argument within its specified width. Normally, the argument is printed right-justified in the specified width. Thus:

printf "%-4s", "foo"

prints ‘foo•’.

space

For numeric conversions, prefix positive values with a space and negative values with a minus sign.

+

The plus sign, used before the width modifier (see later on in this list), says to always supply a sign for numeric conversions, even if the data to format is positive. The ‘+’ overrides the space modifier.

#

Use an “alternative form” for certain control letters. For ‘%o’, supply a leading zero. For ‘%x’ and ‘%X’, supply a leading ‘0x’ or ‘0X’ for a nonzero result. For ‘%e’, ‘%E’, ‘%f’, and ‘%F’, the result always contains a decimal point. For ‘%g’ and ‘%G’, trailing zeros are not removed from the result.

0

A leading ‘0’ (zero) acts as a flag indicating that output should be padded with zeros instead of spaces. This applies only to the numeric output formats. This flag only has an effect when the field width is wider than the value to print.

'

A single quote or apostrophe character is a POSIX extension to ISO C. It indicates that the integer part of a floating-point value, or the entire part of an integer decimal value, should have a thousands-separator character in it. This only works in locales that support such characters. For example:

$ cat thousands.awk          Show source program
-| BEGIN { printf "%'d\n", 1234567 }
$ LC_ALL=C gawk -f thousands.awk
-| 1234567                   Results in "C" locale
$ LC_ALL=en_US.UTF-8 gawk -f thousands.awk
-| 1,234,567                 Results in US English UTF locale

For more information about locales and internationalization issues, see Where You Are Makes a Difference.

NOTE: The ‘'’ flag is a nice feature, but its use complicates things: it becomes difficult to use it in command-line programs. For information on appropriate quoting tricks, see Shell Quoting Issues.

width

This is a number specifying the desired minimum width of a field. Inserting any number between the ‘%’ sign and the format-control character forces the field to expand to this width. The default way to do this is to pad with spaces on the left. For example:

printf "%4s", "foo"

prints ‘•foo’.

The value of width is a minimum width, not a maximum. If the item value requires more than width characters, it can be as wide as necessary. Thus, the following:

printf "%4s", "foobar"

prints ‘foobar’.

Preceding the width with a minus sign causes the output to be padded with spaces on the right, instead of on the left.

.prec

A period followed by an integer constant specifies the precision to use when printing. The meaning of the precision varies by control letter:

%d, %i, %o, %u, %x, %X

Minimum number of digits to print.

%e, %E, %f, %F

Number of digits to the right of the decimal point.

%g, %G

Maximum number of significant digits.

%s

Maximum number of characters from the string that should print.

Thus, the following:

printf "%.4s", "foobar"

prints ‘foob’.

The C library printf’s dynamic width and prec capability (e.g., "%*.*s") is supported. Instead of supplying explicit width and/or prec values in the format string, they are passed in the argument list. For example:

w = 5
p = 3
s = "abcdefg"
printf "%*.*s\n", w, p, s

is exactly equivalent to:

s = "abcdefg"
printf "%5.3s\n", s

Both programs output ‘••abc’. Earlier versions of awk did not support this capability. If you must use such a version, you may simulate this feature by using concatenation to build up the format string, like so:

w = 5
p = 3
s = "abcdefg"
printf "%" w "." p "s\n", s

This is not particularly easy to read, but it does work.

C programmers may be used to supplying additional modifiers (‘h’, ‘j’, ‘l’, ‘L’, ‘t’, and ‘z’) in printf format strings. These are not valid in awk. Most awk implementations silently ignore them. If --lint is provided on the command line (see Command-Line Options), gawk warns about their use. If --posix is supplied, their use is a fatal error.


5.5.4 Examples Using printf

The following simple example shows how to use printf to make an aligned table:

awk '{ printf "%-10s %s\n", $1, $2 }' mail-list

This command prints the names of the people ($1) in the file mail-list as a string of 10 characters that are left-justified. It also prints the phone numbers ($2) next on the line. This produces an aligned two-column table of names and phone numbers, as shown here:

$ awk '{ printf "%-10s %s\n", $1, $2 }' mail-list
-| Amelia     555-5553
-| Anthony    555-3412
-| Becky      555-7685
-| Bill       555-1675
-| Broderick  555-0542
-| Camilla    555-2912
-| Fabius     555-1234
-| Julie      555-6699
-| Martin     555-6480
-| Samuel     555-3430
-| Jean-Paul  555-2127

In this case, the phone numbers had to be printed as strings because the numbers are separated by dashes. Printing the phone numbers as numbers would have produced just the first three digits: ‘555’. This would have been pretty confusing.

It wasn’t necessary to specify a width for the phone numbers because they are last on their lines. They don’t need to have spaces after them.

The table could be made to look even nicer by adding headings to the tops of the columns. This is done using a BEGIN rule (see The BEGIN and END Special Patterns) so that the headers are only printed once, at the beginning of the awk program:

awk 'BEGIN { print "Name      Number"
             print "----      ------" }
           { printf "%-10s %s\n", $1, $2 }' mail-list

The preceding example mixes print and printf statements in the same program. Using just printf statements can produce the same results:

awk 'BEGIN { printf "%-10s %s\n", "Name", "Number"
             printf "%-10s %s\n", "----", "------" }
           { printf "%-10s %s\n", $1, $2 }' mail-list

Printing each column heading with the same format specification used for the column elements ensures that the headings are aligned just like the columns.

The fact that the same format specification is used three times can be emphasized by storing it in a variable, like this:

awk 'BEGIN { format = "%-10s %s\n"
             printf format, "Name", "Number"
             printf format, "----", "------" }
           { printf format, $1, $2 }' mail-list

5.6 Redirecting Output of print and printf

So far, the output from print and printf has gone to the standard output, usually the screen. Both print and printf can also send their output to other places. This is called redirection.

NOTE: When --sandbox is specified (see Command-Line Options), redirecting output to files, pipes, and coprocesses is disabled.

A redirection appears after the print or printf statement. Redirections in awk are written just like redirections in shell commands, except that they are written inside the awk program.

There are four forms of output redirection: output to a file, output appended to a file, output through a pipe to another command, and output to a coprocess. We show them all for the print statement, but they work identically for printf:

print items > output-file

This redirection prints the items into the output file named output-file. The file name output-file can be any expression. Its value is changed to a string and then used as a file name (see Expressions).

When this type of redirection is used, the output-file is erased before the first output is written to it. Subsequent writes to the same output-file do not erase output-file, but append to it. (This is different from how you use redirections in shell scripts.) If output-file does not exist, it is created. For example, here is how an awk program can write a list of peoples’ names to one file named name-list, and a list of phone numbers to another file named phone-list:

$ awk '{ print $2 > "phone-list"
>        print $1 > "name-list" }' mail-list
$ cat phone-list
-| 555-5553
-| 555-3412
…
$ cat name-list
-| Amelia
-| Anthony
…

Each output file contains one name or number per line.

print items >> output-file

This redirection prints the items into the preexisting output file named output-file. The difference between this and the single-‘>’ redirection is that the old contents (if any) of output-file are not erased. Instead, the awk output is appended to the file. If output-file does not exist, then it is created.

print items | command

It is possible to send output to another program through a pipe instead of into a file. This redirection opens a pipe to command, and writes the values of items through this pipe to another process created to execute command.

The redirection argument command is actually an awk expression. Its value is converted to a string whose contents give the shell command to be run. For example, the following produces two files, one unsorted list of peoples’ names, and one list sorted in reverse alphabetical order:

awk '{ print $1 > "names.unsorted"
       command = "sort -r > names.sorted"
       print $1 | command }' mail-list

The unsorted list is written with an ordinary redirection, while the sorted list is written by piping through the sort utility.

The next example uses redirection to mail a message to the mailing list bug-system. This might be useful when trouble is encountered in an awk script run periodically for system maintenance:

report = "mail bug-system"
print("Awk script failed:", $0) | report
print("at record number", FNR, "of", FILENAME) | report
close(report)

The close() function is called here because it’s a good idea to close the pipe as soon as all the intended output has been sent to it. See Closing Input and Output Redirections for more information.

This example also illustrates the use of a variable to represent a file or command—it is not necessary to always use a string constant. Using a variable is generally a good idea, because (if you mean to refer to that same file or command) awk requires that the string value be written identically every time.

print items |& command

This redirection prints the items to the input of command. The difference between this and the single-‘|’ redirection is that the output from command can be read with getline. Thus, command is a coprocess, which works together with but is subsidiary to the awk program.

This feature is a gawk extension, and is not available in POSIX awk. See Using getline from a Coprocess, for a brief discussion. See Two-Way Communications with Another Process, for a more complete discussion.

Redirecting output using ‘>’, ‘>>’, ‘|’, or ‘|&’ asks the system to open a file, pipe, or coprocess only if the particular file or command you specify has not already been written to by your program or if it has been closed since it was last written to.

It is a common error to use ‘>’ redirection for the first print to a file, and then to use ‘>>’ for subsequent output:

# clear the file
print "Don't panic" > "guide.txt"
…
# append
print "Avoid improbability generators" >> "guide.txt"

This is indeed how redirections must be used from the shell. But in awk, it isn’t necessary. In this kind of case, a program should use ‘>’ for all the print statements, because the output file is only opened once. (It happens that if you mix ‘>’ and ‘>>’ output is produced in the expected order. However, mixing the operators for the same file is definitely poor style, and is confusing to readers of your program.)

As mentioned earlier (see Points to Remember About getline), many Many older awk implementations limit the number of pipelines that an awk program may have open to just one! In gawk, there is no such limit. gawk allows a program to open as many pipelines as the underlying operating system permits.

Piping into sh

A particularly powerful way to use redirection is to build command lines and pipe them into the shell, sh. For example, suppose you have a list of files brought over from a system where all the file names are stored in uppercase, and you wish to rename them to have names in all lowercase. The following program is both simple and efficient:

{ printf("mv %s %s\n", $0, tolower($0)) | "sh" }
END { close("sh") }

The tolower() function returns its argument string with all uppercase characters converted to lowercase (see String-Manipulation Functions). The program builds up a list of command lines, using the mv utility to rename the files. It then sends the list to the shell for execution.

See Quoting Strings to Pass to the Shell for a function that can help in generating command lines to be fed to the shell.


5.7 Special Files for Standard Preopened Data Streams

Running programs conventionally have three input and output streams already available to them for reading and writing. These are known as the standard input, standard output, and standard error output. These open streams (and any other open files or pipes) are often referred to by the technical term file descriptors.

These streams are, by default, connected to your keyboard and screen, but they are often redirected with the shell, via the ‘<’, ‘<<’, ‘>’, ‘>>’, ‘>&’, and ‘|’ operators. Standard error is typically used for writing error messages; the reason there are two separate streams, standard output and standard error, is so that they can be redirected separately.

In traditional implementations of awk, the only way to write an error message to standard error in an awk program is as follows:

print "Serious error detected!" | "cat 1>&2"

This works by opening a pipeline to a shell command that can access the standard error stream that it inherits from the awk process. This is far from elegant, and it also requires a separate process. So people writing awk programs often don’t do this. Instead, they send the error messages to the screen, like this:

print "Serious error detected!" > "/dev/tty"

(/dev/tty is a special file supplied by the operating system that is connected to your keyboard and screen. It represents the “terminal,”27 which on modern systems is a keyboard and screen, not a serial console.) This generally has the same effect, but not always: although the standard error stream is usually the screen, it can be redirected; when that happens, writing to the screen is not correct. In fact, if awk is run from a background job, it may not have a terminal at all. Then opening /dev/tty fails.

gawk, BWK awk, and mawk provide special file names for accessing the three standard streams. If the file name matches one of these special names when gawk (or one of the others) redirects input or output, then it directly uses the descriptor that the file name stands for. These special file names work for all operating systems that gawk has been ported to, not just those that are POSIX-compliant:

/dev/stdin

The standard input (file descriptor 0).

/dev/stdout

The standard output (file descriptor 1).

/dev/stderr

The standard error output (file descriptor 2).

With these facilities, the proper way to write an error message then becomes:

print "Serious error detected!" > "/dev/stderr"

Note the use of quotes around the file name. Like with any other redirection, the value must be a string. It is a common error to omit the quotes, which leads to confusing results.

gawk does not treat these file names as special when in POSIX-compatibility mode. However, because BWK awk supports them, gawk does support them even when invoked with the --traditional option (see Command-Line Options).


5.8 Special File names in gawk

Besides access to standard input, standard output, and standard error, gawk provides access to any open file descriptor. Additionally, there are special file names reserved for TCP/IP networking.


5.8.1 Accessing Other Open Files with gawk

Besides the /dev/stdin, /dev/stdout, and /dev/stderr special file names mentioned earlier, gawk provides syntax for accessing any other inherited open file:

/dev/fd/N

The file associated with file descriptor N. Such a file must be opened by the program initiating the awk execution (typically the shell). Unless special pains are taken in the shell from which gawk is invoked, only descriptors 0, 1, and 2 are available.

The file names /dev/stdin, /dev/stdout, and /dev/stderr are essentially aliases for /dev/fd/0, /dev/fd/1, and /dev/fd/2, respectively. However, those names are more self-explanatory.

Note that using close() on a file name of the form "/dev/fd/N", for file descriptor numbers above two, does actually close the given file descriptor.


5.8.2 Special Files for Network Communications

gawk programs can open a two-way TCP/IP connection, acting as either a client or a server. This is done using a special file name of the form:

/net-type/protocol/local-port/remote-host/remote-port

The net-type is one of ‘inet’, ‘inet4’, or ‘inet6’. The protocol is one of ‘tcp’ or ‘udp’, and the other fields represent the other essential pieces of information for making a networking connection. These file names are used with the ‘|&’ operator for communicating with a coprocess (see Two-Way Communications with Another Process). This is an advanced feature, mentioned here only for completeness. Full discussion is delayed until Using gawk for Network Programming.


5.8.3 Special File name Caveats

Here are some things to bear in mind when using the special file names that gawk provides:

  • Recognition of the file names for the three standard preopened files is disabled only in POSIX mode.
  • Recognition of the other special file names is disabled if gawk is in compatibility mode (either --traditional or --posix; see Command-Line Options).
  • gawk always interprets these special file names. For example, using ‘/dev/fd/4’ for output actually writes on file descriptor 4, and not on a new file descriptor that is dup()ed from file descriptor 4. Most of the time this does not matter; however, it is important to not close any of the files related to file descriptors 0, 1, and 2. Doing so results in unpredictable behavior.

5.9 Closing Input and Output Redirections

If the same file name or the same shell command is used with getline more than once during the execution of an awk program (see Explicit Input with getline), the file is opened (or the command is executed) the first time only. At that time, the first record of input is read from that file or command. The next time the same file or command is used with getline, another record is read from it, and so on.

Similarly, when a file or pipe is opened for output, awk remembers the file name or command associated with it, and subsequent writes to the same file or command are appended to the previous writes. The file or pipe stays open until awk exits.

This implies that special steps are necessary in order to read the same file again from the beginning, or to rerun a shell command (rather than reading more output from the same command). The close() function makes these things possible:

close(filename)

or:

close(command)

The argument filename or command can be any expression. Its value must exactly match the string that was used to open the file or start the command (spaces and other “irrelevant” characters included). For example, if you open a pipe with this:

"sort -r names" | getline foo

then you must close it with this:

close("sort -r names")

Once this function call is executed, the next getline from that file or command, or the next print or printf to that file or command, reopens the file or reruns the command. Because the expression that you use to close a file or pipeline must exactly match the expression used to open the file or run the command, it is good practice to use a variable to store the file name or command. The previous example becomes the following:

sortcom = "sort -r names"
sortcom | getline foo
…
close(sortcom)

This helps avoid hard-to-find typographical errors in your awk programs. Here are some of the reasons for closing an output file:

  • To write a file and read it back later on in the same awk program. Close the file after writing it, then begin reading it with getline.
  • To write numerous files, successively, in the same awk program. If the files aren’t closed, eventually awk may exceed a system limit on the number of open files in one process. It is best to close each one when the program has finished writing it.
  • To make a command finish. When output is redirected through a pipe, the command reading the pipe normally continues to try to read input as long as the pipe is open. Often this means the command cannot really do its work until the pipe is closed. For example, if output is redirected to the mail program, the message is not actually sent until the pipe is closed.
  • To run the same program a second time, with the same arguments. This is not the same thing as giving more input to the first run!

    For example, suppose a program pipes output to the mail program. If it outputs several lines redirected to this pipe without closing it, they make a single message of several lines. By contrast, if the program closes the pipe after each line of output, then each line makes a separate message.

If you use more files than the system allows you to have open, gawk attempts to multiplex the available open files among your data files. gawk’s ability to do this depends upon the facilities of your operating system, so it may not always work. It is therefore both good practice and good portability advice to always use close() on your files when you are done with them. In fact, if you are using a lot of pipes, it is essential that you close commands when done. For example, consider something like this:

{
    …
    command = ("grep " $1 " /some/file | my_prog -q " $3)
    while ((command | getline) > 0) {
        process output of command
    }
    # need close(command) here
}

This example creates a new pipeline based on data in each record. Without the call to close() indicated in the comment, awk creates child processes to run the commands, until it eventually runs out of file descriptors for more pipelines.

Even though each command has finished (as indicated by the end-of-file return status from getline), the child process is not terminated;28 more importantly, the file descriptor for the pipe is not closed and released until close() is called or awk exits.

close() silently does nothing if given an argument that does not represent a file, pipe, or coprocess that was opened with a redirection. In such a case, it returns a negative value, indicating an error. In addition, gawk sets ERRNO to a string indicating the error.

Note also that ‘close(FILENAME)’ has no “magic” effects on the implicit loop that reads through the files named on the command line. It is, more likely, a close of a file that was never opened with a redirection, so awk silently does nothing, except return a negative value.

When using the ‘|&’ operator to communicate with a coprocess, it is occasionally useful to be able to close one end of the two-way pipe without closing the other. This is done by supplying a second argument to close(). As in any other call to close(), the first argument is the name of the command or special file used to start the coprocess. The second argument should be a string, with either of the values "to" or "from". Case does not matter. As this is an advanced feature, discussion is delayed until Two-Way Communications with Another Process, which describes it in more detail and gives an example.


5.9.1 Using close()’s Return Value

In many older versions of Unix awk, the close() function is actually a statement. (d.c.) It is a syntax error to try and use the return value from close():

command = "…"
command | getline info
retval = close(command)  # syntax error in many Unix awks

gawk treats close() as a function. The return value is −1 if the argument names something that was never opened with a redirection, or if there is a system problem closing the file or process. In these cases, gawk sets the predefined variable ERRNO to a string describing the problem.

In gawk, starting with version 4.2, when closing a pipe or coprocess (input or output), the return value is the exit status of the command, as described in Table 5.1.29 Otherwise, it is the return value from the system’s close() or fclose() C functions when closing input or output files, respectively. This value is zero if the close succeeds, or −1 if it fails. Recent versions of BWK awk also return the same values from close().

SituationReturn value from close()
Normal exit of commandCommand’s exit status
Death by signal of command256 + number of murderous signal
Death by signal of command with core dump512 + number of murderous signal
Some kind of error−1

Table 5.1: Return values from close() of a pipe

The POSIX standard is very vague; it says that close() returns zero on success and a nonzero value otherwise. In general, different implementations vary in what they report when closing pipes; thus, the return value cannot be used portably. (d.c.) In POSIX mode (see Command-Line Options), gawk just returns zero when closing a pipe.


5.10 Enabling Nonfatal Output

This section describes a gawk-specific feature.

In standard awk, output with print or printf to a nonexistent file, or some other I/O error (such as filling up the disk) is a fatal error.

$ gawk 'BEGIN { print "hi" > "/no/such/file" }'
error→ gawk: cmd. line:1: fatal: can't redirect to `/no/such/file' (No
error→ such file or directory)

gawk makes it possible to detect that an error has occurred, allowing you to possibly recover from the error, or at least print an error message of your choosing before exiting. You can do this in one of two ways:

  • For all output files, by assigning any value to PROCINFO["NONFATAL"].
  • On a per-file basis, by assigning any value to PROCINFO[filename, "NONFATAL"]. Here, filename is the name of the file to which you wish output to be nonfatal.

Once you have enabled nonfatal output, you must check ERRNO after every relevant print or printf statement to see if something went wrong. It is also a good idea to initialize ERRNO to zero before attempting the output. For example:

$ gawk '
> BEGIN {
>     PROCINFO["NONFATAL"] = 1
>     ERRNO = 0
>     print "hi" > "/no/such/file"
>     if (ERRNO) {
>         print("Output failed:", ERRNO) > "/dev/stderr"
>         exit 1
>     }
> }'
error→ Output failed: No such file or directory

Here, gawk did not produce a fatal error; instead it let the awk program code detect the problem and handle it.

This mechanism works also for standard output and standard error. For standard output, you may use PROCINFO["-", "NONFATAL"] or PROCINFO["/dev/stdout", "NONFATAL"]. For standard error, use PROCINFO["/dev/stderr", "NONFATAL"].

When attempting to open a TCP/IP socket (see Using gawk for Network Programming), gawk tries multiple times. The GAWK_SOCK_RETRIES environment variable (see Other Environment Variables) allows you to override gawk’s builtin default number of attempts. However, once nonfatal I/O is enabled for a given socket, gawk only retries once, relying on awk-level code to notice that there was a problem.


5.11 Summary

  • The print statement prints comma-separated expressions. Each expression is separated by the value of OFS and terminated by the value of ORS. OFMT provides the conversion format for numeric values for the print statement.
  • The printf statement provides finer-grained control over output, with format-control letters for different data types and various flags that modify the behavior of the format-control letters.
  • Output from both print and printf may be redirected to files, pipes, and coprocesses.
  • gawk provides special file names for access to standard input, output, and error, and for network communications.
  • Use close() to close open file, pipe, and coprocess redirections. For coprocesses, it is possible to close only one direction of the communications.
  • Normally errors with print or printf are fatal. gawk lets you make output errors be nonfatal either for all files or on a per-file basis. You must then check for errors after every relevant output statement.

5.12 Exercises

  1. Rewrite the program:
    awk 'BEGIN { print "Month Crates"
                 print "----- ------" }
               { print $1, "     ", $2 }' inventory-shipped
    

    from Output Separators, by using a new value of OFS.

  2. Use the printf statement to line up the headings and table data for the inventory-shipped example that was covered in The print Statement.
  3. What happens if you forget the double quotes when redirecting output, as follows:
    BEGIN { print "Serious error detected!" > /dev/stderr }
    

6 Expressions

Expressions are the basic building blocks of awk patterns and actions. An expression evaluates to a value that you can print, test, or pass to a function. Additionally, an expression can assign a new value to a variable or a field by using an assignment operator.

An expression can serve as a pattern or action statement on its own. Most other kinds of statements contain one or more expressions that specify the data on which to operate. As in other languages, expressions in awk can include variables, array references, constants, and function calls, as well as combinations of these with various operators.


6.1 Constants, Variables, and Conversions

Expressions are built up from values and the operations performed upon them. This section describes the elementary objects that provide the values used in expressions.


6.1.1 Constant Expressions

The simplest type of expression is the constant, which always has the same value. There are three types of constants: numeric, string, and regular expression.

Each is used in the appropriate context when you need a data value that isn’t going to change. Numeric constants can have different forms, but are internally stored in an identical manner.


6.1.1.1 Numeric and String Constants

A numeric constant stands for a number. This number can be an integer, a decimal fraction, or a number in scientific (exponential) notation.30 Here are some examples of numeric constants that all have the same value:

105
1.05e+2
1050e-1

A string constant consists of a sequence of characters enclosed in double quotation marks. For example:

"parrot"

represents the string whose contents are ‘parrot’. Strings in gawk can be of any length, and they can contain any of the possible eight-bit ASCII characters, including ASCII NUL (character code zero). Other awk implementations may have difficulty with some character codes.

Some languages allow you to continue long strings across multiple lines by ending the line with a backslash. For example in C:

#include <stdio.h>
int main()
{
    printf("hello, \
world\n");
    return 0;
}

In such a case, the C compiler removes both the backslash and the newline, producing a string as if it had been typed ‘"hello, world\n"’. This is useful when a single string needs to contain a large amount of text.

The POSIX standard says explicitly that newlines are not allowed inside string constants. And indeed, all awk implementations report an error if you try to do so. For example:

$ gawk 'BEGIN { print "hello, 
> world" }'
-| gawk: cmd. line:1: BEGIN { print "hello,
-| gawk: cmd. line:1:               ^ unterminated string
-| gawk: cmd. line:1: BEGIN { print "hello,
-| gawk: cmd. line:1:               ^ syntax error

Although POSIX doesn’t define what happens if you use an escaped newline, as in the previous C example, all known versions of awk allow you to do so. Unfortunately, what each one does with such a string varies. (d.c.) gawk, mawk, and the OpenSolaris POSIX awk (see Other Freely Available awk Implementations) elide the backslash and newline, as in C:

$ gawk 'BEGIN { print "hello, \
> world" }'
-| hello, world

In POSIX mode (see Command-Line Options), gawk does not allow escaped newlines. Otherwise, it behaves as just described.

BWK awk and BusyBox awk remove the backslash but leave the newline intact, as part of the string:

$ nawk 'BEGIN { print "hello, \
> world" }'
-| hello,
-| world

6.1.1.2 Octal and Hexadecimal Numbers

In awk, all numbers are in decimal (i.e., base 10). Many other programming languages allow you to specify numbers in other bases, often octal (base 8) and hexadecimal (base 16). In octal, the numbers go 0, 1, 2, 3, 4, 5, 6, 7, 10, 11, 12, and so on. Just as ‘11’ in decimal is 1 times 10 plus 1, so ‘11’ in octal is 1 times 8 plus 1. This equals 9 in decimal. In hexadecimal, there are 16 digits. Because the everyday decimal number system only has ten digits (‘0’–‘9’), the letters ‘a’ through ‘f’ represent the rest. (Case in the letters is usually irrelevant; hexadecimal ‘a’ and ‘A’ have the same value.) Thus, ‘11’ in hexadecimal is 1 times 16 plus 1, which equals 17 in decimal.

Just by looking at plain ‘11’, you can’t tell what base it’s in. So, in C, C++, and other languages derived from C, there is a special notation to signify the base. Octal numbers start with a leading ‘0’, and hexadecimal numbers start with a leading ‘0x’ or ‘0X’:

11

Decimal value 11

011

Octal 11, decimal value 9

0x11

Hexadecimal 11, decimal value 17

This example shows the difference:

$ gawk 'BEGIN { printf "%d, %d, %d\n", 011, 11, 0x11 }'
-| 9, 11, 17

Being able to use octal and hexadecimal constants in your programs is most useful when working with data that cannot be represented conveniently as characters or as regular numbers, such as binary data of various sorts.

gawk allows the use of octal and hexadecimal constants in your program text. However, such numbers in the input data are not treated differently; doing so by default would break old programs. (If you really need to do this, use the --non-decimal-data command-line option; see Allowing Nondecimal Input Data.) If you have octal or hexadecimal data, you can use the strtonum() function (see String-Manipulation Functions) to convert the data into a number. Most of the time, you will want to use octal or hexadecimal constants when working with the built-in bit-manipulation functions; see Bit-Manipulation Functions for more information.

Unlike in some early C implementations, ‘8’ and ‘9’ are not valid in octal constants. For example, gawk treats ‘018’ as decimal 18:

$ gawk 'BEGIN { print "021 is", 021 ; print 018 }'
-| 021 is 17
-| 18

Octal and hexadecimal source code constants are a gawk extension. If gawk is in compatibility mode (see Command-Line Options), they are not available.

A Constant’s Base Does Not Affect Its Value

Once a numeric constant has been converted internally into a number, gawk no longer remembers what the original form of the constant was; the internal value is always used. This has particular consequences for conversion of numbers to strings:

$ gawk 'BEGIN { printf "0x11 is <%s>\n", 0x11 }'
-| 0x11 is <17>

6.1.1.3 Regular Expression Constants

A regexp constant is a regular expression description enclosed in slashes, such as /^beginning and end$/. Most regexps used in awk programs are constant, but the ‘~’ and ‘!~’ matching operators can also match computed or dynamic regexps (which are typically just ordinary strings or variables that contain a regexp, but could be more complex expressions).


6.1.2 Using Regular Expression Constants

Regular expression constants consist of text describing a regular expression enclosed in slashes (such as /the +answer/). This section describes how such constants work in POSIX awk and gawk, and then goes on to describe strongly typed regexp constants, which are a gawk extension.


6.1.2.1 Standard Regular Expression Constants

When used on the righthand side of the ‘~’ or ‘!~’ operators, a regexp constant merely stands for the regexp that is to be matched. However, regexp constants (such as /foo/) may be used like simple expressions. When a regexp constant appears by itself, it has the same meaning as if it appeared in a pattern (i.e., ‘($0 ~ /foo/)’). (d.c.) See Expressions as Patterns. This means that the following two code segments:

if ($0 ~ /barfly/ || $0 ~ /camelot/)
    print "found"

and:

if (/barfly/ || /camelot/)
    print "found"

are exactly equivalent. One rather bizarre consequence of this rule is that the following Boolean expression is valid, but does not do what its author probably intended:

# Note that /foo/ is on the left of the ~
if (/foo/ ~ $1) print "found foo"

This code is “obviously” testing $1 for a match against the regexp /foo/. But in fact, the expression ‘/foo/ ~ $1’ really means ‘($0 ~ /foo/) ~ $1’. In other words, first match the input record against the regexp /foo/. The result is either zero or one, depending upon the success or failure of the match. That result is then matched against the first field in the record. Because it is unlikely that you would ever really want to make this kind of test, gawk issues a warning when it sees this construct in a program. Another consequence of this rule is that the assignment statement:

matches = /foo/

assigns either zero or one to the variable matches, depending upon the contents of the current input record.

Constant regular expressions are also used as the first argument for the gensub(), sub(), and gsub() functions, as the second argument of the match() function, and as the third argument of the split() and patsplit() functions (see String-Manipulation Functions). Modern implementations of awk, including gawk, allow the third argument of split() to be a regexp constant, but some older implementations do not. (d.c.) Because some built-in functions accept regexp constants as arguments, confusion can arise when attempting to use regexp constants as arguments to user-defined functions (see User-Defined Functions). For example:

function mysub(pat, repl, str, global)
{
    if (global)
        gsub(pat, repl, str)
    else
        sub(pat, repl, str)
    return str
}
{
    …
    text = "hi! hi yourself!"
    mysub(/hi/, "howdy", text, 1)
    …
}

In this example, the programmer wants to pass a regexp constant to the user-defined function mysub(), which in turn passes it on to either sub() or gsub(). However, what really happens is that the pat parameter is assigned a value of either one or zero, depending upon whether or not $0 matches /hi/. gawk issues a warning when it sees a regexp constant used as a parameter to a user-defined function, because passing a truth value in this way is probably not what was intended.


6.1.2.2 Strongly Typed Regexp Constants

This section describes a gawk-specific feature.

As we saw in the previous section, regexp constants (/…/) hold a strange position in the awk language. In most contexts, they act like an expression: ‘$0 ~ /…/’. In other contexts, they denote only a regexp to be matched. In no case are they really a “first class citizen” of the language. That is, you cannot define a scalar variable whose type is “regexp” in the same sense that you can define a variable to be a number or a string:

num = 42        Numeric variable
str = "hi"      String variable
re = /foo/      Wrong! re is the result of $0 ~ /foo/

For a number of more advanced use cases, it would be nice to have regexp constants that are strongly typed; in other words, that denote a regexp useful for matching, and not an expression.

gawk provides this feature. A strongly typed regexp constant looks almost like a regular regexp constant, except that it is preceded by an ‘@’ sign:

re = @/foo/     Regexp variable

Strongly typed regexp constants cannot be used everywhere that a regular regexp constant can, because this would make the language even more confusing. Instead, you may use them only in certain contexts:

  • On the righthand side of the ‘~’ and ‘!~’ operators: ‘some_var ~ @/foo/’ (see How to Use Regular Expressions).
  • In the case part of a switch statement (see The switch Statement).
  • As an argument to one of the built-in functions that accept regexp constants: gensub(), gsub(), match(), patsplit(), split(), and sub() (see String-Manipulation Functions).
  • As a parameter in a call to a user-defined function (see User-Defined Functions).
  • As the return value of a user-defined function.
  • On the righthand side of an assignment to a variable: ‘some_var = @/foo/’. In this case, the type of some_var is regexp. Additionally, some_var can be used with ‘~’ and ‘!~’, passed to one of the built-in functions listed above, or passed as a parameter to a user-defined function.

You may use the -v option (see Command-Line Options) to assign a strongly-typed regexp constant to a variable on the command line, like so:

gawk -v pattern='@/something(interesting)+/' …

You may also make such assignments as regular command-line arguments (see Other Command-Line Arguments).

You may use the typeof() built-in function (see Getting Type Information) to determine if a variable or function parameter is a regexp variable.

The true power of this feature comes from the ability to create variables that have regexp type. Such variables can be passed on to user-defined functions, without the confusing aspects of computed regular expressions created from strings or string constants. They may also be passed through indirect function calls (see Indirect Function Calls) and on to the built-in functions that accept regexp constants.

When used in numeric conversions, strongly typed regexp variables convert to zero. When used in string conversions, they convert to the string value of the original regexp text.

There is an additional, interesting corner case. When used as the third argument to sub() or gsub(), they retain their type. Thus, if you have something like this:

re = @/don't panic/
sub(/don't/, "do", re)
print typeof(re), re

then re retains its type, but now attempts to match the string ‘do panic’. This provides a (very indirect) way to create regexp-typed variables at runtime.


6.1.3 Variables

Variables are ways of storing values at one point in your program for use later in another part of your program. They can be manipulated entirely within the program text, and they can also be assigned values on the awk command line.


6.1.3.1 Using Variables in a Program

Variables let you give names to values and refer to them later. Variables have already been used in many of the examples. The name of a variable must be a sequence of letters, digits, or underscores, and it may not begin with a digit. Here, a letter is any one of the 52 upper- and lowercase English letters. Other characters that may be defined as letters in non-English locales are not valid in variable names. Case is significant in variable names; a and A are distinct variables.

A variable name is a valid expression by itself; it represents the variable’s current value. Variables are given new values with assignment operators, increment operators, and decrement operators (see Assignment Expressions). In addition, the sub() and gsub() functions can change a variable’s value, and the match(), split(), and patsplit() functions can change the contents of their array parameters (see String-Manipulation Functions).

A few variables have special built-in meanings, such as FS (the field separator) and NF (the number of fields in the current input record). See Predefined Variables for a list of the predefined variables. These predefined variables can be used and assigned just like all other variables, but their values are also used or changed automatically by awk. All predefined variables’ names are entirely uppercase.

Variables in awk can be assigned either numeric or string values. The kind of value a variable holds can change over the life of a program. By default, variables are initialized to the empty string, which is zero if converted to a number. There is no need to explicitly initialize a variable in awk, which is what you would do in C and in most other traditional languages.


6.1.3.2 Assigning Variables on the Command Line

Any awk variable can be set by including a variable assignment among the arguments on the command line when awk is invoked (see Other Command-Line Arguments). Such an assignment has the following form:

variable=text

With it, a variable is set either at the beginning of the awk run or in between input files. When the assignment is preceded with the -v option, as in the following:

-v variable=text

the variable is set at the very beginning, even before the BEGIN rules execute. The -v option and its assignment must precede all the file name arguments, as well as the program text. (See Command-Line Options for more information about the -v option.) Otherwise, the variable assignment is performed at a time determined by its position among the input file arguments—after the processing of the preceding input file argument. For example:

awk '{ print $n }' n=4 inventory-shipped n=2 mail-list

prints the value of field number n for all input records. Before the first file is read, the command line sets the variable n equal to four. This causes the fourth field to be printed in lines from inventory-shipped. After the first file has finished, but before the second file is started, n is set to two, so that the second field is printed in lines from mail-list:

$ awk '{ print $n }' n=4 inventory-shipped n=2 mail-list
-| 15
-| 24
…
-| 555-5553
-| 555-3412
…

Command-line arguments are made available for explicit examination by the awk program in the ARGV array (see Using ARGC and ARGV). awk processes the values of command-line assignments for escape sequences (see Escape Sequences). (d.c.)

Normally, variables assigned on the command line (with or without the -v option) are treated as strings. When such variables are used as numbers, awk’s normal automatic conversion of strings to numbers takes place, and everything “just works.”

However, gawk supports variables whose types are “regexp”. You can assign variables of this type using the following syntax:

gawk -v 're1=@/foo|bar/' '…' /path/to/file1 're2=@/baz|quux/' /path/to/file2

Strongly typed regexps are an advanced feature (see Strongly Typed Regexp Constants). We mention them here only for completeness.


6.1.4 Conversion of Strings and Numbers

Number-to-string and string-to-number conversion are generally straightforward. There can be subtleties to be aware of; this section discusses this important facet of awk.


6.1.4.1 How awk Converts Between Strings and Numbers

Strings are converted to numbers and numbers are converted to strings, if the context of the awk program demands it. For example, if the value of either foo or bar in the expression ‘foo + bar’ happens to be a string, it is converted to a number before the addition is performed. If numeric values appear in string concatenation, they are converted to strings. Consider the following:

two = 2; three = 3
print (two three) + 4

This prints the (numeric) value 27. The numeric values of the variables two and three are converted to strings and concatenated together. The resulting string is converted back to the number 23, to which 4 is then added.

If, for some reason, you need to force a number to be converted to a string, concatenate that number with the empty string, "". To force a string to be converted to a number, add zero to that string. A string is converted to a number by interpreting any numeric prefix of the string as numerals: "2.5" converts to 2.5, "1e3" converts to 1,000, and "25fix" has a numeric value of 25. Strings that can’t be interpreted as valid numbers convert to zero.

The exact manner in which numbers are converted into strings is controlled by the awk predefined variable CONVFMT (see Predefined Variables). Numbers are converted using the sprintf() function with CONVFMT as the format specifier (see String-Manipulation Functions).

CONVFMT’s default value is "%.6g", which creates a value with at most six significant digits. For some applications, you might want to change it to specify more precision. On most modern machines, 17 digits is usually enough to capture a floating-point number’s value exactly.31

Strange results can occur if you set CONVFMT to a string that doesn’t tell sprintf() how to format floating-point numbers in a useful way. For example, if you forget the ‘%’ in the format, awk converts all numbers to the same constant string.

As a special case, if a number is an integer, then the result of converting it to a string is always an integer, no matter what the value of CONVFMT may be. Given the following code fragment:

CONVFMT = "%2.2f"
a = 12
b = a ""

b has the value "12", not "12.00". (d.c.)

Pre-POSIX awk Used OFMT for String Conversion

Prior to the POSIX standard, awk used the value of OFMT for converting numbers to strings. OFMT specifies the output format to use when printing numbers with print. CONVFMT was introduced in order to separate the semantics of conversion from the semantics of printing. Both CONVFMT and OFMT have the same default value: "%.6g". In the vast majority of cases, old awk programs do not change their behavior. See The print Statement for more information on the print statement.


6.1.4.2 Locales Can Influence Conversion

Where you are can matter when it comes to converting between numbers and strings. The local character set and language—the locale—can affect numeric formats. In particular, for awk programs, it affects the decimal point character and the thousands-separator character. The "C" locale, and most English-language locales, use the period character (‘.’) as the decimal point and don’t have a thousands separator. However, many (if not most) European and non-English locales use the comma (‘,’) as the decimal point character. European locales often use either a space or a period as the thousands separator, if they have one.

The POSIX standard says that awk always uses the period as the decimal point when reading the awk program source code, and for command-line variable assignments (see Other Command-Line Arguments). However, when interpreting input data, for print and printf output, and for number-to-string conversion, the local decimal point character is used. (d.c.) In all cases, numbers in source code and in input data cannot have a thousands separator. Here are some examples indicating the difference in behavior, on a GNU/Linux system:

$ export POSIXLY_CORRECT=1                        Force POSIX behavior
$ gawk 'BEGIN { printf "%g\n", 3.1415927 }'
-| 3.14159
$ LC_ALL=en_DK.utf-8 gawk 'BEGIN { printf "%g\n", 3.1415927 }'
-| 3,14159
$ echo 4,321 | gawk '{ print $1 + 1 }'
-| 5
$ echo 4,321 | LC_ALL=en_DK.utf-8 gawk '{ print $1 + 1 }'
-| 5,321

The en_DK.utf-8 locale is for English in Denmark, where the comma acts as the decimal point separator. In the normal "C" locale, gawk treats ‘4,321’ as 4, while in the Danish locale, it’s treated as the full number including the fractional part, 4.321.

Some earlier versions of gawk fully complied with this aspect of the standard. However, many users in non-English locales complained about this behavior, because their data used a period as the decimal point, so the default behavior was restored to use a period as the decimal point character. You can use the --use-lc-numeric option (see Command-Line Options) to force gawk to use the locale’s decimal point character. (gawk also uses the locale’s decimal point character when in POSIX mode, either via --posix or the POSIXLY_CORRECT environment variable, as shown previously.)

Table 6.1 describes the cases in which the locale’s decimal point character is used and when a period is used. Some of these features have not been described yet.

FeatureDefault--posix or --use-lc-numeric
%'gUse localeUse locale
%gUse periodUse locale
InputUse periodUse locale
strtonum()Use periodUse locale

Table 6.1: Locale decimal point versus a period

Finally, modern-day formal standards and the IEEE standard floating-point representation can have an unusual but important effect on the way gawk converts some special string values to numbers. The details are presented in Standards Versus Existing Practice.


6.2 Operators: Doing Something with Values

This section introduces the operators that make use of the values provided by constants and variables.


6.2.1 Arithmetic Operators

The awk language uses the common arithmetic operators when evaluating expressions. All of these arithmetic operators follow normal precedence rules and work as you would expect them to.

The following example uses a file named grades, which contains a list of student names as well as three test scores per student (it’s a small class):

Pat   100 97 58
Sandy  84 72 93
Chris  72 92 89

This program takes the file grades and prints the average of the scores:

$ awk '{ sum = $2 + $3 + $4 ; avg = sum / 3
>        print $1, avg }' grades
-| Pat 85
-| Sandy 83
-| Chris 84.3333

The following list provides the arithmetic operators in awk, in order from the highest precedence to the lowest:

x ^ y
x ** y

Exponentiation; x raised to the y power. ‘2 ^ 3’ has the value eight; the character sequence ‘**’ is equivalent to ‘^’. (c.e.)

- x

Negation.

+ x

Unary plus; the expression is converted to a number.

x * y

Multiplication.

x / y

Division; because all numbers in awk are floating-point numbers, the result is not rounded to an integer—‘3 / 4’ has the value 0.75. (It is a common mistake, especially for C programmers, to forget that all numbers in awk are floating point, and that division of integer-looking constants produces a real number, not an integer.)

x % y

Remainder; further discussion is provided in the text, just after this list.

x + y

Addition.

x - y

Subtraction.

Unary plus and minus have the same precedence, the multiplication operators all have the same precedence, and addition and subtraction have the same precedence.

When computing the remainder of ‘x % y’, the quotient is rounded toward zero to an integer and multiplied by y. This result is subtracted from x; this operation is sometimes known as “trunc-mod.” The following relation always holds:

b * int(a / b) + (a % b) == a

One possibly undesirable effect of this definition of remainder is that ‘x % y’ is negative if x is negative. Thus:

-17 % 8 = -1

This definition is compliant with the POSIX standard, which says that the % operator produces results equivalent to using the standard C fmod() function, and that function in turn works as just described.

In other awk implementations, the signedness of the remainder may be machine-dependent.

NOTE: The POSIX standard only specifies the use of ‘^’ for exponentiation. For maximum portability, do not use the ‘**’ operator.


6.2.2 String Concatenation

It seemed like a good idea at the time.

Brian Kernighan

There is only one string operation: concatenation. It does not have a specific operator to represent it. Instead, concatenation is performed by writing expressions next to one another, with no operator. For example:

$ awk '{ print "Field number one: " $1 }' mail-list
-| Field number one: Amelia
-| Field number one: Anthony
…

Without the space in the string constant after the ‘:’, the line runs together. For example:

$ awk '{ print "Field number one:" $1 }' mail-list
-| Field number one:Amelia
-| Field number one:Anthony
…

Because string concatenation does not have an explicit operator, it is often necessary to ensure that it happens at the right time by using parentheses to enclose the items to concatenate. For example, you might expect that the following code fragment concatenates file and name:

file = "file"
name = "name"
print "something meaningful" > file name

This produces a syntax error with some versions of Unix awk.32 It is necessary to use the following:

print "something meaningful" > (file name)

Parentheses should be used around concatenation in all but the most common contexts, such as on the righthand side of ‘=’. Be careful about the kinds of expressions used in string concatenation. In particular, the order of evaluation of expressions used for concatenation is undefined in the awk language. Consider this example:

BEGIN {
    a = "don't"
    print (a " " (a = "panic"))
}

It is not defined whether the second assignment to a happens before or after the value of a is retrieved for producing the concatenated value. The result could be either ‘don't panic’, or ‘panic panic’.

The precedence of concatenation, when mixed with other operators, is often counter-intuitive. Consider this example:

$ awk 'BEGIN { print -12 " " -24 }'
-| -12-24

This “obviously” is concatenating −12, a space, and −24. But where did the space disappear to? The answer lies in the combination of operator precedences and awk’s automatic conversion rules. To get the desired result, write the program this way:

$ awk 'BEGIN { print -12 " " (-24) }'
-| -12 -24

This forces awk to treat the ‘-’ on the ‘-24’ as unary. Otherwise, it’s parsed as follows:

    −12 (" " − 24)
⇒ −12 (0 − 24)
⇒ −12 (−24)
⇒ −12−24

As mentioned earlier, when mixing concatenation with other operators, parenthesize. Otherwise, you’re never quite sure what you’ll get.


6.2.3 Assignment Expressions

An assignment is an expression that stores a (usually different) value into a variable. For example, let’s assign the value one to the variable z:

z = 1

After this expression is executed, the variable z has the value one. Whatever old value z had before the assignment is forgotten.

Assignments can also store string values. For example, the following stores the value "this food is good" in the variable message:

thing = "food"
predicate = "good"
message = "this " thing " is " predicate

This also illustrates string concatenation. The ‘=’ sign is called an assignment operator. It is the simplest assignment operator because the value of the righthand operand is stored unchanged. Most operators (addition, concatenation, and so on) have no effect except to compute a value. If the value isn’t used, there’s no reason to use the operator. An assignment operator is different; it does produce a value, but even if you ignore it, the assignment still makes itself felt through the alteration of the variable. We call this a side effect.

The lefthand operand of an assignment need not be a variable (see Variables); it can also be a field (see Changing the Contents of a Field) or an array element (see Arrays in awk). These are all called lvalues, which means they can appear on the lefthand side of an assignment operator. The righthand operand may be any expression; it produces the new value that the assignment stores in the specified variable, field, or array element. (Such values are called rvalues.)

It is important to note that variables do not have permanent types. A variable’s type is simply the type of whatever value was last assigned to it. In the following program fragment, the variable foo has a numeric value at first, and a string value later on:

foo = 1
print foo
foo = "bar"
print foo

When the second assignment gives foo a string value, the fact that it previously had a numeric value is forgotten.

String values that do not begin with a digit have a numeric value of zero. After executing the following code, the value of foo is five:

foo = "a string"
foo = foo + 5

NOTE: Using a variable as a number and then later as a string can be confusing and is poor programming style. The previous two examples illustrate how awk works, not how you should write your programs!

An assignment is an expression, so it has a value—the same value that is assigned. Thus, ‘z = 1’ is an expression with the value one. One consequence of this is that you can write multiple assignments together, such as:

x = y = z = 5

This example stores the value five in all three variables (x, y, and z). It does so because the value of ‘z = 5’, which is five, is stored into y and then the value of ‘y = z = 5’, which is five, is stored into x.

Assignments may be used anywhere an expression is called for. For example, it is valid to write ‘x != (y = 1)’ to set y to one, and then test whether x equals one. But this style tends to make programs hard to read; such nesting of assignments should be avoided, except perhaps in a one-shot program.

Aside from ‘=’, there are several other assignment operators that do arithmetic with the old value of the variable. For example, the operator ‘+=’ computes a new value by adding the righthand value to the old value of the variable. Thus, the following assignment adds five to the value of foo:

foo += 5

This is equivalent to the following:

foo = foo + 5

Use whichever makes the meaning of your program clearer.

There are situations where using ‘+=’ (or any assignment operator) is not the same as simply repeating the lefthand operand in the righthand expression. For example:

# Thanks to Pat Rankin for this example
BEGIN  {
    foo[rand()] += 5
    for (x in foo)
       print x, foo[x]
    bar[rand()] = bar[rand()] + 5
    for (x in bar)
       print x, bar[x]
}

The indices of bar are practically guaranteed to be different, because rand() returns different values each time it is called. (Arrays and the rand() function haven’t been covered yet. See Arrays in awk, and see Numeric Functions for more information.) This example illustrates an important fact about assignment operators: the lefthand expression is only evaluated once.

It is up to the implementation as to which expression is evaluated first, the lefthand or the righthand. Consider this example:

i = 1
a[i += 2] = i + 1

The value of a[3] could be either two or four.

Table 6.2 lists the arithmetic assignment operators. In each case, the righthand operand is an expression whose value is converted to a number.

OperatorEffect
lvalue += incrementAdd increment to the value of lvalue.
lvalue -= decrementSubtract decrement from the value of lvalue.
lvalue *= coefficientMultiply the value of lvalue by coefficient.
lvalue /= divisorDivide the value of lvalue by divisor.
lvalue %= modulusSet lvalue to its remainder by modulus.
lvalue ^= powerRaise lvalue to the power power.
lvalue **= powerRaise lvalue to the power power. (c.e.)

Table 6.2: Arithmetic assignment operators

NOTE: Only the ‘^=’ operator is specified by POSIX. For maximum portability, do not use the ‘**=’ operator.

Syntactic Ambiguities Between ‘/=’ and Regular Expressions

There is a syntactic ambiguity between the /= assignment operator and regexp constants whose first character is an ‘=’. (d.c.) This is most notable in some commercial awk versions. For example:

$ awk /==/ /dev/null
error→ awk: syntax error at source line 1
error→  context is
error→         >>> /= <<<
error→ awk: bailing out at source line 1

A workaround is:

awk '/[=]=/' /dev/null

gawk does not have this problem; BWK awk and mawk also do not.


6.2.4 Increment and Decrement Operators

Increment and decrement operators increase or decrease the value of a variable by one. An assignment operator can do the same thing, so the increment operators add no power to the awk language; however, they are convenient abbreviations for very common operations.

The operator used for adding one is written ‘++’. It can be used to increment a variable either before or after taking its value. To pre-increment a variable v, write ‘++v’. This adds one to the value of v—that new value is also the value of the expression. (The assignment expression ‘v += 1’ is completely equivalent.) Writing the ‘++’ after the variable specifies post-increment. This increments the variable value just the same; the difference is that the value of the increment expression itself is the variable’s old value. Thus, if foo has the value four, then the expression ‘foo++’ has the value four, but it changes the value of foo to five. In other words, the operator returns the old value of the variable, but with the side effect of incrementing it.

The post-increment ‘foo++’ is nearly the same as writing ‘(foo += 1) - 1’. It is not perfectly equivalent because all numbers in awk are floating point—in floating point, ‘foo + 1 - 1’ does not necessarily equal foo. But the difference is minute as long as you stick to numbers that are fairly small (less than 1012).

Fields and array elements are incremented just like variables. (Use ‘$(i++)’ when you want to do a field reference and a variable increment at the same time. The parentheses are necessary because of the precedence of the field reference operator ‘$’.)

The decrement operator ‘--’ works just like ‘++’, except that it subtracts one instead of adding it. As with ‘++’, it can be used before the lvalue to pre-decrement or after it to post-decrement. Following is a summary of increment and decrement expressions:

++lvalue

Increment lvalue, returning the new value as the value of the expression.

lvalue++

Increment lvalue, returning the old value of lvalue as the value of the expression.

--lvalue

Decrement lvalue, returning the new value as the value of the expression. (This expression is like ‘++lvalue’, but instead of adding, it subtracts.)

lvalue--

Decrement lvalue, returning the old value of lvalue as the value of the expression. (This expression is like ‘lvalue++’, but instead of adding, it subtracts.)

Operator Evaluation Order

Doctor, it hurts when I do this!
Then don’t do that!

Groucho Marx

What happens for something like the following?

b = 6
print b += b++

Or something even stranger?

b = 6
b += ++b + b++
print b

In other words, when do the various side effects prescribed by the postfix operators (‘b++’) take effect? When side effects happen is implementation-defined. In other words, it is up to the particular version of awk. The result for the first example may be 12 or 13, and for the second, it may be 22 or 23.

In short, doing things like this is not recommended and definitely not anything that you can rely upon for portability. You should avoid such things in your own programs.


6.3 Truth Values and Conditions

In certain contexts, expression values also serve as “truth values”; i.e., they determine what should happen next as the program runs. This section describes how awk defines “true” and “false” and how values are compared.


6.3.1 True and False in awk

Many programming languages have a special representation for the concepts of “true” and “false.” Such languages usually use the special constants true and false, or perhaps their uppercase equivalents. However, awk is different. It borrows a very simple concept of true and false from C. In awk, any nonzero numeric value or any nonempty string value is true. Any other value (zero or the null string, "") is false. The following program prints ‘A strange truth value’ three times:

BEGIN {
   if (3.1415927)
       print "A strange truth value"
   if ("Four Score And Seven Years Ago")
       print "A strange truth value"
   if (j = 57)
       print "A strange truth value"
}

There is a surprising consequence of the “nonzero or non-null” rule: the string constant "0" is actually true, because it is non-null. (d.c.)


6.3.2 Variable Typing and Comparison Expressions

The Guide is definitive. Reality is frequently inaccurate.

Douglas Adams, The Hitchhiker’s Guide to the Galaxy

Unlike in other programming languages, in awk variables do not have a fixed type. Instead, they can be either a number or a string, depending upon the value that is assigned to them. We look now at how variables are typed, and how awk compares variables.


6.3.2.1 String Type versus Numeric Type

Scalar objects in awk (variables, array elements, and fields) are dynamically typed. This means their type can change as the program runs, from untyped before any use,33 to string or number, and then from string to number or number to string, as the program progresses. (gawk also provides regexp-typed scalars, but let’s ignore that for now; see Strongly Typed Regexp Constants.)

You can’t do much with untyped variables, other than tell that they are untyped. The following program tests a against "" and 0; the test succeeds when a has never been assigned a value. It also uses the built-in typeof() function (not presented yet; see Getting Type Information) to show a’s type:

$ gawk 'BEGIN { print (a == "" && a == 0 ?
> "a is untyped" : "a has a type!") ; print typeof(a) }'
-| a is untyped
-| unassigned

A scalar has numeric type when assigned a numeric value, such as from a numeric constant, or from another scalar with numeric type:

$ gawk 'BEGIN { a = 42 ; print typeof(a)
> b = a ; print typeof(b) }'
number
number

Similarly, a scalar has string type when assigned a string value, such as from a string constant, or from another scalar with string type:

$ gawk 'BEGIN { a = "forty two" ; print typeof(a)
> b = a ; print typeof(b) }'
string
string

So far, this is all simple and straightforward. What happens, though, when awk has to process data from a user? Let’s start with field data. What should the following command produce as output?

echo hello | awk '{ printf("%s %s < 42\n", $1,
                           ($1 < 42 ? "is" : "is not")) }'

Since ‘hello’ is alphabetic data, awk can only do a string comparison. Internally, it converts 42 into "42" and compares the two string values "hello" and "42". Here’s the result:

$ echo hello | awk '{ printf("%s %s < 42\n", $1,
>                            ($1 < 42 ? "is" : "is not")) }'
-| hello is not < 42

However, what happens when data from a user looks like a number? On the one hand, in reality, the input data consists of characters, not binary numeric values. But, on the other hand, the data looks numeric, and awk really ought to treat it as such. And indeed, it does:

$ echo 37 | awk '{ printf("%s %s < 42\n", $1,
>                         ($1 < 42 ? "is" : "is not")) }'
-| 37 is < 42

Here are the rules for when awk treats data as a number, and for when it treats data as a string.

The POSIX standard uses the term numeric string for input data that looks numeric. The ‘37’ in the previous example is a numeric string. So what is the type of a numeric string? Answer: numeric.

The type of a variable is important because the types of two variables determine how they are compared. Variable typing follows these definitions and rules:

  • A numeric constant or the result of a numeric operation has the numeric attribute.
  • A string constant or the result of a string operation has the string attribute.
  • Fields, getline input, FILENAME, ARGV elements, ENVIRON elements, and the elements of an array created by match(), split(), and patsplit() that are numeric strings have the strnum attribute.34 Otherwise, they have the string attribute. Uninitialized variables also have the strnum attribute.
  • Attributes propagate across assignments but are not changed by any use.

The last rule is particularly important. In the following program, a has numeric type, even though it is later used in a string operation:

BEGIN {
     a = 12.345
     b = a " is a cute number"
     print b
}

When two operands are compared, either string comparison or numeric comparison may be used. This depends upon the attributes of the operands, according to the following symmetric matrix:

        +----------------------------------------------
        |       STRING          NUMERIC         STRNUM
--------+----------------------------------------------
        |
STRING  |       string          string          string
        |
NUMERIC |       string          numeric         numeric
        |
STRNUM  |       string          numeric         numeric
--------+----------------------------------------------

The basic idea is that user input that looks numeric—and only user input—should be treated as numeric, even though it is actually made of characters and is therefore also a string. Thus, for example, the string constant " +3.14", when it appears in program source code, is a string—even though it looks numeric—and is never treated as a number for comparison purposes.

In short, when one operand is a “pure” string, such as a string constant, then a string comparison is performed. Otherwise, a numeric comparison is performed. (The primary difference between a number and a strnum is that for strnums gawk preserves the original string value that the scalar had when it came in.)

This point bears additional emphasis: Input that looks numeric is numeric. All other input is treated as strings.

Thus, the six-character input string ‘ +3.14 receives the strnum attribute. In contrast, the eight characters " +3.14" appearing in program text comprise a string constant. The following examples print ‘1’ when the comparison between the two different constants is true, and ‘0’ otherwise:

$ echo ' +3.14' | awk '{ print($0 == " +3.14") }'    True
-| 1
$ echo ' +3.14' | awk '{ print($0 == "+3.14") }'     False
-| 0
$ echo ' +3.14' | awk '{ print($0 == "3.14") }'      False
-| 0
$ echo ' +3.14' | awk '{ print($0 == 3.14) }'        True
-| 1
$ echo ' +3.14' | awk '{ print($1 == " +3.14") }'    False
-| 0
$ echo ' +3.14' | awk '{ print($1 == "+3.14") }'     True
-| 1
$ echo ' +3.14' | awk '{ print($1 == "3.14") }'      False
-| 0
$ echo ' +3.14' | awk '{ print($1 == 3.14) }'        True
-| 1

You can see the type of an input field (or other user input) using typeof():

$ echo hello 37 | gawk '{ print typeof($1), typeof($2) }'
-| string strnum

6.3.2.2 Comparison Operators

Comparison expressions compare strings or numbers for relationships such as equality. They are written using relational operators, which are a superset of those in C. Table 6.3 describes them.

ExpressionResult
x < yTrue if x is less than y
x <= yTrue if x is less than or equal to y
x > yTrue if x is greater than y
x >= yTrue if x is greater than or equal to y
x == yTrue if x is equal to y
x != yTrue if x is not equal to y
x ~ yTrue if the string x matches the regexp denoted by y
x !~ yTrue if the string x does not match the regexp denoted by y
subscript in arrayTrue if the array array has an element with the subscript subscript

Table 6.3: Relational operators

Comparison expressions have the value one if true and zero if false. When comparing operands of mixed types, numeric operands are converted to strings using the value of CONVFMT (see Conversion of Strings and Numbers).

Strings are compared by comparing the first character of each, then the second character of each, and so on. Thus, "10" is less than "9". If there are two strings where one is a prefix of the other, the shorter string is less than the longer one. Thus, "abc" is less than "abcd".

It is very easy to accidentally mistype the ‘==’ operator and leave off one of the ‘=’ characters. The result is still valid awk code, but the program does not do what is intended:

if (a = b)   # oops! should be a == b
   …
else
   …

Unless b happens to be zero or the null string, the if part of the test always succeeds. Because the operators are so similar, this kind of error is very difficult to spot when scanning the source code.

The following list of expressions illustrates the kinds of comparisons awk performs, as well as what the result of each comparison is:

1.5 <= 2.0

Numeric comparison (true)

"abc" >= "xyz"

String comparison (false)

1.5 != " +2"

String comparison (true)

"1e2" < "3"

String comparison (true)

a = 2; b = "2"
a == b

String comparison (true)

a = 2; b = " +2"
a == b

String comparison (false)

In this example:

$ echo 1e2 3 | awk '{ print ($1 < $2) ? "true" : "false" }'
-| false

the result is ‘false’ because both $1 and $2 are user input. They are numeric strings—therefore both have the strnum attribute, dictating a numeric comparison. The purpose of the comparison rules and the use of numeric strings is to attempt to produce the behavior that is “least surprising,” while still “doing the right thing.”

String comparisons and regular expression comparisons are very different. For example:

x == "foo"

has the value one, or is true if the variable x is precisely ‘foo’. By contrast:

x ~ /foo/

has the value one if x contains ‘foo’, such as "Oh, what a fool am I!".

The righthand operand of the ‘~’ and ‘!~’ operators may be either a regexp constant (//) or an ordinary expression. In the latter case, the value of the expression as a string is used as a dynamic regexp (see How to Use Regular Expressions; also see Using Dynamic Regexps).

A constant regular expression in slashes by itself is also an expression. /regexp/ is an abbreviation for the following comparison expression:

$0 ~ /regexp/

One special place where /foo/ is not an abbreviation for ‘$0 ~ /foo/’ is when it is the righthand operand of ‘~’ or ‘!~’. See Using Regular Expression Constants, where this is discussed in more detail.


6.3.2.3 String Comparison Based on Locale Collating Order

The POSIX standard used to say that all string comparisons are performed based on the locale’s collating order. This is the order in which characters sort, as defined by the locale (for more discussion, see Where You Are Makes a Difference). This order is usually very different from the results obtained when doing straight byte-by-byte comparison.35

Because this behavior differs considerably from existing practice, gawk only implemented it when in POSIX mode (see Command-Line Options). Here is an example to illustrate the difference, in an en_US.UTF-8 locale:

$ gawk 'BEGIN { printf("ABC < abc = %s\n",
>                     ("ABC" < "abc" ? "TRUE" : "FALSE")) }'
-| ABC < abc = TRUE
$ gawk --posix 'BEGIN { printf("ABC < abc = %s\n",
>                             ("ABC" < "abc" ? "TRUE" : "FALSE")) }'
-| ABC < abc = FALSE

Fortunately, as of August 2016, comparison based on locale collating order is no longer required for the == and != operators.36 However, comparison based on locales is still required for <, <=, >, and >=. POSIX thus recommends as follows:

Since the == operator checks whether strings are identical, not whether they collate equally, applications needing to check whether strings collate equally can use:

a <= b && a >= b

As of version 4.2, gawk continues to use locale collating order for <, <=, >, and >= only in POSIX mode.


6.3.3 Boolean Expressions

A Boolean expression is a combination of comparison expressions or matching expressions, using the Boolean operators “or” (‘||’), “and” (‘&&’), and “not” (‘!’), along with parentheses to control nesting. The truth value of the Boolean expression is computed by combining the truth values of the component expressions. Boolean expressions are also referred to as logical expressions. The terms are equivalent.

Boolean expressions can be used wherever comparison and matching expressions can be used. They can be used in if, while, do, and for statements (see Control Statements in Actions). They have numeric values (one if true, zero if false) that come into play if the result of the Boolean expression is stored in a variable or used in arithmetic.

In addition, every Boolean expression is also a valid pattern, so you can use one as a pattern to control the execution of rules. The Boolean operators are:

boolean1 && boolean2

True if both boolean1 and boolean2 are true. For example, the following statement prints the current input record if it contains both ‘edu’ and ‘li’:

if ($0 ~ /edu/ && $0 ~ /li/) print

The subexpression boolean2 is evaluated only if boolean1 is true. This can make a difference when boolean2 contains expressions that have side effects. In the case of ‘$0 ~ /foo/ && ($2 == bar++)’, the variable bar is not incremented if there is no substring ‘foo’ in the record.

boolean1 || boolean2

True if at least one of boolean1 or boolean2 is true. For example, the following statement prints all records in the input that contain eitheredu’ or ‘li’:

if ($0 ~ /edu/ || $0 ~ /li/) print

The subexpression boolean2 is evaluated only if boolean1 is false. This can make a difference when boolean2 contains expressions that have side effects. (Thus, this test never really distinguishes records that contain both ‘edu’ and ‘li’—as soon as ‘edu’ is matched, the full test succeeds.)

! boolean

True if boolean is false. For example, the following program prints ‘no home!’ in the unusual event that the HOME environment variable is not defined:

BEGIN { if (! ("HOME" in ENVIRON))
            print "no home!" }

(The in operator is described in Referring to an Array Element.)

The ‘&&’ and ‘||’ operators are called short-circuit operators because of the way they work. Evaluation of the full expression is “short-circuited” if the result can be determined partway through its evaluation.

Statements that end with ‘&&’ or ‘||’ can be continued simply by putting a newline after them. But you cannot put a newline in front of either of these operators without using backslash continuation (see awk Statements Versus Lines).

The actual value of an expression using the ‘!’ operator is either one or zero, depending upon the truth value of the expression it is applied to. The ‘!’ operator is often useful for changing the sense of a flag variable from false to true and back again. For example, the following program is one way to print lines in between special bracketing lines:

$1 == "START"   { interested = ! interested; next }
interested      { print }
$1 == "END"     { interested = ! interested; next }

The variable interested, as with all awk variables, starts out initialized to zero, which is also false. When a line is seen whose first field is ‘START’, the value of interested is toggled to true, using ‘!’. The next rule prints lines as long as interested is true. When a line is seen whose first field is ‘END’, interested is toggled back to false.37

Most commonly, the ‘!’ operator is used in the conditions of if and while statements, where it often makes more sense to phrase the logic in the negative:

if (! some condition || some other condition) {
    … do whatever processing …
}

NOTE: The next statement is discussed in The next Statement. next tells awk to skip the rest of the rules, get the next record, and start processing the rules over again at the top. The reason it’s there is to avoid printing the bracketing ‘START’ and ‘END’ lines.


6.3.4 Conditional Expressions

A conditional expression is a special kind of expression that has three operands. It allows you to use one expression’s value to select one of two other expressions. The conditional expression in awk is the same as in the C language, as shown here:

selector ? if-true-exp : if-false-exp

There are three subexpressions. The first, selector, is always computed first. If it is “true” (not zero or not null), then if-true-exp is computed next, and its value becomes the value of the whole expression. Otherwise, if-false-exp is computed next, and its value becomes the value of the whole expression. For example, the following expression produces the absolute value of x:

x >= 0 ? x : -x

Each time the conditional expression is computed, only one of if-true-exp and if-false-exp is used; the other is ignored. This is important when the expressions have side effects. For example, this conditional expression examines element i of either array a or array b, and increments i:

x == y ? a[i++] : b[i++]

This is guaranteed to increment i exactly once, because each time only one of the two increment expressions is executed and the other is not. See Arrays in awk, for more information about arrays.

As a minor gawk extension, a statement that uses ‘?:’ can be continued simply by putting a newline after either character. However, putting a newline in front of either character does not work without using backslash continuation (see awk Statements Versus Lines). If --posix is specified (see Command-Line Options), this extension is disabled.


6.4 Function Calls

A function is a name for a particular calculation. This enables you to ask for it by name at any point in the program. For example, the function sqrt() computes the square root of a number.

A fixed set of functions are built in, which means they are available in every awk program. The sqrt() function is one of these. See Built-in Functions for a list of built-in functions and their descriptions. In addition, you can define functions for use in your program. See User-Defined Functions for instructions on how to do this. Finally, gawk lets you write functions in C or C++ that may be called from your program (see Writing Extensions for gawk).

The way to use a function is with a function call expression, which consists of the function name followed immediately by a list of arguments in parentheses. The arguments are expressions that provide the raw materials for the function’s calculations. When there is more than one argument, they are separated by commas. If there are no arguments, just write ‘()’ after the function name. The following examples show function calls with and without arguments:

sqrt(x^2 + y^2)        one argument
atan2(y, x)            two arguments
rand()                 no arguments

CAUTION: Do not put any space between the function name and the opening parenthesis! A user-defined function name looks just like the name of a variable—a space would make the expression look like concatenation of a variable with an expression inside parentheses. With built-in functions, space before the parenthesis is harmless, but it is best not to get into the habit of using space to avoid mistakes with user-defined functions.

Each function expects a particular number of arguments. For example, the sqrt() function must be called with a single argument, the number of which to take the square root:

sqrt(argument)

Some of the built-in functions have one or more optional arguments. If those arguments are not supplied, the functions use a reasonable default value. See Built-in Functions for full details. If arguments are omitted in calls to user-defined functions, then those arguments are treated as local variables. Such local variables act like the empty string if referenced where a string value is required, and like zero if referenced where a numeric value is required (see User-Defined Functions).

As an advanced feature, gawk provides indirect function calls, which is a way to choose the function to call at runtime, instead of when you write the source code to your program. We defer discussion of this feature until later; see Indirect Function Calls.

Like every other expression, the function call has a value, often called the return value, which is computed by the function based on the arguments you give it. In this example, the return value of ‘sqrt(argument)’ is the square root of argument. The following program reads numbers, one number per line, and prints the square root of each one:

$ awk '{ print "The square root of", $1, "is", sqrt($1) }'
1
-| The square root of 1 is 1
3
-| The square root of 3 is 1.73205
5
-| The square root of 5 is 2.23607
Ctrl-d

A function can also have side effects, such as assigning values to certain variables or doing I/O. This program shows how the match() function (see String-Manipulation Functions) changes the variables RSTART and RLENGTH:

{
    if (match($1, $2))
        print RSTART, RLENGTH
    else
        print "no match"
}

Here is a sample run:

$ awk -f matchit.awk
aaccdd  c+
-| 3 2
foo     bar
-| no match
abcdefg e
-| 5 1

6.5 Operator Precedence (How Operators Nest)

Operator precedence determines how operators are grouped when different operators appear close by in one expression. For example, ‘*’ has higher precedence than ‘+’; thus, ‘a + b * c’ means to multiply b and c, and then add a to the product (i.e., ‘a + (b * c)’).

The normal precedence of the operators can be overruled by using parentheses. Think of the precedence rules as saying where the parentheses are assumed to be. In fact, it is wise to always use parentheses whenever there is an unusual combination of operators, because other people who read the program may not remember what the precedence is in this case. Even experienced programmers occasionally forget the exact rules, which leads to mistakes. Explicit parentheses help prevent any such mistakes.

When operators of equal precedence are used together, the leftmost operator groups first, except for the assignment, conditional, and exponentiation operators, which group in the opposite order. Thus, ‘a - b + c’ groups as ‘(a - b) + c’ and ‘a = b = c’ groups as ‘a = (b = c)’.

Normally the precedence of prefix unary operators does not matter, because there is only one way to interpret them: innermost first. Thus, ‘$++i’ means ‘$(++i)’ and ‘++$x’ means ‘++($x)’. However, when another operator follows the operand, then the precedence of the unary operators can matter. ‘$x^2’ means ‘($x)^2’, but ‘-x^2’ means ‘-(x^2)’, because ‘-’ has lower precedence than ‘^’, whereas ‘$’ has higher precedence. Also, operators cannot be combined in a way that violates the precedence rules; for example, ‘$$0++--’ is not a valid expression because the first ‘$’ has higher precedence than the ‘++’; to avoid the problem the expression can be rewritten as ‘$($0++)--’.

This list presents awk’s operators, in order of highest to lowest precedence:

()

Grouping.

$

Field reference.

++ --

Increment, decrement.

^ **

Exponentiation. These operators group right to left.

+ - !

Unary plus, minus, logical “not.”

* / %

Multiplication, division, remainder.

+ -

Addition, subtraction.

String concatenation

There is no special symbol for concatenation. The operands are simply written side by side (see String Concatenation).

< <= == != > >= >> | |&

Relational and redirection. The relational operators and the redirections have the same precedence level. Characters such as ‘>’ serve both as relationals and as redirections; the context distinguishes between the two meanings.

Note that the I/O redirection operators in print and printf statements belong to the statement level, not to expressions. The redirection does not produce an expression that could be the operand of another operator. As a result, it does not make sense to use a redirection operator near another operator of lower precedence without parentheses. Such combinations (e.g., ‘print foo > a ? b : c’) result in syntax errors. The correct way to write this statement is ‘print foo > (a ? b : c)’.

~ !~

Matching, nonmatching.

in

Array membership.

&&

Logical “and.”

||

Logical “or.”

?:

Conditional. This operator groups right to left.

= += -= *= /= %= ^= **=

Assignment. These operators group right to left.

NOTE: The ‘|&’, ‘**’, and ‘**=’ operators are not specified by POSIX. For maximum portability, do not use them.


6.6 Where You Are Makes a Difference

Modern systems support the notion of locales: a way to tell the system about the local character set and language. The ISO C standard defines a default "C" locale, which is an environment that is typical of what many C programmers are used to.

Once upon a time, the locale setting used to affect regexp matching, but this is no longer true (see Regexp Ranges and Locales: A Long Sad Story).

Locales can affect record splitting. For the normal case of ‘RS = "\n"’, the locale is largely irrelevant. For other single-character record separators, setting ‘LC_ALL=C’ in the environment will give you much better performance when reading records. Otherwise, gawk has to make several function calls, per input character, to find the record terminator.

Locales can affect how dates and times are formatted (see Time Functions). For example, a common way to abbreviate the date September 4, 2015, in the United States is “9/4/15.” In many countries in Europe, however, it is abbreviated “4.9.15.” Thus, the ‘%x’ specification in a "US" locale might produce ‘9/4/15’, while in a "EUROPE" locale, it might produce ‘4.9.15’.

According to POSIX, string comparison is also affected by locales (similar to regular expressions). The details are presented in String Comparison Based on Locale Collating Order.

Finally, the locale affects the value of the decimal point character used when gawk parses input data. This is discussed in detail in Conversion of Strings and Numbers.


6.7 Summary

  • Expressions are the basic elements of computation in programs. They are built from constants, variables, function calls, and combinations of the various kinds of values with operators.
  • awk supplies three kinds of constants: numeric, string, and regexp. gawk lets you specify numeric constants in octal and hexadecimal (bases 8 and 16) as well as decimal (base 10). In certain contexts, a standalone regexp constant such as /foo/ has the same meaning as ‘$0 ~ /foo/’.
  • Variables hold values between uses in computations. A number of built-in variables provide information to your awk program, and a number of others let you control how awk behaves.
  • Numbers are automatically converted to strings, and strings to numbers, as needed by awk. Numeric values are converted as if they were formatted with sprintf() using the format in CONVFMT. Locales can influence the conversions.
  • awk provides the usual arithmetic operators (addition, subtraction, multiplication, division, modulus), and unary plus and minus. It also provides comparison operators, Boolean operators, an array membership testing operator, and regexp matching operators. String concatenation is accomplished by placing two expressions next to each other; there is no explicit operator. The three-operand ‘?:’ operator provides an “if-else” test within expressions.
  • Assignment operators provide convenient shorthands for common arithmetic operations.
  • In awk, a value is considered to be true if it is nonzero or non-null. Otherwise, the value is false.
  • A variable’s type is set upon each assignment and may change over its lifetime. The type determines how it behaves in comparisons (string or numeric).
  • Function calls return a value that may be used as part of a larger expression. Expressions used to pass parameter values are fully evaluated before the function is called. awk provides built-in and user-defined functions; this is described in Functions.
  • Operator precedence specifies the order in which operations are performed, unless explicitly overridden by parentheses. awk’s operator precedence is compatible with that of C.
  • Locales can affect the format of data as output by an awk program, and occasionally the format for data read as input.

7 Patterns, Actions, and Variables

As you have already seen, each awk statement consists of a pattern with an associated action. This chapter describes how you build patterns and actions, what kinds of things you can do within actions, and awk’s predefined variables.

The pattern–action rules and the statements available for use within actions form the core of awk programming. In a sense, everything covered up to here has been the foundation that programs are built on top of. Now it’s time to start building something useful.


7.1 Pattern Elements

Patterns in awk control the execution of rules—a rule is executed when its pattern matches the current input record. The following is a summary of the types of awk patterns:

/regular expression/

A regular expression. It matches when the text of the input record fits the regular expression. (See Regular Expressions.)

expression

A single expression. It matches when its value is nonzero (if a number) or non-null (if a string). (See Expressions as Patterns.)

begpat, endpat

A pair of patterns separated by a comma, specifying a range of records. The range includes both the initial record that matches begpat and the final record that matches endpat. (See Specifying Record Ranges with Patterns.)

BEGIN
END

Special patterns for you to supply startup or cleanup actions for your awk program. (See The BEGIN and END Special Patterns.)

BEGINFILE
ENDFILE

Special patterns for you to supply startup or cleanup actions to be done on a per-file basis. (See The BEGINFILE and ENDFILE Special Patterns.)

empty

The empty pattern matches every input record. (See The Empty Pattern.)


7.1.1 Regular Expressions as Patterns

Regular expressions are one of the first kinds of patterns presented in this book. This kind of pattern is simply a regexp constant in the pattern part of a rule. Its meaning is ‘$0 ~ /pattern/’. The pattern matches when the input record matches the regexp. For example:

/foo|bar|baz/  { buzzwords++ }
END            { print buzzwords, "buzzwords seen" }

7.1.2 Expressions as Patterns

Any awk expression is valid as an awk pattern. The pattern matches if the expression’s value is nonzero (if a number) or non-null (if a string). The expression is reevaluated each time the rule is tested against a new input record. If the expression uses fields such as $1, the value depends directly on the new input record’s text; otherwise, it depends on only what has happened so far in the execution of the awk program.

Comparison expressions, using the comparison operators described in Variable Typing and Comparison Expressions, are a very common kind of pattern. Regexp matching and nonmatching are also very common expressions. The left operand of the ‘~’ and ‘!~’ operators is a string. The right operand is either a constant regular expression enclosed in slashes (/regexp/), or any expression whose string value is used as a dynamic regular expression (see Using Dynamic Regexps). The following example prints the second field of each input record whose first field is precisely ‘li’:

$ awk '$1 == "li" { print $2 }' mail-list

(There is no output, because there is no person with the exact name ‘li’.) Contrast this with the following regular expression match, which accepts any record with a first field that contains ‘li’:

$ awk '$1 ~ /li/ { print $2 }' mail-list
-| 555-5553
-| 555-6699

A regexp constant as a pattern is also a special case of an expression pattern. The expression /li/ has the value one if ‘li’ appears in the current input record. Thus, as a pattern, /li/ matches any record containing ‘li’.

Boolean expressions are also commonly used as patterns. Whether the pattern matches an input record depends on whether its subexpressions match. For example, the following command prints all the records in mail-list that contain both ‘edu’ and ‘li’:

$ awk '/edu/ && /li/' mail-list
-| Samuel       555-3430     [email protected]        A

The following command prints all records in mail-list that contain eitheredu’ or ‘li’ (or both, of course):

$ awk '/edu/ || /li/' mail-list
-| Amelia       555-5553     [email protected]    F
-| Broderick    555-0542     [email protected] R
-| Fabius       555-1234     [email protected]    F
-| Julie        555-6699     [email protected]   F
-| Samuel       555-3430     [email protected]        A
-| Jean-Paul    555-2127     [email protected]     R

The following command prints all records in mail-list that do not contain the string ‘li’:

$ awk '! /li/' mail-list
-| Anthony      555-3412     [email protected]   A
-| Becky        555-7685     [email protected]      A
-| Bill         555-1675     [email protected]       A
-| Camilla      555-2912     [email protected]     R
-| Fabius       555-1234     [email protected]    F
-| Martin       555-6480     [email protected]    A
-| Jean-Paul    555-2127     [email protected]     R

The subexpressions of a Boolean operator in a pattern can be constant regular expressions, comparisons, or any other awk expressions. Range patterns are not expressions, so they cannot appear inside Boolean patterns. Likewise, the special patterns BEGIN, END, BEGINFILE, and ENDFILE, which never match any input record, are not expressions and cannot appear inside Boolean patterns.

The precedence of the different operators that can appear in patterns is described in Operator Precedence (How Operators Nest).


7.1.3 Specifying Record Ranges with Patterns

A range pattern is made of two patterns separated by a comma, in the form ‘begpat, endpat’. It is used to match ranges of consecutive input records. The first pattern, begpat, controls where the range begins, while endpat controls where the pattern ends. For example, the following:

awk '$1 == "on", $1 == "off"' myfile

prints every record in myfile between ‘on’/‘off’ pairs, inclusive.

A range pattern starts out by matching begpat against every input record. When a record matches begpat, the range pattern is turned on, and the range pattern matches this record as well. As long as the range pattern stays turned on, it automatically matches every input record read. The range pattern also matches endpat against every input record; when this succeeds, the range pattern is turned off again for the following record. Then the range pattern goes back to checking begpat against each record.

The record that turns on the range pattern and the one that turns it off both match the range pattern. If you don’t want to operate on these records, you can write if statements in the rule’s action to distinguish them from the records you are interested in.

It is possible for a pattern to be turned on and off by the same record. If the record satisfies both conditions, then the action is executed for just that record. For example, suppose there is text between two identical markers (e.g., the ‘%’ symbol), each on its own line, that should be ignored. A first attempt would be to combine a range pattern that describes the delimited text with the next statement (not discussed yet, see The next Statement). This causes awk to skip any further processing of the current record and start over again with the next input record. Such a program looks like this:

/^%$/,/^%$/    { next }
               { print }

This program fails because the range pattern is both turned on and turned off by the first line, which just has a ‘%’ on it. To accomplish this task, write the program in the following manner, using a flag:

/^%$/     { skip = ! skip; next }
skip == 1 { next } # skip lines with `skip' set

In a range pattern, the comma (‘,’) has the lowest precedence of all the operators (i.e., it is evaluated last). Thus, the following program attempts to combine a range pattern with another, simpler test:

echo Yes | awk '/1/,/2/ || /Yes/'

The intent of this program is ‘(/1/,/2/) || /Yes/’. However, awk interprets this as ‘/1/, (/2/ || /Yes/)’. This cannot be changed or worked around; range patterns do not combine with other patterns:

$ echo Yes | gawk '(/1/,/2/) || /Yes/'
error→ gawk: cmd. line:1: (/1/,/2/) || /Yes/
error→ gawk: cmd. line:1:           ^ syntax error

As a minor point of interest, although it is poor style, POSIX allows you to put a newline after the comma in a range pattern. (d.c.)


7.1.4 The BEGIN and END Special Patterns

All the patterns described so far are for matching input records. The BEGIN and END special patterns are different. They supply startup and cleanup actions for awk programs. BEGIN and END rules must have actions; there is no default action for these rules because there is no current record when they run. BEGIN and END rules are often referred to as “BEGIN and END blocks” by longtime awk programmers.


7.1.4.1 Startup and Cleanup Actions

A BEGIN rule is executed once only, before the first input record is read. Likewise, an END rule is executed once only, after all the input is read. For example:

$ awk '
> BEGIN { print "Analysis of \"li\"" }
> /li/  { ++n }
> END   { print "\"li\" appears in", n, "records." }' mail-list
-| Analysis of "li"
-| "li" appears in 4 records.

This program finds the number of records in the input file mail-list that contain the string ‘li’. The BEGIN rule prints a title for the report. There is no need to use the BEGIN rule to initialize the counter n to zero, as awk does this automatically (see Variables). The second rule increments the variable n every time a record containing the pattern ‘li’ is read. The END rule prints the value of n at the end of the run.

The special patterns BEGIN and END cannot be used in ranges or with Boolean operators (indeed, they cannot be used with any operators). An awk program may have multiple BEGIN and/or END rules. They are executed in the order in which they appear: all the BEGIN rules at startup and all the END rules at termination.

BEGIN and END rules may be intermixed with other rules. This feature was added in the 1987 version of awk and is included in the POSIX standard. The original (1978) version of awk required the BEGIN rule to be placed at the beginning of the program, the END rule to be placed at the end, and only allowed one of each. This is no longer required, but it is a good idea to follow this template in terms of program organization and readability.

Multiple BEGIN and END rules are useful for writing library functions, because each library file can have its own BEGIN and/or END rule to do its own initialization and/or cleanup. The order in which library functions are named on the command line controls the order in which their BEGIN and END rules are executed. Therefore, you have to be careful when writing such rules in library files so that the order in which they are executed doesn’t matter. See Command-Line Options for more information on using library functions. See A Library of awk Functions, for a number of useful library functions.

If an awk program has only BEGIN rules and no other rules, then the program exits after the BEGIN rules are run.38 However, if an END rule exists, then the input is read, even if there are no other rules in the program. This is necessary in case the END rule checks the FNR and NR variables, or the fields.


7.1.4.2 Input/Output from BEGIN and END Rules

There are several (sometimes subtle) points to be aware of when doing I/O from a BEGIN or END rule. The first has to do with the value of $0 in a BEGIN rule. Because BEGIN rules are executed before any input is read, there simply is no input record, and therefore no fields, when executing BEGIN rules. References to $0 and the fields yield a null string or zero, depending upon the context. One way to give $0 a real value is to execute a getline command without a variable (see Explicit Input with getline). Another way is simply to assign a value to $0.

The second point is similar to the first, but from the other direction. Traditionally, due largely to implementation issues, $0 and NF were undefined inside an END rule. The POSIX standard specifies that NF is available in an END rule. It contains the number of fields from the last input record. Most probably due to an oversight, the standard does not say that $0 is also preserved, although logically one would think that it should be. In fact, all of BWK awk, mawk, and gawk preserve the value of $0 for use in END rules. Be aware, however, that some other implementations and many older versions of Unix awk do not.

The third point follows from the first two. The meaning of ‘print’ inside a BEGIN or END rule is the same as always: ‘print $0’. If $0 is the null string, then this prints an empty record. Many longtime awk programmers use an unadorned ‘print’ in BEGIN and END rules to mean ‘print ""’, relying on $0 being null. Although one might generally get away with this in BEGIN rules, it is a very bad idea in END rules, at least in gawk. It is also poor style, because if an empty line is needed in the output, the program should print one explicitly.

Finally, the next and nextfile statements are not allowed in a BEGIN rule, because the implicit read-a-record-and-match-against-the-rules loop has not started yet. Similarly, those statements are not valid in an END rule, because all the input has been read. (See The next Statement and see The nextfile Statement.)


7.1.5 The BEGINFILE and ENDFILE Special Patterns

This section describes a gawk-specific feature.

Two special kinds of rule, BEGINFILE and ENDFILE, give you “hooks” into gawk’s command-line file processing loop. As with the BEGIN and END rules (see The BEGIN and END Special Patterns), BEGINFILE rules in a program execute in the order they are read by gawk. Similarly, all ENDFILE rules also execute in the order they are read.

The bodies of the BEGINFILE rules execute just before gawk reads the first record from a file. FILENAME is set to the name of the current file, and FNR is set to zero.

Prior to version 5.1.1 of gawk, as an accident of the implementation, $0 and the fields retained any previous values they had in BEGINFILE rules. Starting with version 5.1.1, $0 and the fields are cleared, since no record has been read yet from the file that is about to be processed.

The BEGINFILE rule provides you the opportunity to accomplish two tasks that would otherwise be difficult or impossible to perform:

  • You can test if the file is readable. Normally, it is a fatal error if a file named on the command line cannot be opened for reading. However, you can bypass the fatal error and move on to the next file on the command line.

    You do this by checking if the ERRNO variable is not the empty string; if so, then gawk was not able to open the file. In this case, your program can execute the nextfile statement (see The nextfile Statement). This causes gawk to skip the file entirely. Otherwise, gawk exits with the usual fatal error.

  • If you have written extensions that modify the record handling (by inserting an “input parser”; see Customized Input Parsers), you can invoke them at this point, before gawk has started processing the file. (This is a very advanced feature, currently used only by the gawkextlib project.)

The ENDFILE rule is called when gawk has finished processing the last record in an input file. For the last input file, it will be called before any END rules. The ENDFILE rule is executed even for empty input files.

Normally, when an error occurs when reading input in the normal input-processing loop, the error is fatal. However, if a BEGINFILE rule is present, the error becomes non-fatal, and instead ERRNO is set. This makes it possible to catch and process I/O errors at the level of the awk program.

The next statement (see The next Statement) is not allowed inside either a BEGINFILE or an ENDFILE rule. The nextfile statement is allowed only inside a BEGINFILE rule, not inside an ENDFILE rule.

The getline statement (see Explicit Input with getline) is restricted inside both BEGINFILE and ENDFILE: only redirected forms of getline are allowed.

BEGINFILE and ENDFILE are gawk extensions. In most other awk implementations, or if gawk is in compatibility mode (see Command-Line Options), they are not special.


7.1.6 The Empty Pattern

An empty (i.e., nonexistent) pattern is considered to match every input record. For example, the program:

awk '{ print $1 }' mail-list

prints the first field of every record.


7.2 Using Shell Variables in Programs

awk programs are often used as components in larger programs written in shell. For example, it is very common to use a shell variable to hold a pattern that the awk program searches for. There are two ways to get the value of the shell variable into the body of the awk program.

A common method is to use shell quoting to substitute the variable’s value into the program inside the script. For example, consider the following program:

printf "Enter search pattern: "
read pattern
awk "/$pattern/ "'{ nmatches++ }
     END { print nmatches, "found" }' /path/to/data

The awk program consists of two pieces of quoted text that are concatenated together to form the program. The first part is double-quoted, which allows substitution of the pattern shell variable inside the quotes. The second part is single-quoted.

Variable substitution via quoting works, but can potentially be messy. It requires a good understanding of the shell’s quoting rules (see Shell Quoting Issues), and it’s often difficult to correctly match up the quotes when reading the program.

A better method is to use awk’s variable assignment feature (see Assigning Variables on the Command Line) to assign the shell variable’s value to an awk variable. Then use dynamic regexps to match the pattern (see Using Dynamic Regexps). The following shows how to redo the previous example using this technique:

printf "Enter search pattern: "
read pattern
awk -v pat="$pattern" '$0 ~ pat { nmatches++ }
       END { print nmatches, "found" }' /path/to/data

Now, the awk program is just one single-quoted string. The assignment ‘-v pat="$pattern"’ still requires double quotes, in case there is whitespace in the value of $pattern. The awk variable pat could be named pattern too, but that would be more confusing. Using a variable also provides more flexibility, as the variable can be used anywhere inside the program—for printing, as an array subscript, or for any other use—without requiring the quoting tricks at every point in the program.


7.3 Actions

An awk program or script consists of a series of rules and function definitions interspersed. (Functions are described later. See User-Defined Functions.) A rule contains a pattern and an action, either of which (but not both) may be omitted. The purpose of the action is to tell awk what to do once a match for the pattern is found. Thus, in outline, an awk program generally looks like this:

[pattern]  { action }
 pattern  [{ action }]
…
function name(args) { … }

An action consists of one or more awk statements, enclosed in braces (‘{}’). Each statement specifies one thing to do. The statements are separated by newlines or semicolons. The braces around an action must be used even if the action contains only one statement, or if it contains no statements at all. However, if you omit the action entirely, omit the braces as well. An omitted action is equivalent to ‘{ print $0 }’:

/foo/  { }     match foo, do nothing --- empty action
/foo/          match foo, print the record --- omitted action

The following types of statements are supported in awk:

Expressions

Call functions or assign values to variables (see Expressions). Executing this kind of statement simply computes the value of the expression. This is useful when the expression has side effects (see Assignment Expressions).

Control statements

Specify the control flow of awk programs. The awk language gives you C-like constructs (if, for, while, and do) as well as a few special ones (see Control Statements in Actions).

Compound statements

Enclose one or more statements in braces. A compound statement is used in order to put several statements together in the body of an if, while, do, or for statement.

Input statements

Use the getline command (see Explicit Input with getline). Also supplied in awk are the next statement (see The next Statement) and the nextfile statement (see The nextfile Statement).

Output statements

Such as print and printf. See Printing Output.

Deletion statements

For deleting array elements. See The delete Statement.


7.4 Control Statements in Actions

Control statements, such as if, while, and so on, control the flow of execution in awk programs. Most of awk’s control statements are patterned after similar statements in C.

All the control statements start with special keywords, such as if and while, to distinguish them from simple expressions. Many control statements contain other statements. For example, the if statement contains another statement that may or may not be executed. The contained statement is called the body. To include more than one statement in the body, group them into a single compound statement with braces, separating them with newlines or semicolons.


7.4.1 The if-else Statement

The if-else statement is awk’s decision-making statement. It looks like this:

if (condition) then-body [else else-body]

The condition is an expression that controls what the rest of the statement does. If the condition is true, then-body is executed; otherwise, else-body is executed. The else part of the statement is optional. The condition is considered false if its value is zero or the null string; otherwise, the condition is true. Refer to the following:

if (x % 2 == 0)
    print "x is even"
else
    print "x is odd"

In this example, if the expression ‘x % 2 == 0’ is true (i.e., if the value of x is evenly divisible by two), then the first print statement is executed; otherwise, the second print statement is executed. If the else keyword appears on the same line as then-body and then-body is not a compound statement (i.e., not surrounded by braces), then a semicolon must separate then-body from the else. To illustrate this, the previous example can be rewritten as:

if (x % 2 == 0) print "x is even"; else
        print "x is odd"

If the ‘;’ is left out, awk can’t interpret the statement and it produces a syntax error. Don’t actually write programs this way, because a human reader might fail to see the else if it is not the first thing on its line.


7.4.2 The while Statement

In programming, a loop is a part of a program that can be executed two or more times in succession. The while statement is the simplest looping statement in awk. It repeatedly executes a statement as long as a condition is true. For example:

while (condition)
  body

body is a statement called the body of the loop, and condition is an expression that controls how long the loop keeps running. The first thing the while statement does is test the condition. If the condition is true, it executes the statement body. After body has been executed, condition is tested again, and if it is still true, body executes again. This process repeats until the condition is no longer true. If the condition is initially false, the body of the loop never executes and awk continues with the statement following the loop. This example prints the first three fields of each record, one per line:

awk '
{
    i = 1
    while (i <= 3) {
        print $i
        i++
    }
}' inventory-shipped

The body of this loop is a compound statement enclosed in braces, containing two statements. The loop works in the following manner: first, the value of i is set to one. Then, the while statement tests whether i is less than or equal to three. This is true when i equals one, so the ith field is printed. Then the ‘i++’ increments the value of i and the loop repeats. The loop terminates when i reaches four.

A newline is not required between the condition and the body; however, using one makes the program clearer unless the body is a compound statement or else is very simple. The newline after the open brace that begins the compound statement is not required either, but the program is harder to read without it.


7.4.3 The do-while Statement

The do loop is a variation of the while looping statement. The do loop executes the body once and then repeats the body as long as the condition is true. It looks like this:

do
  body
while (condition)

Even if the condition is false at the start, the body executes at least once (and only once, unless executing body makes condition true). Contrast this with the corresponding while statement:

while (condition)
    body

This statement does not execute the body even once if the condition is false to begin with. The following is an example of a do statement:

{
    i = 1
    do {
        print $0
        i++
    } while (i <= 10)
}

This program prints each input record 10 times. However, it isn’t a very realistic example, because in this case an ordinary while would do just as well. This situation reflects actual experience; only occasionally is there a real use for a do statement.


7.4.4 The for Statement

The for statement makes it more convenient to count iterations of a loop. The general form of the for statement looks like this:

for (initialization; condition; increment)
  body

The initialization, condition, and increment parts are arbitrary awk expressions, and body stands for any awk statement.

The for statement starts by executing initialization. Then, as long as the condition is true, it repeatedly executes body and then increment. Typically, initialization sets a variable to either zero or one, increment adds one to it, and condition compares it against the desired number of iterations. For example:

awk '
{
    for (i = 1; i <= 3; i++)
        print $i
}' inventory-shipped

This prints the first three fields of each input record, with one input field per output line.

C and C++ programmers might expect to be able to use the comma operator to set more than one variable in the initialization part of the for loop, or to increment multiple variables in the increment part of the loop, like so:

for (i = 0, j = length(a); i < j; i++, j--) …   C/C++, not awk!

You cannot do this; the comma operator is not supported in awk. There are workarounds, but they are nonobvious and can lead to code that is difficult to read and understand. It is best, therefore, to simply write additional initializations as separate statements preceding the for loop and to place additional increment statements at the end of the loop’s body.

Most often, increment is an increment expression, as in the earlier example. But this is not required; it can be any expression whatsoever. For example, the following statement prints all the powers of two between 1 and 100:

for (i = 1; i <= 100; i *= 2)
    print i

If there is nothing to be done, any of the three expressions in the parentheses following the for keyword may be omitted. Thus, ‘for (; x > 0;) is equivalent to ‘while (x > 0). If the condition is omitted, it is treated as true, effectively yielding an infinite loop (i.e., a loop that never terminates).

In most cases, a for loop is an abbreviation for a while loop, as shown here:

initialization
while (condition) {
  body
  increment
}

The only exception is when the continue statement (see The continue Statement) is used inside the loop. Changing a for statement to a while statement in this way can change the effect of the continue statement inside the loop.

The awk language has a for statement in addition to a while statement because a for loop is often both less work to type and more natural to think of. Counting the number of iterations is very common in loops. It can be easier to think of this counting as part of looping rather than as something to do inside the loop.

There is an alternative version of the for loop, for iterating over all the indices of an array:

for (i in array)
    do something with array[i]

See Scanning All Elements of an Array for more information on this version of the for loop.


7.4.5 The switch Statement

This section describes a gawk-specific feature. If gawk is in compatibility mode (see Command-Line Options), it is not available.

The switch statement allows the evaluation of an expression and the execution of statements based on a case match. Case statements are checked for a match in the order they are defined. If no suitable case is found, the default section is executed, if supplied.

Each case contains a single constant, be it numeric, string, or regexp. The switch expression is evaluated, and then each case’s constant is compared against the result in turn. The type of constant determines the comparison: numeric or string do the usual comparisons. A regexp constant (either regular, /foo/, or strongly typed, @/foo/) does a regular expression match against the string value of the original expression. The general form of the switch statement looks like this:

switch (expression) {
case value or regular expression:
    case-body
default:
    default-body
}

Control flow in the switch statement works as it does in C. Once a match to a given case is made, the case statement bodies execute until a break, continue, next, nextfile, or exit is encountered, or the end of the switch statement itself. For example:

while ((c = getopt(ARGC, ARGV, "aksx")) != -1) {
    switch (c) {
    case "a":
        # report size of all files
        all_files = TRUE;
        break
    case "k":
        BLOCK_SIZE = 1024       # 1K block size
        break
    case "s":
        # do sums only
        sum_only = TRUE
        break
    case "x":
        # don't cross filesystems
        fts_flags = or(fts_flags, FTS_XDEV)
        break
    case "?":
    default:
        usage()
        break
    }
}

Note that if none of the statements specified here halt execution of a matched case statement, execution falls through to the next case until execution halts. In this example, the case for "?" falls through to the default case, which is to call a function named usage(). (The getopt() function being called here is described in Processing Command-Line Options.)


7.4.6 The break Statement

The break statement jumps out of the innermost for, while, or do loop that encloses it. The following example finds the smallest divisor of any integer, and also identifies prime numbers:

# find smallest divisor of num
{
    num = $1
    for (divisor = 2; divisor * divisor <= num; divisor++) {
        if (num % divisor == 0)
            break
    }
    if (num % divisor == 0)
        printf "Smallest divisor of %d is %d\n", num, divisor
    else
        printf "%d is prime\n", num
}

When the remainder is zero in the first if statement, awk immediately breaks out of the containing for loop. This means that awk proceeds immediately to the statement following the loop and continues processing. (This is very different from the exit statement, which stops the entire awk program. See The exit Statement.)

The following program illustrates how the condition of a for or while statement could be replaced with a break inside an if:

# find smallest divisor of num
{
    num = $1
    for (divisor = 2; ; divisor++) {
        if (num % divisor == 0) {
            printf "Smallest divisor of %d is %d\n", num, divisor
            break
        }
        if (divisor * divisor > num) {
            printf "%d is prime\n", num
            break
        }
    }
}

The break statement is also used to break out of the switch statement. This is discussed in The switch Statement.

The break statement has no meaning when used outside the body of a loop or switch. However, although it was never documented, historical implementations of awk treated the break statement outside of a loop as if it were a next statement (see The next Statement). (d.c.) Recent versions of BWK awk no longer allow this usage, nor does gawk.


7.4.7 The continue Statement

Similar to break, the continue statement is used only inside for, while, and do loops. It skips over the rest of the loop body, causing the next cycle around the loop to begin immediately. Contrast this with break, which jumps out of the loop altogether.

The continue statement in a for loop directs awk to skip the rest of the body of the loop and resume execution with the increment-expression of the for statement. The following program illustrates this fact:

BEGIN {
     for (x = 0; x <= 20; x++) {
         if (x == 5)
             continue
         printf "%d ", x
     }
     print ""
}

This program prints all the numbers from 0 to 20—except for 5, for which the printf is skipped. Because the increment ‘x++’ is not skipped, x does not remain stuck at 5. Contrast the for loop from the previous example with the following while loop:

BEGIN {
     x = 0
     while (x <= 20) {
         if (x == 5)
             continue
         printf "%d ", x
         x++
     }
     print ""
}

This program loops forever once x reaches 5, because the increment (‘x++’) is never reached.

The continue statement has no special meaning with respect to the switch statement, nor does it have any meaning when used outside the body of a loop. Historical versions of awk treated a continue statement outside a loop the same way they treated a break statement outside a loop: as if it were a next statement (see The next Statement). (d.c.) Recent versions of BWK awk no longer work this way, nor does gawk.


7.4.8 The next Statement

The next statement forces awk to immediately stop processing the current record and go on to the next record. This means that no further rules are executed for the current record, and the rest of the current rule’s action isn’t executed.

Contrast this with the effect of the getline function (see Explicit Input with getline). That also causes awk to read the next record immediately, but it does not alter the flow of control in any way (i.e., the rest of the current action executes with a new input record).

At the highest level, awk program execution is a loop that reads an input record and then tests each rule’s pattern against it. If you think of this loop as a for statement whose body contains the rules, then the next statement is analogous to a continue statement. It skips to the end of the body of this implicit loop and executes the increment (which reads another record).

For example, suppose an awk program works only on records with four fields, and it shouldn’t fail when given bad input. To avoid complicating the rest of the program, write a “weed out” rule near the beginning, in the following manner:

NF != 4 {
    printf("%s:%d: skipped: NF != 4\n", FILENAME, FNR) > "/dev/stderr"
    next
}

Because of the next statement, the program’s subsequent rules won’t see the bad record. The error message is redirected to the standard error output stream, as error messages should be. For more detail, see Special File names in gawk.

If the next statement causes the end of the input to be reached, then the code in any END rules is executed. See The BEGIN and END Special Patterns.

The next statement is not allowed inside BEGINFILE and ENDFILE rules. See The BEGINFILE and ENDFILE Special Patterns.

According to the POSIX standard, the behavior is undefined if the next statement is used in a BEGIN or END rule. gawk treats it as a syntax error. Although POSIX does not disallow it, most other awk implementations don’t allow the next statement inside function bodies (see User-Defined Functions). Just as with any other next statement, a next statement inside a function body reads the next record and starts processing it with the first rule in the program.


7.4.9 The nextfile Statement

The nextfile statement is similar to the next statement. However, instead of abandoning processing of the current record, the nextfile statement instructs awk to stop processing the current data file.

Upon execution of the nextfile statement, FILENAME is updated to the name of the next data file listed on the command line, FNR is reset to one, and processing starts over with the first rule in the program. If the nextfile statement causes the end of the input to be reached, then the code in any END rules is executed. An exception to this is when nextfile is invoked during execution of any statement in an END rule; in this case, it causes the program to stop immediately. See The BEGIN and END Special Patterns.

The nextfile statement is useful when there are many data files to process but it isn’t necessary to process every record in every file. Without nextfile, in order to move on to the next data file, a program would have to continue scanning the unwanted records. The nextfile statement accomplishes this much more efficiently.

In gawk, execution of nextfile causes additional things to happen: any ENDFILE rules are executed if gawk is not currently in an END rule, ARGIND is incremented, and any BEGINFILE rules are executed. (ARGIND hasn’t been introduced yet. See Predefined Variables.)

There is an additional, special, use case with gawk. nextfile is useful inside a BEGINFILE rule to skip over a file that would otherwise cause gawk to exit with a fatal error. In this special case, ENDFILE rules are not executed. See The BEGINFILE and ENDFILE Special Patterns.

Although it might seem that ‘close(FILENAME)’ would accomplish the same as nextfile, this isn’t true. close() is reserved for closing files, pipes, and coprocesses that are opened with redirections. It is not related to the main processing that awk does with the files listed in ARGV.

NOTE: For many years, nextfile was a common extension. In September 2012, it was accepted for inclusion into the POSIX standard. See the Austin Group website.

The current version of BWK awk and mawk also support nextfile. However, they don’t allow the nextfile statement inside function bodies (see User-Defined Functions). gawk does; a nextfile inside a function body reads the first record from the next file and starts processing it with the first rule in the program, just as any other nextfile statement.


7.4.10 The exit Statement

The exit statement causes awk to immediately stop executing the current rule and to stop processing input; any remaining input is ignored. The exit statement is written as follows:

exit [return code]

When an exit statement is executed from a BEGIN rule, the program stops processing everything immediately. No input records are read. However, if an END rule is present, as part of executing the exit statement, the END rule is executed (see The BEGIN and END Special Patterns). If exit is used in the body of an END rule, it causes the program to stop immediately.

An exit statement that is not part of a BEGIN or END rule stops the execution of any further automatic rules for the current record, skips reading any remaining input records, and executes the END rule if there is one. gawk also skips any ENDFILE rules; they do not execute.

In such a case, if you don’t want the END rule to do its job, set a variable to a nonzero value before the exit statement and check that variable in the END rule. See Assertions for an example that does this.

If an argument is supplied to exit, its value is used as the exit status code for the awk process. If no argument is supplied, exit causes awk to return a “success” status. In the case where an argument is supplied to a first exit statement, and then exit is called a second time from an END rule with no argument, awk uses the previously supplied exit value. (d.c.) See gawk’s Exit Status for more information.

For example, suppose an error condition occurs that is difficult or impossible to handle. Conventionally, programs report this by exiting with a nonzero status. An awk program can do this using an exit statement with a nonzero argument, as shown in the following example:

BEGIN {
    if (("date" | getline date_now) <= 0) {
        print "Can't get system date" > "/dev/stderr"
        exit 1
    }
    print "current date is", date_now
    close("date")
}

NOTE: For full portability, exit values should be between zero and 126, inclusive. Negative values, and values of 127 or greater, may not produce consistent results across different operating systems.


7.5 Predefined Variables

Most awk variables are available to use for your own purposes; they never change unless your program assigns values to them, and they never affect anything unless your program examines them. However, a few variables in awk have special built-in meanings. awk examines some of these automatically, so that they enable you to tell awk how to do certain things. Others are set automatically by awk, so that they carry information from the internal workings of awk to your program.

This section documents all of gawk’s predefined variables, most of which are also documented in the chapters describing their areas of activity.


7.5.1 Built-in Variables That Control awk

The following is an alphabetical list of variables that you can change to control how awk does certain things.

The variables that are specific to gawk are marked with a pound sign (‘#’). These variables are gawk extensions. In other awk implementations or if gawk is in compatibility mode (see Command-Line Options), they are not special. (Any exceptions are noted in the description of each variable.)

BINMODE #

On non-POSIX systems, this variable specifies use of binary mode for all I/O. Numeric values of one, two, or three specify that input files, output files, or all files, respectively, should use binary I/O. A numeric value less than zero is treated as zero, and a numeric value greater than three is treated as three. Alternatively, string values of "r" or "w" specify that input files and output files, respectively, should use binary I/O. A string value of "rw" or "wr" indicates that all files should use binary I/O. Any other string value is treated the same as "rw", but causes gawk to generate a warning message. BINMODE is described in more detail in Using gawk on PC Operating Systems. mawk (see Other Freely Available awk Implementations) also supports this variable, but only using numeric values.

CONVFMT

A string that controls the conversion of numbers to strings (see Conversion of Strings and Numbers). It works by being passed, in effect, as the first argument to the sprintf() function (see String-Manipulation Functions). Its default value is "%.6g". CONVFMT was introduced by the POSIX standard.

FIELDWIDTHS #

A space-separated list of columns that tells gawk how to split input with fixed columnar boundaries. Starting in version 4.2, each field width may optionally be preceded by a colon-separated value specifying the number of characters to skip before the field starts. Assigning a value to FIELDWIDTHS overrides the use of FS and FPAT for field splitting. See Reading Fixed-Width Data for more information.

FPAT #

A regular expression (as a string) that tells gawk to create the fields based on text that matches the regular expression. Assigning a value to FPAT overrides the use of FS and FIELDWIDTHS for field splitting. See Defining Fields by Content for more information.

FS

The input field separator (see Specifying How Fields Are Separated). The value is a single-character string or a multicharacter regular expression that matches the separations between fields in an input record. If the value is the null string (""), then each character in the record becomes a separate field. (This behavior is a gawk extension. POSIX awk does not specify the behavior when FS is the null string. Nonetheless, some other versions of awk also treat "" specially.)

The default value is " ", a string consisting of a single space. As a special exception, this value means that any sequence of spaces, TABs, and/or newlines is a single separator. It also causes spaces, TABs, and newlines at the beginning and end of a record to be ignored.

You can set the value of FS on the command line using the -F option:

awk -F, 'program' input-files

If gawk is using FIELDWIDTHS or FPAT for field splitting, assigning a value to FS causes gawk to return to the normal, FS-based field splitting. An easy way to do this is to simply say ‘FS = FS’, perhaps with an explanatory comment.

IGNORECASE #

If IGNORECASE is nonzero or non-null, then all string comparisons and all regular expression matching are case-independent. This applies to regexp matching with ‘~’ and ‘!~’, the gensub(), gsub(), index(), match(), patsplit(), split(), and sub() functions, record termination with RS, and field splitting with FS and FPAT. However, the value of IGNORECASE does not affect array subscripting and it does not affect field splitting when using a single-character field separator. See Case Sensitivity in Matching.

LINT #

When this variable is true (nonzero or non-null), gawk behaves as if the --lint command-line option is in effect (see Command-Line Options). With a value of "fatal", lint warnings become fatal errors. With a value of "invalid", only warnings about things that are actually invalid are issued. (This is not fully implemented yet.) Any other true value prints nonfatal warnings. Assigning a false value to LINT turns off the lint warnings.

This variable is a gawk extension. It is not special in other awk implementations. Unlike with the other special variables, changing LINT does affect the production of lint warnings, even if gawk is in compatibility mode. Much as the --lint and --traditional options independently control different aspects of gawk’s behavior, the control of lint warnings during program execution is independent of the flavor of awk being executed.

OFMT

A string that controls conversion of numbers to strings (see Conversion of Strings and Numbers) for printing with the print statement. It works by being passed as the first argument to the sprintf() function (see String-Manipulation Functions). Its default value is "%.6g". Earlier versions of awk used OFMT to specify the format for converting numbers to strings in general expressions; this is now done by CONVFMT.

OFS

The output field separator (see Output Separators). It is output between the fields printed by a print statement. Its default value is " ", a string consisting of a single space.

ORS

The output record separator. It is output at the end of every print statement. Its default value is "\n", the newline character. (See Output Separators.)

PREC #

The working precision of arbitrary-precision floating-point numbers, 53 bits by default (see Setting the Precision).

ROUNDMODE #

The rounding mode to use for arbitrary-precision arithmetic on numbers, by default "N" (roundTiesToEven in the IEEE 754 standard; see Setting the Rounding Mode).

RS

The input record separator. Its default value is a string containing a single newline character, which means that an input record consists of a single line of text. It can also be the null string, in which case records are separated by runs of blank lines. If it is a regexp, records are separated by matches of the regexp in the input text. (See How Input Is Split into Records.)

The ability for RS to be a regular expression is a gawk extension. In most other awk implementations, or if gawk is in compatibility mode (see Command-Line Options), just the first character of RS’s value is used.

SUBSEP

The subscript separator. It has the default value of "\034" and is used to separate the parts of the indices of a multidimensional array. Thus, the expression ‘foo["A", "B"]’ really accesses foo["A\034B"] (see Multidimensional Arrays).

TEXTDOMAIN #

Used for internationalization of programs at the awk level. It sets the default text domain for specially marked string constants in the source text, as well as for the dcgettext(), dcngettext(), and bindtextdomain() functions (see Internationalization with gawk). The default value of TEXTDOMAIN is "messages".


7.5.2 Built-in Variables That Convey Information

The following is an alphabetical list of variables that awk sets automatically on certain occasions in order to provide information to your program.

The variables that are specific to gawk are marked with a pound sign (‘#’). These variables are gawk extensions. In other awk implementations or if gawk is in compatibility mode (see Command-Line Options), they are not special:

ARGC, ARGV

The command-line arguments available to awk programs are stored in an array called ARGV. ARGC is the number of command-line arguments present. See Other Command-Line Arguments. Unlike most awk arrays, ARGV is indexed from 0 to ARGC − 1. In the following example:

$ awk 'BEGIN {
>         for (i = 0; i < ARGC; i++)
>             print ARGV[i]
>      }' inventory-shipped mail-list
-| awk
-| inventory-shipped
-| mail-list

ARGV[0] contains ‘awk’, ARGV[1] contains ‘inventory-shipped’, and ARGV[2] contains ‘mail-list’. The value of ARGC is three, one more than the index of the last element in ARGV, because the elements are numbered from zero.

The names ARGC and ARGV, as well as the convention of indexing the array from 0 to ARGC − 1, are derived from the C language’s method of accessing command-line arguments.

The value of ARGV[0] can vary from system to system. Also, you should note that the program text is not included in ARGV, nor are any of awk’s command-line options. See Using ARGC and ARGV for information about how awk uses these variables. (d.c.)

ARGIND #

The index in ARGV of the current file being processed. Every time gawk opens a new data file for processing, it sets ARGIND to the index in ARGV of the file name. When gawk is processing the input files, ‘FILENAME == ARGV[ARGIND]’ is always true.

This variable is useful in file processing; it allows you to tell how far along you are in the list of data files as well as to distinguish between successive instances of the same file name on the command line.

While you can change the value of ARGIND within your awk program, gawk automatically sets it to a new value when it opens the next file.

ENVIRON

An associative array containing the values of the environment. The array indices are the environment variable names; the elements are the values of the particular environment variables. For example, ENVIRON["HOME"] might be /home/arnold.

For POSIX awk, changing this array does not affect the environment passed on to any programs that awk may spawn via redirection or the system() function.

However, beginning with version 4.2, if not in POSIX compatibility mode, gawk does update its own environment when ENVIRON is changed, thus changing the environment seen by programs that it creates. You should therefore be especially careful if you modify ENVIRON["PATH"], which is the search path for finding executable programs.

This can also affect the running gawk program, since some of the built-in functions may pay attention to certain environment variables. The most notable instance of this is mktime() (see Time Functions), which pays attention the value of the TZ environment variable on many systems.

Some operating systems may not have environment variables. On such systems, the ENVIRON array is empty (except for ENVIRON["AWKPATH"] and ENVIRON["AWKLIBPATH"]; see The AWKPATH Environment Variable and see The AWKLIBPATH Environment Variable).

ERRNO #

If a system error occurs during a redirection for getline, during a read for getline, or during a close() operation, then ERRNO contains a string describing the error.

In addition, gawk clears ERRNO before opening each command-line input file. This enables checking if the file is readable inside a BEGINFILE pattern (see The BEGINFILE and ENDFILE Special Patterns).

Otherwise, ERRNO works similarly to the C variable errno. Except for the case just mentioned, gawk never clears it (sets it to zero or ""). Thus, you should only expect its value to be meaningful when an I/O operation returns a failure value, such as getline returning −1. You are, of course, free to clear it yourself before doing an I/O operation.

If the value of ERRNO corresponds to a system error in the C errno variable, then PROCINFO["errno"] will be set to the value of errno. For non-system errors, PROCINFO["errno"] will be zero.

FILENAME

The name of the current input file. When no data files are listed on the command line, awk reads from the standard input and FILENAME is set to "-". FILENAME changes each time a new file is read (see Reading Input Files). Inside a BEGIN rule, the value of FILENAME is "", because there are no input files being processed yet.39 (d.c.) Note, though, that using getline (see Explicit Input with getline) inside a BEGIN rule can give FILENAME a value.

FNR

The current record number in the current file. awk increments FNR each time it reads a new record (see How Input Is Split into Records). awk resets FNR to zero each time it starts a new input file.

NF

The number of fields in the current input record. NF is set each time a new record is read, when a new field is created, or when $0 changes (see Examining Fields).

Unlike most of the variables described in this subsection, assigning a value to NF has the potential to affect awk’s internal workings. In particular, assignments to NF can be used to create fields in or remove fields from the current record. See Changing the Contents of a Field.

FUNCTAB #

An array whose indices and corresponding values are the names of all the built-in, user-defined, and extension functions in the program.

NOTE: Attempting to use the delete statement with the FUNCTAB array causes a fatal error. Any attempt to assign to an element of FUNCTAB also causes a fatal error.

NR

The number of input records awk has processed since the beginning of the program’s execution (see How Input Is Split into Records). awk increments NR each time it reads a new record.

PROCINFO #

The elements of this array provide access to information about the running awk program. The following elements (listed alphabetically) are guaranteed to be available:

PROCINFO["argv"]

The PROCINFO["argv"] array contains all of the command-line arguments (after glob expansion and redirection processing on platforms where that must be done manually by the program) with subscripts ranging from 0 through argc − 1. For example, PROCINFO["argv"][0] will contain the name by which gawk was invoked. Here is an example of how this feature may be used:

gawk '
BEGIN {
        for (i = 0; i < length(PROCINFO["argv"]); i++)
                print i, PROCINFO["argv"][i]
}'

Please note that this differs from the standard ARGV array which does not include command-line arguments that have already been processed by gawk (see Using ARGC and ARGV).

PROCINFO["egid"]

The value of the getegid() system call.

PROCINFO["errno"]

The value of the C errno variable when ERRNO is set to the associated error message.

PROCINFO["euid"]

The value of the geteuid() system call.

PROCINFO["FS"]

This is "FS" if field splitting with FS is in effect, "FIELDWIDTHS" if field splitting with FIELDWIDTHS is in effect, "FPAT" if field matching with FPAT is in effect, or "API" if field splitting is controlled by an API input parser.

PROCINFO["gid"]

The value of the getgid() system call.

PROCINFO["identifiers"]

A subarray, indexed by the names of all identifiers used in the text of the awk program. An identifier is simply the name of a variable (be it scalar or array), built-in function, user-defined function, or extension function. For each identifier, the value of the element is one of the following:

"array"

The identifier is an array.

"builtin"

The identifier is a built-in function.

"extension"

The identifier is an extension function loaded via @load or -l.

"scalar"

The identifier is a scalar.

"untyped"

The identifier is untyped (could be used as a scalar or an array; gawk doesn’t know yet).

"user"

The identifier is a user-defined function.

The values indicate what gawk knows about the identifiers after it has finished parsing the program; they are not updated while the program runs.

PROCINFO["platform"]

This element gives a string indicating the platform for which gawk was compiled. The value will be one of the following:

"mingw"

Microsoft Windows, using MinGW.

"os390"

OS/390 (also known as z/OS).

"posix"

GNU/Linux, Cygwin, macOS, and legacy Unix systems.

"vms"

OpenVMS.

PROCINFO["pgrpid"]

The process group ID of the current process.

PROCINFO["pid"]

The process ID of the current process.

PROCINFO["pma"]

The version of the PMA memory allocator compiled into gawk. This element will not be present if the PMA allocator is not available for use. See Preserving Data Between Runs.

PROCINFO["ppid"]

The parent process ID of the current process.

PROCINFO["strftime"]

The default time format string for strftime(). Assigning a new value to this element changes the default. See Time Functions.

PROCINFO["uid"]

The value of the getuid() system call.

PROCINFO["version"]

The version of gawk.

The following additional elements in the array are available to provide information about the MPFR and GMP libraries if your version of gawk supports arbitrary-precision arithmetic (see Arithmetic and Arbitrary-Precision Arithmetic with gawk):

PROCINFO["gmp_version"]

The version of the GNU MP library.

PROCINFO["mpfr_version"]

The version of the GNU MPFR library.

PROCINFO["prec_max"]

The maximum precision supported by MPFR.

PROCINFO["prec_min"]

The minimum precision required by MPFR.

The following additional elements in the array are available to provide information about the version of the extension API, if your version of gawk supports dynamic loading of extension functions (see Writing Extensions for gawk):

PROCINFO["api_major"]

The major version of the extension API.

PROCINFO["api_minor"]

The minor version of the extension API.

On some systems, there may be elements in the array, "group1" through "groupN" for some N. N is the number of supplementary groups that the process has. Use the in operator to test for these elements (see Referring to an Array Element).

The following elements allow you to change gawk’s behavior:

PROCINFO["NONFATAL"]

If this element exists, then I/O errors for all redirections become nonfatal. See Enabling Nonfatal Output.

PROCINFO["name", "NONFATAL"]

Make I/O errors for name be nonfatal. See Enabling Nonfatal Output.

PROCINFO["command", "pty"]

For two-way communication to command, use a pseudo-tty instead of setting up a two-way pipe. See Two-Way Communications with Another Process for more information.

PROCINFO["input_name", "READ_TIMEOUT"]

Set a timeout for reading from input redirection input_name. See Reading Input with a Timeout for more information.

PROCINFO["input_name", "RETRY"]

If an I/O error that may be retried occurs when reading data from input_name, and this array entry exists, then getline returns −2 instead of following the default behavior of returning −1 and configuring input_name to return no further data. An I/O error that may be retried is one where errno has the value EAGAIN, EWOULDBLOCK, EINTR, or ETIMEDOUT. This may be useful in conjunction with PROCINFO["input_name", "READ_TIMEOUT"] or situations where a file descriptor has been configured to behave in a non-blocking fashion. See Retrying Reads After Certain Input Errors for more information.

PROCINFO["sorted_in"]

If this element exists in PROCINFO, its value controls the order in which array indices will be processed by ‘for (indx in array)’ loops. This is an advanced feature, so we defer the full description until later; see Using Predefined Array Scanning Orders with gawk.

RLENGTH

The length of the substring matched by the match() function (see String-Manipulation Functions). RLENGTH is set by invoking the match() function. Its value is the length of the matched string, or −1 if no match is found.

RSTART

The start index in characters of the substring that is matched by the match() function (see String-Manipulation Functions). RSTART is set by invoking the match() function. Its value is the position of the string where the matched substring starts, or zero if no match was found.

RT #

The input text that matched the text denoted by RS, the record separator. It is set every time a record is read.

SYMTAB #

An array whose indices are the names of all defined global variables and arrays in the program. SYMTAB makes gawk’s symbol table visible to the awk programmer. It is built as gawk parses the program and is complete before the program starts to run.

The array may be used for indirect access to read or write the value of a variable:

foo = 5
SYMTAB["foo"] = 4
print foo    # prints 4

The isarray() function (see Getting Type Information) may be used to test if an element in SYMTAB is an array. Also, you may not use the delete statement with the SYMTAB array.

Prior to version 5.0 of gawk, you could use an index for SYMTAB that was not a predefined identifier:

SYMTAB["xxx"] = 5
print SYMTAB["xxx"]

This no longer works, instead producing a fatal error, as it led to rampant confusion.

The SYMTAB array is more interesting than it looks. Andrew Schorr points out that it effectively gives awk data pointers. Consider his example:

# Indirect multiply of any variable by amount, return result
function multiply(variable, amount)
{
    return SYMTAB[variable] *= amount
}

You would use it like this:

BEGIN {
    answer = 10.5
    multiply("answer", 4)
    print "The answer is", answer
}

When run, this produces:

$ gawk -f answer.awk
-| The answer is 42

NOTE: In order to avoid severe time-travel paradoxes,40 neither FUNCTAB nor SYMTAB is available as an element within the SYMTAB array.

Changing NR and FNR

awk increments NR and FNR each time it reads a record, instead of setting them to the absolute value of the number of records read. This means that a program can change these variables and their new values are incremented for each record. (d.c.) The following example shows this:

$ echo '1
> 2
> 3
> 4' | awk 'NR == 2 { NR = 17 }
> { print NR }'
-| 1
-| 17
-| 18
-| 19

Before FNR was added to the awk language (see Major Changes Between V7 and SVR3.1), many awk programs used this feature to track the number of records in a file by resetting NR to zero when FILENAME changed.


7.5.3 Using ARGC and ARGV

Built-in Variables That Convey Information presented the following program describing the information contained in ARGC and ARGV:

$ awk 'BEGIN {
>        for (i = 0; i < ARGC; i++)
>            print ARGV[i]
>      }' inventory-shipped mail-list
-| awk
-| inventory-shipped
-| mail-list

In this example, ARGV[0] contains ‘awk’, ARGV[1] contains ‘inventory-shipped’, and ARGV[2] contains ‘mail-list’. Notice that the awk program is not entered in ARGV. The other command-line options, with their arguments, are also not entered. This includes variable assignments done with the -v option (see Command-Line Options). Normal variable assignments on the command line are treated as arguments and do show up in the ARGV array. Given the following program in a file named showargs.awk:

BEGIN {
    printf "A=%d, B=%d\n", A, B
    for (i = 0; i < ARGC; i++)
        printf "\tARGV[%d] = %s\n", i, ARGV[i]
}
END   { printf "A=%d, B=%d\n", A, B }

Running it produces the following:

$ awk -v A=1 -f showargs.awk B=2 /dev/null
-| A=1, B=0
-|        ARGV[0] = awk
-|        ARGV[1] = B=2
-|        ARGV[2] = /dev/null
-| A=1, B=2

A program can alter ARGC and the elements of ARGV. Each time awk reaches the end of an input file, it uses the next element of ARGV as the name of the next input file. By storing a different string there, a program can change which files are read. Use "-" to represent the standard input. Storing additional elements and incrementing ARGC causes additional files to be read.

If the value of ARGC is decreased, that eliminates input files from the end of the list. By recording the old value of ARGC elsewhere, a program can treat the eliminated arguments as something other than file names.

To eliminate a file from the middle of the list, store the null string ("") into ARGV in place of the file’s name. As a special feature, awk ignores file names that have been replaced with the null string. Another option is to use the delete statement to remove elements from ARGV (see The delete Statement).

All of these actions are typically done in the BEGIN rule, before actual processing of the input begins. See Splitting a Large File into Pieces and see Duplicating Output into Multiple Files for examples of each way of removing elements from ARGV.

To actually get options into an awk program, end the awk options with -- and then supply the awk program’s options, in the following manner:

awk -f myprog.awk -- -v -q file1 file2 …

The following fragment processes ARGV in order to examine, and then remove, the previously mentioned command-line options:

BEGIN {
    for (i = 1; i < ARGC; i++) {
        if (ARGV[i] == "-v")
            verbose = 1
        else if (ARGV[i] == "-q")
            debug = 1
        else if (ARGV[i] ~ /^-./) {
            e = sprintf("%s: unrecognized option -- %c",
                    ARGV[0], substr(ARGV[i], 2, 1))
            print e > "/dev/stderr"
        } else
            break
        delete ARGV[i]
    }
}

Ending the awk options with -- isn’t necessary in gawk. Unless --posix has been specified, gawk silently puts any unrecognized options into ARGV for the awk program to deal with. As soon as it sees an unknown option, gawk stops looking for other options that it might otherwise recognize. The previous command line with gawk would be:

gawk -f myprog.awk -q -v file1 file2 …

Because -q is not a valid gawk option, it and the following -v are passed on to the awk program. (See Processing Command-Line Options for an awk library function that parses command-line options.)

When designing your program, you should choose options that don’t conflict with gawk’s, because it will process any options that it accepts before passing the rest of the command line on to your program. Using ‘#!’ with the -E option may help (see Executable awk Programs and see Command-Line Options).


7.6 Summary

  • Pattern–action pairs make up the basic elements of an awk program. Patterns are either normal expressions, range expressions, or regexp constants; one of the special keywords BEGIN, END, BEGINFILE, or ENDFILE; or empty. The action executes if the current record matches the pattern. Empty (missing) patterns match all records.
  • I/O from BEGIN and END rules has certain constraints. This is also true, only more so, for BEGINFILE and ENDFILE rules. The latter two give you “hooks” into gawk’s file processing, allowing you to recover from a file that otherwise would cause a fatal error (such as a file that cannot be opened).
  • Shell variables can be used in awk programs by careful use of shell quoting. It is easier to pass a shell variable into awk by using the -v option and an awk variable.
  • Actions consist of statements enclosed in curly braces. Statements are built up from expressions, control statements, compound statements, input and output statements, and deletion statements.
  • The control statements in awk are if-else, while, for, and do-while. gawk adds the switch statement. There are two flavors of for statement: one for performing general looping, and the other for iterating through an array.
  • break and continue let you exit early or start the next iteration of a loop (or get out of a switch).
  • next and nextfile let you read the next record and start over at the top of your program or skip to the next input file and start over, respectively.
  • The exit statement terminates your program. When executed from an action (or function body), it transfers control to the END statements. From an END statement body, it exits immediately. You may pass an optional numeric value to be used as awk’s exit status.
  • Some predefined variables provide control over awk, mainly for I/O. Other variables convey information from awk to your program.
  • ARGC and ARGV make the command-line arguments available to your program. Manipulating them from a BEGIN rule lets you control how awk will process the provided data files.

8 Arrays in awk

An array is a table of values called elements. The elements of an array are distinguished by their indices. Indices may be either numbers or strings.

This chapter describes how arrays work in awk, how to use array elements, how to scan through every element in an array, and how to remove array elements. It also describes how awk simulates multidimensional arrays, as well as some of the less obvious points about array usage. The chapter moves on to discuss gawk’s facility for sorting arrays, and ends with a brief description of gawk’s ability to support true arrays of arrays.


8.1 The Basics of Arrays

This section presents the basics: working with elements in arrays one at a time, and traversing all of the elements in an array.


8.1.1 Introduction to Arrays

Doing linear scans over an associative array is like trying to club someone to death with a loaded Uzi.

Larry Wall

The awk language provides one-dimensional arrays for storing groups of related strings or numbers. Every awk array must have a name. Array names have the same syntax as variable names; any valid variable name would also be a valid array name. But one name cannot be used in both ways (as an array and as a variable) in the same awk program.

Arrays in awk superficially resemble arrays in other programming languages, but there are fundamental differences. In awk, it isn’t necessary to specify the size of an array before starting to use it. Additionally, any number or string, not just consecutive integers, may be used as an array index.

In most other languages, arrays must be declared before use, including a specification of how many elements or components they contain. In such languages, the declaration causes a contiguous block of memory to be allocated for that many elements. Usually, an index in the array must be a nonnegative integer. For example, the index zero specifies the first element in the array, which is actually stored at the beginning of the block of memory. Index one specifies the second element, which is stored in memory right after the first element, and so on. It is impossible to add more elements to the array, because it has room only for as many elements as given in the declaration. (Some languages allow arbitrary starting and ending indices—e.g., ‘15 .. 27’—but the size of the array is still fixed when the array is declared.)

A contiguous array of four elements might look like Figure 8.1, conceptually, if the element values are eight, "foo", "", and 30.

A Contiguous Array

Figure 8.1: A contiguous array

Only the values are stored; the indices are implicit from the order of the values. Here, eight is the value at index zero, because eight appears in the position with zero elements before it.

Arrays in awk are different—they are associative. This means that each array is a collection of pairs—an index and its corresponding array element value:

IndexValue
330
1"foo"
08
2""

The pairs are shown in jumbled order because their order is irrelevant.41

One advantage of associative arrays is that new pairs can be added at any time. For example, suppose a tenth element is added to the array whose value is "number ten". The result is:

IndexValue
10"number ten"
330
1"foo"
08
2""

Now the array is sparse, which just means some indices are missing. It has elements 0–3 and 10, but doesn’t have elements 4, 5, 6, 7, 8, or 9.

Another consequence of associative arrays is that the indices don’t have to be nonnegative integers. Any number, or even a string, can be an index. For example, the following is an array that translates words from English to French:

IndexValue
"dog""chien"
"cat""chat"
"one""un"
1"un"

Here we decided to translate the number one in both spelled-out and numeric form—thus illustrating that a single array can have both numbers and strings as indices. (In fact, array subscripts are always strings. There are some subtleties to how numbers work when used as array subscripts; this is discussed in more detail in Using Numbers to Subscript Arrays.) Here, the number 1 isn’t double-quoted, because awk automatically converts it to a string.

The value of IGNORECASE has no effect upon array subscripting. The identical string value used to store an array element must be used to retrieve it. When awk creates an array (e.g., with the split() built-in function), that array’s indices are consecutive integers starting at one. (See String-Manipulation Functions.)

awk’s arrays are efficient—the time to access an element is independent of the number of elements in the array.


8.1.2 Referring to an Array Element

The principal way to use an array is to refer to one of its elements. An array reference is an expression as follows:

array[index-expression]

Here, array is the name of an array. The expression index-expression is the index of the desired element of the array.

The value of the array reference is the current value of that array element. For example, foo[4.3] is an expression referencing the element of array foo at index ‘4.3’.

A reference to an array element that has no recorded value yields a value of "", the null string. This includes elements that have not been assigned any value as well as elements that have been deleted (see The delete Statement).

NOTE: A reference to an element that does not exist automatically creates that array element, with the null string as its value. (In some cases, this is unfortunate, because it might waste memory inside awk.)

Novice awk programmers often make the mistake of checking if an element exists by checking if the value is empty:

# Check if "foo" exists in a:         Incorrect!
if (a["foo"] != "") …

This is incorrect for two reasons. First, it creates a["foo"] if it didn’t exist before! Second, it is valid (if a bit unusual) to set an array element equal to the empty string.

To determine whether an element exists in an array at a certain index, use the following expression:

indx in array

This expression tests whether the particular index indx exists, without the side effect of creating that element if it is not present. The expression has the value one (true) if array[indx] exists and zero (false) if it does not exist. (We use indx here, because ‘index’ is the name of a built-in function.) For example, this statement tests whether the array frequencies contains the index ‘2’:

if (2 in frequencies)
    print "Subscript 2 is present."

Note that this is not a test of whether the array frequencies contains an element whose value is two. There is no way to do that except to scan all the elements. Also, this does not create frequencies[2], while the following (incorrect) alternative does:

if (frequencies[2] != "")
    print "Subscript 2 is present."

8.1.3 Assigning Array Elements

Array elements can be assigned values just like awk variables:

array[index-expression] = value

array is the name of an array. The expression index-expression is the index of the element of the array that is assigned a value. The expression value is the value to assign to that element of the array.


8.1.4 Basic Array Example

The following program takes a list of lines, each beginning with a line number, and prints them out in order of line number. The line numbers are not in order when they are first read—instead, they are scrambled. This program sorts the lines by making an array using the line numbers as subscripts. The program then prints out the lines in sorted order of their numbers. It is a very simple program and gets confused upon encountering repeated numbers, gaps, or lines that don’t begin with a number:

{
    if ($1 > max)
        max = $1
    arr[$1] = $0
}
END {
    for (x = 1; x <= max; x++)
        print arr[x]
}

The first rule keeps track of the largest line number seen so far; it also stores each line into the array arr, at an index that is the line’s number. The second rule runs after all the input has been read, to print out all the lines. When this program is run with the following input:

5  I am the Five man
2  Who are you?  The new number two!
4  . . . And four on the floor
1  Who is number one?
3  I three you.

Its output is:

1  Who is number one?
2  Who are you?  The new number two!
3  I three you.
4  . . . And four on the floor
5  I am the Five man

If a line number is repeated, the last line with a given number overrides the others. Gaps in the line numbers can be handled with an easy improvement to the program’s END rule, as follows:

END {
    for (x = 1; x <= max; x++)
        if (x in arr)
            print arr[x]
}

As mentioned, the program is simplistic. It can be easily confused; for example, by using negative or nonalphabetic line numbers. The point here is merely to demonstrate basic array usage.


8.1.5 Scanning All Elements of an Array

In programs that use arrays, it is often necessary to use a loop that executes once for each element of an array. In other languages, where arrays are contiguous and indices are limited to nonnegative integers, this is easy: all the valid indices can be found by counting from the lowest index up to the highest. This technique won’t do the job in awk, because any number or string can be an array index. So awk has a special kind of for statement for scanning an array:

for (var in array)
    body

This loop executes body once for each index in array that the program has previously used, with the variable var set to that index.

The following program uses this form of the for statement. The first rule scans the input records and notes which words appear (at least once) in the input, by storing a one into the array used with the word as the index. The second rule scans the elements of used to find all the distinct words that appear in the input. It prints each word that is more than 10 characters long and also prints the number of such words. See String-Manipulation Functions for more information on the built-in function length().

# Record a 1 for each word that is used at least once
{
    for (i = 1; i <= NF; i++)
        used[$i] = 1
}
# Find number of distinct words more than 10 characters long
END {
    for (x in used) {
        if (length(x) > 10) {
            ++num_long_words
            print x
        }
    }
    print num_long_words, "words longer than 10 characters"
}

See Generating Word-Usage Counts for a more detailed example of this type.

The order in which elements of the array are accessed by this statement is determined by the internal arrangement of the array elements within awk and in standard awk cannot be controlled or changed. This can lead to problems if new elements are added to array by statements in the loop body; it is not predictable whether the for loop will reach them. Similarly, changing var inside the loop may produce strange results. It is best to avoid such things.

As a point of information, gawk sets up the list of elements to be iterated over before the loop starts, and does not change it. But not all awk versions do so. Consider this program, named loopcheck.awk:

BEGIN {
    a["here"] = "here"
    a["is"] = "is"
    a["a"] = "a"
    a["loop"] = "loop"
    for (i in a) {
        j++
        a[j] = j
        print i
    }
}

Here is what happens when run with gawk (and mawk):

$ gawk -f loopcheck.awk
-| here
-| loop
-| a
-| is

Contrast this to BWK awk:

$ nawk -f loopcheck.awk
-| loop
-| here
-| is
-| a
-| 1

8.1.6 Using Predefined Array Scanning Orders with gawk

This subsection describes a feature that is specific to gawk.

By default, when a for loop traverses an array, the order is undefined, meaning that the awk implementation determines the order in which the array is traversed. This order is usually based on the internal implementation of arrays and will vary from one version of awk to the next.

Often, though, you may wish to do something simple, such as “traverse the array by comparing the indices in ascending order,” or “traverse the array by comparing the values in descending order.” gawk provides two mechanisms that give you this control:

  • Set PROCINFO["sorted_in"] to one of a set of predefined values. We describe this now.
  • Set PROCINFO["sorted_in"] to the name of a user-defined function to use for comparison of array elements. This advanced feature is described later in Controlling Array Traversal and Array Sorting.

The following special values for PROCINFO["sorted_in"] are available:

"@unsorted"

Array elements are processed in arbitrary order, which is the default awk behavior.

"@ind_str_asc"

Order by indices in ascending order compared as strings; this is the most basic sort. (Internally, array indices are always strings, so with ‘a[2*5] = 1’ the index is "10" rather than numeric 10.)

"@ind_num_asc"

Order by indices in ascending order but force them to be treated as numbers in the process. Any index with a non-numeric value will end up positioned as if it were zero.

"@val_type_asc"

Order by element values in ascending order (rather than by indices). Ordering is by the type assigned to the element (see Variable Typing and Comparison Expressions). All numeric values come before all string values, which in turn come before all subarrays. (Subarrays have not been described yet; see Arrays of Arrays.)

If you choose to use this feature in traversing FUNCTAB (see Built-in Variables That Convey Information), then the order is built-in functions first (see Built-in Functions), then user-defined functions (see User-Defined Functions) next, and finally functions loaded from an extension (see Writing Extensions for gawk).

"@val_str_asc"

Order by element values in ascending order (rather than by indices). Scalar values are compared as strings. If the string values are identical, the index string values are compared instead. When comparing non-scalar values, "@val_type_asc" sort ordering is used, so subarrays, if present, come out last.

"@val_num_asc"

Order by element values in ascending order (rather than by indices). Scalar values are compared as numbers. Non-scalar values are compared using "@val_type_asc" sort ordering, so subarrays, if present, come out last. When numeric values are equal, the string values are used to provide an ordering: this guarantees consistent results across different versions of the C qsort() function,42 which gawk uses internally to perform the sorting. If the string values are also identical, the index string values are compared instead.

"@ind_str_desc"

Like "@ind_str_asc", but the string indices are ordered from high to low.

"@ind_num_desc"

Like "@ind_num_asc", but the numeric indices are ordered from high to low.

"@val_type_desc"

Like "@val_type_asc", but the element values, based on type, are ordered from high to low. Subarrays, if present, come out first.

"@val_str_desc"

Like "@val_str_asc", but the element values, treated as strings, are ordered from high to low. If the string values are identical, the index string values are compared instead. When comparing non-scalar values, "@val_type_desc" sort ordering is used, so subarrays, if present, come out first.

"@val_num_desc"

Like "@val_num_asc", but the element values, treated as numbers, are ordered from high to low. If the numeric values are equal, the string values are compared instead. If they are also identical, the index string values are compared instead. Non-scalar values are compared using "@val_type_desc" sort ordering, so subarrays, if present, come out first.

The array traversal order is determined before the for loop starts to run. Changing PROCINFO["sorted_in"] in the loop body does not affect the loop. For example:

$ gawk '
> BEGIN {
>    a[4] = 4
>    a[3] = 3
>    for (i in a)
>        print i, a[i]
> }'
-| 4 4
-| 3 3
$ gawk '
> BEGIN {
>    PROCINFO["sorted_in"] = "@ind_str_asc"
>    a[4] = 4
>    a[3] = 3
>    for (i in a)
>        print i, a[i]
> }'
-| 3 3
-| 4 4

When sorting an array by element values, if a value happens to be a subarray then it is considered to be greater than any string or numeric value, regardless of what the subarray itself contains, and all subarrays are treated as being equal to each other. Their order relative to each other is determined by their index strings.

Here are some additional things to bear in mind about sorted array traversal:

  • The value of PROCINFO["sorted_in"] is global. That is, it affects all array traversal for loops. If you need to change it within your own code, you should see if it’s defined and save and restore the value:
    …
    if ("sorted_in" in PROCINFO)
        save_sorted = PROCINFO["sorted_in"]
    PROCINFO["sorted_in"] = "@val_str_desc" # or whatever
    …
    if (save_sorted)
        PROCINFO["sorted_in"] = save_sorted
    
  • As already mentioned, the default array traversal order is represented by "@unsorted". You can also get the default behavior by assigning the null string to PROCINFO["sorted_in"] or by just deleting the "sorted_in" element from the PROCINFO array with the delete statement. (The delete statement hasn’t been described yet; see The delete Statement.)

In addition, gawk provides built-in functions for sorting arrays; see Sorting Array Values and Indices with gawk.


8.2 Using Numbers to Subscript Arrays

An important aspect to remember about arrays is that array subscripts are always strings. When a numeric value is used as a subscript, it is converted to a string value before being used for subscripting (see Conversion of Strings and Numbers). This means that the value of the predefined variable CONVFMT can affect how your program accesses elements of an array. For example:

xyz = 12.153
data[xyz] = 1
CONVFMT = "%2.2f"
if (xyz in data)
    printf "%s is in data\n", xyz
else
    printf "%s is not in data\n", xyz

This prints ‘12.15 is not in data’. The first statement gives xyz a numeric value. Assigning to data[xyz] subscripts data with the string value "12.153" (using the default conversion value of CONVFMT, "%.6g"). Thus, the array element data["12.153"] is assigned the value one. The program then changes the value of CONVFMT. The test ‘(xyz in data)’ generates a new string value from xyz—this time "12.15"—because the value of CONVFMT only allows two significant digits. This test fails, because "12.15" is different from "12.153".

According to the rules for conversions (see Conversion of Strings and Numbers), integer values always convert to strings as integers, no matter what the value of CONVFMT may happen to be. So the usual case of the following works:

for (i = 1; i <= maxsub; i++)
    do something with array[i]

The “integer values always convert to strings as integers” rule has an additional consequence for array indexing. Octal and hexadecimal constants (see Octal and Hexadecimal Numbers) are converted internally into numbers, and their original form is forgotten. This means, for example, that array[17], array[021], and array[0x11] all refer to the same element!

As with many things in awk, the majority of the time things work as you would expect them to. But it is useful to have a precise knowledge of the actual rules, as they can sometimes have a subtle effect on your programs.


8.3 Using Uninitialized Variables as Subscripts

Suppose it’s necessary to write a program to print the input data in reverse order. A reasonable attempt to do so (with some test data) might look like this:

$ echo 'line 1
> line 2
> line 3' | awk '{ l[lines] = $0; ++lines }
> END {
>     for (i = lines - 1; i >= 0; i--)
>        print l[i]
> }'
-| line 3
-| line 2

Unfortunately, the very first line of input data did not appear in the output!

Upon first glance, we would think that this program should have worked. The variable lines is uninitialized, and uninitialized variables have the numeric value zero. So, awk should have printed the value of l[0].

The issue here is that subscripts for awk arrays are always strings. Uninitialized variables, when used as strings, have the value "", not zero. Thus, ‘line 1’ ends up stored in l[""]. The following version of the program works correctly:

{ l[lines++] = $0 }
END {
    for (i = lines - 1; i >= 0; i--)
       print l[i]
}

Here, the ‘++’ forces lines to be numeric, thus making the “old value” numeric zero. This is then converted to "0" as the array subscript.

Even though it is somewhat unusual, the null string ("") is a valid array subscript. (d.c.) gawk warns about the use of the null string as a subscript if --lint is provided on the command line (see Command-Line Options).


8.4 The delete Statement

To remove an individual element of an array, use the delete statement:

delete array[index-expression]

Once an array element has been deleted, any value the element once had is no longer available. It is as if the element had never been referred to or been given a value. The following is an example of deleting elements in an array:

for (i in frequencies)
    delete frequencies[i]

This example removes all the elements from the array frequencies. Once an element is deleted, a subsequent for statement to scan the array does not report that element and using the in operator to check for the presence of that element returns zero (i.e., false):

delete foo[4]
if (4 in foo)
    print "This will never be printed"

It is important to note that deleting an element is not the same as assigning it a null value (the empty string, ""). For example:

foo[4] = ""
if (4 in foo)
  print "This is printed, even though foo[4] is empty"

It is not an error to delete an element that does not exist. However, if --lint is provided on the command line (see Command-Line Options), gawk issues a warning message when an element that is not in the array is deleted.

All the elements of an array may be deleted with a single statement by leaving off the subscript in the delete statement, as follows:

delete array

Using this version of the delete statement is about three times more efficient than the equivalent loop that deletes each element one at a time.

This form of the delete statement is also supported by BWK awk and mawk, as well as by a number of other implementations.

NOTE: For many years, using delete without a subscript was a common extension. In September 2012, it was accepted for inclusion into the POSIX standard. See the Austin Group website.

The following statement provides a portable but nonobvious way to clear out an array:43

split("", array)

The split() function (see String-Manipulation Functions) clears out the target array first. This call asks it to split apart the null string. Because there is no data to split out, the function simply clears the array and then returns.

CAUTION: Deleting all the elements from an array does not change its type; you cannot clear an array and then use the array’s name as a scalar (i.e., a regular variable). For example, the following does not work:

a[1] = 3
delete a
a = 3

8.5 Multidimensional Arrays

A multidimensional array is an array in which an element is identified by a sequence of indices instead of a single index. For example, a two-dimensional array requires two indices. The usual way (in many languages, including awk) to refer to an element of a two-dimensional array named grid is with grid[x,y].

Multidimensional arrays are supported in awk through concatenation of indices into one string. awk converts the indices into strings (see Conversion of Strings and Numbers) and concatenates them together, with a separator between them. This creates a single string that describes the values of the separate indices. The combined string is used as a single index into an ordinary, one-dimensional array. The separator used is the value of the built-in variable SUBSEP.

For example, suppose we evaluate the expression ‘foo[5,12] = "value"’ when the value of SUBSEP is "@". The numbers 5 and 12 are converted to strings and concatenated with an ‘@’ between them, yielding "5@12"; thus, the array element foo["5@12"] is set to "value".

Once the element’s value is stored, awk has no record of whether it was stored with a single index or a sequence of indices. The two expressions ‘foo[5,12]’ and ‘foo[5 SUBSEP 12] are always equivalent.

The default value of SUBSEP is the string "\034", which contains a nonprinting character that is unlikely to appear in an awk program or in most input data. The usefulness of choosing an unlikely character comes from the fact that index values that contain a string matching SUBSEP can lead to combined strings that are ambiguous. Suppose that SUBSEP is "@"; then ‘foo["a@b", "c"] and ‘foo["a", "b@c"] are indistinguishable because both are actually stored as ‘foo["a@b@c"]’.

To test whether a particular index sequence exists in a multidimensional array, use the same operator (in) that is used for single-dimensional arrays. Write the whole sequence of indices in parentheses, separated by commas, as the left operand:

if ((subscript1, subscript2, …) in array)
    …

Here is an example that treats its input as a two-dimensional array of fields; it rotates this array 90 degrees clockwise and prints the result. It assumes that all lines have the same number of elements:

{
     if (max_nf < NF)
          max_nf = NF
     max_nr = NR
     for (x = 1; x <= NF; x++)
          vector[x, NR] = $x
}
END {
     for (x = 1; x <= max_nf; x++) {
          for (y = max_nr; y >= 1; --y)
               printf("%s ", vector[x, y])
          printf("\n")
     }
}

When given the input:

1 2 3 4 5 6
2 3 4 5 6 1
3 4 5 6 1 2
4 5 6 1 2 3

the program produces the following output:

4 3 2 1
5 4 3 2
6 5 4 3
1 6 5 4
2 1 6 5
3 2 1 6

8.5.1 Scanning Multidimensional Arrays

There is no special for statement for scanning a “multidimensional” array. There cannot be one, because, in truth, awk does not have multidimensional arrays or elements—there is only a multidimensional way of accessing an array.

However, if your program has an array that is always accessed as multidimensional, you can get the effect of scanning it by combining the scanning for statement (see Scanning All Elements of an Array) with the built-in split() function (see String-Manipulation Functions). It works in the following manner:

for (combined in array) {
    split(combined, separate, SUBSEP)
    …
}

This sets the variable combined to each concatenated combined index in the array, and splits it into the individual indices by breaking it apart where the value of SUBSEP appears. The individual indices then become the elements of the array separate.

Thus, if a value is previously stored in array[1, "foo"], then an element with index "1\034foo" exists in array. (Recall that the default value of SUBSEP is the character with code 034.) Sooner or later, the for statement finds that index and does an iteration with the variable combined set to "1\034foo". Then the split() function is called as follows:

split("1\034foo", separate, "\034")

The result is to set separate[1] to "1" and separate[2] to "foo". Presto! The original sequence of separate indices is recovered.


8.6 Arrays of Arrays

gawk goes beyond standard awk’s multidimensional array access and provides true arrays of arrays. Elements of a subarray are referred to by their own indices enclosed in square brackets, just like the elements of the main array. For example, the following creates a two-element subarray at index 1 of the main array a:

a[1][1] = 1
a[1][2] = 2

This simulates a true two-dimensional array. Each subarray element can contain another subarray as a value, which in turn can hold other arrays as well. In this way, you can create arrays of three or more dimensions. The indices can be any awk expressions, including scalars separated by commas (i.e., a regular awk simulated multidimensional subscript). So the following is valid in gawk:

a[1][3][1, "name"] = "barney"

Each subarray and the main array can be of different length. In fact, the elements of an array or its subarray do not all have to have the same type. This means that the main array and any of its subarrays can be nonrectangular, or jagged in structure. You can assign a scalar value to the index 4 of the main array a, even though a[1] is itself an array and not a scalar:

a[4] = "An element in a jagged array"

The terms dimension, row, and column are meaningless when applied to such an array, but we will use “dimension” henceforth to imply the maximum number of indices needed to refer to an existing element. The type of any element that has already been assigned cannot be changed by assigning a value of a different type. You have to first delete the current element, which effectively makes gawk forget about the element at that index:

delete a[4]
a[4][5][6][7] = "An element in a four-dimensional array"

This removes the scalar value from index 4 and then inserts a three-level nested subarray containing a scalar. You can also delete an entire subarray or subarray of subarrays:

delete a[4][5]
a[4][5] = "An element in subarray a[4]"

But recall that you can not delete the main array a and then use it as a scalar.

The built-in functions that take array arguments can also be used with subarrays. For example, the following code fragment uses length() (see String-Manipulation Functions) to determine the number of elements in the main array a and its subarrays:

print length(a), length(a[1]), length(a[1][3])

This results in the following output for our main array a:

2, 3, 1

The ‘subscript in array’ expression (see Referring to an Array Element) works similarly for both regular awk-style arrays and arrays of arrays. For example, the tests ‘1 in a’, ‘3 in a[1]’, and ‘(1, "name") in a[1][3]’ all evaluate to one (true) for our array a.

The ‘for (item in array)’ statement (see Scanning All Elements of an Array) can be nested to scan all the elements of an array of arrays if it is rectangular in structure. In order to print the contents (scalar values) of a two-dimensional array of arrays (i.e., in which each first-level element is itself an array, not necessarily of the same length), you could use the following code:

for (i in array)
    for (j in array[i])
        print array[i][j]

The isarray() function (see Getting Type Information) lets you test if an array element is itself an array:

for (i in array) {
    if (isarray(array[i])) {
        for (j in array[i]) {
            print array[i][j]
        }
    }
    else
        print array[i]
}

If the structure of a jagged array of arrays is known in advance, you can often devise workarounds using control statements. For example, the following code prints the elements of our main array a:

for (i in a) {
    for (j in a[i]) {
        if (j == 3) {
            for (k in a[i][j])
                print a[i][j][k]
        } else
            print a[i][j]
    }
}

See Traversing Arrays of Arrays for a user-defined function that “walks” an arbitrarily dimensioned array of arrays.

Recall that a reference to an uninitialized array element yields a value of "", the null string. This has one important implication when you intend to use a subarray as an argument to a function, as illustrated by the following example:

$ gawk 'BEGIN { split("a b c d", b[1]); print b[1][1] }'
error→ gawk: cmd. line:1: fatal: split: second argument is not an array

The way to work around this is to first force b[1] to be an array by creating an arbitrary index:

$ gawk 'BEGIN { b[1][1] = ""; split("a b c d", b[1]); print b[1][1] }'
-| a

8.7 Summary

  • Standard awk provides one-dimensional associative arrays (arrays indexed by string values). All arrays are associative; numeric indices are converted automatically to strings.
  • Array elements are referenced as array[indx]. Referencing an element creates it if it did not exist previously.
  • The proper way to see if an array has an element with a given index is to use the in operator: ‘indx in array’.
  • Use ‘for (indx in array) …’ to scan through all the individual elements of an array. In the body of the loop, indx takes on the value of each element’s index in turn.
  • The order in which a ‘for (indx in array)’ loop traverses an array is undefined in POSIX awk and varies among implementations. gawk lets you control the order by assigning special predefined values to PROCINFO["sorted_in"].
  • Use ‘delete array[indx]’ to delete an individual element. To delete all of the elements in an array, use ‘delete array’. This latter feature has been a common extension for many years and is now standard, but may not be supported by all commercial versions of awk.
  • Standard awk simulates multidimensional arrays by separating subscript values with commas. The values are concatenated into a single string, separated by the value of SUBSEP. The fact that such a subscript was created in this way is not retained; thus, changing SUBSEP may have unexpected consequences. You can use ‘(sub1, sub2, …) in array’ to see if such a multidimensional subscript exists in array.
  • gawk provides true arrays of arrays. You use a separate set of square brackets for each dimension in such an array: data[row][col], for example. Array elements may thus be either scalar values (number or string) or other arrays.
  • Use the isarray() built-in function to determine if an array element is itself a subarray.

9 Functions

This chapter describes awk’s built-in functions, which fall into three categories: numeric, string, and I/O. gawk provides additional groups of functions to work with values that represent time, do bit manipulation, sort arrays, provide type information, and internationalize and localize programs.

Besides the built-in functions, awk has provisions for writing new functions that the rest of a program can use. The second half of this chapter describes these user-defined functions. Finally, we explore indirect function calls, a gawk-specific extension that lets you determine at runtime what function is to be called.


9.1 Built-in Functions

Built-in functions are always available for your awk program to call. This section defines all the built-in functions in awk; some of these are mentioned in other sections but are summarized here for your convenience.


9.1.1 Calling Built-in Functions

To call one of awk’s built-in functions, write the name of the function followed by arguments in parentheses. For example, ‘atan2(y + z, 1)’ is a call to the function atan2() and has two arguments.

Whitespace is ignored between the built-in function name and the opening parenthesis, but nonetheless it is good practice to avoid using whitespace there. User-defined functions do not permit whitespace in this way, and it is easier to avoid mistakes by following a simple convention that always works—no whitespace after a function name.

Each built-in function accepts a certain number of arguments. In some cases, arguments can be omitted. The defaults for omitted arguments vary from function to function and are described under the individual functions. In some awk implementations, extra arguments given to built-in functions are ignored. However, in gawk, it is a fatal error to give extra arguments to a built-in function.

When a function is called, expressions that create the function’s actual parameters are evaluated completely before the call is performed. For example, in the following code fragment:

i = 4
j = sqrt(i++)

the variable i is incremented to the value five before sqrt() is called with a value of four for its actual parameter. The order of evaluation of the expressions used for the function’s parameters is undefined. Thus, avoid writing programs that assume that parameters are evaluated from left to right or from right to left. For example:

i = 5
j = atan2(++i, i *= 2)

If the order of evaluation is left to right, then i first becomes six, and then 12, and atan2() is called with the two arguments six and 12. But if the order of evaluation is right to left, i first becomes 10, then 11, and atan2() is called with the two arguments 11 and 10.


9.1.2 Generating Boolean Values

This function is specific to gawk. It is not available in compatibility mode (see Command-Line Options):

mkbool(expression)

Return a Boolean-typed value based on the regular Boolean value of expression. Boolean “true” values have numeric value one. Boolean “false” values have numeric zero. This is discussed in more detail in Boolean Typed Values.


9.1.3 Numeric Functions

The following list describes all of the built-in functions that work with numbers. Optional parameters are enclosed in square brackets ([ ]):

atan2(y, x)

Return the arctangent of y / x in radians. You can use ‘pi = atan2(0, -1)’ to retrieve the value of pi.

cos(x)

Return the cosine of x, with x in radians.

exp(x)

Return the exponential of x (e ^ x) or report an error if x is out of range. The range of values x can have depends on your machine’s floating-point representation.

int(x)

Return the nearest integer to x, located between x and zero and truncated toward zero. For example, int(3) is 3, int(3.9) is 3, int(-3.9) is −3, and int(-3) is −3 as well.

log(x)

Return the natural logarithm of x, if x is positive; otherwise, return NaN (“not a number”) on IEEE 754 systems. Additionally, gawk prints a warning message when x is negative.

rand()

Return a random number. The values of rand() are uniformly distributed between zero and one. The value could be zero but is never one.44

Often random integers are needed instead. Following is a user-defined function that can be used to obtain a random nonnegative integer less than n:

function randint(n)
{
    return int(n * rand())
}

The multiplication produces a random number greater than or equal to zero and less than n. Using int(), this result is made into an integer between zero and n − 1, inclusive.

The following example uses a similar function to produce random integers between one and n. This program prints a new random number for each input record:

# Function to roll a simulated die.
function roll(n) { return 1 + int(rand() * n) }
# Roll 3 six-sided dice and
# print total number of points.
{
    printf("%d points\n", roll(6) + roll(6) + roll(6))
}

CAUTION: In most awk implementations, including gawk, rand() starts generating numbers from the same starting number, or seed, each time you run awk.45 Thus, a program generates the same results each time you run it. The numbers are random within one awk run but predictable from run to run. This is convenient for debugging, but if you want a program to do different things each time it is used, you must change the seed to a value that is different in each run. To do this, use srand().

sin(x)

Return the sine of x, with x in radians.

sqrt(x)

Return the positive square root of x. gawk prints a warning message if x is negative. Thus, sqrt(4) is 2.

srand([x])

Set the starting point, or seed, for generating random numbers to the value x.

Each seed value leads to a particular sequence of random numbers.46 Thus, if the seed is set to the same value a second time, the same sequence of random numbers is produced again.

CAUTION: Different awk implementations use different random-number generators internally. Don’t expect the same awk program to produce the same series of random numbers when executed by different versions of awk.

If the argument x is omitted, as in ‘srand()’, then the current date and time of day are used for a seed. This is the way to get random numbers that are truly unpredictable.

The return value of srand() is the previous seed. This makes it easy to keep track of the seeds in case you need to consistently reproduce sequences of random numbers.

POSIX does not specify the initial seed; it differs among awk implementations.


9.1.4 String-Manipulation Functions

The functions in this section look at or change the text of one or more strings.

gawk understands locales (see Where You Are Makes a Difference) and does all string processing in terms of characters, not bytes. This distinction is particularly important to understand for locales where one character may be represented by multiple bytes. Thus, for example, length() returns the number of characters in a string, and not the number of bytes used to represent those characters. Similarly, index() works with character indices, and not byte indices.

CAUTION: A number of functions deal with indices into strings. For these functions, the first character of a string is at position (index) one. This is different from C and the languages descended from it, where the first character is at position zero. You need to remember this when doing index calculations, particularly if you are used to C.

In the following list, optional parameters are enclosed in square brackets ([ ]). Several functions perform string substitution; the full discussion is provided in the description of the sub() function, which comes toward the end, because the list is presented alphabetically.

Those functions that are specific to gawk are marked with a pound sign (‘#’). They are not available in compatibility mode (see Command-Line Options):

asort(source [, dest [, how ] ]) #
asorti(source [, dest [, how ] ]) #

These two functions are similar in behavior, so they are described together.

NOTE: The following description ignores the third argument, how, as it requires understanding features that we have not discussed yet. Thus, the discussion here is a deliberate simplification. (We do provide all the details later on; see Sorting Array Values and Indices with gawk for the full story.)

Both functions return the number of elements in the array source. For asort(), gawk sorts the values of source and replaces the indices of the sorted values of source with sequential integers starting with one. If the optional array dest is specified, then source is duplicated into dest. dest is then sorted, leaving the indices of source unchanged.

When comparing strings, IGNORECASE affects the sorting (see Sorting Array Values and Indices with gawk). If the source array contains subarrays as values (see Arrays of Arrays), they will come last, after all scalar values. Subarrays are not recursively sorted.

For example, if the contents of a are as follows:

a["last"] = "de"
a["first"] = "sac"
a["middle"] = "cul"

A call to asort():

asort(a)

results in the following contents of a:

a[1] = "cul"
a[2] = "de"
a[3] = "sac"

The asorti() function works similarly to asort(); however, the indices are sorted, instead of the values. Thus, in the previous example, starting with the same initial set of indices and values in a, calling ‘asorti(a)’ would yield:

a[1] = "first"
a[2] = "last"
a[3] = "middle"

NOTE: You may not use either SYMTAB or FUNCTAB as the second argument to these functions. Attempting to do so produces a fatal error. You may use them as the first argument, but only if providing a second array to use for the actual sorting.

You are allowed to use the same array for both the source and dest arguments, but doing so only makes sense if you’re also supplying the third argument.

gensub(regexp, replacement, how [, target]) #

Search the target string target for matches of the regular expression regexp. If how is a string beginning with ‘g’ or ‘G’ (short for “global”), then replace all matches of regexp with replacement. Otherwise, treat how as a number indicating which match of regexp to replace. Treat numeric values less than one as if they were one. If no target is supplied, use $0. Return the modified string as the result of the function. The original target string is not changed.

The returned value is always a string, even if the original target was a number or a regexp value.

gensub() is a general substitution function. Its purpose is to provide more features than the standard sub() and gsub() functions.

gensub() provides an additional feature that is not available in sub() or gsub(): the ability to specify components of a regexp in the replacement text. This is done by using parentheses in the regexp to mark the components and then specifying ‘\N’ in the replacement text, where N is a digit from 1 to 9. For example:

$ gawk '
> BEGIN {
>      a = "abc def"
>      b = gensub(/(.+) (.+)/, "\\2 \\1", "g", a)
>      print b
> }'
-| def abc

As with sub(), you must type two backslashes in order to get one into the string. In the replacement text, the sequence ‘\0’ represents the entire matched text, as does the character ‘&’.

The following example shows how you can use the third argument to control which match of the regexp should be changed:

$ echo a b c a b c |
> gawk '{ print gensub(/a/, "AA", 2) }'
-| a b c AA b c

In this case, $0 is the default target string. gensub() returns the new string as its result, which is passed directly to print for printing.

If the how argument is a string that does not begin with ‘g’ or ‘G’, or if it is a number that is less than or equal to zero, only one substitution is performed. If how is zero, gawk issues a warning message.

If regexp does not match target, gensub()’s return value is the original unchanged value of target. Note that, as mentioned above, the returned value is a string, even if target was not.

gsub(regexp, replacement [, target])

Search target for all of the longest, leftmost, nonoverlapping matching substrings it can find and replace them with replacement. The ‘g’ in gsub() stands for “global,” which means replace everywhere. For example:

{ gsub(/Britain/, "United Kingdom"); print }

replaces all occurrences of the string ‘Britain’ with ‘United Kingdom’ for all input records.

The gsub() function returns the number of substitutions made. If the variable to search and alter (target) is omitted, then the entire input record ($0) is used. As in sub(), the characters ‘&’ and ‘\’ are special, and the third argument must be assignable.

index(in, find)

Search the string in for the first occurrence of the string find, and return the position in characters where that occurrence begins in the string in. Consider the following example:

$ awk 'BEGIN { print index("peanut", "an") }'
-| 3

If find is not found, index() returns zero.

With BWK awk and gawk, it is a fatal error to use a regexp constant for find. Other implementations allow it, simply treating the regexp constant as an expression meaning ‘$0 ~ /regexp/’. (d.c.)

length([string])

Return the number of characters in string. If string is a number, the length of the digit string representing that number is returned. For example, length("abcde") is five. By contrast, length(15 * 35) works out to three. In this example, 15 * 35 = 525, and 525 is then converted to the string "525", which has three characters.

If no argument is supplied, length() returns the length of $0.

NOTE: In older versions of awk, the length() function could be called without any parentheses. Doing so is considered poor practice, although the 2008 POSIX standard explicitly allows it, to support historical practice. For programs to be maximally portable, always supply the parentheses.

If length() is called with a variable that has not been used, gawk forces the variable to be a scalar. Other implementations of awk leave the variable without a type. (d.c.) Consider:

$ gawk 'BEGIN { print length(x) ; x[1] = 1 }'
-| 0
error→ gawk: fatal: attempt to use scalar `x' as array
$ nawk 'BEGIN { print length(x) ; x[1] = 1 }'
-| 0

If --lint has been specified on the command line, gawk issues a warning about this.

With gawk and several other awk implementations, when given an array argument, the length() function returns the number of elements in the array. (c.e.) This is less useful than it might seem at first, as the array is not guaranteed to be indexed from one to the number of elements in it. If --lint is provided on the command line (see Command-Line Options), gawk warns that passing an array argument is not portable. If --posix is supplied, using an array argument is a fatal error (see Arrays in awk).

match(string, regexp [, array])

Search string for the longest, leftmost substring matched by the regular expression regexp and return the character position (index) at which that substring begins (one, if it starts at the beginning of string). If no match is found, return zero.

The regexp argument may be either a regexp constant (//) or a string constant (""). In the latter case, the string is treated as a regexp to be matched. See Using Dynamic Regexps for a discussion of the difference between the two forms, and the implications for writing your program correctly.

The order of the first two arguments is the opposite of most other string functions that work with regular expressions, such as sub() and gsub(). It might help to remember that for match(), the order is the same as for the ‘~’ operator: ‘string ~ regexp’.

The match() function sets the predefined variable RSTART to the index. It also sets the predefined variable RLENGTH to the length in characters of the matched substring. If no match is found, RSTART is set to zero, and RLENGTH to −1.

For example:

{
    if ($1 == "FIND")
        regex = $2
    else {
        where = match($0, regex)
        if (where != 0)
            print "Match of", regex, "found at", where, "in", $0
       }
}

This program looks for lines that match the regular expression stored in the variable regex. This regular expression can be changed. If the first word on a line is ‘FIND’, regex is changed to be the second word on that line. Therefore, if given:

FIND ru+n
My program runs
but not very quickly
FIND Melvin
JF+KM
This line is property of Reality Engineering Co.
Melvin was here.

awk prints:

Match of ru+n found at 12 in My program runs
Match of Melvin found at 1 in Melvin was here.

If array is present, it is cleared, and then the zeroth element of array is set to the entire portion of string matched by regexp. If regexp contains parentheses, the integer-indexed elements of array are set to contain the portion of string matching the corresponding parenthesized subexpression. For example:

$ echo foooobazbarrrrr |
> gawk '{ match($0, /(fo+).+(bar*)/, arr)
>         print arr[1], arr[2] }'
-| foooo barrrrr

In addition, multidimensional subscripts are available providing the start index and length of each matched subexpression:

$ echo foooobazbarrrrr |
> gawk '{ match($0, /(fo+).+(bar*)/, arr)
>           print arr[1], arr[2]
>           print arr[1, "start"], arr[1, "length"]
>           print arr[2, "start"], arr[2, "length"]
> }'
-| foooo barrrrr
-| 1 5
-| 9 7

There may not be subscripts for the start and index for every parenthesized subexpression, because they may not all have matched text; thus, they should be tested for with the in operator (see Referring to an Array Element).

The array argument to match() is a gawk extension. In compatibility mode (see Command-Line Options), using a third argument is a fatal error.

patsplit(string, array [, fieldpat [, seps ] ]) #

Divide string into pieces (or “fields”) defined by fieldpat and store the pieces in array and the separator strings in the seps array. The first piece is stored in array[1], the second piece in array[2], and so forth. The third argument, fieldpat, is a regexp describing the fields in string (just as FPAT is a regexp describing the fields in input records). It may be either a regexp constant or a string. If fieldpat is omitted, the value of FPAT is used. patsplit() returns the number of elements created. seps[i] is the possibly null separator string after array[i]. The possibly null leading separator will be in seps[0]. So a non-null string with n fields will have n+1 separators. A null string has no fields or separators.

The patsplit() function splits strings into pieces in a manner similar to the way input lines are split into fields using FPAT (see Defining Fields by Content).

Before splitting the string, patsplit() deletes any previously existing elements in the arrays array and seps.

split(string, array [, fieldsep [, seps ] ])

Divide string into pieces separated by fieldsep and store the pieces in array and the separator strings in the seps array. The first piece is stored in array[1], the second piece in array[2], and so forth. The string value of the third argument, fieldsep, is a regexp describing where to split string (much as FS can be a regexp describing where to split input records). If fieldsep is omitted, the value of FS is used. split() returns the number of elements created. seps is a gawk extension, with seps[i] being the separator string between array[i] and array[i+1]. If fieldsep is a single space, then any leading whitespace goes into seps[0] and any trailing whitespace goes into seps[n], where n is the return value of split() (i.e., the number of elements in array).

The split() function splits strings into pieces in the same way that input lines are split into fields. For example:

split("cul-de-sac", a, "-", seps)

splits the string "cul-de-sac" into three fields using ‘-’ as the separator. It sets the contents of the array a as follows:

a[1] = "cul"
a[2] = "de"
a[3] = "sac"

and sets the contents of the array seps as follows:

seps[1] = "-"
seps[2] = "-"

The value returned by this call to split() is three.

As with input field-splitting, when the value of fieldsep is " ", leading and trailing whitespace is ignored in values assigned to the elements of array but not in seps, and the elements are separated by runs of whitespace. Also, as with input field splitting, if fieldsep is the null string, each individual character in the string is split into its own array element. (c.e.) Additionally, if fieldsep is a single-character string, that string acts as the separator, even if its value is a regular expression metacharacter.

Note, however, that RS has no effect on the way split() works. Even though ‘RS = ""’ causes the newline character to also be an input field separator, this does not affect how split() splits strings.

Modern implementations of awk, including gawk, allow the third argument to be a regexp constant (//) as well as a string. (d.c.) The POSIX standard allows this as well. See Using Dynamic Regexps for a discussion of the difference between using a string constant or a regexp constant, and the implications for writing your program correctly.

Before splitting the string, split() deletes any previously existing elements in the arrays array and seps.

If string is null, the array has no elements. (So this is a portable way to delete an entire array with one statement. See The delete Statement.)

If string does not match fieldsep at all (but is not null), array has one element only. The value of that element is the original string.

In POSIX mode (see Command-Line Options), the fourth argument is not allowed.

sprintf(format, expression1, …)

Return (without printing) the string that printf would have printed out with the same arguments (see Using printf Statements for Fancier Printing). For example:

pival = sprintf("pi = %.2f (approx.)", 22/7)

assigns the string ‘pi = 3.14 (approx.) to the variable pival.

strtonum(str) #

Examine str and return its numeric value. If str begins with a leading ‘0’, strtonum() assumes that str is an octal number. If str begins with a leading ‘0x’ or ‘0X’, strtonum() assumes that str is a hexadecimal number. For example:

$ echo 0x11 |
> gawk '{ printf "%d\n", strtonum($1) }'
-| 17

Using the strtonum() function is not the same as adding zero to a string value; the automatic coercion of strings to numbers works only for decimal data, not for octal or hexadecimal.47

Note also that strtonum() uses the current locale’s decimal point for recognizing numbers (see Where You Are Makes a Difference).

sub(regexp, replacement [, target])

Search target, which is treated as a string, for the leftmost, longest substring matched by the regular expression regexp. Modify the entire string by replacing the matched text with replacement. The modified string becomes the new value of target. Return the number of substitutions made (zero or one).

The regexp argument may be either a regexp constant (//) or a string constant (""). In the latter case, the string is treated as a regexp to be matched. See Using Dynamic Regexps for a discussion of the difference between the two forms, and the implications for writing your program correctly.

This function is peculiar because target is not simply used to compute a value, and not just any expression will do—it must be a variable, field, or array element so that sub() can store a modified value there. If this argument is omitted, then the default is to use and alter $0.48 For example:

str = "water, water, everywhere"
sub(/at/, "ith", str)

sets str to ‘wither, water, everywhere, by replacing the leftmost longest occurrence of ‘at’ with ‘ith’.

If the special character ‘&’ appears in replacement, it stands for the precise substring that was matched by regexp. (If the regexp can match more than one string, then this precise substring may vary.) For example:

{ sub(/candidate/, "& and his wife"); print }

changes the first occurrence of ‘candidate’ to ‘candidate and his wife’ on each input line. Here is another example:

$ awk 'BEGIN {
>         str = "daabaaa"
>         sub(/a+/, "C&C", str)
>         print str
> }'
-| dCaaCbaaa

This shows how ‘&’ can represent a nonconstant string and also illustrates the “leftmost, longest” rule in regexp matching (see How Much Text Matches?).

The effect of this special character (‘&’) can be turned off by putting a backslash before it in the string. As usual, to insert one backslash in the string, you must write two backslashes. Therefore, write ‘\\&’ in a string constant to include a literal ‘&’ in the replacement. For example, the following shows how to replace the first ‘|’ on each line with an ‘&’:

{ sub(/\|/, "\\&"); print }

As mentioned, the third argument to sub() must be a variable, field, or array element. Some versions of awk allow the third argument to be an expression that is not an lvalue. In such a case, sub() still searches for the pattern and returns zero or one, but the result of the substitution (if any) is thrown away because there is no place to put it. Such versions of awk accept expressions like the following:

sub(/USA/, "United States", "the USA and Canada")

For historical compatibility, gawk accepts such erroneous code. However, using any other nonchangeable object as the third parameter causes a fatal error and your program will not run.

Finally, if the regexp is not a regexp constant, it is converted into a string, and then the value of that string is treated as the regexp to match.

substr(string, start [, length ])

Return a length-character-long substring of string, starting at character number start. The first character of a string is character number one.49 For example, substr("washington", 5, 3) returns "ing".

If length is not present, substr() returns the whole suffix of string that begins at character number start. For example, substr("washington", 5) returns "ington". The whole suffix is also returned if length is greater than the number of characters remaining in the string, counting from character start.

If start is less than one, substr() treats it as if it was one. (POSIX doesn’t specify what to do in this case: BWK awk acts this way, and therefore gawk does too.) If start is greater than the number of characters in the string, substr() returns the null string. Similarly, if length is present but less than or equal to zero, the null string is returned.

The string returned by substr() cannot be assigned. Thus, it is a mistake to attempt to change a portion of a string, as shown in the following example:

string = "abcdef"
# try to get "abCDEf", won't work
substr(string, 3, 3) = "CDE"

It is also a mistake to use substr() as the third argument of sub() or gsub():

gsub(/xyz/, "pdq", substr($0, 5, 20))  # WRONG

(Some commercial versions of awk treat substr() as assignable, but doing so is not portable.)

If you need to replace bits and pieces of a string, combine substr() with string concatenation, in the following manner:

string = "abcdef"
…
string = substr(string, 1, 2) "CDE" substr(string, 6)
tolower(string)

Return a copy of string, with each uppercase character in the string replaced with its corresponding lowercase character. Nonalphabetic characters are left unchanged. For example, tolower("MiXeD cAsE 123") returns "mixed case 123".

toupper(string)

Return a copy of string, with each lowercase character in the string replaced with its corresponding uppercase character. Nonalphabetic characters are left unchanged. For example, toupper("MiXeD cAsE 123") returns "MIXED CASE 123".

At first glance, the split() and patsplit() functions appear to be mirror images of each other. But there are differences:

  • split() treats its third argument like FS, with all the special rules involved for FS.
  • Matching of null strings differs. This is discussed in FS Versus FPAT: A Subtle Difference.
Matching the Null String

In awk, the ‘*’ operator can match the null string. This is particularly important for the sub(), gsub(), and gensub() functions. For example:

$ echo abc | awk '{ gsub(/m*/, "X"); print }'
-| XaXbXcX

Although this makes a certain amount of sense, it can be surprising.


9.1.4.1 More about ‘\’ and ‘&’ with sub(), gsub(), and gensub()

CAUTION: This subsubsection has been reported to cause headaches. You might want to skip it upon first reading.

When using sub(), gsub(), or gensub(), and trying to get literal backslashes and ampersands into the replacement text, you need to remember that there are several levels of escape processing going on.

First, there is the lexical level, which is when awk reads your program and builds an internal copy of it to execute. Then there is the runtime level, which is when awk actually scans the replacement string to determine what to generate.

At both levels, awk looks for a defined set of characters that can come after a backslash. At the lexical level, it looks for the escape sequences listed in Escape Sequences. Thus, for every ‘\’ that awk processes at the runtime level, you must type two backslashes at the lexical level. When a character that is not valid for an escape sequence follows the ‘\’, BWK awk and gawk both simply remove the initial ‘\’ and put the next character into the string. Thus, for example, "a\qb" is treated as "aqb".

At the runtime level, the various functions handle sequences of ‘\’ and ‘&’ differently. The situation is (sadly) somewhat complex. Historically, the sub() and gsub() functions treated the two-character sequence ‘\&’ specially; this sequence was replaced in the generated text with a single ‘&’. Any other ‘\’ within the replacement string that did not precede an ‘&’ was passed through unchanged. This is illustrated in Table 9.1.

 You type         sub() sees          sub() generates
 ——–         ———-          —————
     \&              &            The matched text
    \\&             \&            A literal ‘&\\\&             \&            A literal ‘&\\\\&            \\&            A literal ‘\&\\\\\&            \\&            A literal ‘\&\\\\\\&           \\\&            A literal ‘\\&\\q             \q            A literal ‘\q

Table 9.1: Historical escape sequence processing for sub() and gsub()

This table shows the lexical-level processing, where an odd number of backslashes becomes an even number at the runtime level, as well as the runtime processing done by sub(). (For the sake of simplicity, the rest of the following tables only show the case of even numbers of backslashes entered at the lexical level.)

The problem with the historical approach is that there is no way to get a literal ‘\’ followed by the matched text.

Several editions of the POSIX standard attempted to fix this problem but weren’t successful. The details are irrelevant at this point in time.

At one point, the gawk maintainer submitted proposed text for a revised standard that reverts to rules that correspond more closely to the original existing practice. The proposed rules have special cases that