2017-05-09 17:18:46 +03:00
|
|
|
---
|
|
|
|
language: awk
|
|
|
|
filename: learnawk.awk
|
|
|
|
contributors:
|
2017-08-25 11:19:05 +03:00
|
|
|
- ["Marshall Mason", "http://github.com/marshallmason"]
|
|
|
|
|
2017-05-09 17:18:46 +03:00
|
|
|
---
|
|
|
|
|
|
|
|
AWK is a standard tool on every POSIX-compliant UNIX system. It's like a
|
|
|
|
stripped-down Perl, perfect for text-processing tasks and other scripting
|
|
|
|
needs. It has a C-like syntax, but without semicolons, manual memory
|
|
|
|
management, or static typing. It excels at text processing. You can call to it
|
|
|
|
from a shell script, or you can use it as a stand-alone scripting language.
|
|
|
|
|
|
|
|
Why use AWK instead of Perl? Mostly because AWK is part of UNIX. You can always
|
|
|
|
count on it, whereas Perl's future is in question. AWK is also easier to read
|
|
|
|
than Perl. For simple text-processing scripts, particularly ones that read
|
|
|
|
files line by line and split on delimiters, AWK is probably the right tool for
|
|
|
|
the job.
|
|
|
|
|
|
|
|
```awk
|
|
|
|
#!/usr/bin/awk -f
|
|
|
|
|
|
|
|
# Comments are like this
|
|
|
|
|
|
|
|
# AWK programs consist of a collection of patterns and actions. The most
|
|
|
|
# important pattern is called BEGIN. Actions go into brace blocks.
|
|
|
|
BEGIN {
|
|
|
|
|
|
|
|
# BEGIN will run at the beginning of the program. It's where you put all
|
|
|
|
# the preliminary set-up code, before you process any text files. If you
|
|
|
|
# have no text files, then think of BEGIN as the main entry point.
|
|
|
|
|
|
|
|
# Variables are global. Just set them or use them, no need to declare..
|
|
|
|
count = 0
|
|
|
|
|
|
|
|
# Operators just like in C and friends
|
|
|
|
a = count + 1
|
|
|
|
b = count - 1
|
|
|
|
c = count * 1
|
2017-10-14 16:01:05 +03:00
|
|
|
d = count / 1 # integer division
|
2017-05-09 17:18:46 +03:00
|
|
|
e = count % 1 # modulus
|
|
|
|
f = count ^ 1 # exponentiation
|
|
|
|
|
|
|
|
a += 1
|
|
|
|
b -= 1
|
|
|
|
c *= 1
|
|
|
|
d /= 1
|
|
|
|
e %= 1
|
|
|
|
f ^= 1
|
|
|
|
|
|
|
|
# Incrementing and decrementing by one
|
|
|
|
a++
|
|
|
|
b--
|
|
|
|
|
|
|
|
# As a prefix operator, it returns the incremented value
|
|
|
|
++a
|
|
|
|
--b
|
|
|
|
|
|
|
|
# Notice, also, no punctuation such as semicolons to terminate statements
|
|
|
|
|
|
|
|
# Control statements
|
|
|
|
if (count == 0)
|
|
|
|
print "Starting with count of 0"
|
|
|
|
else
|
|
|
|
print "Huh?"
|
|
|
|
|
|
|
|
# Or you could use the ternary operator
|
|
|
|
print (count == 0) ? "Starting with count of 0" : "Huh?"
|
|
|
|
|
|
|
|
# Blocks consisting of multiple lines use braces
|
|
|
|
while (a < 10) {
|
|
|
|
print "String concatenation is done" " with a series" " of"
|
|
|
|
" space-separated strings"
|
|
|
|
print a
|
|
|
|
|
|
|
|
a++
|
|
|
|
}
|
|
|
|
|
|
|
|
for (i = 0; i < 10; i++)
|
|
|
|
print "Good ol' for loop"
|
|
|
|
|
|
|
|
# As for comparisons, they're the standards:
|
|
|
|
a < b # Less than
|
|
|
|
a <= b # Less than or equal
|
|
|
|
a != b # Not equal
|
|
|
|
a == b # Equal
|
|
|
|
a > b # Greater than
|
|
|
|
a >= b # Greater than or equal
|
|
|
|
|
|
|
|
# Logical operators as well
|
|
|
|
a && b # AND
|
|
|
|
a || b # OR
|
|
|
|
|
|
|
|
# In addition, there's the super useful regular expression match
|
|
|
|
if ("foo" ~ "^fo+$")
|
|
|
|
print "Fooey!"
|
|
|
|
if ("boo" !~ "^fo+$")
|
|
|
|
print "Boo!"
|
|
|
|
|
|
|
|
# Arrays
|
|
|
|
arr[0] = "foo"
|
|
|
|
arr[1] = "bar"
|
|
|
|
# Unfortunately, there is no other way to initialize an array. Ya just
|
|
|
|
# gotta chug through every value line by line like that.
|
|
|
|
|
|
|
|
# You also have associative arrays
|
|
|
|
assoc["foo"] = "bar"
|
|
|
|
assoc["bar"] = "baz"
|
|
|
|
|
|
|
|
# And multi-dimensional arrays, with some limitations I won't mention here
|
|
|
|
multidim[0,0] = "foo"
|
|
|
|
multidim[0,1] = "bar"
|
|
|
|
multidim[1,0] = "baz"
|
|
|
|
multidim[1,1] = "boo"
|
|
|
|
|
|
|
|
# You can test for array membership
|
|
|
|
if ("foo" in assoc)
|
|
|
|
print "Fooey!"
|
|
|
|
|
|
|
|
# You can also use the 'in' operator to traverse the keys of an array
|
|
|
|
for (key in assoc)
|
|
|
|
print assoc[key]
|
|
|
|
|
|
|
|
# The command line is in a special array called ARGV
|
|
|
|
for (argnum in ARGV)
|
|
|
|
print ARGV[argnum]
|
|
|
|
|
|
|
|
# You can remove elements of an array
|
|
|
|
# This is particularly useful to prevent AWK from assuming the arguments
|
|
|
|
# are files for it to process
|
|
|
|
delete ARGV[1]
|
|
|
|
|
|
|
|
# The number of command line arguments is in a variable called ARGC
|
|
|
|
print ARGC
|
|
|
|
|
|
|
|
# AWK has several built-in functions. They fall into three categories. I'll
|
|
|
|
# demonstrate each of them in their own functions, defined later.
|
|
|
|
|
|
|
|
return_value = arithmetic_functions(a, b, c)
|
|
|
|
string_functions()
|
|
|
|
io_functions()
|
|
|
|
}
|
|
|
|
|
|
|
|
# Here's how you define a function
|
2017-10-14 16:01:05 +03:00
|
|
|
function arithmetic_functions(a, b, c, d) {
|
2017-05-09 17:18:46 +03:00
|
|
|
|
|
|
|
# Probably the most annoying part of AWK is that there are no local
|
|
|
|
# variables. Everything is global. For short scripts, this is fine, even
|
|
|
|
# useful, but for longer scripts, this can be a problem.
|
|
|
|
|
|
|
|
# There is a work-around (ahem, hack). Function arguments are local to the
|
|
|
|
# function, and AWK allows you to define more function arguments than it
|
|
|
|
# needs. So just stick local variable in the function declaration, like I
|
|
|
|
# did above. As a convention, stick in some extra whitespace to distinguish
|
|
|
|
# between actual function parameters and local variables. In this example,
|
|
|
|
# a, b, and c are actual parameters, while d is merely a local variable.
|
|
|
|
|
|
|
|
# Now, to demonstrate the arithmetic functions
|
|
|
|
|
|
|
|
# Most AWK implementations have some standard trig functions
|
|
|
|
localvar = sin(a)
|
|
|
|
localvar = cos(a)
|
|
|
|
localvar = atan2(a, b) # arc tangent of b / a
|
|
|
|
|
|
|
|
# And logarithmic stuff
|
|
|
|
localvar = exp(a)
|
|
|
|
localvar = log(a)
|
|
|
|
|
|
|
|
# Square root
|
|
|
|
localvar = sqrt(a)
|
|
|
|
|
|
|
|
# Truncate floating point to integer
|
|
|
|
localvar = int(5.34) # localvar => 5
|
|
|
|
|
|
|
|
# Random numbers
|
|
|
|
srand() # Supply a seed as an argument. By default, it uses the time of day
|
|
|
|
localvar = rand() # Random number between 0 and 1.
|
|
|
|
|
|
|
|
# Here's how to return a value
|
|
|
|
return localvar
|
|
|
|
}
|
|
|
|
|
|
|
|
function string_functions( localvar, arr) {
|
|
|
|
|
|
|
|
# AWK, being a string-processing language, has several string-related
|
|
|
|
# functions, many of which rely heavily on regular expressions.
|
|
|
|
|
|
|
|
# Search and replace, first instance (sub) or all instances (gsub)
|
|
|
|
# Both return number of matches replaced
|
|
|
|
localvar = "fooooobar"
|
|
|
|
sub("fo+", "Meet me at the ", localvar) # localvar => "Meet me at the bar"
|
|
|
|
gsub("e+", ".", localvar) # localvar => "m..t m. at th. bar"
|
|
|
|
|
|
|
|
# Search for a string that matches a regular expression
|
|
|
|
# index() does the same thing, but doesn't allow a regular expression
|
|
|
|
match(localvar, "t") # => 4, since the 't' is the fourth character
|
|
|
|
|
|
|
|
# Split on a delimiter
|
|
|
|
split("foo-bar-baz", arr, "-") # a => ["foo", "bar", "baz"]
|
|
|
|
|
|
|
|
# Other useful stuff
|
|
|
|
sprintf("%s %d %d %d", "Testing", 1, 2, 3) # => "Testing 1 2 3"
|
|
|
|
substr("foobar", 2, 3) # => "oob"
|
|
|
|
substr("foobar", 4) # => "bar"
|
|
|
|
length("foo") # => 3
|
|
|
|
tolower("FOO") # => "foo"
|
|
|
|
toupper("foo") # => "FOO"
|
|
|
|
}
|
|
|
|
|
|
|
|
function io_functions( localvar) {
|
|
|
|
|
|
|
|
# You've already seen print
|
|
|
|
print "Hello world"
|
|
|
|
|
|
|
|
# There's also printf
|
|
|
|
printf("%s %d %d %d\n", "Testing", 1, 2, 3)
|
|
|
|
|
|
|
|
# AWK doesn't have file handles, per se. It will automatically open a file
|
|
|
|
# handle for you when you use something that needs one. The string you used
|
|
|
|
# for this can be treated as a file handle, for purposes of I/O. This makes
|
|
|
|
# it feel sort of like shell scripting:
|
|
|
|
|
|
|
|
print "foobar" >"/tmp/foobar.txt"
|
|
|
|
|
|
|
|
# Now the string "/tmp/foobar.txt" is a file handle. You can close it:
|
|
|
|
close("/tmp/foobar.txt")
|
|
|
|
|
|
|
|
# Here's how you run something in the shell
|
|
|
|
system("echo foobar") # => prints foobar
|
|
|
|
|
|
|
|
# Reads a line from standard input and stores in localvar
|
|
|
|
getline localvar
|
|
|
|
|
|
|
|
# Reads a line from a pipe
|
|
|
|
"echo foobar" | getline localvar # localvar => "foobar"
|
|
|
|
close("echo foobar")
|
|
|
|
|
|
|
|
# Reads a line from a file and stores in localvar
|
|
|
|
getline localvar <"/tmp/foobar.txt"
|
|
|
|
close("/tmp/foobar.txt")
|
|
|
|
}
|
|
|
|
|
|
|
|
# As I said at the beginning, AWK programs consist of a collection of patterns
|
|
|
|
# and actions. You've already seen the all-important BEGIN pattern. Other
|
|
|
|
# patterns are used only if you're processing lines from files or standard
|
|
|
|
# input.
|
|
|
|
#
|
|
|
|
# When you pass arguments to AWK, they are treated as file names to process.
|
|
|
|
# It will process them all, in order. Think of it like an implicit for loop,
|
|
|
|
# iterating over the lines in these files. these patterns and actions are like
|
|
|
|
# switch statements inside the loop.
|
|
|
|
|
|
|
|
/^fo+bar$/ {
|
|
|
|
|
|
|
|
# This action will execute for every line that matches the regular
|
|
|
|
# expression, /^fo+bar$/, and will be skipped for any line that fails to
|
|
|
|
# match it. Let's just print the line:
|
|
|
|
|
|
|
|
print
|
|
|
|
|
|
|
|
# Whoa, no argument! That's because print has a default argument: $0.
|
|
|
|
# $0 is the name of the current line being processed. It is created
|
|
|
|
# automatically for you.
|
|
|
|
|
|
|
|
# You can probably guess there are other $ variables. Every line is
|
2017-08-23 11:14:39 +03:00
|
|
|
# implicitly split before every action is called, much like the shell
|
2017-05-09 17:18:46 +03:00
|
|
|
# does. And, like the shell, each field can be access with a dollar sign
|
|
|
|
|
|
|
|
# This will print the second and fourth fields in the line
|
|
|
|
print $2, $4
|
|
|
|
|
|
|
|
# AWK automatically defines many other variables to help you inspect and
|
|
|
|
# process each line. The most important one is NF
|
|
|
|
|
|
|
|
# Prints the number of fields on this line
|
|
|
|
print NF
|
|
|
|
|
|
|
|
# Print the last field on this line
|
|
|
|
print $NF
|
|
|
|
}
|
|
|
|
|
|
|
|
# Every pattern is actually a true/false test. The regular expression in the
|
|
|
|
# last pattern is also a true/false test, but part of it was hidden. If you
|
|
|
|
# don't give it a string to test, it will assume $0, the line that it's
|
|
|
|
# currently processing. Thus, the complete version of it is this:
|
|
|
|
|
|
|
|
$0 ~ /^fo+bar$/ {
|
|
|
|
print "Equivalent to the last pattern"
|
|
|
|
}
|
|
|
|
|
|
|
|
a > 0 {
|
|
|
|
# This will execute once for each line, as long as a is positive
|
|
|
|
}
|
|
|
|
|
|
|
|
# You get the idea. Processing text files, reading in a line at a time, and
|
|
|
|
# doing something with it, particularly splitting on a delimiter, is so common
|
|
|
|
# in UNIX that AWK is a scripting language that does all of it for you, without
|
|
|
|
# you needing to ask. All you have to do is write the patterns and actions
|
|
|
|
# based on what you expect of the input, and what you want to do with it.
|
|
|
|
|
|
|
|
# Here's a quick example of a simple script, the sort of thing AWK is perfect
|
|
|
|
# for. It will read a name from standard input and then will print the average
|
|
|
|
# age of everyone with that first name. Let's say you supply as an argument the
|
|
|
|
# name of a this data file:
|
|
|
|
#
|
|
|
|
# Bob Jones 32
|
|
|
|
# Jane Doe 22
|
|
|
|
# Steve Stevens 83
|
|
|
|
# Bob Smith 29
|
|
|
|
# Bob Barker 72
|
|
|
|
#
|
|
|
|
# Here's the script:
|
|
|
|
|
|
|
|
BEGIN {
|
|
|
|
|
|
|
|
# First, ask the user for the name
|
|
|
|
print "What name would you like the average age for?"
|
|
|
|
|
|
|
|
# Get a line from standard input, not from files on the command line
|
|
|
|
getline name <"/dev/stdin"
|
|
|
|
}
|
|
|
|
|
|
|
|
# Now, match every line whose first field is the given name
|
|
|
|
$1 == name {
|
|
|
|
|
|
|
|
# Inside here, we have access to a number of useful variables, already
|
|
|
|
# pre-loaded for us:
|
|
|
|
# $0 is the entire line
|
|
|
|
# $3 is the third field, the age, which is what we're interested in here
|
|
|
|
# NF is the number of fields, which should be 3
|
|
|
|
# NR is the number of records (lines) seen so far
|
|
|
|
# FILENAME is the name of the file being processed
|
|
|
|
# FS is the field separator being used, which is " " here
|
|
|
|
# ...etc. There are plenty more, documented in the man page.
|
|
|
|
|
|
|
|
# Keep track of a running total and how many lines matched
|
|
|
|
sum += $3
|
|
|
|
nlines++
|
|
|
|
}
|
|
|
|
|
|
|
|
# Another special pattern is called END. It will run after processing all the
|
|
|
|
# text files. Unlike BEGIN, it will only run if you've given it input to
|
|
|
|
# process. It will run after all the files have been read and processed
|
|
|
|
# according to the rules and actions you've provided. The purpose of it is
|
|
|
|
# usually to output some kind of final report, or do something with the
|
|
|
|
# aggregate of the data you've accumulated over the course of the script.
|
|
|
|
|
|
|
|
END {
|
|
|
|
if (nlines)
|
|
|
|
print "The average age for " name " is " sum / nlines
|
|
|
|
}
|
|
|
|
|
|
|
|
```
|
|
|
|
Further Reading:
|
|
|
|
|
|
|
|
* [Awk tutorial](http://www.grymoire.com/Unix/Awk.html)
|
|
|
|
* [Awk man page](https://linux.die.net/man/1/awk)
|
|
|
|
* [The GNU Awk User's Guide](https://www.gnu.org/software/gawk/manual/gawk.html) GNU Awk is found on most Linux systems.
|