Help language development. Donate to The Perl Foundation

AttrX::Mooish zef:vrurg last updated on 2022-08-09

README.md
NAME
====

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

SYNOPSIS
========

    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 {
            rand;
        }

        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 = Foo.new;

    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
    UNDEF
    0.08662089602505263
    0.49049512098324255
    0.49049512098324255
    0.5983833081770437
    0.9367804461546302
    0.9367804461546302
    0.9367804461546302
    0.9367804461546302
    0.9367804461546302
    0.9367804461546302

DESCRIPTION
===========

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](#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 = Foo.new
    say $foo.has-bar; # False
    say $foo.bar;     # 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 = Notifier.new;
        }

        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 { Notifier.new( :$!failed-object ) }

        method report-failure {
            $.notifier.alert;
        }

        ...
    }

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.

USAGE
=====

The [SYNOPSIS](#SYNOPSIS) is a very good example of how to use the trait `mooish`.

Trait parameters
----------------

  * *`lazy`*

    `Bool`, defines wether attribute is lazy. Can have `Bool`, `Str`, or `Callable` value. The later two have the same meaning, as for *`builder`* parameter.

  * *`builder`*

    Defines builder method for a lazy attribute. The value returned by the method will be used to initialize the attribute.

    This parameter can have `Str` or `Callable` values or be not defined at all. In the latter case we expect a method with a name composed of "*build-*" prefix followed by attribute name to be defined in our class. For example, for a attribute named `$!bar` the method name is expected to be *build-bar*.

    A string value defines builder's method name.

    A callable value is used as-is and invoked as an object method. For example:

        class Foo {
            has $.bar is mooish(:lazy, :builder( -> $,*% {"in-place"} );
        }

        $inst = Foo.new;
        say $inst.bar;

    This would output '*in-place*'.

    *Note* the use of slurpy `*%` in the pointy block. Read about callback parameters below.

  * *`predicate`*

    Could be `Bool` or `Str`. When defined trait will add a method to determine if attribute is set or not. Note that it doesn't matter wether it was set with a builder or by an assignment.

    If parameter is `Bool` *True* then method name is made of attribute name prefixed with _has-_. See [What is "lazy attribute"](#What is "lazy attribute") section for example.

    If parameter is `Str` then the string contains predicate method name:

                has $.bar is mooish(:lazy, :predicate<bar-is-ready>);
                ...
                method baz {
                    if self.bar-is-ready {
                        ...
                    }
                }

  * *`clearer`*

    Could be `Bool` or `Str`. When defined trait will add a method to reset the attribute to uninitialzed state. This is not equivalent to *undefined* because, as was stated above, *Nil* is a valid value of initialized attribute.

    Similarly to *`predicate`*, when *True* the method name is formed with _clear-_ prefix followed by attribute's name. A `Str` value defines method name:

                has $.bar is mooish(:lazy, :clearer<reset-bar>, :predicate);
                ...
                method baz {
                    $.bar = "a value";
                    say self.has-bar;  # True
                    self.reset-bar;
                    say self.has-bar;  # False
                }

  * *`filter`*

    A filter is a method which is executed right before storing a value to an attribute. What is returned by the method will actually be stored into the attribute. This allows us to manipulate with a user-supplied value in any necessary way.

    The parameter can have values of `Bool`, `Str`, `Callable`. All values are treated similarly to the `builder` parameter except that prefix '*filter-*' is used when value is *True*.

    The filter method is passed with user-supplied value and the following named parameters:

    `attribute` - contains full attribute name.

    `builder` - passed if filter is called as a stage of attribute building.

    `old-value` - passed with the previous attribute value if it had one; i.e. if attribute has been initialized.

    **Note** that it is not recommended for a filter method to use the corresponding attribute directly as it may cause unforseen side-effects like deep recursion. The `old-value` parameter is the right way to do it.

  * *`trigger`*

    A trigger is a method which is executed right after attribute value has been changed.

    Allowed values for this parameter are `Bool`, `Str`, `Callable`. All values are treated similarly to the `builder` parameter except that prefix '*trigger-*' is used when value is *True*.

    Trigger method gets passed with the stored value as first positional parameter. If there is also a `filter` defined for the attribute then trigger receives the value returned by the filter, not the initial. I.e. it always get what's eventually stored in the attribute. It also receives the same named parameters as `filter` method: `attribute`, `builder`, `old-value`.

  * *`alias`, `aliases`, `init-arg`, `init-args`*

    Those are four different names for the same parameter which allows defining attribute aliases. So, whereas Internally you would have single container for an attribute that container would be accessible via different names. And it means not only attribute accessors but also clearer and predicate methods:

        class Foo {
            has $.bar is rw is mooish(:clearer, :lazy, :aliases<fubar baz>);

            method build-bar { "The Answer" }
        }

        my $inst = Foo.new( fubar => 42 );
        say $inst.bar; # 42
        $inst.clear-baz;
        say $inst.bar; # The Answer
        $inst.fubar = pi;
        say $inst.baz; # 3.1415926

    Aliases are not applicable to methods called by the module like builders, triggers, etc.

  * *`no-init`*

    This parameter will prevent the attribute from being initialized by the constructor:

        class Foo {
            has $.bar is mooish(:lazy, :no-init);

            method build-bar { 42 }
        }

        my $inst = Foo.new( bar => "wrong answer" );
        note $inst.bar; # 42

  * *`composer`*

    This is a very specific option mostly useful until role `COMPOSE` phaser is implemented. Method of this option is called upon class composition time.

Public/Private
--------------

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 {
                    self!reset-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 = Foo.new;
        $inst.bar = "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 = Foo.new;
        $inst.foo = "ABC";
        $inst.bar = "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, *% {...}));
        }

### Options

  * *`attribute`*

    Full attribute name with twigil. Passed to all callbacks.

  * *`builder`*

    Only set to *True* for `filter` and `trigger` methods when attribute value is generated by lazy builder. Otherwise no this parameter is not passed to the method.

  * *`old-value`*

    Set for `filter` only. See its description above.

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](#SYNOPSIS).

Performance
-----------

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.

Multi-threading
---------------

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

  * build operations are safe among themselves

  * clear operations are safe among themselves

  * anything else, including mix of build/clear operations, is unsafe

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.

CAVEATS
=======

  * Due to the "magical" nature of attribute behaviour conflicts with other traits are possible. In particular, mixing up with `is built` trait is not recommended.

  * Use of `Proxy` as the container may have unexpected side effects in some use cases like passing it as a parameter. Multiple calls of `Proxy`'s `FETCH` are possible, for example. While generally harmless this may result in performance issues of affected application. To workaround the problem attribute value can be temporarily assigned into a variable.

  * Another surprising side effect happens when a "mooified" array or hash attribute is used with a loop. Since `Proxy` is a container, loops are considering such attributes as itemized, no matter what their final value is. Consider the following:

        class Foo {
            has @.a is mooish(:lazy);
            method build-a { 1,2,3 }
            method dump {
                for @!a -> $val { say $val.raku }
            }
        }
        Foo.new.dump; # $[1, 2, 3]

    Note that this only happens when attribute is accessed privately as `for Foo.new.a {...}` would behave as expected. Also, for non-filtering and non-triggering attributes this only happens when the attribute is not initialized yet.

    The problem could be workarounded either by using `@.a` notation, or with explicit decontainerization `@!a<>`.

SEE
===

ALSO

[ChangeLog](ChangeLog.md)

AUTHOR
======

Vadim Belman <[email protected]>

LICENSE
=======

Artistic License 2.0

See the LICENSE file in this distribution.