Associative arrays in shell scripts

Another option, if portability is not your main concern, is to use associative arrays that are built in to the shell. This should work in bash 4.0 (available now on most major distros, though not on OS X unless you install it yourself), ksh, and zsh:

declare -A newmap
newmap[name]="Irfan Zulfiqar"
newmap[designation]=SSE
newmap[company]="My Own Company"

echo ${newmap[company]}
echo ${newmap[name]}

Depending on the shell, you may need to do a typeset -A newmap instead of declare -A newmap, or in some it may not be necessary at all.

Another non-bash 4 way.

#!/bin/bash

# A pretend Python dictionary with bash 3 
ARRAY=( "cow:moo"
        "dinosaur:roar"
        "bird:chirp"
        "bash:rock" )

for animal in "${ARRAY[@]}" ; do
    KEY=${animal%%:*}
    VALUE=${animal#*:}
    printf "%s likes to %s.\n" "$KEY" "$VALUE"
done

echo -e "${ARRAY[1]%%:*} is an extinct animal which likes to ${ARRAY[1]#*:}\n"

You could throw an if statement for searching in there as well. if [[ $var =~ /blah/ ]]. or whatever.

I think that you need to step back and think about what a map, or associative array, really is. All it is is a way to store a value for a given key, and get that value back quickly and efficiently. You may also want to be able to iterate over the keys to retrieve every key value pair, or delete keys and their associated values.

Now, think about a data structure you use all the time in shell scripting, and even just in the shell without writing a script, that has these properties. Stumped? It's the filesystem.

Really, all you need to have an associative array in shell programming is a temp directory. mktemp -d is your associative array constructor:

prefix=$(basename -- "$0")
map=$(mktemp -dt ${prefix})
echo >${map}/key somevalue
value=$(cat ${map}/key)

If you don't feel like using echo and cat, you can always write some little wrappers; these ones are modelled off of Irfan's, though they just output the value rather than setting arbitrary variables like $value:

#!/bin/sh

prefix=$(basename -- "$0")
mapdir=$(mktemp -dt ${prefix})
trap 'rm -r ${mapdir}' EXIT

put() {
  [ "$#" != 3 ] && exit 1
  mapname=$1; key=$2; value=$3
  [ -d "${mapdir}/${mapname}" ] || mkdir "${mapdir}/${mapname}"
  echo $value >"${mapdir}/${mapname}/${key}"
}

get() {
  [ "$#" != 2 ] && exit 1
  mapname=$1; key=$2
  cat "${mapdir}/${mapname}/${key}"
}

put "newMap" "name" "Irfan Zulfiqar"
put "newMap" "designation" "SSE"
put "newMap" "company" "My Own Company"

value=$(get "newMap" "company")
echo $value

value=$(get "newMap" "name")
echo $value

edit: This approach is actually quite a bit faster than the linear search using sed suggested by the questioner, as well as more robust (it allows keys and values to contain -, =, space, qnd ":SP:"). The fact that it uses the filesystem does not make it slow; these files are actually never guaranteed to be written to the disk unless you call sync; for temporary files like this with a short lifetime, it's not unlikely that many of them will never be written to disk.

I did a few benchmarks of Irfan's code, Jerry's modification of Irfan's code, and my code, using the following driver program:

#!/bin/sh

mapimpl=$1
numkeys=$2
numvals=$3

. ./${mapimpl}.sh    #/ <- fix broken stack overflow syntax highlighting

for (( i = 0 ; $i < $numkeys ; i += 1 ))
do
    for (( j = 0 ; $j < $numvals ; j += 1 ))
    do
        put "newMap" "key$i" "value$j"
        get "newMap" "key$i"
    done
done

The results:

    $ time ./driver.sh irfan 10 5

    real    0m0.975s
    user    0m0.280s
    sys     0m0.691s

    $ time ./driver.sh brian 10 5

    real    0m0.226s
    user    0m0.057s
    sys     0m0.123s

    $ time ./driver.sh jerry 10 5

    real    0m0.706s
    user    0m0.228s
    sys     0m0.530s

    $ time ./driver.sh irfan 100 5

    real    0m10.633s
    user    0m4.366s
    sys     0m7.127s

    $ time ./driver.sh brian 100 5

    real    0m1.682s
    user    0m0.546s
    sys     0m1.082s

    $ time ./driver.sh jerry 100 5

    real    0m9.315s
    user    0m4.565s
    sys     0m5.446s

    $ time ./driver.sh irfan 10 500

    real    1m46.197s
    user    0m44.869s
    sys     1m12.282s

    $ time ./driver.sh brian 10 500

    real    0m16.003s
    user    0m5.135s
    sys     0m10.396s

    $ time ./driver.sh jerry 10 500

    real    1m24.414s
    user    0m39.696s
    sys     0m54.834s

    $ time ./driver.sh irfan 1000 5

    real    4m25.145s
    user    3m17.286s
    sys     1m21.490s

    $ time ./driver.sh brian 1000 5

    real    0m19.442s
    user    0m5.287s
    sys     0m10.751s

    $ time ./driver.sh jerry 1000 5

    real    5m29.136s
    user    4m48.926s
    sys     0m59.336s

To add to Irfan's answer, here is a shorter and faster version of get() since it requires no iteration over the map contents:

get() {
    mapName=$1; key=$2

    map=${!mapName}
    value="$(echo $map |sed -e "s/.*--${key}=\([^ ]*\).*/\1/" -e 's/:SP:/ /g' )"
}

Yet another non-bash-4 (i.e., bash 3, Mac-compatible) way:

val_of_key() {
    case $1 in
        'A1') echo 'aaa';;
        'B2') echo 'bbb';;
        'C3') echo 'ccc';;
        *) echo 'zzz';;
    esac
}

for x in 'A1' 'B2' 'C3' 'D4'; do
    y=$(val_of_key "$x")
    echo "$x => $y"
done

Prints:

A1 => aaa
B2 => bbb
C3 => ccc
D4 => zzz

The function with the case acts like an associative array. Unfortunately it cannot use return, so it has to echo its output, but this is not a problem, unless you are a purist that shuns forking subshells.

Now answering this question.

Following scripts simulates associative arrays in shell scripts. Its simple and very easy to understand.

Map is nothing but a never ending string that has keyValuePair saved as --name=Irfan --designation=SSE --company=My:SP:Own:SP:Company

spaces are replaced with ':SP:' for values

put() {
    if [ "$#" != 3 ]; then exit 1; fi
    mapName=$1; key=$2; value=`echo $3 | sed -e "s/ /:SP:/g"`
    eval map="\"\$$mapName\""
    map="`echo "$map" | sed -e "s/--$key=[^ ]*//g"` --$key=$value"
    eval $mapName="\"$map\""
}

get() {
    mapName=$1; key=$2; valueFound="false"

    eval map=\$$mapName

    for keyValuePair in ${map};
    do
        case "$keyValuePair" in
            --$key=*) value=`echo "$keyValuePair" | sed -e 's/^[^=]*=//'`
                      valueFound="true"
        esac
        if [ "$valueFound" == "true" ]; then break; fi
    done
    value=`echo $value | sed -e "s/:SP:/ /g"`
}

put "newMap" "name" "Irfan Zulfiqar"
put "newMap" "designation" "SSE"
put "newMap" "company" "My Own Company"

get "newMap" "company"
echo $value

get "newMap" "name"
echo $value

edit: Just added another method to fetch all keys.

getKeySet() {
    if [ "$#" != 1 ]; 
    then 
        exit 1; 
    fi

    mapName=$1; 

    eval map="\"\$$mapName\""

    keySet=`
           echo $map | 
           sed -e "s/=[^ ]*//g" -e "s/\([ ]*\)--/\1/g"
          `
}

Adding another option, if jq is available:

export NAMES="{
  \"Mary\":\"100\",
  \"John\":\"200\",
  \"Mary\":\"50\",
  \"John\":\"300\",
  \"Paul\":\"100\",
  \"Paul\":\"400\",
  \"David\":\"100\"
}"
export NAME=David
echo $NAMES | jq --arg v "$NAME" '.[$v]' | tr -d '"' 

I've found it true, as already mentioned, that the best performing method is to write out key/vals to a file, and then use grep/awk to retrieve them. It sounds like all sorts of unnecessary IO, but disk cache kicks in and makes it extremely efficient -- much faster than trying to store them in memory using one of the above methods (as the benchmarks show).

Here's a quick, clean method I like:

hinit() {
    rm -f /tmp/hashmap.$1
}

hput() {
    echo "$2 $3" >> /tmp/hashmap.$1
}

hget() {
    grep "^$2 " /tmp/hashmap.$1 | awk '{ print $2 };'
}

hinit capitols
hput capitols France Paris
hput capitols Netherlands Amsterdam
hput capitols Spain Madrid

echo `hget capitols France` and `hget capitols Netherlands` and `hget capitols Spain`

If you wanted to enforce single-value per key, you could also do a little grep/sed action in hput().