Util::H2O - Hash to Object: turns hashrefs into objects with accessors for keys


 use Util::H2O;
 my $hash = h2o { foo => "bar", x => "y" }, qw/ more keys /;
 print $hash->foo, "\n";           # accessor
 $hash->x("z");                    # change value
 $hash->more("cowbell");           # additional keys
 my $struct = { hello => { perl => "world!" } };
 h2o -recurse, $struct;            # objectify nested hashrefs as well
 print $struct->hello->perl, "\n";
 my $obj = h2o -meth, {            # code references become methods
     what => "beans",
     cool => sub {
         my $self = shift;
         print $self->what, "\n";
     } };
 $obj->cool;                       # prints "beans"
 h2o -classify=>'Point', {         # whip up a class
         angle => sub { my $self = shift; atan2($self->y, $self->x) }
     }, qw/ x y /;
 my $one = Point->new(x=>1, y=>2);
 my $two = Point->new(x=>3, y=>4);
 printf "%.3f\n", $two->angle;     # prints 0.927


This module allows you to turn hashrefs into objects, so that instead of $hash->{key} you can write $hash->key, plus you get protection from typos. In addition, options are provided that allow you to whip up really simple classes.

You can still use the hash like a normal hashref as well, as in $hash->{key}, keys %$hash, and so on, but note that by default this function also locks the hash's keyset to prevent typos there too.

This module exports a single function by default.

h2o @opts, $hashref, @additional_keys


If you specify an option with a value multiple times, only the last one will take effect.


Nested hashes are objectified as well. The only options that are passed down to nested hashes are -lock and -ro. None of the other options will be applied to the nested hashes, including @additional_keys. Nested arrayrefs are not recursed into, but see the -arrays option for that.

Versions of this module before v0.12 did not pass down the -lock option, meaning that if you used -nolock, -recurse on those versions, the nested hashes would still be locked.


Like -recurse, but additionally, h2o is applied to elements of nested arrays as well. The same options as with -recurse are passed down to nested hashes and arrayrefs. Takes precedence over the -pass option, i.e. if you use these two options together, arrayrefs are still descended into. Like hashrefs, the original arrays are modified!

This option implies -recurse. This option was added in v0.20.


Any code references present in the hash at the time of this function call will be turned into methods. Because these methods are installed into the object's package, they can't be changed later by modifying the hash.

To avoid confusion when iterating over the hash, the hash entries that were turned into methods are removed from the hash. The key is also removed from the "allowed keys" (see the -lock option), unless you specify it in @additional_keys. In that case, you can change the value of that key completely independently of the method with the same name.

-class => classname

Specify the class name into which to bless the object (as opposed to the default: a generated, unique package name in Util::H2O::).

Note: If you use this option, -clean defaults to false, meaning that the package will stay in Perl's symbol table and use memory accordingly, and since this function installs the accessors in the package every time it is called, if you re-use the same package name, you will get "redefined" warnings. Therefore, if you want to create multiple objects in the same package, you should probably use -new or -classify.

If you wanted to generate a unique package name in a different package, you could use: h2o -class => sprintf('My::Class::Name::_%x', $hash+0), $hash, perhaps even in combination with -isa => 'My::Class::Name'. However, keep in mind that you shouldn't step into another class' namespace without knowing that this won't cause conflicts, and also that not using the default class names means that functions like o2h will no longer identify the objects as coming from h2o.

-classify => classname_string or $hashref

In the form -classify => classname_string, this is simply the short form of the options -new, -meth, -class => classname_string.

As of v0.16, in the special form -classify => $hashref, where the -classify must be the last option in @opts before the $hashref, it is the same as -new, -meth, -class => __PACKAGE__, $hashref - that is, the current package's name is used as the custom class name. It does not make sense to use this outside of an explicit package, since your class will be named main. With this option, the Point example in the "Synopsis" can be written like the following, which can be useful if you want to add more things to the package, or perhaps if you want to write your methods as regular subs:

     package Point;
     use Util::H2O;
     h2o -classify, {
          angle => sub { my $self = shift; atan2($self->y, $self->x) }
     }, qw/ x y /;

Note h2o will remain in the package's namespace, one possibility is that you could load namespace::clean after you load this module.

You might also note that in the above example, one could write angle as a regular sub in the package. And at that point, one might recongize the similarity between the code and what one can do with e.g. Class::Tiny or even Moo.

-isa => arrayref or scalar

Convenience option to set the @ISA variable in the package of the object, so that the object inherits from that/those package(s). This option was added in v0.14.

Warning: The methods created by h2o will not call superclass methods. This means the parent class' DESTROY method(s) are not called, and any accessors generated from hash keys are blindly overriden.


Generates a constructor named new in the package. The constructor works as a class and instance method, and dies if it is given any arguments that it doesn't know about. If you want more advanced features, like required arguments, validation, or other initialization, you should probably switch to something like Moo instead.

-destroy => coderef

Allows you to specify a custom destructor. This coderef will be called from the object's actual DESTROY in void context with the first argument being the same as the first argument to the DESTROY method. Errors will be converted to warnings. This option was added in v0.14.

-clean => bool

Whether or not to clean up the generated package when the object is destroyed. Defaults to false when -class is specified, true otherwise. If this is false, be aware that the packages will stay in Perl's symbol table and use memory accordingly, and any subs/methods in those packages may cause "redefined" warnings if the package name is re-used.

As of v0.16, this module will refuse to delete the package if it is named main.

-lock => bool

Whether or not to use Hash::Util's lock_ref_keys to prevent modifications to the hash's keyset. Defaults to true. The -nolock option is provided as a short form of -lock=>0.

Keysets of objects created by the constructor generated by the -new option are also locked. Versions of this module before v0.12 did not lock the keysets of new objects.

Note that on really old Perls, that is, before Perl v5.8.9, Hash::Util and its lock_ref_keys are not available, so the hash is never locked on those versions of Perl. Versions of this module before v0.06 did not lock the keyset. Versions of this module as of v0.12 issue a warning on old Perls.


Short form of the option -lock=>0.


Makes the entire hash read-only using Hash::Util's lock_hashref and the generated accessors will also throw an error if you try to change values. In other words, this makes the object and the underlying hash immutable.

You cannot specify any @additional_keys with this option enabled unless you also use the -new option - the additional keys will then only be useful as arguments to the constructor. This option can't be used with -nolock or -lock=>0.

This option was added in v0.12. Using this option will not work and cause a warning when used on really old Perls (before v5.8.9), because this functionality was not yet available there.

-pass => "ref" or "undef"

When this option is set to "undef" (that's the string "undef", not undef itself!), then passing a value of undef for the $hashref will not result in a fatal error, the value will simply be passed through.

When this option is set to the string "ref", then any value other than a plain hashref that is a reference, including objects, plus undef as above, will be passed through without modification. Any hashes nested inside of these references will not be descended into, even when -recurse is specified. However, -arrays takes precedence over this option, see its documentation.

This option was added in v0.18.


You must supply a plain (unblessed) hash reference here, unless you've specified the -pass and/or -arrays options. Be aware that this function does modify the original hashref(s) by blessing it and locking its keyset (the latter can be disabled with the -lock option), and if you use -meth or -classify, keys whose values are code references will be removed. If you use -arrays, the elements of those arrays may also be modified.

An accessor will be set up for each key in the hash(es); note that the keys must of course be valid Perl identifiers for you to be able to call the method normally (see also the "Cookbook").

The following keys will be treated specially by this module. Please note that there are further keys that are treated specially by Perl and/or that other code may expect to be special, such as UNIVERSAL's isa. See also perlsub and the references therein.


This key is not allowed in the hash if the -new option is on.


This key is not allowed except if all of the following apply:

  • -destroy is not used,

  • -clean is off (which happens by default when you use -class),

  • -meth is on, and

  • the value of the key DESTROY is a coderef.

Versions of this module before v0.14 allowed a DESTROY key in more circumstances (whenever -clean was off).


If your hash contains a key named AUTOLOAD, or this key is present in @additional_keys, this module will set up a method called AUTOLOAD, which is subject to Perl's normal autoloading behavior - see "Autoloading" in perlsub and "AUTOLOAD" in perlobj. Without the -meth option, you will get a "catch-all" accessor to which all method calls to unknown method names will go, and with -meth enabled (which is implied by -classify), you can install your own custom AUTOLOAD handler by passing a coderef as the value for this key - see "An Autoloading Example". However, it is important to note that enabling autoloading removes any typo protection on method names!


Methods will be set up for these keys even if they do not exist in the hash.

Please see the list of keys that are treated specially above.


The (now blessed and optionally locked) $hashref.

o2h @opts, $h2object

This function takes an object as created by h2o and turns it back into a hashref by making shallow copies of the object hash and any nested objects that may have been created via -recurse, -arrays, or created manually. This function is recursive by default because for a non-recursive operation you can simply write: {%$h2object} (making a shallow copy).

Unlike h2o, this function returns a new hashref instead of modifying the given variable in place (unless what you give this function is not an h2o object, in which case it will just be returned unchanged). Similarly, if you specify the -arrays option, shallow copies of arrays will be returned in place of the original ones, with o2h applied to the elements.

Note that this function operates only on objects in the default package - it does not step into plain hashrefs, it does not step into arrayrefs unless you specify -arrays, nor does it operate on objects created with the -class or -classify options. Also be aware that because methods created via -meth are removed from the object hash, these will disappear in the resulting hashref.

This function was added in v0.18.


If you specify an option with a value multiple times, only the last one will take effect.


If you specify this option, nested arrayrefs are descended into as well.

This option was added in v0.20.


Keys with Spaces, Dashes, or Other Non-Identifier Characters

If the hash you want to pass to h2o contains keys that are not usable as method names, such as keys containing spaces or dashes, you can transform the hash before passing it to h2o. There are several ways to achieve this, including in plain Perl, but one of the easier ways is with pairmap from the core module List::Util.

 use List::Util 'pairmap';
 my $hash = { "foo bar" => 123, "quz-ba%z" => 456 };
 my $obj = h2o { pairmap { $a=~tr/a-zA-Z0-9/_/c; ($a,$b) } %$hash };
 print $obj->foo_bar, $obj->quz_ba_z, "\n";  # prints "123456"

Using with Config::Tiny

One common use case for this module is to make accessing hashes nicer, like for example those you get from Config::Tiny. Here's how you can create a new h2o object from a configuration file:

 use Util::H2O 0.18 qw/ h2o o2h /;  # v0.18 for o2h
 use Config::Tiny 2.27;             # v2.27 for writing file back out
 my $config = h2o -recurse, {%{ Config::Tiny->read($config_filename) }};
 say $config->foo->bar;  # prints the value of "bar" in section "[foo]"
 $config->foo->bar("Hello, World!");  # change value
 # write file back out
 Config::Tiny->new(o2h $config)->write($config_filename);


Because the packages generated by h2o are dynamic, note that any debugging dumps of these objects will be somewhat incomplete because they won't show the methods. However, if you'd like somewhat nicer looking dumps of the data contained in the objects, one way you can do that is with Data::Dump::Filtered:

 use Util::H2O;
 use Data::Dump qw/dd/;
 use Data::Dump::Filtered qw/add_dump_filter/;
 add_dump_filter( sub {
     my ($ctx, $obj) = @_;
     return { bless=>'', comment=>'Util::H2O::h2o()' }
         if $ctx->class=~/^Util::H2O::/;
     return undef; # normal Data::Dump processing for all other objects
 my $x = h2o -recurse, { foo => "bar", quz => { abc => 123 } };
 dd $x;


 # Util::H2O::h2o()
   foo => "bar",
   quz => # Util::H2O::h2o()
          { abc => 123 },

An Autoloading Example

If you wanted to create a class where (almost!) every method call is automatically translated to a hash access of the corresponding key, here's how you could do that:

 h2o -classify=>'HashLikeObj', -nolock, {
     AUTOLOAD => sub {
         my $self = shift;
         our $AUTOLOAD;
         ( my $key = $AUTOLOAD ) =~ s/.*:://;
         $self->{$key} = shift if @_;
         return $self->{$key};
     } };

Upgrading to Moo

Let's say you've used this module to whip up two simple classes:

 h2o -classify => 'My::Class', {}, qw/ foo bar details /;
 h2o -classify => 'My::Class::Details', {}, qw/ a b /;

But now you need more features and would like to upgrade to an actual OO system like Moo. Here's how you'd write the above code using that, with some Type::Tiny thrown in:

 package My::Class2 {
     use Moo;
     use Types::Standard qw/ InstanceOf /;
     use namespace::clean; # optional but recommended
     has foo     => (is=>'rw');
     has bar     => (is=>'rw');
     has details => (is=>'rw', isa=>InstanceOf['My::Class2::Details']);
 package My::Class2::Details {
     use Moo;
     use namespace::clean;
     has a => (is=>'rw');
     has b => (is=>'rw');

See Also

Inspired in part by lock_keys from Hash::Util.

Many, many other modules exist to simplify object creation in Perl. This one is mine ;-P

Similar modules include Object::Adhoc, Object::Anon, Hash::AsObject, Object::Result, and Hash::Wrap, the latter of which also contains a comprehensive list of similar modules. Also, see Class::Tiny for another minimalistic class generation module.

For real OO work, I like Moo and Type::Tiny (see "Upgrading to Moo").

Further modules that might be useful in combination with this one: Hash::Merge for merging hashes before using this module (for example, to supply default values for keys); Role::Tiny for applying roles.

See also Util::H2O::More by OODLER, a module with additional functionality on top of this module.

Special Thanks

Thanks to oodler577 on GitHub (OODLER on CPAN), whose many suggestions have inspired a lot of the features in this module!

Author, Copyright, and License

Copyright (c) 2020-2023 Hauke Daempfling (

This library is free software; you can redistribute it and/or modify it under the same terms as Perl 5 itself.

For more information see the Perl Artistic License, which should have been distributed with your copy of Perl. Try the command perldoc perlartistic or see