Let's say I have an associative array in bash,

declare -A hash

where both keys and values are unknown to me (the actual data is read from external sources).

How may I create an array of the keys corresponding to the same value, so that I may, in a loop over all unique values, do

printf 'Value "%s" is present with the following keys: %s\n' "$value" "${keys[*]}"

and get the output (not necessarily in this order)

Value "aa" is present with the following keys: foo baz wobble
Value "bb" is present with the following keys: bar quux
Value "cc" is present with the following keys: wibble

The important bit is that the keys are stored as separate elements in the keys array and that they therefore do not need to be parsed out of a text string.

I could do something like

declare -A seen
for value in "${hash[@]}"; do
    if [ -n "${seen[$value]}" ]; then

    for key in "${!hash[@]}"; do
        if [ "${hash[$key]}" = "$value" ]; then
            keys+=( "$key" )

    printf 'Value "%s" is present with the following keys: %s\n' \
        "$value" "${keys[*]}"


But it seems a bit inefficient with that double loop.

Is there a piece of array syntax that I've missed for bash?

Would doing this in e.g. zsh give me access to more powerful array manipulation tools?

In Perl, I would do

my %hash = (
    'foo'    => 'aa',
    'bar'    => 'bb',
    'baz'    => 'aa',
    'quux'   => 'bb',
    'wibble' => 'cc',
    'wobble' => 'aa'

my %keys;
while ( my ( $key, $value ) = each(%hash) ) {
    push( @{ $keys{$value} }, $key );

foreach my $value ( keys(%keys) ) {
    printf( "Value \"%s\" is present with the following keys: %s\n",
        $value, join( " ", @{ $keys{$value} } ) );

But bash associative arrays can't hold arrays...

I'd also be interested in any old school solution possibly using some form of indirect indexing (building a set of index array(s) when reading the values that I said I had in hash above?). It feels like there ought to be a way to do this in linear time.



to reverse keys <=> values

In zsh, where the primary syntax for defining a hash is hash=(k1 v1 k2 v2...) like in perl (newer versions also support the awkward ksh93/bash syntax for compatibility though with variations when it comes to quoting the keys)


typeset -A reversed
reversed=("${(@)values:^keys}") # array zipping operator

or using a loop:

for k v ("${(@kv}hash}") reversed[$v]=$k

The @ and double quotes is to preserve empty keys and values (note that bash associative arrays don't support empty keys). As the expansion of elements in associative arrays is in no particular order, if several elements of $hash have the same value (which will end up being a key in $reversed), you can't tell which key will be used as the value in $reversed.

for your loop

You'd use the R hash subscript flag to get elements based on value instead of key, combined with e for exact (as opposed to wildcard) match, and then get the keys for those elements with the k parameter expansion flag:

for value ("${(@u)hash}")
  print -r "elements with '$value' as value: ${(@k)hash[(Re)$value]}"

your perl approach

zsh (contrary to ksh93) doesn't support arrays of arrays, but its variables can contain the NUL byte, so you could use that to separate elements if the elements don't otherwise contain NUL bytes, or use the ${(q)var} / ${(Q)${(z)var}} to encode/decode a list using quoting.

typeset -A seen
for k v ("${(@kv)hash}")
  seen[$v]+=" ${(q)k}"

for k v ("${(@kv)seen}")
  print -r "elements with '$k' as value: ${(Q@)${(z)v}}"


ksh93 was the first shell to introduce associative arrays in 1993. The syntax for assigning values as a whole means it's very difficult to do it programmatically contrary to zsh, but at least it's somewhat justified in ksh93 in that ksh93 supports complex nested data structures.

In particular, here ksh93 supports arrays as values for hash elements, so you can do:

typeset -A seen
for k in "${!hash[@]}"; do

for k in "${!seen[@]}"; do
  print -r "elements with '$k' as value ${x[$k][@]}"


bash added support for associative arrays decades later, copied the ksh93 syntax, but not the other advanced data structures, and doesn't have any of the advanced parameter expansion operators of zsh.

In bash, you could use the quoted list approach mentioned in the zsh using printf %q or with newer versions ${var@Q}.

typeset -A seen
for k in "${!hash[@]}"; do
  printf -v quoted_k %q "$k"
  seen[${hash[$k]}]+=" $quoted_k"

for k in "${!seen[@]}"; do
  eval "elements=(${seen[$k]})"
  echo -E "elements with '$k' as value: ${elements[@]}"

As noted earlier however, bash associative arrays don't support the empty value as a key, so it won't work if some of $hash's values are empty. You could choose to replace the empty string with some place holder like <EMPTY> or prefix the key with some character that you'd later strip for display.

| improve this answer | |
  • Nice. You've got a mistake in the bash example: last seen (in the echo) should be elements. I suppose though, each ${seen[$k]} is a string with space separated elements, which thus needs that extra parsing the OP didn't want(?) – Ralph Rönnquist Mar 18 '19 at 11:41

The stumbling block, as I'm sure you know, is to get the whole value of an indexed array when having its name as value of a (another) variable. I couldn't do it with less than having an intermediate whose value becomes of format ${v[@]} and then use eval on that. So, here's that approach:

declare -A keys
N=0 # counter for the index variables IX1, IX2, IX3, ...
for key in "${!hash[@]}"; do
    if [ -z "${keys[$value]}" ] ; then N=$((N+1)) ; keys[$value]=IX$N ; fi
    index="${keys[$value]}" # 'index' is now name of index variable
    eval "$index=( $X $key )" # adding next key to it

for value in "${!keys[@]}" ; do
    printf "Value %s is present with the following keys: %s\n" \
       "$value" "$(eval echo "$X")"

This is for Linux bash. It creates indexed arrays IX1, IX2, etc., for the various values it encounters, and holds those names in the keys associative array for the values. Thus, ${keys[$value]} is the name of the indexed array that holds the keys for that value. Then X is set up to be the variable "access phrase" for the collection of values, allowing eval echo "$X" to translate into those values with space separation. For example, if a value has indexed array IX2, then X will be the string ${IX2[@]}.

I believe zsh is similar in not supporting arrays of arrays, so it'd probably require a similar solution. IMHO though, the access phrases in zsh are slightly clearer.

| improve this answer | |

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.