# Class attribute unique to subclasses but used in parent clas

**URL:** <https://www.ruby-forum.com/t/class-attribute-unique-to-subclasses-but-used-in-parent-clas/73961>\
**Category:** Ruby\
**Created:** [October 28, 2006, 3:37pm UTC](https://www.ruby-forum.com/t/class-attribute-unique-to-subclasses-but-used-in-parent-clas/73961 "2006-10-28T15:37:28Z")\
**Posts on this page:** 4\
**Page:** 1

<div class="post-metadata">

**Author:** ![casper\_the\_ghost](https://www.ruby-forum.com/letter_avatar_proxy/v4/letter/c/58956e/32.png) [@casper\_the\_ghost](https://www.ruby-forum.com/u/casper_the_ghost)\
**Post date:** [October 28, 2006, 3:37pm UTC](https://www.ruby-forum.com/t/class-attribute-unique-to-subclasses-but-used-in-parent-clas/73961/1 "2006-10-28T15:37:28Z")

</div>

Hello,

I’m new in the Ruby World and I have a question.

Let’s say I have a class with a class attribute (I call this class  
Parent), this class attribute is used in instance methods of Parent.  
Parent class has many subsclasses. Each subsclass should have it’s own  
value of the class attribute.

Example:

class Parent  
@@test = nil

def Parent.create(value)  
@@test=value  
new  
end

def print\_test  
puts @@test  
end  
end

class Child1 \< Parent  
end

class Child2 \< Parent  
end

c1 = Parent.create ‘1’  
c1.print\_test # =\> 1  
c2 = Parent.create ‘2’  
c2.print\_test # =\> 2  
c1.print\_test # =\> 2

What I’ll like, is that Child1 and Child2 have their own values of  
@@test, but @@test should be initialized and used in Parent.  
Also, @@test cannot be an instance variable because it is heavy to  
initialize and will be common to all instances of Child1 or Child2.

Thank you

---

<div class="post-metadata">

**Author:** ![casper\_the\_ghost](https://www.ruby-forum.com/letter_avatar_proxy/v4/letter/c/58956e/32.png) [@casper\_the\_ghost](https://www.ruby-forum.com/u/casper_the_ghost)\
**Post date:** [October 28, 2006, 3:45pm UTC](https://www.ruby-forum.com/t/class-attribute-unique-to-subclasses-but-used-in-parent-clas/73961/2 "2006-10-28T15:45:28Z")

</div>

Hi –

On Sat, 28 Oct 2006, removed\_email\_address@domain.invalid wrote:

> end  
> c2 = Parent.create ‘2’  
> c2.print\_test # =\> 2  
> c1.print\_test # =\> 2
> 
> What I’ll like, is that Child1 and Child2 have their own values of  
> @@test, but @@test should be initialized and used in Parent.  
> Also, @@test cannot be an instance variable because it is heavy to  
> initialize and will be common to all instances of Child1 or Child2.

Actually it should be an instance variable, but an instance variable  
of the class object.

What you’re trying to do is to maintain state on a per-object basis,  
and for that you need an instance variable. In fact, you can give  
your class objects their own accessor methods, by creating those  
methods in the singleton class of the first one:

class C  
class \<\< self  
attr\_accessor :n  
end  
end

class D \< C  
end

C.n = 1  
p C.n # 1  
D.n = 2  
p D.n # 2  
p C.n # 1

Note that if you do this, you have to use the accessor methods to get  
at the values from outside (as is the case with all objects):

class C  
def my\_class\_n  
self.class.n  
end  
end

c = C.new  
C.n = 100  
puts c.my\_class\_n # 100

But that’s actually good. Since the object © is maintaining its own  
state, it makes perfect sense to have to query it. Class variables  
demolish that logic, which is one of a number of reasons I dislike  
them.

David

---

<div class="post-metadata">

**Author:** ![casper\_the\_ghost](https://www.ruby-forum.com/letter_avatar_proxy/v4/letter/c/58956e/32.png) [@casper\_the\_ghost](https://www.ruby-forum.com/u/casper_the_ghost)\
**Post date:** [October 28, 2006, 6:11pm UTC](https://www.ruby-forum.com/t/class-attribute-unique-to-subclasses-but-used-in-parent-clas/73961/3 "2006-10-28T18:11:35Z")

</div>

removed\_email\_address@domain.invalid wrote:

> What I’ll like, is that Child1 and Child2 have their own values of  
> @@test, but @@test should be initialized and used in Parent.  
> Also, @@test cannot be an instance variable because it is heavy to  
> initialize and will be common to all instances of Child1 or Child2.

Class variables and class instance variables sound way too much like  
global state coming into the system through a backdoor, I’d avoid both.

Personally, I’d replace lookup through a class variable or class  
instance variable with injecting the shared object into instances. The  
less shared state, the more flexibility you gain in being able to rewire  
your objects.

Disclaimer: I am biased in favour of IoC and DI based “objects floating  
in vacuum” designs, especially for scenarios that involve reusing an  
expensive-to-create component like this one. If nothing else,  
declarative dependency resolution lets you plug in testing components  
easier / clearer than imperative (even if Ruby makes the latter easier  
in that you’re allowed to molest any object in arbitrary ways.)

David V.

---

<div class="post-metadata">

**Author:** ![casper\_the\_ghost](https://www.ruby-forum.com/letter_avatar_proxy/v4/letter/c/58956e/32.png) [@casper\_the\_ghost](https://www.ruby-forum.com/u/casper_the_ghost)\
**Post date:** [October 28, 2006, 6:17pm UTC](https://www.ruby-forum.com/t/class-attribute-unique-to-subclasses-but-used-in-parent-clas/73961/4 "2006-10-28T18:17:35Z")

</div>

Hi –

On Sun, 29 Oct 2006, David V. wrote:

> Personally, I’d replace lookup through a class variable or class  
> instance variable with injecting the shared object into instances. The  
> less shared state, the more flexibility you gain in being able to rewire  
> your objects.

Instance variables don’t involve shared state, though; they’re  
strictly per object (Class object or otherwise).

David