MooseX
A postmodern object DSL for Ruby
Introduction
This is another DSL for object creation, aspects, method delegation and much more. It is based on Perl Moose and Moo, two important modules who add a better way of Object Orientation development (and I enjoy A LOT). Using a declarative style, using Moose/Moo you can create attributes, methods, the entire constructor and much more. But I can't find something similar in Ruby world, so I decide port a small subset of Moose to create a powerfull DSL for object construction.
Want to help? Install the gem moosex
and start to use. If you find some issue, please assign to me :)
Of course, there is few similar projects in ruby like
But the objetive of MooseX is different: this is a toolbox to create Classes based on DSL, with unique features like
- method delegation and currying ( see 'handles')
- lazy attributes
- roles / abstract classes / interfaces
- traits / monads
- plugins
- parameterized roles
- composable type check
- events
and much more.
This rubygem is based on this modules:
See also:
- Reindeer, another Moose port to Ruby (still on 0.0.1 version)
- Joose, a javascript port of Moose.
- Perl 6, Perl 6 OO programming style.
- Elk, Elk is an object system for Python inspired by Moose.
Why MooseX? Because the namespace MooseX/MooX is open to third-party projects/plugins/extensions. You can upgrade your Moo(se) class using other components if you want. And there is one gem called 'moose' :/
THIS MODULE IS EXPERIMENTAL YET! BE CAREFUL!
Talk is cheap. Show me the code!
require 'moosex'
class Point
include MooseX
has x: {
is: :rw, # read-write (mandatory)
isa: Integer, # should be Integer
default: 0, # default value is 0 (constant)
}
has y: {
is: :rw,
isa: Integer,
default: -> { 0 }, # you should specify a lambda
}
def clear!
self.x= 0 # to run with type-check you must
self.y= 0 # use the setter instad @x=
end
def to_s
"Point[x=#{self.x}, y=#{self.y}]"
end
end
# now you have a generic constructor
p1 = Point.new # x and y will be 0
p2 = Point.new( x: 5 ) # y will be 0
p3 = Point.new( x: 5, y: 4)
Installation
Add this line to your application's Gemfile:
gem 'moosex'
And then execute:
$ bundle
Or install it yourself as:
$ gem install moosex
You need ruby 2.0.x or superior.
Description
MooseX is an extension of Ruby object system. The main goal of MooseX is to make Ruby Object Oriented programming easier, more consistent, and less tedious. With MooseX you can think more about what you want to do and less about the mechanics of OOP. It is a port of Moose/Moo from Perl to Ruby world.
Read more about Moose on http://moose.iinteractive.com/en/
Motivation
It is fun
Usage
You just need include the MooseX module in your class and start to describe the attributes with our DSL. This module will inject one smart constructor, acessor and other necessary methods.
Instead the normal way of add accessors, constructor, validation, etc
class Foo
attr_accessor :bar, :baz, :bam
def initialize(bar=0, baz=0, bam=0)
unless [bar, baz, bam].all? {|x| x.is_a? Integer }
raise "you should use only Integers to build Foo"
end
@bar = bar
@baz = baz
@bam = bam
end
end
you can do this:
class Foo
include MooseX
has [:bar, :baz, :bam], {
is: :rw,
isa: Integer,
default: 0
}
# you can declare inline too
has :another, is: :rw, isa: Integer, default: ->{ Object.new }
end
DSL: the 'has' method
The key of the DSL is the 'has' method injected in your class. You should use this method do describe your class and define the behavior like this:
has :attribute_name, { hash of properties }
to describe one new attribute you shoud specify some properties inside a Hash. The only mandatory property is the ':is', to specify how we should create the acessors (if public or private).
The options for "has" are as follows:
is => ro|rw|rwp|private|lazy
Important, may be :ro, :rw, :rwp, :private or :lazy. If you not specify, we will consider :rw, with all acessors with public visibility (NEW).
"ro" specify a read-only attribute - generate only the reader method - you should specify the value in the constructor or using "default".
"rw" specify a read-write attribute - generate both reader and writter methods.
"rwp" specify a read-write private attribute. Similar to "rw" but the writter is a private method.
"private" will generate both reader and writter as private methods
"lazy" similar to "ro", but also sets "lazy" to true and "builder" to "build_#{attribute_name}".
isa => Class|lambda
You can specify an optional type check for the attribute. Accepts a lambda, and it must raise one exception if the type check fails. If you provides a Class or Module, we will call the 'is_a?' method in the new value againt the Class/Module. We call the type check routine on the constructor and in each call of the writter method.
You can specify your own kind of type validation.
isa: ->(value) do
unless value.respond_to? :to_sym
raise "bar should respond to to_sym method!"
end
end,
Important: if you access the attribute instance name using @attribute_name= you loose the type check feature. You need always set/get the attribute value using the acessors generated by MooseX.
default => Constant|lambda
You can specify an optional default value to one attribute. If we don't specify in the constructor, we will initialize the attribute with this value. You also can specify one lambda to force object creation.
default: 0,
or
default: -> { MyObject.new },
required => true|false
if true, the constructor will raise error if this attribute was not present.
required: true,
if this attribute has a default value, we will initialize with this value and no exception will be raised.
Optional.
coerce => method name|lambda
You can try to coerce the attribute value by a lambda/method before the type check phase. For example you can do
coerce: ->(value) { value.to_i },
or just
coerce: :to_i,
to force a convertion to integer. Or flatten one array, convert to symbol, etc. Optional.
handles => Array|Hash|Class|Module
One of the greatest features in MooseX: you can inject methods and delegate the method calling to the attribute. For example, instead do this:
def some_method(a,b,c)
@attribute.some_method(a,b,c)
end
you simply specify one or more methods to inject.
handles: [:some_method],
If you specify one Module or Class, we will handle all public instance methods defined in that Module or Class ( if Class, we will consider all methods except the methods declared in the superclass). The only limitation is BasicObject (forbidden).
If you need rename the method, you can specify a Hash:
handles: {
my_method_1: :method1,
my_method_2: :method2,
},
handles is similar to Forwardable module, the difference is the currying support and it is integrate with the has method/endpoint. If you want to use Forwardable, please use the reader method name instead the attribute name with @.
Optional.
Currying
It is possible curry constant values declaring a pair/hash and set one or more constant values / lambdas
handles: {
my_method_1: {
method1: 1
}
},
this will curry the constant 1 to the argument list. In other words:
obj.target.method1(1,2,3) # OR
obj.my_method_1(2,3)
are equivalent. You can curry as many arguments as you can.
handles: {
my_method_2: {
method2: [1, ->{ 2 } ]
}
},
will generate
obj.target.method1(1,2,3) # OR
obj.my_method_2(3)
are equivalent. if we find one lambda we will call on runtime.
Important: if you need do something more complex ( like manipulate the argument list, etc ) consider use the hook 'around'.
But how we can curry arrays?
Use Double arrays
handles: {
my_method_1: {
method1: [ [1,2,3] ]
}
},
this will curry the array [1,2,3] to the argument list. In other words:
obj.target.method1([1,2,3],2,3) # OR
obj.my_method_1(2,3)
are equivalent.
trigger => method name|lambda
You can specify one lambda or method name to be executed in each writter ( if coerce and type check does not raise any exception ). The trigger will be called in each setter and in the constructor if we do not use the default value. Useful to add a logging operation or some complex validation.
trigger: ->(this, new_value) do
this.logger.log "change the attribute value to #{new_value}"
end
or
has a: { is: :rw }
has b: {
is: :rw,
trigger: :verify_if_a_and_b_are_different,
}
...
def verify_if_a_and_b_are_different(new_value_of_b)
if self.a.eql? new_value_of_b
raise "a and b should be different!"
end
end
Optional.
writter => true|method name
You can specify the name of the attribute acessor, default is "#{attribute_name}=" (if true).
reader => true|method name
You can specify the name of the attribute acessor, default is "attribute_name" (if true).
predicate => true|method name
Creates a method who returns a boolean value if the attribute is defined. If true, will create one public "has_#{attribute_name}?" method by default.
For example
class Foo
include MooseX
has x: {
is: :rw,
predicate: true,
}
end
foo = Foo.new
foo.has_x? # returns false
foo.x= 10
foo.has_x? # returns true
Important: nil is different than undefined. If you do not initialize one attribute, you will receive one 'nil' if you try to fetch the value, but the state of this attribute is 'undefined'. If you set any value (even nil), the attribute will be considered 'defined' and the predicate will return true.
Optional.
clearer => true| attribute name
Creates a method who will unset the attribute. If true, will create one public "clear_#{attribute_name}!" method by default. Unset in this case is not 'nil', we will remove the instance variable. For example:
class Foo
include MooseX
has x: {
is: :rw,
predicate: true,
clearer: true,
}
end
foo = Foo.new
foo.has_x? # returns false
foo.x= 10
foo.has_x? # returns true
foo.clear_x! # will unset the attribute x
foo.has_x? # returns false
Optional.
init_arg => "new attribute name"
You can rename the attribute name in the constructor. For example:
class Foo
include MooseX
has secret: {
is: :rw,
writter: :x=,
reader: :x,
init_arg: :x,
}
end
foo = Foo.new(x: 1) # specify the value of secret in the constructor
foo.x # return 1
foo.x= 2 # will set 'secret' to 2
lazy => true|false
Another great feature: lazy attributes. If you this to true, we will wait until the first reader accessor be called to create the object using the builder method, then store the value. For example:
class Foo
include MooseX
has x: {
is: :rw,
lazy: :true,
predicate: true, # predicate and clearer are just
clearer: true, # to show a better example
}
def builder_x
Some::Class.new
end
end
foo = Foo.new # normal...
foo.has_x? # returns false
foo.x # will call the builder_x method and store the value
foo.has_x? # returns true
foo.x # returns the stored value
foo.clear_x! # will unset the attribute x
foo.has_x? # returns false
foo.x # will call the builder again and store the value
A lazy attribute needs a builder method or lambda. By default you should implement the "builder_#{attribute_name}" method. Using lazy you should initialize one attribute when you really need.
Optional.
builder => method name|lambda
You can specify the builder name if the attribute is lazy, or you can specity one lambda. If true, the default name of the builder will be "builder_#{attribute_name}". This attribute will be ignored if the attribute is not lazy.
class Foo
include MooseX
has x: {
is: :rw,
lazy: :true,
builder: ->(this) { Some::Class.new } # you can ignore foo, or use it!
}
end
The difference between builder and default is: default is for initialization if you omit the value in the constructor and you can't access the object (it is not "created" yet), builder is for lazy attributes, when you call the reader method for the first time ( or after clear it ) we will initialize the attribute and you have access to the object (in this example using the parameter 'this' in the lambda or using a normal method).
Optional.
weak => true|false
If true, we will always coerce the value from default, constructor or writter to a WeakRef. Weak Reference class that allows a referenced object to be garbage-collected.
class Foo
include MooseX
has x: { is: :rw, weak: true }
end
f = Foo.new(x: Object.new)
puts f.x.class # will be WeakRef
GC.start
puts f.x # may raise exception (recycled)
You should verify with weakref_alive?
method to avoid exceptions.
Optional.
doc => String
You can add a string metadata about the attribute. If you include MooseX with meta: true
you can inspect the list of attributes and documentation.
Optional.
override => true|false
If you need override one attribute, you should use override: true
, or MooseX will raise one exception.
traits => Trait | [Array of Traits]
The objective of use traits is extends the original attribute, using delegators (think in Monads). We support few traits at this moment and, Important, if you set a list of Traits we will apply each trait in sequence. Have a Suggestion? Open an Issue on Github!
Trait Counter
class MyHomePage
include MooseX::Types
has :counter, {
is: :ro,
isa: Integer,
default: 0,
traits: MooseX::Traits::Counter,
handles: {
increase_counter: :inc,
decrease_counter: :dec,
reset_counter_to_zero!: :reset,
}
}
In this example, the class MyHomePage has one attribute counter, Integer, and you can increase and decrease the counter value (or reset to zero) using increase_counter
, decrease_counter
and reset_counter_to_zero!
methods. Without the trait Counter
you should create this three methods and set the visibility to :rwp
to be able to access the value, do some math and save. In this example, MooseX::Traits::Counter
is a SimpleDelegator who act as a wrapper to the original value and act as a proxy for all methods except inc
, dec
and reset
.
We apply the traits list on default values, constructor and writter, after type check and coerce, and it is useful to extend the behavior of the attribute. In this case, 0 is a Immutable object, we can't increase the value using some method (different than String, when you use methods like capitalize!
) but we can use a Delegator to act as a proxy and increase/decrease the value.
Trait Bool
Another example, using mutable boolean values:
has bit: {
is: :ro,
default: true,
traits: MooseX::Traits::Bool,
handles: [ :toggle!, :value ],
}
In this case, we have a bit, and this bit is turn on (true). It is read-only so you can't set the bit. But using toggle!
we can turn true => false and false => true. To access the original boolean value we should use !! ( to coerce to a true|false value ) or we can access the value
method. It is mandatory if you want to use in if/unless statements. Of course you can change the name of each method.
Trait Pair
Sometimes we need store an array of fixed size and each element has an identity. For example, we should store arrays as tuples, or pairs. Of course we can create an object to be more clear, but we have this option and, sometimes, it is useful. Now, consider this example:
has array_with_surname_name: {
is: :private,
isa: Array, # or isTuple(String, String) if you include MooseX::Types
traits: [ MooseX::Traits::Pair ],
handles: {
:surname => :first,
:surname= => :first=,
:name => :second,
:name= => :second=,
:surname_and_name => { join: ->{","} }
},
required: true,
}
We store surname and name as a tuple of Strings. Instead access array_with_surname_name[0]
or array_with_surname_name[1]
, we can apply the Trait Pair
, and access (or save!) each component of this pair, but you can't change the pair itself. Look the example of currying in surname_and_name
, calling join with "," as an argument.
Trait Expires
This trait will wrap the original value and set one expiration time (in seconds, -1 to never expires). Accepts one tuple (array) of [ value, expiration ].
requires 'moosex'
requires 'moosex/traits'
class MyHomePage
include MooseX
has session: {
is: :rw,
default: -> { {} },
coerce: ->(value) { ((value.is_a?(Array))? value : [value, 3]) },
traits: MooseX::Traits::Expires,
}
end
page = MyHomePage.new
page.session.valid? # => true
page.session # => {}
sleep(3)
page.session.valid? # => false
page.session= { bar: 5 } # will be coerce to [ {bar: 5}, 3 ]
page.session.valid? # => true
page.session # => { bar: 5 }
sleep(3)
page.session.valid? # => false
If you want avoid the ugly coerce, you can specify one expired time using the method 'with'.
requires 'moosex'
requires 'moosex/traits'
class MyHomePage
include MooseX
has session: {
is: :rw,
default: -> { {} },
traits: MooseX::Traits::Expires.with(3),
}
end
See plugin "ExpiredAttribute" to see how you can specify one expire time and mix with lazy mechanism to auto build in case of not valid value.
Create your own trait
You should create a Class with a constructor who will receive a reference to the value. For example, the trait Counter is using SimpleDelegator:
require 'delegate' # to use SimpleDelegator
module MooseX
module Traits
class Counter < SimpleDelegator
def initialize(value)
@value = value
super(@value)
end
def inc(by=1)
@value += by
__setobj__(@value)
@value
end
...
You can create or extend your own Traits too. It is easy.
Traits Removed
IMPORTANT RescueToNil, RescueToZero and RescueToEmptyString traits are removed since 0.0.20 version
Plugins
You can extend MooseX using Plugins, and you can create your own plugins. The only kind of plugin supported is attribute plugin
, but we can accept other kinds of plugins in the future. To enable one ( or more ) plugins you should specify the list of plugins in the init method when you include the MooseX module.
Plugin Chained
The original behavior of the writter method is return the attribute value. If you want return self
to continue calling other methods - to create one fluent interface, for example - you can use the plugin MooseX::Plugins::Chained
and avoid writters with = in the end. For example:
require 'moosex'
require 'moosex/plugins'
class EmailMessage
include MooseX.init(
with_plugins: MooseX::Plugins::Chained
)
has :_from, writter: :from, chained: true
has :_to, writter: :to, chained: true
has :_subject, writter: :withSubject, chained: true
has :_body , writter: :withBody, chained: true
def send!
# add logic
end
end
EmailMessage.new.
from("foo@bar.com").
to("me@baz.com").
withSubject("test").
withBody("hi!").
send!
Plugin ExpiredAttribute
It is a easy way to apply the trait Expired in lazy attributes, and auto build the attribute if it is not valid!
require 'moosex'
require 'moosex/plugins'
class MyClass
include MooseX.init(with_plugins: MooseX::Plugins::ExpiredAttribute)
has config: {
is: :lazy,
clearer: true, # optional, if not specify, we will force the default clearer method
expires: 60, # seconds
}
def build_config
# read configuration...
end
end
You need enable the clearer to read the configuration, in this case.
Build your own Plugin
You should create one Class who accepts one parameter in the constructor (it is a reference for the MooseX::Attribute class) and one method 'process' who will be invoked against the argument hash ( in the constructor ), and another method prepare. The reference for the attribute can be used to change the original behavoir and you must delete the used arguments from the hash (in process). See the file lib/moosex/plugins.rb
for more examples.
module MooseX
module Plugins
class Chained
def prepare(options); end
def initialize(this)
@this = this
end
def process(options)
chained = !! options.delete(:chained)
if chained
writter = @this.attribute_map[:writter]
old_proc = @this.methods[ writter ]
@this.methods[writter] = ->(this, value) { old_proc.call(this, value); this }
end
@this.attribute_map[:chained] = chained
end
end
...
Important the public API for MooseX::Attribute is under development and can change in any moment. This will be true until the first stable release.
Enable more than one plugin
You can pass the list of plugins as an array.
require 'moosex'
require 'moosex/plugins'
class MyClass
include MooseX.init(with_plugins: [ MooseX::Plugins::ExpiredAttribute, MooseX::Plugins::Chained ])
Hooks: after/before/around
Another great feature imported from Moose are the hooks after/before/around one method. You can run an arbitrary code, for example:
class Point
include MooseX
has [:x, :y ], { is: :rw, required: true }
def clear!
self.x = 0
self.y = 0
end
end
class Point3D < Point
has z: { is: :rw, required: true }
after :clear! ->(this) { this.z = 0 }
end
instead redefine the 'clear!' method in the subclass, we just add a piece of code, a lambda, and it will be executed after the normal 'clear!' method.
Each after/before/around will redefine the original method in order. If you want override the method in some subclass, you will loose all hooks. The best way to preserve all hooks is using after/before/around to modify the behavior if possible.
Roles and Hooks
If you try to add one role to a method who does not exists yet, this will be added in the next class. BE CAREFUL, THIS IS EXPERIMENTAL! PLEASE REPORT ANY BUG IF YOU FIND!!!
after (method| ARRAY) => lambda
The after hook should receive the name of the method as a Symbol and a lambda. This lambda will, in the argument list, one reference for the object (self) and the rest of the arguments. This will redefine the the original method, add the code to run after the method. The after does not affect the return value of the original method, if you need this, use the 'around' hook.
before (method| ARRAY) => lambda
The before hook should receive the name of the method as a Symbol and a lambda. This lambda will, in the argument list, one reference for the object (self) and the rest of the arguments. This will redefine the the original method, add the code to run before the method.
A good example should be logging:
class Point
include MooseX
def my_method(x)
# do something
end
before :my_method ->(this, x) do
puts "#{Time.now} before my_method(#{x})"
end
after :my_method ->(this, x) do
puts "#{Time.now} after my_method(#{x})"
end
end
around (method| ARRAY) => lambda
The around hook is agressive: it will substitute the original method for a lambda. This lambda will receive the original method as a lambda, a reference for the object and the argument list, you shuld call the method_lambda using object + arguments
around(:sum) ->(method_lambda, this, a,b,c) do
c = 0
result = method_lambda.call(this,a,b,c)
result + 1
end
it is useful to manipulate the return value or argument list, add a begin/rescue block, aditional validations, etc, if you need.
MooseX.init
The init method is useful to change some behavior of MooseX on the fly.
warnings => true|false
will disable all warnings from MooseX. It is Global. Default is true.
class X
include MooseX.init(warnings: false)
fatal => true|false
all warnings will raise exceptions. It is Global. Default is false
class X
include MooseX.init(fatal: true)
meta => Symbol|String|true|false
will expose an alias for the class metadata in the class. If you use true, we will generate a method 'meta', if you provide a Symbol or String we will create a method with same name.
class X
include MooseX.init(meta: true)
has :foo, { is: :rw, doc => "this is foo"}
end
X.meta.info # will return { :foo => "this is foo" }
The metadata has two public methods: attrs and info. Be careful.
MooseX::Types
WARNING This will be extract for a parallel project/ gem!
MooseX has a built-in type system to be helpful in many circunstances. How many times you need check if some argument is_a? Something? Or it respond_to? :some_method ? Now it is over. If you include the MooseX::Types module in your MooseX class you can use:
isAny
will accept any type. Useful to combine with other types.
class Example
include MooseX
include MooseX::Types
has x: { is: :rw, isa: isAny }
isConstant
will verify using :=== if the value is equal to some contant
has x: { is: :rw, isa: isConstant(1) }
isType, isInstanceOf and isConsumerOf
will verify the type using is_a? method. should receive a Class. isInstanceOf is an alias, to be used with Classes and isConsumerOf is for Modules.
has x: { is: :rw, isa: isConsumerOf(Enumerable) }
hasMethods
will verify if the value respond_to? for one or more methods.
has x: { is: :rw, isa: hasMethods(:to_s) }
isEnum
verify if the value is part of one enumeration of constants.
has x: { is: :rw, isa: isEnum(:black, :white, :red, :green, :yellow) }
isMaybe(type)
verify if the value isa type or is nil. You can combine with other types.
has x: { is: :rw, isa: isMaybe(Integer) } # accepts 1,2,3... or nil
isNot(type)
will revert the type check. Useful to combine with other types
has x: {
is: :rw, # x will accept any values EXCEPT :black, :white...
isa: isNot(isEnum(:black, :white, :red, :green, :yellow))
}
isArray(type)
Will verify if the value is an Array. Can receive one extra type, and we will verify each element inside the array againt this type, or Any if we not specify the type.
has x: {
is: :rw,
isa: isArray() # will accept any array
}
has y: {
is: :rw, # this is a more complex type
isa: isArray(isArray(isMaybe(Integer)))
}
isHash(type=>type)
similar to isArray. if you do not specify a pair of types, it will check only if the value is_a? Hash. Otherwise we will verify each pair key/value.
has x: {
is: :rw,
isa: isHash() # will accept any Hash
}
has y: {
is: :rw, # this is a more complex type
isa: isHash(Integer => isArray(isMaybe(Integer)))
}
isSet(type)
similar to isArray. the difference is: it will raise one exception if there are non unique elements in this array.
has x: {
is: :rw,
isa: isSet(Integer) # will accept [1,2,3] but not [1,1,1]
}
isTuple(types)
similar to isArray, Tuples are Arrays with fixed size. We will verify the type of each element.
For example, to specify one tuple with three elements, the first is Integer, the second is a Symbol and the las should be a String or nil:
has x: {
is: :rw,
isa: isTuple(Integer, Symbol, isMaybe(String))
}
isAllOf(types)
will combine all types and will fail if one of the condition fails.
has x: {
is: :rw,
isa: isAllOf(
hasMethods(:foo, :bar), # x should has foo and bar methods
isConsumerOf(Enumerable), # AND should be an Enumerable
isNot(SomeForbiddenClass), # AND and should not be an SomeForbiddenClass
)
}
isAnyOf(types)
will combine all types and will fail if all of the condition fails.
has x: {
is: :rw,
isa: isAnyOf(
hasMethods(:foo, :bar), # x should has foo and bar methods
isEnum(1,2,3,4), # OR be 1,2,3 or 4
isHash(isAny => Integer) # OR be an Hash of any type => Integers
)
}
Roles
Roles are Modules with steroids. If you include the MooseX module in another module, this module became a "Role", capable of store attributes to be reuse in other MooseX classes. For example:
require 'moosex'
module Eq
include MooseX
requires :equal
def no_equal(other)
! self.equal(other)
end
end
module Valuable
include MooseX
has value: { is: :ro, required: true }
end
class Currency
include Valuable.init(warnings: false) # default is true!
include Eq # default will warn about equal missing.
# alternative: include after equal definition!
def equal(other)
self.value == other.value
end
# include Eq # warning 'safe' include, after equal declaration
end
c1 = Currency.new( value: 12 )
c2 = Currency.new( value: 12 )
c1.equal(c2) # true, they have the same value
c1.no_equal(c2) # will return false
Roles can support has to describe attributes, and you can reuse code easily.
requires
You can also mark one or more methods as 'required'. When you do this, we will raise one exception if you try to create a new instance and the class does not implement it. It is a safe way to create interfaces or abstract classes. It uses respond_to? to verify.
Parameterized Roles
Parameterized roles is a good way of reuse code based on roles. For example, to create one or more attributes in the class who includes our role, we just add the code to be executed in the on_init hook.
module EasyCrud
include MooseX
on_init ->(*attributes) do
attributes.each ->(attr) do
has attr, { is: :rw, predicate: "has_attr_#{attr}_or_not?" }
end
end
end
class LogTest
include EasyCrud.init(:a, :b)
...
when we call init
with arguments, we will call all on_init blocks defined in the role. In this example we inject attributes 'a' and 'b' with reader/writter and a predicate based on the name ex: has_attr_a_or_not?
composable parameterized roles
To combine one or more parameterized roles to another parameterized role you should do something like this:
module Logabble2
include MooseX
on_init ->(args) do
args[:klass] = self
include Logabble.init(args)
end
end
module Logabble
include MooseX
on_init ->(args) do
klass = args[:klass] || self # <= THIS will guarantee you will
methods = args[:methods] || [] # modify the right class
methods.each do |method|
klass.around(method) do |original, object, *args|
# add log around method
# call original method
# return
...
BUILD
If you need run some code after the creation of the object (like some extra validation), you should implement the BUILD method.
class BuildExample
include MooseX
has [:x, :y], {
is: :rw,
required: true,
}
def BUILD
if self.x == self.y
raise "invalid: you should use x != y"
end
end
end
b1 = BuildExample.new(x: 1, y: 2) # will create the object
b2 = BuildExample.new(x: 1, y: 1) # will raise the exception!
BUILDARGS
If you need manupulate the constructor argument list, you should implement the method BUILDARGS. You MUST return one Hash of attribute_name => value.
class BuildArgsExample2
include MooseX
has [:x, :y], {
is: :rw,
required: true,
}
def BUILDARGS(x=4,y=8)
args = {}
args[:x] = x
args[:y] = y
args
end
end
ex1 = BuildArgsExample2.new(1,2) # x == 1, y == 2
ex2 = BuildArgsExample2.new(1) # x == 1, y == 8
ex3 = BuildArgsExample2.new() # x == 4, y == 8
EVENTS
WARNING This will be extract for a parallel project/ gem!
MooseX has a built-in event system, and it should be useful if you want to avoid after/before hooks ( depends of what is your problem ).
require 'moosex'
require 'moosex/event'
class Example
include MooseX
include MooseX::Event
def ping
self.emit(:pinged)
end
end
e = Example.new
e.on(:pinged) ->(this) do
puts "Ping!"
end
e.once(:pinged) ->(this) do
puts "Ping Once!"
end
e.ping # will print two messages, "Ping!" and "Ping Once!"
e.ping # will print just "Ping!"
e.remove_all_listeners(:pinged)
e.ping # will no longer print nothing
# you can use arguments
# consider you have one logger attribute in this example
listener = e.on(:error) ->(obj, message) do
obj.logger.fatal("Error: #{message}")
end
e.emit(:error, "ops") # will log, as fatal, "Error: ops"
e.remove_listener( error: listener )
e.emit(:error, "...") # will no longer log nothing
If you want to restrict how many different events you can handle, you should overload the has_events
method and return one array of valid events. If you want accept all, you should return nil (default).
For example, your method should emit many events, in many points, and you can add/remove listeners easily. And Much More!
Events + Handles / Currying
Look this good example:
require 'moosex'
require 'moosex/event'
class EventHandler
include MooseX
include MooseX::Event
def has_events ; [ :pinged, :ponged ]; end
end
class EventProcessor
include MooseX
has event_handler: {
is: :ro,
isa: EventHandler,
default: -> { EventHandler.new }, # EventProcessor HAS ONE EventHandler
handles: { # Now, lets start to delegate and currying:
ping: { emit: :pinged }, # ping() is the same of event_handler.emit(:pinged)
pong: { emit: :ponged }, # pong(x) is the same of event_handler.emit(:pinged,x)
on_ping: { on: :pinged }, #
on_pong: { on: :ponged }, # same thing for on_ping / on_pong
},
}
end
ep = EventProcessor.new()
ep.on_ping ->(obj) do
puts "receive ping!"
end
ep.on_pong ->(obj, message) do
puts "receive pong with #{message}!"
end
ep.ping # will print "receive ping!"
ep.pong 1 # will print "receive pong with 1!"
Now, imagine what you can do with a parameterized role: we can create all handles based on event names!
IMPORTANT
This module is experimental. I should test more and more to be possible consider this "production ready". If you find some issue/bug please add here: https://github.com/peczenyj/MooseX/issues
Until the first 0.1 version I can change anything without warning.
I am open to suggestions too.
Mailing List
https://groups.google.com/d/forum/moosex-ruby-dev-list
TODO
- Support to Roles ( it is a Module on Steroids ) [done]
- Support to after/before/around [done]
- Improve the typecheck system (we should specify: we need an array of positive integers) [done]
- Improve the exception and warning system [in progress]
- Profit!
Limitations
Experimental module, be careful.
Now has limited support to subclassing.
Contributing
- Fork it ( http://github.com/peczenyj/MooseX/fork )
- Create your feature branch (
git checkout -b my-new-feature
) - Commit your changes (
git commit -am 'Add some feature'
) - Push to the branch (
git push origin my-new-feature
) - Create new Pull Request