The London Perl and Raku Workshop takes place on 26th Oct 2024. If your company depends on Perl, please consider sponsoring and/or attending.


Data::Rmap - recursive map, apply a block to a data structure


 $ perl -MData::Rmap -e 'print rmap { $_ } 1, [2,3], \\4, "\n"'

 $ perl -MData::Rmap=:all
 rmap_all { print (ref($_) || "?") ,"\n" } \@array, \%hash, \*glob;

 # OUTPUT (Note: a GLOB always has a SCALAR, hence the last two items)
 # ?

 # Upper-case your leaves in-place
 $array = [ "a", "b", "c" ];
 $hash  = { key => "a value" };
 rmap { $_ = uc $_; } $array, $hash;

 use Data::Dumper; $Data::Dumper::Terse=1; $Data::Dumper::Indent=0;
 print Dumper($array), " ", Dumper($hash), "\n";

 # ['A','B','C'] {'key' => 'A VALUE'}

 # Simple array dumper.
 # Uses $self->recurse method to alter traversal order
 ($dump) = rmap_to {

    return "'$_'" unless ref($_); # scalars are quoted and returned

    my $self = shift;
    # use $self->recurse to grab results and wrap them
    return '[ ' . join(', ', $self->recurse() ) . ' ]';

  } ARRAY|VALUE,  [ 1, [ 2, [ [ 3 ], 4 ] ], 5 ];

 print "$dump\n";
 # [ '1', [ '2', [ [ '3' ], '4' ] ], '5' ]



Recursively evaluate a BLOCK over a list of data structures (locally setting $_ to each element) and return the list composed of the results of such evaluations. $_ can be used to modify the elements.

Data::Rmap currently traverses HASH, ARRAY, SCALAR and GLOB reference types and ignores others. Depending on which rmap_* wrapper is used, the BLOCK is called for only scalar values, arrays, hashes, references, all elements or a customizable combination.

The list of data structures is traversed pre-order in a depth-first fashion. That is, the BLOCK is called for the container reference before is it called for it's elements (although see "recurse" below for post-order). The values of a hash are traversed in the usual "values" order which may affect some applications.

If the "cut" subroutine is called in the BLOCK then the traversal stops for that branch, say if you "cut" an array then the code is never called for it's elements (or their sub-elements). To simultaneously return values and cut, simply pass the return list to cut: cut('add','to','returned');

The first parameter to the BLOCK is an object which maintains the state of the traversal. Methods available on this object are described in "State Object" below.


By default:

 rmap, rmap_all, cut


 rmap_scalar rmap_hash rmap_array rmap_code rmap_ref rmap_to
 :all => ... # everything


The various names are just wrappers which select when to call the code BLOCK. rmap_all always calls it, the others are more selective while rmap_to takes an extra parameter permitting you to provide selection criteria. Furthermore, you can always just rmap_all and skip nodes which are not of interest.

rmap_to { ... } $want, @data_structures;

Most general first.

Recurse the @data_structures and apply the BLOCK to elements selected by $want. The $want parameter is the bitwise "or" of whatever types you choose (imported with :types):

 VALUE  - non-reference scalar, eg. 1
 HASH   - hash reference
 ARRAY  - array reference
 SCALAR - scalar refernce, eg. \1
 REF    - higher-level reference, eg. \\1, \\{}
          B<NOT> any reference type, see <Scalar::Util>'s reftype:
          perl -MScalar::Util=reftype -le 'print map reftype($_), \1, \\1'
 GLOB   - glob reference, eg. \*x
          (scalar, hash and array recursed, code too as of 0.63)
 ALL    - all of the above (not CODE)
 CODE   - code references (as of 0.63)
 NONE   - none of the above

So to call the block for arrays and scalar values do:

 use Data::Rmap ':all';         # or qw(:types rmap_to)
 rmap { ... } ARRAY|VALUE, @data_structures;

(ALL | CODE) and (ALL & !GLOB) might also be handy.

The remainder of the wrappers are given in terms of the $want for rmap_to.

rmap { ... } @list;

Recurse and call the BLOCK on non-reference scalar values. $want = VALUE

rmap_all BLOCK LIST

Recurse and call the BLOCK on everything. $want = ALL

rmap_scalar { ... } @list

Recurse and call the BLOCK on non-collection scalars. $want = VALUE|SCALAR|REF


Recurse and call the BLOCK on hash refs. $want = HASH


Recurse and call the BLOCK on array refs. $want = ARRAY


Recurse and call the BLOCK on code refs. $want = CODE


Recurse and call the BLOCK on all "normal" references: $want = HASH|ARRAY|SCALAR|REF

Note: rmap_ref isn't the same as rmap_to {} REF


Don't traverse sub-elements and return the @list immediately. For example, if $_ is an ARRAY reference, then the array's elements are not traversed.

If there's two paths to an element, both will need to be cut.

State Object

The first parameter to the BLOCK is an object which maintains most of the traversal state (except current node, which is $_). You will ignore it most of the time. The "recurse" method may be useful. Other methods should only be used in throw away tools, see TODO



Process child nodes of $_ now and return the result.

This makes it easier to perform post-order and in-order processing of a structure. Note that since the same "seen list" is used, the child nodes aren't reprocessed.


The code reference of the BLOCK itself. Possible useful in some situations.


Reference to the HASH used to track where we have visited. You may want to modify it in some situations (though I haven't yet). Beware circular references. The (current) convention used for the key is in the source.


The $want state described in rmap_to.


 # command-line play
 $ perl -MData::Rmap -le 'print join ":", rmap { $_ } 1,2,[3..5],\\6'

 # Linearly number questions on a set of pages
 my $qnum = 1;
 rmap_hash {
     $_->{qnum} = $qnum++ if($_->{qn});
 } @pages;

 # Grep recursively, finding ALL objects
 use Scalar::Util qw(blessed);
 my @objects = rmap_ref {
     blessed($_) ? $_ : ();
 } $data_structure;

 # Grep recursively, finding public objects (note the cut)
 use Scalar::Util qw(blessed);
 my @objects = rmap_ref {
     blessed($_) ?  cut($_) : ();
 } $data_structure;

 # Return a modified structure
 # (result flattening means we must cheat by cloning then modifying)
 use Storable qw(dclone);
 use Lingua::EN::Numbers::Easy;

 $words = [ 1, \2, { key => 3 } ];
 $nums = dclone $words;
 rmap { $_ = $N{$_} || $_ } $nums;

 # Make an assertion about a structure
 use Data::Dump;
 rmap_ref {
    blessed($_) && $_->isa('Question') && defined($_->name)
        or die "Question doesn't have a name:", dump($_);
 } @pages;

 # Traverse a tree using localize state
 $tree = [
     one =>
     two =>
         three_one =>
         three_two =>
             three_three_one =>
         three_four =>
     four =>
             five_one_one =>

 @path = ('q');
 rmap_to {
     if(ref $_) {
         local(@path) = (@path, 1); # ARRAY adds a new level to the path
         $_[0]->recurse(); # does stuff within local(@path)'s scope
     } else {
         print join('.', @path), " = $_ \n"; # show the scalar's path
     $path[-1]++; # bump last element (even when it was an aref)
 } ARRAY|VALUE, $tree;

 # q.1 = one
 # q.2 = two
 # q.3.1 = three_one
 # q.3.2 = three_two
 # q.3.3.1 = three_three_one
 # q.3.4 = three_four
 # q.4 = four
 # q.5.1.1 = five_one_one

 # replace CODE with "<CODE>"
 $ perl -MData::Rmap=:all -E 'say join ":", rmap_code { "<CODE>" } sub{},sub{}'

 # look inside code refs with PadWalker
 $ perl -MData::Rmap=:all -MSub::Identify=:all -MPadWalker=:all -MSub::Name
   use 5.10.0;
   my $s = sub {}; sub A::a { $s };
   say join ", ",
    rmap_code {
        sub_fullname($_),                       # name string
        map { $_[0]->recurse } closed_over($_)  # then recurse the sub innards
    } \*A::a, subname b => sub { $s };
   # A::a, main::__ANON__, main::b


Beware comma after block:

 rmap { print }, 1..3;
               ^-------- bad news, you get an empty list:
 rmap(sub { print $_; }), 1..3;

If you don't import a function, perl's confusion may produce:

 $ perl -MData::Rmap -le 'rmap_scalar { print } 1'
 Can't call method "rmap_scalar" without a package or object reference...

 $ perl -MData::Rmap -le 'rmap_scalar { $_++ } 1'
 Can't call method "rmap_scalar" without a package or object reference...

If there's two paths to an element, both will need to be cut.

If there's two paths to an element, one will be taken randomly when there is an intervening hash.

Autovivification can lead to "Deep recursion" warnings if you test exists $_->{this}{that} instead of exists $_->{this} && exists $_->{this}{that} as you may follow a long chain of "this"s Alternatively use the "no autovivification" pragma to avoid this problem.


put for @_ in wrapper to allow parameters in a different wrapper, solve localizing problem.

Store custom localized data about the traversal. Seems too difficult and ugly when compare to doing it at the call site. Should support multiple reentrancy so avoid the symbol table.

rmap_args { } $data_structure, @args form to pass parameters. Could potentially help localizing needs. (Maybe only recurse last item)

Benchmark. Use array based object and/or direct access internally.

Think about permitting different callback for different types. The prototype syntax is a bit too flaky....

Ensure that no memory leaks are possible, leaking the closure.


map, grep, Storable's dclone, Scalar::Util's reftype and blessed

Faint traces of treemap:

Update: various alternatives have appear over the years, Data::Visitor has a list.


Brad Bowman <>


Copyright (c) 2004- Brad Bowman (<>). All rights reserved.

This module is free software; you can redistribute it and/or modify it under the same terms as Perl itself. See perlartistic and perlgpl.

This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.