v1.1.1 / chapter 3 of 3 / 01 aug 08 / greg goebel / public domain
* This chapter polishes off the discussion by covering a number of topics.
* The Awk programming language was designed to be simple but powerful. It allows a user to perform relatively sophisticated text-manipulation operations through Awk programs written on the command line.
For example, suppose I want to turn a document with single-spacing into a
document with double-spacing. I could easily do that with the following Awk
program:
awk '{print ; print ""}' infile > outfile
Notice how single-quotes (' ') are used to allow using double-quotes (" ")
within the Awk expression. This "hides" special characters from the shell.
We could also do this as follows:
awk "{print ; print \"\"}" infile > outfile
-- but the single-quote method is simpler.
This program does what it supposed to, but it also doubles every blank line
in the input file, which leaves a lot of empty space in the output. That's
easy to fix, just tell Awk to print an extra blank line if the current line
is not blank:
awk '{print ; if (NF != 0) print ""}' infile > outfile
* One of the problems with Awk is that it is ingenious enough to make a user
want to tinker with it, and use it for tasks for which it isn't really
appropriate. For example, we could use Awk to count the number of lines in
a file:
awk 'END {print NR}' infile
-- but this is dumb, because the "wc (word count)" utility gives the same
answer with less bother: "Use the right tool for the job."
Awk is the right tool for slightly more complicated tasks. Once I had a file
containing an email distribution list. The email addresses of various
different groups were placed on consecutive lines in the file, with the
different groups separated by blank lines. If I wanted to quickly and
reliably determine how many people were on the distribution list, I couldn't
use "wc", since, it counts blank lines, but Awk handled it easily:
awk 'NF != 0 {++count} END {print count}' list
* Another problem I ran into was determining the average size of a number of
files. I was creating a set of bitmaps with a scanner and storing them on a
disk. The disk started getting full and I was curious to know just how many
more bitmaps I could store on the disk.
I could obtain the file sizes in bytes using "wc -c" or the "list" utility ("ls -l" or "ll"). A few tests showed that "ll" was faster. Since "ll" lists the file size in the fifth field, all I had to do was sum up the fifth field and divide by NR. There was one slight problem, however: the first line of the output of "ll" listed the total number of sectors used, and had to be skipped.
No problem. I simply entered:
ll | awk 'NR!=1 {s+=$5} END {print "Average: " s/(NR-1)}'
This gave me the average as about 40 KB per file.
* Awk is useful for performing simple iterative computations for which a more
sophisticated language like C might prove overkill. Consider the Fibonacci
sequence:
1 1 2 3 5 8 13 21 34 ...
Each element in the sequence is constructed by adding the two previous
elements together, with the first two elements defined as both "1". It's a
discrete formula for exponential growth. It is very easy to use Awk to
generate this sequence:
awk 'BEGIN {a=1;b=1; while(++x<=10){print a; t=a;a=a+b;b=t}; exit}'
This generates the following output data:
1
2
3
5
8
13
21
34
55
89
BACK_TO_TOP
* Sometimes an Awk program needs to be used repeatedly. In that case, it's
simple to execute the Awk program from a shell script. For example, consider
an Awk script to print each word in a file on a separate line. This could be
done with a script named "words" containing:
awk '{c=split($0, s); for(n=1; n<=c; ++n) print s[n] }' $1
"Words" could them be made executable (using "chmod +x words") and the
resulting shell "program" invoked just like any other command. For example,
"words" could be invoked from the "vi" text editor as follows:
:%!words
This would turn all the text into a list of single words.
* For another example, consider the double-spacing program mentioned
previously. This could be slightly changed to accept standard input, using a
"-" as described earlier, then copied into a file named "double":
awk '{print; if (NF != 0) print ""}' -
-- and then could be invoked from "vi" to double-space all the text in the
editor.
* The next step would be to also allow "double" to perform the reverse
operation: To take a double-spaced file and return it to single-spaced,
using the option:
undouble
The first part of the task is, of course, to design a way of stripping out
the extra blank lines, without destroying the spacing of the original
single-spaced file by taking out all the blank lines. The simplest
approach would be to delete every other blank line in a continuous block of
such blank lines. This won't necessarily preserve the original spacing, but
it will preserve spacing in some form.
The method for achieving this is also simple, and involves using a variable
named "skip". This variable is set to "1" every time a blank line is
skipped, to tell the Awk program NOT to skip the next one. The scheme is as
follows:
BEGIN {set skip to 0}
scan the input:
if skip == 0 if line is blank
skip = 1
else
print the line
get next line of input
if skip == 1 print the line
skip = 0
get next line of input
This translates directly into the following Awk program:
BEGIN {skip = 0}
skip == 0 {if (NF == 0)
{skip = 1}
else
{print};
next}
skip == 1 {print;
skip = 0;
next}
This program could be placed in a separate file, named, say, "undouble.awk",
with the shell script "undouble" written as:
awk -f undouble.awk
It could also be embedded directly in the shell script, using single-quotes
to enclose the program and backslashes ("\") to allow for multiple lines:
awk 'BEGIN {skip = 0} \
skip == 0 {if (NF == 0)
{skip = 1} \
else
{print}; \
next} \
skip == 1 {print; \
skip = 0; \
next}'
Remember that when "\" is used to embed an Awk program in a script file, the
program appears as one line to Awk. A semicolon must be used to
separate commands.
For a more sophisticated example, I have a problem that when I write text
documents, occasionally I'll somehow end up with the same word typed in
twice: "And the result was also also that ... " Such duplicate words are
hard to spot on proofreading, but it is straightforward to write an Awk
program to do the job, scanning through a text file to find duplicate;
printing the duplicate word and the line it is found on if a duplicate is
found; or otherwise printing "no duplicates found".
BEGIN { dups=0; w="xy-zzy" }
{ for( n=1; n<=NF; n++)
{ if ( w == $n ) { print w, "::", $0 ; dups = 1 } ; w = $n }
}
END { if (dups == 0) print "No duplicates found." }
The "w" variable stores each word in the file, comparing it to the next word
in the file; w is initialized to "xy-zzy" since that is unlikely to be a word
in the file. The "dup" variable is initialized to 0 and set to 1 if a
duplicate is found; if it's still 0 at the end of the end, the program prints
the "no duplicate found" message. As with the previous example, we could
put this into a separate file or embed it into a script file.
* These last examples use variables to allow an Awk program to keep track of what it has been doing. Awk, as repeatedly mentioned, operates in a cycle: get a line, process it, get the next line, process it, and so on; to have an Awk program remember things between cycles, it needs to leave a little message for itself in a variable.
For example, say we want to match on a line whose first field has the value
1,000 -- but then print the next line. We could do that as follows:
BEGIN {flag = 0}
$1 == 1000 {flag = 1;
next}
flag == 1 {print;
flag = 0;
next}
This program sets a variable named "flag" when it finds a line starting with
1,000, and then goes and gets the next line of input. The next line of input
is printed, and then "flag" is cleared so the line after that won't be
printed.
If we wanted to print the next five lines, we could do that in much the
same way using a variable named, say, "counter":
BEGIN {counter = 0}
$1 == 1000 {counter = 5;
next}
counter > 0 {print;
counter--;
next}
This program initializes a variable named "counter" to 5 when it finds a line
starting with 1,000; for each of the following 5 lines of input, it prints
them and decrements "counter" until it is zero.
This approach can be taken to as great a level of elaboration as needed.
Suppose we have a list of, say, five different actions on five lines of
input, to be taken after matching a line of input; we can then create a
variable named, say, "state", that stores which item in the list to perform
next. The scheme is generally as follows:
BEGIN {set state to 0}
scan the input:
if match set state to 1
get next line of input
if state == 1 do the first thing in the list
state = 2
get next line of input
if state == 2 do the second thing in the list
state = 3
get next line of input
if state == 3 do the third thing in the list
state = 4
get next line of input
if state == 4 do the fourth thing in the list
state = 5
get next line of input
if state == 5 do the fifth (and last) thing in the list
state = 0
get next line of input
This is called a "state machine". In this case, it's performing a simple
list of actions, but the same approach could also be used to perform a more
complicated branching sequence of actions, such as we might have in a
flowchart instead of a simple list.
We could assign state numbers to the blocks in the flowchart and then use if-then tests for the decision-making blocks to set the state variable to indicate which of the alternate actions should be performed next. However, few Awk programs require such complexities, and going into more elaborate examples here would probably be more confusing than it's worth. The essential thing to remember is that an awk program can leave messages for itself in a variable on one line-scan cycle to tell it what to do on later line-scan cycles.
* Awk is an excellent tool for building UNIX/Linux shell scripts, but there
are potential pitfalls. Say we have a scriptfile named "testscript", and it
takes two filenames as parameters:
testscript myfile1 myfile2
If we're executing Awk commands from a file, handling the two filenames isn't
very difficult. We can initialize variables on the command line as follows:
cat $1 $2 | awk -f testscript.awk f1=$1 f2=$2 > tmpfile
The Awk program will use two variables, "f1" and "f2", that are initialized
from the script command line variables "$1" and "$2".
Where this measure gets obnoxious is when we are specifying Awk commands directly, which is preferable if possible since it reduces the number of files needed to implement a script. The problem is that "$1" and "$2" have different meanings to the scriptfile and to Awk. To the scriptfile, they are command-line parameters, but to Awk they indicate text fields in the input.
The handling of these variables depends on how Awk print fields are defined
-- either enclosed in double-quotes (" ") or in single-quotes (' '). If we
invoke Awk as follows:
awk "{ print \"This is a test: \" $1 }" $1
-- we won't get anything printed for the "$1" variable. If we instead
use single-quotes to ensure that the scriptfile leaves the Awk positional
variables alone, we can insert scriptfile variables by initializing them to
variables on the command line:
awk '{ print "This is a test: " $1 " / parm2 = " f }' f=$2 < $1
This provides the first field in "myfile1" as the first parameter and the
name of "myfile2" as the second parameter.
Remember that Awk is relatively slow and clumsy and should not be regarded as the default tool for all scriptfile jobs. We can use "cat" to append to files, "head" and "tail" to cut off a given number of lines of text from the front or back of a file, "grep" or "fgrep" to find lines in a particular file, and "sed" to do search-replaces on the stream in the file.
* The original version of Awk was developed in 1977. It was optimized for throwing together "one-liners" or short, quick-and-dirty programs. However, some users liked Awk so much that they used it for much more complicated tasks. To quote the language's authors: "Our first reaction to a program that didn't fit on one page was shock and amazement." Some users regarded Awk as their primary programming tool, and many had in fact learned programming using Awk.
After the authors got over their initial consternation, they decided to accept the fact, and enhance Awk to make it a better general-purpose programming tool. The new version of Awk was released in 1985. The new version is often, if not always, known as Nawk ("New Awk") to distinguish it from the old one.
* Nawk incorporates several major improvements. The most important
improvement is that users can define their own functions. For example, the
following Nawk program implements the "signum" function:
{for (field=1; field<=NF; ++field) {print signum($field)}};
function signum(n) {
if (n<0) return -1
else if (n==0) return 0
else return 1}
Function declarations can be placed in a program wherever a match-action
clause can. All parameters are local to the function. Local variables can
be defined inside the function.
* A second improvement is a new function, "getline", that allows input from
files other than those specified in the command line at invocation (as well
as input from pipes). "Getline" can be used in a number of ways:
getline Loads $0 from current input.
getline myvar Loads "myvar" from current input.
getline myfile Loads $0 from "myfile".
getline myvar myfile Loads "myvar" from "myfile".
command | getline Loads $0 from output of "command".
command | getline myvar Loads "myvar" from output of "command".
* A related function, "close", allows a file to be closed so it can be read
from the beginning again:
close("myfile")
* A new function, "system", allows Awk programs to invoke system commands:
system("rm myfile")
* Command-line parameters can be interpreted using two new predefined
variables, ARGC and ARGV, a mechanism instantly familiar to C programmers.
ARGC ("argument count") gives the number of command-line elements, and ARGV
("argument vector") is an array whose entries store the elements
individually.
* There is a new conditional-assignment expression, known as "?:", which is
used as follows:
status = (condition == "green")? "go" : "stop"
This translates to:
if (condition=="green") {status = "go"} else {status = "stop"}
This construct should also be familiar to C programmers.
* There are new math functions, such as trig and random-number functions:
sin(x) Sine, with x in radians.
cos(x) Cosine, with x in radians.
atan2(y,z) Arctangent of y/x, in range -PI to PI.
rand() Random number, with 0 <= number < 1.
srand() Seed for random-number generator.
* There are new string functions, such as match and substitution functions:
Search the target string for the search string; return 0 if no match, return starting index of search string if match. Also sets built-in variable RSTART to the starting index, and sets built-in variable RLENGTH to the matched string's length.
Search for first match of regular expression in $0 and substitute replacement string. This function returns the number of substitutions made, as do the other substitution functions.
Search for first match of regular expression in target string and substitute replacement string.
Search for all matches of regular expression in $0 and substitute replacement string.
Search for all matches of regular expression in target string and substitute replacement string.
* There is a mechanism for handling multidimensional arrays. For example,
the following program creates and prints a matrix, and then prints the
transposition of the matrix:
BEGIN {count = 1;
for (row = 1; row <= 5; ++row) {
for (col = 1; col <= 3; ++col) {
printf("%4d",count);
array[row,col] = count++; }
printf("\n"); }
printf("\n");
for (col = 1; col <= 3; ++col) {
for (row = 1; row <= 5; ++row) {
printf("%4d",array[row,col]); }
printf("\n"); }
exit; }
This yields:
1 2 3
4 5 6
7 8 9
10 11 12
13 14 15
1 4 7 10 13
2 5 8 11 14
3 6 9 12 15
Nawk also includes a new "delete" function, which deletes array elements:
delete(array[count])
* Characters can be expressed as octal codes. "\033", for example, can be
used to define an "escape" character.
* A new built-in variable, FNR, keeps track of the record number of the current file, as opposed to NR, which keeps track of the record number of the current line of input, regardless of how many files have contributed to that input. Its behavior is otherwise exactly identical to that of NR.
* While Nawk does have useful refinements, they are generally intended to support the development of complicated programs. My feeling is that Nawk represents overkill for all but the most dedicated Awk users, and in any case would require a substantial document of its own to do its capabilities justice. Those who would like to know more about Nawk are encouraged to read THE AWK PROGRAMMING LANGUAGE by Aho / Weinberger / Kernighan. This short, terse, detailed book outlines the capabilities of Nawk and provides sophisticated examples of its use.
* This final section provides a convenient lookup reference for Awk programming.
* Invoking Awk:
awk [-F<ch>] {pgm} | {-f <pgm file>} [<vars>] [-|<data file>]
-- where:
ch: Field-separator character.
pgm: Awk command-line program.
pgm file: File containing an Awk program.
vars: Awk variable initializations.
data file: Input data file.
* General form of Awk program:
BEGIN {<initializations>}
<search pattern 1> {<program actions>}
<search pattern 2> {<program actions>}
...
END {<final actions>}
* Search patterns:
/<string>/ Search for string.
/^<string>/ Search for string at beginning of line.
/<string>$/ Search for string at end of line.
The search can be constrained to particular fields:
$<field> ~ /<string>/ Search for string in specified field.
$<field> !~ /<string>/ Search for string \Inot\i in specified field.
Strings can be ORed in a search:
/(<string1>)|(<string2>)/
The search can be for an entire range of lines, bounded by two strings:
/<string1>/,/<string2>/
The search can be for any condition, such as line number, and can use the
following comparison operators:
== != < > <= >=
Different conditions can be ORed with "||" or ANDed with "&&".
[<charlist or range>] Match on any character in list or range.
[^<charlist or range>] Match on any character not in list or range.
. Match any single character.
* Match 0 or more occurrences of preceding string.
? Match 0 or 1 occurrences of preceding string.
+ Match 1 or more occurrences of preceding string.
If a metacharacter is part of the search string, it can be "escaped" by
preceding it with a "\".
* Special characters:
\n Newline (line feed).
Backspace.
\r Carriage return.
\f Form feed.
A "\" can be embedded in a string by entering it twice: "\\".
* Built-in variables:
$0; $1,$2,$3,... Field variables.
NR Number of records (lines).
NF Number of fields.
FILENAME Current input filename.
FS Field separator character (default: " ").
RS Record separator character (default: "\n").
OFS Output field separator (default: " ").
ORS Output record separator (default: "\n").
OFMT Output format (default: "%.6g").
* Arithmetic operations:
+ Addition.
- Subtraction.
* Multiplication.
/ Division.
% Mod.
++ Increment.
-- Decrement.
Shorthand assignments:
x += 2 -- is the same as: x = x + 2
x -= 2 -- is the same as: x = x - 2
x *= 2 -- is the same as: x = x * 2
x /= 2 -- is the same as: x = x / 2
x %= 2 -- is the same as: x = x % 2
* The only unique string operation is concatenation, which is performed simply
by listing two strings connected by a blank space.
* Arithmetic functions:
sqrt() Square root.
log() Base \Ie\i log.
exp() Power of \Ie\i.
int() Integer part of argument.
* String functions:
Length of string.
Get substring.
Split string into array, with initial array index being 1.
Find index of search string in target string.
Perform formatted print into string.
* Control structures:
if (<condition>) <action 1> [else <action 2>]
while (<condition>) <action>
for (<initial action>;<condition>;<end-of-loop action>) <action>
Scanning through an associative array with "for":
for (<variable> in <array>) <action>
Unconditional control statements:
break Break out of "while" or "for" loop.
continue Perform next iteration of "while" or "for" loop.
next Get and scan next line of input.
exit Finish reading input and perform END statements.
* Print:
print <i1>, <i2>, ... Print items separated by OFS; end with newline.
print <i1> <i2> ... Print items concatenated; end with newline.
* Printf():
General format:
printf(<string with format codes>,[<parameters>])
Newlines must be explicitly specified with a "\n".
General form of format code:
%[<number>]<format code>
The optional "number" can consist of:
The format codes are:
d Prints a number in decimal format.
o Prints a number in octal format.
x Prints a number in hexadecimal format.
c Prints a character, given its numeric code.
s Prints a string.
e Prints a number in exponential format.
f Prints a number in floating-point format.
g Prints a number in exponential or floating-point format.
* Awk can perform output redirection (using ">" and ">>") and piping (using
"|") from both "print" and "printf".
* Revision history:
v1.0 / 11 mar 90 / gvg
v1.1 / 29 nov 94 / gvg / Cosmetic rewrite.
v1.2 / 12 oct 95 / gvg / Web rewrite, added stuff on shell scripts.
v1.3 / 15 jan 99 / gvg / Minor cosmetic update.
v1.0.4 / 01 jan 02 / gvg / Minor cosmetic update.
v1.0.5 / 01 jan 04 / gvg / Minor cosmetic update.
v1.0.6 / 01 may 04 / gvg / Added comments on state variables.
v1.0.7 / 01 jun 04 / gvg / Added comments on numeric / string comparisons.
v1.0.8 / 01 jul 04 / gvg / Corrected an obnoxious typo error.
v1.0.9 / 01 oct 04 / gvg / Corrected another tweaky error.
v1.1.0 / 01 sep 06 / gvg / Minor update, changed title to "Primer".
v1.1.1 / 01 aug 08 / gvg / Minor cosmetic update.
BACK_TO_TOP