Set variable in current shell from awk - bash

Question

Is there a way to set a variable in my current shell from within awk?
I'd like to do some processing on a file and print out some data; since I'll read the whole file through, I'd like to save the number of lines -- in this case, FNR.
Happens though I can't seem to find a way to set a shell variable with FNR value; if not this, I'd have to read the FNR from my output file, to set, say num_lines, with FNR value.
I've tried some combinations using awk 'END{system(...)}', but could not manage it to work. Any way around this?

Answer 1

Here's another way.
This is especially useful when when you've got the values of your variables in a single variable and you want split them up. For example, you have a list of values from a single row in a database that you want to create variables out of.
val="hello|beautiful|world" # assume this string comes from a database query
read a b c <<< $( echo ${val} | awk -F"|" '{print $1" "$2" "$3}' )
echo $a #hello
echo $b #beautiful
echo $c #world
We need the 'here string' i.e <<< in this case, because the read command does not read from a pipe and instead reads from stdin

Answer 2

$ echo "$var"
$ declare $( awk 'BEGIN{print "var=17"}' )
$ echo "$var"
17
Here's why you should use declare instead of eval:
$ eval $( awk 'BEGIN{print "echo \"removing all of your files, ha ha ha....\""}' )
removing all of your files, ha ha ha....
$ declare $( awk 'BEGIN{print "echo \"removing all of your files\""}' )
bash: declare: `"removing': not a valid identifier
bash: declare: `files"': not a valid identifier
Note in the first case that eval executes whatever string awk prints, which could accidentally be a very bad thing!

Answer 3

You can't export variables from a subshell to its parent shell. You have some other choices, though, including:
Make another pass of the file using AWK to count records, and use command substitution to capture the result. For example:
FNR=$(awk 'END {print FNR}' filename)
Print FNR in the subshell, and parse the output in your other process.
If FNR is the same as number of lines, you can call wc -l < filename to get your count.

Answer 4

A warning for anyone trying to use declare as suggested by several answers.
eval does not have this problem.
If the awk (or other expression) provided to declare results in an empty string then declare will dump the current environment.
This is almost certainly not what you would want.
eg: if your awk pattern doesn't exist in the input you will never print an output, therefore you will end up with unexpected behaviour.
An example of this....
unset var
var=99
declare $( echo "foobar" | awk '/fail/ {print "var=17"}' )
echo "var=$var"
var=99
The current environment as seen by declare is printed
and $var is not changed
A minor change to store the value to set in an awk variable and print it at the end solves this....
unset var
var=99
declare $( echo "foobar" | awk '/fail/ {tmp="17"} END {print "var="tmp}' )
echo "var=$var"
var=
This time $var is unset ie: set to the null string var=''
and there is no unwanted output.
To show this working with a matching pattern
unset var
var=99
declare $( echo "foobar" | awk '/foo/ {tmp="17"} END {print "var="tmp}' )
echo "var=$var"
var=
This time $var is unset ie: set to the null string var=''
and there is no unwanted output.

Answer 5

Make awk print out the assignment statement:
MYVAR=NewValue
Then in your shell script, eval the output of your awk script:
eval $(awk ....)
# then use $MYVAR
EDIT: people recommend using declare instead of eval, to be slightly less error-prone if something other than the assignment is printed by the inner script. It's bash-only, but it's okay when the shell is bash and the script has #!/bin/bash, correctly stating this dependency.
The eval $(...) variant is widely used, with existing programs generating output suitable for eval but not for declare (lesspipe is an example); that's why it's important to understand it, and the bash-only variant is "too localized".

Answer 6

To synthesize everything here so far I'll share what I find is useful to set a shell environment variable from a script that reads a one-line file using awk. Obviously a /pattern/ could be used instead of NR==1 to find the needed variable.
# export a variable from a script (such as in a .dotfile)
declare $( awk 'NR==1 {tmp=$1} END {print "SHELL_VAR=" tmp}' /path/to/file )
export SHELL_VAR
This will avoid a massive output of variables if a declare command is issued with no argument, as well as the security risks of a blind eval.

Answer 7

echo "First arg: $1"
for ((i=0 ; i < $1 ; i++)); do
echo "inside"
echo "Welcome $i times."
cat man.xml | awk '{ x[NR] = $0 } END { for ( i=2 ; i<=NR ; i++ ) { if (x[i] ~ // ) {x[i+1]=" '$i'"}print x[i] }} ' > $i.xml
done
echo "compleated"