Help language development. Donate to The Perl Foundation

AttrX::Mooish zef:vrurg last updated on 2022-10-31



AttrX::Mooish - extend attributes with ideas from Moo/Moose (laziness!)


use AttrX::Mooish;
class Foo {
    has $.bar1 is mooish(:lazy, :clearer, :predicate) is rw;
    has $!bar2 is mooish(:lazy, :clearer, :predicate, :trigger);
    has Num $.bar3 is rw is mooish(:lazy, :filter);

    method build-bar1 {
        "lazy init value"

    method !build-bar2 {
        "this is private mana!"

    method !trigger-bar2 ( $value ) {
        # do something after attribute changed.

    method build-bar3 {

    method filter-bar3 ( $value, *%params ) {
        if %params<old-value>:exists {
            # Only allow the value to grow
            return ( !%params<old-value>.defined || $value > %params<old-value> ) ?? $value !! %params<old-value>;
        # Only allow inital values from 0.5 and higher
        return $value < 0.5 ?? Nil !! $value;

    method baz {
        # Yes, works with private too! Isn't it magical? ;)
        "Take a look at the magic: «{ $!bar2 }»";

my $foo =;

say $foo.bar1;
say $foo.bar3.defined ?? "DEF" !! "UNDEF";
for 1..10 { $foo.bar3 = rand; say $foo.bar3 }

The above would generate a output similar to the following:

lazy init value


This module is aiming at providing some functionality we're all missing from Moo/Moose. It implements laziness, accompanying methods and adds attribute value filter on top of what standard Moo/Moose provide.

What makes this module different from previous versions one could find in the Raku modules repository is that it implements true laziness allowing Nil to be a first-class value of a lazy attribute. In other words, if you look at the SYNOPSIS section, $.bar3 value could randomly be either undefined or 3.1415926.

Laziness for beginners

This section is inteded for beginners and could be skipped by experienced lazybones.

What is "lazy attribute"

As always, more information could be found by Google. In few simple words: a lazy attribute is the one which gets its first value on demand, i.e. – on first read operation. Consider the following code:

class Foo {
    has $.bar is mooish(:lazy, :predicate);

    method build-bar { π }

my $foo =
say $foo.has-bar; # False
say $;     # 3.1415926...
say $foo.has-bar; # True

When is it useful?

Laziness becomes very handy in cases where intializing an attribute is very expensive operation yet it is not certain if attribute is gonna be used later or not. For example, imagine a monitoring code which raises an alert when a failure is detected:

class Monitor {
    has $.notifier;
    has $!failed-object;

    submethod BUILD {
        $!notifier =;

    method report-failure {
        $.notifier.alert( :$!failed-object );


Now, imagine that notifier is a memory-consuming object, which is capable of sending notification over different kinds of media (SMTP, SMS, messengers, etc...). Besides, preparing handlers for all those media takes time. Yet, failures are rare and we may need the object, say, once in 10000 times. So, here is the solution:

class Monitor {
    has $.notifier is mooish(:lazy);
    has $!failed-object;

    method build-notifier { :$!failed-object ) }

    method report-failure {


Now, it would only be created when we really need it.

Such approach also works well in interactive code where many wuch objects are created only the moment a user action requires them. This way overall responsiveness of a program could be significally incresed so that instead of waiting long once a user would experience many short delays which sometimes are even hard to impossible to be aware of.

Laziness has another interesting application in the area of taking care of attribute dependency. Say, $.bar1 value depend on $.bar2, which, in turn, depends either on $.bar3 or $.bar4. In this case instead of manually defining the order of initialization in a BUILD submethod, we just have the following code in our attribute builders:

method build-bar2 {
    if $some-condition {
        return self.prepare( $.bar3 );
    self.prepare( $.bar4 );

This module would take care of the rest.


The SYNOPSIS is a very good example of how to use the trait mooish.

Trait parameters


For all the trait parameters, if it is applied to a private attribute then all auto-generated methods will be private too.

The call-back style options such as builder, trigger, filter are expected to share the privace mode of their respective attribute:

    class Foo {
        has $!bar is rw is mooish(:lazy, :clearer<reset-bar>, :predicate, :filter<wrap-filter>);

        method !build-bar { "a private value" }
        method baz {
            if self!has-bar {
        method !wrap-filter ( $value, :$attribute ) {
            "filtered $attribute: ($value)"

Though if a callback option is defined with method name instead of Bool True then if method wit the same privacy mode is not found then opposite mode would be tried before failing:

    class Foo {
        has $.bar is mooish( :trigger<on_change> );
        has $!baz is mooish( :trigger<on_change> );
        has $!fubar is mooish( :lazy<set-fubar> );

        method !on_change ( $val ) { say "changed! ({$val})"; }
        method set-baz { $!baz = "new pvt" }
        method use-fubar { $!fubar }

    $inst =;
    $ = "new";  # changed! (new)
    $inst.set-baz;      # changed! (new pvt)
    $inst.use-fubar;    # Dies with "No such private method '!set-fubar' for invocant of type 'Foo'" message

User method's (callbacks) options

User defined (callback-type) methods receive additional named parameters (options) to help them understand their context. For example, a class might have a couple of attributes for which it's ok to have same trigger method if only it knows what attribute it is applied to:

    class Foo {
        has $.foo is rw is mooish(:trigger('on_fubar'));
        has $.bar is rw is mooish(:trigger('on_fubar'));

        method on_fubar ( $value, *%opt ) {
            say "Triggered for {%opt<attribute>} with {$value}";

    my $inst =;
    $ = "ABC";
    $ = "123";

The expected output would be:

    Triggered for $!foo with with ABC
    Triggered for $!bar with with 123

NOTE: If a method doesn't care about named parameters it may only have positional arguments in its signature. This doesn't work for pointy blocks where anonymous slurpy hash would be required:

    class Foo {
        has $.bar is rw is mooish(:trigger(-> $, $val, *% {...}));


Definite Types

The module allows for lazy attributes to have a definite type. By default the compiler will throw an error unless such attribute has is required trait or a default value. But for lazy attributes the default value must be provided by their builders:

class Foo {
    has Int:D $.the-answer is mooish(:lazy);
    method build-the-answer { 42 }

Apparently, the value returned by a builder must pass the typecheck. Therefore the following code will fail a run time:

class Foo {
    has Str:D $.bad is mooish(:lazy);
    method build-bad { Str }

Some magic

Note that use of this trait doesn't change attribute accessors. More than that, accessors are not required for private attributes. Consider the $!bar2 attribute from SYNOPSIS.


Module versions prior to v0.5.0 were pretty much costly perfomance-wise. This was happening due to use of Proxy to handle all attribute read/writes. Since v0.5.0 only the first read/write operation would be handled by this module unless filter or trigger parameters are used. When AttrX::Mooish is assured that the attribute is properly initialized it steps aside and lets the Raku core to do its job without intervention.

The only exception takes place if clearer parameter is used and clear-<attribute> method is called. In this case the attribute state is reverted back to uninitialized state and Proxy is getting installed again – until the next read/write operation.

filter and trigger are special with this respect because they require permanent monitoring of attribute operations making it effectively impossible to strip off Proxy from attribute's value. For this reason use of these parameters must be very carefully considered. One is highly discouraged from using them for any code where performance is important.


This module provides partial thread-safety and must be used with care with this respect. This means that the following conditions are guaranteed:

Consider it the way we normally consider working with an attribute in a concurrent environment, where reads and writes must be mutually protected to ensure data safety.

To sum up the above stated, what would be guaranteed is that a read-only attribute would provide robust results in a multi-threaded environment, as it is expected from a read-only pre-initialized attribute.

Predicates are considered read operations and as such are not protected either. Think of testing a non-mooified attribute for definedness, for example.



Cloning of an object with mooified attributes is a non-trivial case resulting from the use of Proxy. Again. The root of the problem lies in the fact that Proxy is using closures for its FETCH/STORE. This is how it knows what object is to be updated when necessary: the object is captured by the closures. But what help us under normal circumstances becomes our problem after cloning because the new object attributes would still be bound to proxies referring the original instance!

AttrX::Mooish handles this situation starting with v1.0.0 release by implementing post-clone fix up procedure where all proxies are getting replaced using new closures. This is implemented by installing special clone method into each class containing mooified attributes.

So far, so good until it comes down to lazy attributes which were already initialized using other attributes of the class. Simple cloning doesn't affect them, but if their dependency has been changed using twiddles then we're in trouble:

class Foo {
    has $.a1 = 1;
    has $.a2 is mooish(:lazy);
    method build-a2 { $!a1 * 2 }
my $obj =;
say $obj.a2; # 2
my $copy = $obj.clone(:a1(3));
say $copy.a2; # Oops, it's still 2!

In such situations AttrX::Mooish resets the lazy attributes to unitialized state so they would get re-initialized again using new values.

The only exception is for writable (is rw) lazy attributes. When they hold a value there is no way for us to know where the value came from because it could have been assigned by code external to our class in which case resetting it might be not so smart.

IMPORTTANT! Also remember that a non-writable attribute can still be assigned via private accessor ($!attr = $value;) by class code. These cases cannot be detected either but disreprected by the module. Taking care of them is developer's responsibility!

Generally speaking, there is no good strategy to handle all possible cases what it comes to cloning lazy entities. Sometimes it would be better to resolve edge cases manually. If this is your case then metamodel method post-clone can be used to do the standard fixup job:

class Foo {
    method clone(*%twiddles) {
        my \cloned = ::?CLASS.^post-clone: self, callsame(), %twiddles;
        ... # Do specific fixups here

Basically, the first line is what AttrX::Mooish installs for you by default. Consider the use of ::?CLASS instead of self. This is mandatory because self can be an instance of a child class.





Vadim Belman [email protected]


Artistic License 2.0

See the LICENSE file in this distribution.