URLS http://raa.ruby-lang.org/search.rhtml?search=traits http://codeforpeople.com/lib/ruby/traits ABOUT traits.rb is set of attr_* like methods on steroids, caffeine, and botox. it encourages better living through meta-programming and uniform access priciples. traits.rb supercedes attributes.rb. why? the name is shorter ;-) VERSION 0.8.0 HISTORY 0.8.0 - traits now supports a whole slew of hooks that can be registered to fire pre or post setting an attribute, to cast a value to another type, to munge a value destructively, to require only certain types, to require a certain ducktype signature, and to validate arguments passed. check out sample/m.rb, sample/n.rb, or sample.o.rb to see it in action. the mechanism is quite flexible allowing method names, lambdas of varying arity, and lists of either/or to be passed to any hook. - you can find a gem for trais on codeforpeople - but i've still not coded up automated updating from codeforpeople to rubyforge so it won't show up as a remote gem yet. 0.7.0 - patched in the support i had written eariler for 'hooks' to be called pre/post setting a trait. plus shortcut to 'validate' traits which simply sets up a 'pre' hook which is used as a predicate. eg: class C; trait 'number', 'validate' => proc{|n| Numeric === n} pre and post hooks are used in the same way, eg: class C trait 'a', 'pre' => proc{|val| p "#{ val } to set with"}, 'post' => proc{|val| p "#{ val } set"}, end but the really cool thing is that all of these blocks are both passed the value in question but also evaluate with 'self' set appropriately. eg class Car positive_int = lambda{|n| Fixnum === n and n > 0} legal = proc{|s| s < speed_limit} trait 'speed_limit', 'validate' => positive_int, 'default' => 42 trait 'speed', 'validate' => legal end c = Car::new c.speed = 115 works as you'd expect: (eval):14:in `speed=': validation of speed=(115) failed! (ArgumentError) from a.rb:13 0.6.0 - fixed bug in where a default trait given as an empty array, eg: class C; has 'a' => []; end was exploded into the empty list when passed to the setter to initialize the default value. 0.5.0 - general code cleanup 0.4.0 - tweaked writer code so multiple values can be passed to setters - tweaked method of running blocks to use instance_eval so explicit 'this' arg is no longer needed (though it can still be used) 0.3.0 added ability of default values to be specified with block for deferred context sensitive initialization (see sample/c.rb) 0.1.0 completely reworked impl so NO parsing of inspect strings is required - it's all straight methods (albeit quite confusing ones) now. the interface is unchanged. 0.0.0 initial version AUTHOR ara [dot] t [dot] howard [at] noaa [dot] gov SAMPLES <========< sample/a.rb >========> ~ > cat sample/a.rb require 'traits' # # defining a trait is like attr_accessor in the simple case # class C trait :t end o = C::new o.t = 42 p o.t # # and can be made even shorter # class B; has :x; end o = B::new o.x = 42 p o.x ~ > ruby sample/a.rb 42 42 <========< sample/b.rb >========> ~ > cat sample/b.rb require 'traits' # # multiple traits can be defined at once using a list/array of string/sybmol # arguments # class C has :t0, :t1 has %w( t2 t3 ) end obj = C::new obj.t0 = 4 obj.t3 = 2 print obj.t0, obj.t3, "\n" ~ > ruby sample/b.rb 42 <========< sample/c.rb >========> ~ > cat sample/c.rb require 'traits' # # a hash argument can be used to specify default values # class C has 'a' => 4, :b => 2 end o = C::new print o.a, o.b, "\n" # # and these traits are smartly inherited # class K < C; end o = K::new o.a = 40 p( o.a + o.b ) # note that we pick up a default b from C class here since it # has not been set o.a = 42 o.b = nil p( o.b || o.a ) # but not here since we've explicitly set it to nil # # if a block is specifed as the default the initialization of the default value # is deferred until needed which makes for quite natural trait definitions. the # block is passed 'self' so references to the current object can be made. (if # this were not done 'self' in the block would be bound to the class!) # class C class << self has('classname'){ name.upcase } end has('classname'){ self.class.classname.downcase } end class B < C; end o = C::new p C::classname p o.classname o = B::new p B::classname p o.classname ~ > ruby sample/c.rb 42 42 42 "C" "c" "B" "b" <========< sample/d.rb >========> ~ > cat sample/d.rb require 'traits' # # all behaviours work within class scope (metal/singleton-class) to define # class methods # class C class << self traits 'a' => 4, 'b' => 2 end end print C::a, C::b, "\n" # # singleton methods can even be defined on objects # class << (a = %w[dog cat ostrich]) has 'category' => 'pets' end p a.category # # and modules # module Mmmm class << self; trait 'good' => 'bacon'; end end p Mmmm.good ~ > ruby sample/d.rb 42 "pets" "bacon" <========< sample/e.rb >========> ~ > cat sample/e.rb require 'traits' # # shorhands exit to enter 'class << self' in order to define class traits # class C class_trait 'a' => 4 c_has :b => 2 end print C::a, C::b, "\n" ~ > ruby sample/e.rb 42 <========< sample/f.rb >========> ~ > cat sample/f.rb require 'traits' # # as traits are defined they are remembered and can be accessed # class C class_trait :first_class_method trait :first_instance_method end class C class_trait :second_class_method trait :second_instance_method end # # readers and writers are remembered separatedly # p C::class_reader_traits p C::instance_writer_traits # # and can be gotten together at class or instance level # p C::class_traits p C::traits ~ > ruby sample/f.rb ["first_class_method", "second_class_method"] ["first_instance_method=", "second_instance_method="] [["first_class_method", "first_class_method="], ["second_class_method", "second_class_method="]] [["first_instance_method", "first_instance_method="], ["second_instance_method", "second_instance_method="]] <========< sample/g.rb >========> ~ > cat sample/g.rb require 'traits' # # another neat feature is that they are remembered per hierarchy # class C class_traits :base_class_method trait :base_instance_method end class K < C class_traits :derived_class_method trait :derived_instance_method end p C::class_traits p K::class_traits ~ > ruby sample/g.rb [["base_class_method", "base_class_method="]] [["derived_class_method", "derived_class_method="], ["base_class_method", "base_class_method="]] <========< sample/h.rb >========> ~ > cat sample/h.rb require 'traits' # # a depth first search path is used to find defaults # class C has 'a' => 42 end class K < C; end k = K::new p k.a # # once assigned this is short-circuited # k.a = 'forty-two' p k.a ~ > ruby sample/h.rb 42 "forty-two" <========< sample/i.rb >========> ~ > cat sample/i.rb require 'traits' # # getters and setters can be defined separately # class C has_r :r end class D has_w :w end # # defining a reader trait still defines __public__ query and __private__ writer # methods # class C def using_private_writer_and_query p r? self.r = 42 p r end end C::new.using_private_writer_and_query # # defining a writer trait still defines __private__ query and __private__ reader # methods # class D def using_private_reader p w? self.w = 'forty-two' p w end end D::new.using_private_reader ~ > ruby sample/i.rb false 42 false "forty-two" <========< sample/j.rb >========> ~ > cat sample/j.rb require 'traits' # # getters delegate to setters iff called with arguments # class AbstractWidget class_trait 'color' => 'pinky-green' class_trait 'size' => 42 class_trait 'shape' => 'square' %w( color size shape ).each{|t| trait(t){self.class.send t}} def inspect "color <#{ color }> size <#{ size }> shape <#{ shape }>" end end class BlueWidget < AbstractWidget color 'blue' size 420 end p BlueWidget::new ~ > ruby sample/j.rb color size <420> shape <========< sample/k.rb >========> ~ > cat sample/k.rb require 'traits' # # the rememberance of traits can make generic intializers pretty slick # class C # # define class traits with defaults # class_traits( 'a' => 40, 'b' => 1, 'c' => 0 ) # # define instance traits whose defaults come from readable class ones # class_rtraits.each{|ct| instance_trait ct => send(ct)} # # any option we respond_to? clobbers defaults # def initialize opts = {} opts.each{|k,v| send(k,v) if respond_to? k} end # # show anything we can read # def inspect self.class.rtraits.inject(0){|n,t| n += send(t)} end end c = C::new 'c' => 1 p c ~ > ruby sample/k.rb 42 <========< sample/l.rb >========> ~ > cat sample/l.rb require 'traits' # # even defining single methods on object behaves # a = [] class << a trait 'singleton_class' => class << self;self;end class << self class_trait 'x' => 42 end end p a.singleton_class.x ~ > ruby sample/l.rb 42 <========< sample/m.rb >========> ~ > cat sample/m.rb require 'traits' # # pre and post hooks can be passed a proc or the name of a method, the arity is # detected and the proc/method sent either the value, or the name/value pair # class C HOOK_A = lambda{|value| puts "HOOK_A : #{ value }"} HOOK_B = lambda{|name, value| puts "HOOK_B : #{ name } = #{ value }"} def hook_a value puts "hook_a : #{ value }" end def hook_b name, value puts "hook_b : #{ name } = #{ value }" end trait 'x', 'pre' => HOOK_A, 'post' => 'hook_b' trait 'y', 'pre' => HOOK_B, 'post' => 'hook_a' end c = C::new c.x = 42 c.y = 'forty-two' ~ > ruby sample/m.rb HOOK_A : 42 hook_b : x = 42 HOOK_B : y = forty-two hook_a : forty-two <========< sample/n.rb >========> ~ > cat sample/n.rb require 'traits' # # two kinds of in-place modifications are supported : casting and munging. # casting is a hook that requires either a proc or the name of a method that # will be used to convert the objects type. munging is similar execpt the # method is called on the object itself. like all hooks, lists may be provided # instead of a single argument # # you'll notice that the hooks and methods defined here are not strictly needed, # but are for illustration purposes only. note that all hooks operate in the # context of self - they have access to instance vars, etc., like instance_eval # class C INT = lambda{|i| int i} def int i Integer i end trait 'a', 'cast' => 'int' trait 'b', 'cast' => INT trait 'c', 'munge' => 'to_i' trait 'd', 'cast' => 'Integer' trait 'e', 'munge' => %w( to_i abs ) end c = C::new c.a = '42' p c.a c.b = '42' p c.b c.c = '42' p c.c c.d = '42' p c.d c.e = '-42' p c.e ~ > ruby sample/n.rb 42 42 42 42 42 <========< sample/o.rb >========> ~ > cat sample/o.rb require 'traits' # # three kinds of validation hooks are supported, flat out typing, ducktyping, # and generic validation. all hooks are evaluated in the context of self and # can consist of lists of strings/symbols and/or lambdas. any string/symbol is # taken to be a method name of the current object. # # # type hooks must be an object, which will be compared with '===', or a lambda # or method name that returns an object to be compared using '==='. 'case' is # an alias for 'type'. # class C trait 'a', 'types' => [Fixnum, Numeric] trait 'b', 'case' => %r/^4#{ 1 + 1}$/, 'munge' => 'to_s' trait 'c', 'type' => lambda{rand > 0.42 ? Fixnum : Numeric} end c = C::new c.a = 42 p c.a c.b = 42 p c.b c.c = 42 p c.c # # ducktyping is supported as a list of methods an object must support. # class C trait 'a', 'ducktypes' => %w( abs zero? ) trait 'b', 'ducktype' => %w( abs floor ) end c = C::new c.a = 42 p c.a c.b = 42.0 p c.b # # finally, generic true/false validation is supported # class C def forty_two n n.to_i == 42 end trait 'a', 'validate' => lambda{|x| x == 42} trait 'b', 'validate' => 'forty_two' end c = C::new c.a = 42 p c.a c.b = 42.0 p c.b ~ > ruby sample/o.rb 42 "42" 42 42 42.0 42 42.0 CAVEATS this library is experimental and subject to (eg. will) change - though it has not for several version and much of my code hinges is on it now so you can expect it to be stable-ish in the future - the only changes would be ones to fix bugs.