Ruby "good practice" using constants?

Dan Rathbun

I did the first all in one line, at the console, and it did not work:

` module Dan; OUTER = self; module Plugin; PLUGIN = self; class Special; CLASS = self; end; end; end

$ Dan::Plugin::Special::OUTER
%(#008000)[Error: #<NameError: (eval): uninitialized constant Dan::Plugin::Special::OUTER>
(eval)]`

Dan Rathbun

@thomthom said:

@dan rathbun said:

This is what include and extend are for. Shared library modules are best written as "mixin" modules.

Does that duplicate the methods?

NO.

@thomthom said:

Or do they all refer to the source module?

YES... it's like a proxy lookup.

So if you define two sub-modules that include a library mixin module of constants, and sub-module two changes a value of a constant, it changes IN the mixin module, and any other sub-module that has included the mixin module, will see the changes.

Dan Rathbun

@thomthom said:

My snippet works. Try it.

Reason, because it's a Ruby interpreter block scope issue (Easter egg like thing.)

The interpreter reads the humanized script, and translates this:

module TT;;Plugins;;Example

  PLUGIN = self

  module FooBar
    def self.magic_button
       puts PLUGIN.name
    end
  end

end

.. to (something similar on the C-side):

TT;;Plugins;;Example = Module.new {

  PLUGIN = self

  FooBar = Module.new {
    def self.magic_button
       puts PLUGIN.name
    end
  }

}

So the Foobar definition block has access to the object references in the wrapping scope.

IMHO: Although it works.. explaining why is complicated and will likely confuse newbies, and it not really following good organizational form, (which is what we want to teach newbies.)

thomthom

@dan rathbun said:

So if you define two sub-modules that include a library mixin module of constants, and sub-module two changes a value of a constant, it changes IN the mixin module, and any other sub-module that has included the mixin module, will see the changes.

That's interesting. So one can then add all common methods and constants to the mixin... and no extra overhead.

Dan Rathbun

@thomthom said:

That's interesting. So one can then add all common methods and constants to the mixin... and no extra overhead.

Yea! That is the whole point ... and it also works with the module vars in the mixin module.

However sometimes evaluation may actually occur with in the Mixin module, rather than the "mixee" module. SO .. test.

thomthom

@dan rathbun said:

However sometimes evaluation may actually occur with in the Mixin module, rather than the "mixee" module. SO .. test.

What evaluation?

Dan Rathbun

of methods... so example you'd have to pass in references from the mixee modules. (Don't expect the mixin method to have access to objects in the mixee module. ... From what I remember.)

However (as always,) mixing into classes is different. If the mixin instance method has a ref to a @var, the object used is the one in the mixee class. Which is what you would expect.

It's just that including modules into are a bit more "funky" then mixing modules into a class.

Dan Rathbun

But we are getting off-topic here ... let's get back on constants.

thomthom

@dan rathbun said:

of methods... so example you'd have to pass in references from the mixee modules. (Don't expect the mixin method to have access to objects in the mixee module. ... From what I remember.)

I'm not quite following you here..
Got a sample code to illustrate?

brewsky

Thanks guys!

I have to do some reading to be able to follow the discussion
I kinda lost it somewhere along the way...

TT's first post came close to what I want to do.
This is more what I mean:

module TT;;Plugins

  PLUGIN = QuadFaceTools.new

end

@dan rathbun said:

Can you give us a code shell showing the namespace nesting, and how you wish to share the reference to your classes ??

This is an example of what I'm doing now.
I make some sort of "root-plugin" object(class-instance, not a module) that holds all other plugin objects/data.
And pass this on to all nested objects to be able to access the embedded data.

module Brewsky;;BimTools

  class BimTools
    attr_accessor ;project_list, ;web_dialog
    def initialize
      btProject = BtProject.new(self)
    end
  end
  
  class BtProject
    attr_reader ;model, ;guid, ;name, ;description
    def initialize(bt)
      @bt = bt
    end
    def update_dialog(value)
      dialog = @bt.web_dialog
      
      # do something with "dialog" using "value"
      
    end
  end
  
  BimTools.new

end

Dan Rathbun

Sometimes I pass in the outer namespace via the constructor (as you show in your example,) but usually it is into Observer instances. I normally give the "handle" the name @parent, @outer or similar.

Access to data objects in the parent namespace, (if they are few,) can be done with more clarity, by also passing them in via the constructor call.
See my simple plugin example: [Code] AnimateSelection Example
In this example, the references are passed in as individual arguments.
But if there are many, it may make more sense to collect the objects into a Hash or a Struct, and pass that in.

brewsky

I guess the way I'm passing the object into every sub-object is "structurally" the best way to go.
But in this way I'm constantly making new pointers to always the same old base-object.
And because there is only one instance, would it not be clearer to use some sort of "almost-global" object, such as a module-constant, and have access to it anywhere in the program?

Like TT does?
And after thinking on this, maybe TT's way of just making a module as a base-object for the plugin is a better approach than my BimTools-class. Because is't only used once, the module-approach seems more fitting...

Something like:

module Brewsky

  class BimTools
    attr_accessor ;project_list, ;web_dialog
    def initialize
      btProject = BtProject.new
    end
  end
  
  class BtProject
    attr_reader ;model, ;guid, ;name, ;description
    def update_dialog(value)
      dialog = PLUGIN.web_dialog
      
      # do something with "dialog" using "value"
      
    end
  end
  
  PLUGIN = BimTools.new

end

Dan Rathbun

Here is a SketchUp specific example of using a Constant Library mixin module so an author's various plugin modules (and/or submodules,) can share the references to the author's menu and submenu objects.

See: One Submenu for many plugins

brewsky

@dan rathbun said:

So for example, your nested module Brewsky::BimTools::Manager is actually an instance object, and the preceeding identifier is the reference to the instance.

If you remember this... then it can be easier to understand how using an anonymous singleton proxy class instance inside your Module class instance, makes sense.

Thank you very much for this very helpful post!
I completely missed it untill now

My plugin is in need of "some" improvement...

Dan Rathbun

@brewsky said:

And after thinking on this, maybe TT's way of just making a module as a base-object for the plugin is a better approach than my BimTools-class. Because is't only used once, the module-approach seems more fitting...

Oh hell yes.

If you need only one copy of a code object, then generally it should be a module.

If you need multiple copies of a code object (usually because the code must adapt to many other instance objects,) then you make it a class, and instantiate instances that are syncronized to a particular instance object.

Often coders try to avoid using a module, because they think it's a static kind of object, and they believe it is harder to use than a class instance.
What they miss is, that a module definition, is really an instance of class Module. So for example, your nested module Brewsky::BimTools::Manager is actually an instance object, and the preceeding identifier is the reference to the instance.

If you remember this... then it can be easier to understand how using an anonymous singleton proxy class instance inside your Module class instance, makes sense.

Imagine the Module class is "the hen".

It lays an egg, which is your nested Manager module instance, that could be created thus:
` Brewsky::BimTools::Manager = Module.new {

plugin managerial code here

}But the Ruby interpreter calls the new()` constructor for you on the C-side of things. (Ie, the defintion block syntax for scripts, was invented for human happiness and readability; Ruby itself does not really need it.)

But having methods in modules communicate (call each other,) works a bit different than in a class definition. Instance method definitions in a module, are meant for Library Mixin modules. (Read up on the include and extend methods.) They become different kinds of methods, depending on whether they are mixed into a class or module, and whether include and extend is used to do the mixing.)

So at first blush, the coder thinks they must define all methods in a module as module functions that must be called with self.method_name(), ... they find this cumbersome, and they switch back to using a class defintion, and using only one instance of it. (A sort of psuedo-singleton class.)

BUT.. the egg can have a membrane inside it's shell. This membrane analogy is the anonymous singleton proxy class instance created with the following syntax:

module Brewsky
  module BimTools
  end
end

module Brewsky;;BimTools;;Manager

  # module variables and constants declared here
  MGR_VERSION ||= '1.2.3'
  
  @@bimmenu ||= nil

  class << self # self evaluates to the module instance

    # Everything in here acts like a class
    #  instance BECAUSE IT ACTUALLY IS !

    # In here we can access the module @@vars directly.

    # In here we define instance methods, not module methods.
    # But if they are public, they can be called like module
    # functions, from anywhere outside the module.
    
    private
    
    def get_version()
      MGR_VERSION
    end

    # In here we can call any other method in here,
    #  without module qualification.

    public
    
    def version()
      get_version()
    end

  end # proxy class

  # Out here we can call any of the methods inside
  #  the proxy class directly, without qualification.
  unless @@bimmenu
    @@bimmenu = UI.menu('Plugins').add_submenu('BIMTools')
    @@bimmenu.add_item('Version') { UI.messagebox("BIMTools ver #{version()}") }
  end

end # module Brewsky;;BimTools;;Manager

# 99.9% of the time, there is no good reason to
# have any executing code outside your namespace.

EDIT(2012-12-16): changed post title to "Why use a module instead of a class ?"

Ruby &quot;good practice&quot; using constants?

plugin managerial code here

Ruby "good practice" using constants?