Subclassing Class

ruby-talk
1

Ok. This is a wild idea.

I don't like the factory pattern.

We don't need no stinkin Factories. We have the Class object.

All a Factory is really Class.

However, a traditional Factory object can have...
   * State.
   * Richer behaviour than the Class class.

Ok, this gets a little hairy so lets get concrete.

class Foo
end

fo = Foo.new

What is the class of fo? Foo

What is the class of Foo? Class

Now a FooFactory object is just an object with a bit more intelligence
and state that the traditional Foo class object for making Foo
instances in a better smarter way.

So the obvious thing to do is instead of making a new class FooFactory
< Object, it should inherit from Class.

Well, since Foo is already a Class, we really just want a smarter Class.

ie..

Instead of

class FooFactory
   def initialize( foo_state)
    @foo_state = foo_state
   end

   def foo_fu( bah)
     # Knowledge of Foo making
   end

   def create( needs_fu, stuff)
     Foo.new( foo_fu( needs_fu), @foo_state, stuff)
   end
end

factory = FooFactory.new( 'foo juice')
foo = factory.create( 'footile')

We really want

class Foo < Class
   def initialize( foo_state)
    @foo_state = foo_state
   end

   def foo_fu( bah)
     # Knowledge of Foo making
   end

   def new( needs_fu, stuff)
     super( foo_fu( needs_fu), @foo_state, stuff)
   end
end

foo = Foo.new( 'fooless', 'stuffed')

Except ruby doesn't like that
    class Foo < Class
bit.

It gets quite grumpy in fact.

Ruby says "can't make subclass of Class (TypeError)"

Rats.

We can however say...

FooClass = Class.new(Class)

Unfortunately we can't then say....

FooClass.new( Object)

Ruby says...
:in `new': wrong instance allocation (TypeError)

Ahh! I'm just going around in circles fighting Ruby.

We need a way of saying...

class FooClass < Class
   # Add state and behaviour to Class
end

And then instead of saying...
   class Foo < FooParent
     # Add state and behaviour to FooParent
   end
we need to be able to say...

FooClass Foo < FooParent
   # Add state and behaviour to FooParent
end

Aargh! I'm chasing my tail.

Is there anyway, (short of the traditional Factory Pattern) for me to catch my tail?

John Carter Phone : (64)(3) 358 6639
Tait Electronics Fax : (64)(3) 359 4632
PO Box 1645 Christchurch Email : john.carter@tait.co.nz
New Zealand

Carter's Clarification of Murphy's Law.

"Things only ever go right so that they may go more spectacularly wrong later."

From this principle, all of life and physics may be deduced.

2

We really want

class Foo < Class
   def initialize( foo_state)
    @foo_state = foo_state
   end

   def foo_fu( bah)
     # Knowledge of Foo making
   end

   def new( needs_fu, stuff)
     super( foo_fu( needs_fu), @foo_state, stuff)
   end
end

foo = Foo.new( 'fooless', 'stuffed')

Except ruby doesn't like that
    class Foo < Class
bit.

It gets quite grumpy in fact.

Ruby says "can't make subclass of Class (TypeError)"

Yes, there is a way: Singleton classes or "eigenclasses":

class Foo
  def self.new(needs_fu, stuff) # Could be Foo.new, too, but self keeps
us warm and DRY.
    super(foo_fu(needs_fu), @foo_state, stuff)
  end
end

Each Class (and indeed, every other object) has its own private class,
for just that object alone, the eigenclass. It can have methods added to
with the def instance.method syntax -- self.new, in this case, or
Foo.new would work just as well -- and so, Foo.new is different than
another class's new method. Since it's a "virtual" child class of Class,
it works as you'd expect with regards to super.

Aria

3

Why are you trying so hard?

class Foo
    # state and behavior for Foo the Class
    # using class instance variables and singleton methods
   # eg:
   @foo_state = foo_state # this works believe it or not
   # etc.
@old_new = method(:new)
     def self.new(needs_fi, stuff)
            @old_new.call(foo_fu(needs_fu, @foo_state, stuff)
     end
end

What does subclassing class get you over this approach? Your class can be your factory (in ruby anyway).

···

On Dec 5, 2005, at 7:57 PM, John Carter wrote:

Ok. This is a wild idea.

I don't like the factory pattern.

We don't need no stinkin Factories. We have the Class object.

All a Factory is really Class.

However, a traditional Factory object can have...
  * State.
  * Richer behaviour than the Class class.

Ok, this gets a little hairy so lets get concrete.

class Foo
end

fo = Foo.new

What is the class of fo? Foo

What is the class of Foo? Class

Now a FooFactory object is just an object with a bit more intelligence
and state that the traditional Foo class object for making Foo
instances in a better smarter way.

So the obvious thing to do is instead of making a new class FooFactory
< Object, it should inherit from Class.

Well, since Foo is already a Class, we really just want a smarter Class.

ie..

Instead of

class FooFactory
  def initialize( foo_state)
   @foo_state = foo_state
  end

  def foo_fu( bah)
    # Knowledge of Foo making
  end

  def create( needs_fu, stuff)
    Foo.new( foo_fu( needs_fu), @foo_state, stuff)
  end
end

factory = FooFactory.new( 'foo juice')
foo = factory.create( 'footile')

We really want

class Foo < Class
  def initialize( foo_state)
   @foo_state = foo_state
  end

  def foo_fu( bah)
    # Knowledge of Foo making
  end

  def new( needs_fu, stuff)
    super( foo_fu( needs_fu), @foo_state, stuff)
  end
end

foo = Foo.new( 'fooless', 'stuffed')

Except ruby doesn't like that
   class Foo < Class
bit.

It gets quite grumpy in fact.

Ruby says "can't make subclass of Class (TypeError)"

Rats.

We can however say...

FooClass = Class.new(Class)

Unfortunately we can't then say....

FooClass.new( Object)

Ruby says...
:in `new': wrong instance allocation (TypeError)

Ahh! I'm just going around in circles fighting Ruby.

We need a way of saying...

class FooClass < Class
  # Add state and behaviour to Class
end

And then instead of saying...
  class Foo < FooParent
    # Add state and behaviour to FooParent
  end
we need to be able to say...

FooClass Foo < FooParent
  # Add state and behaviour to FooParent
end

Aargh! I'm chasing my tail.

Is there anyway, (short of the traditional Factory Pattern) for me to catch my tail?

John Carter Phone : (64)(3) 358 6639
Tait Electronics Fax : (64)(3) 359 4632
PO Box 1645 Christchurch Email : john.carter@tait.co.nz
New Zealand

Carter's Clarification of Murphy's Law.

"Things only ever go right so that they may go more spectacularly wrong later."

From this principle, all of life and physics may be deduced.

4

John Carter ha scritto:

Ok. This is a wild idea.

I don't like the factory pattern.

We don't need no stinkin Factories. We have the Class object.

given that people already told you about overriding Class#new, I won't replicate.
But I wonder what is the problem you're tying to solve, if you just want to add state and behaviour to an object, maybe you could just include/extend a module?

5

Hi --

Each Class (and indeed, every other object) has its own private class,
for just that object alone, the eigenclass. It can have methods added to

                                  ^^^^^^^^^^

Is there any chance of stemming the tide of nicknames and coined synonyms
for singleton classes? Or may I at least put in a plea for flagging them
as nicknames and coinages? I know there's a kind of ongoing discussion
about what the best eventual official name might be, if "singleton class"
is ever replaced, but I think that having a lot of unofficial/unflagged
names in circulation is potentially very confusing and therefore probably
not an ideal interim measure.

David

···

On Tue, 6 Dec 2005, Aredridel wrote:
__
David A. Black
dblack@wobblini.net

"Ruby for Rails", forthcoming from Manning Publications, April 2006!

6

Yes, there is a way: Singleton classes or "eigenclasses".

In this context it is generally called the metaclass. But I'm really
digging this new "adhoc". As in, "He completely rewrote the method
_adhoc_."

BTW, factories come into play when the class of the object instantiated
is unknown. I don't see that. It looks like you (John) are talking more
about alternate instantation signitures --but maybe I've overlooked
something in what you said. Nonetheless, it is true that a class method
(class methods are adhoc) can be used in either case.

T.

7

Add state and polymorphic behaviour, not to the object, but the object's class. And that is where the wheels fall off.

@@variables give classes state.

ClassName.method_name functions given classes behaviour.

But it is all static, no polymorphism.

(Perhaps I should have called it the factory method pattern. When I tried to find a definitive answer on that I find I'm certainly not alone in stirring that particular pot of confusion...)

John Carter Phone : (64)(3) 358 6639
Tait Electronics Fax : (64)(3) 359 4632
PO Box 1645 Christchurch Email : john.carter@tait.co.nz
New Zealand

Carter's Clarification of Murphy's Law.

"Things only ever go right so that they may go more spectacularly wrong later."

From this principle, all of life and physics may be deduced.

···

On Tue, 6 Dec 2005, gabriele renzi wrote:

John Carter ha scritto:

Ok. This is a wild idea.

I don't like the factory pattern.

We don't need no stinkin Factories. We have the Class object.

But I wonder what is the problem you're tying to solve, if you just want to add state and behaviour to an object, maybe you could just include/extend a module?

8

Err. Right.

On to the next step. Let's rename Foo to AbstractFoo

Now I wanted
   class AbstractFoo < Class
   end

and you have given me that. But that was merely the first step in my agenda.

I actually also want....
   class ConcreteFooClass < AbstractFooClass
   end

   class OtherConcreteFooClass < AbstractFooClass
   end
that polymorphically inherits behaviour from AbstractFooClass.

I'm starting to really hate the Builder, Factory, AbstractFactory, FactoryMethod names. They are not explicit enough to distinguish which is what from the name, and the web seems to pretty confused on the subject as well. Anyhoo, factory method pattern is what trying to do.

John Carter Phone : (64)(3) 358 6639
Tait Electronics Fax : (64)(3) 359 4632
PO Box 1645 Christchurch Email : john.carter@tait.co.nz
New Zealand

Carter's Clarification of Murphy's Law.

"Things only ever go right so that they may go more spectacularly wrong later."

From this principle, all of life and physics may be deduced.

···

On Tue, 6 Dec 2005, Aredridel wrote:

class Foo
def self.new(needs_fu, stuff) # Could be Foo.new, too, but self keeps
us warm and DRY.
   super(foo_fu(needs_fu), @foo_state, stuff)
end
end

Each Class (and indeed, every other object) has its own private class,
for just that object alone, the eigenclass. It can have methods added to
with the def instance.method syntax -- self.new, in this case, or
Foo.new would work just as well -- and so, Foo.new is different than
another class's new method. Since it's a "virtual" child class of Class,
it works as you'd expect with regards to super.

9

Hmm. It's perfectly standard mathematical terminology for an operator A, if you can find a limited subset of items e_i such that A operating on item e_i results in the same item e_i (upto a scalar factor), then the e_i are called an eigenstate of the operator A and the scalar factor is the eigenvalue.

This usage of the term "eigen" is far far wider than ruby and far far older.

If this usage maps in some fairly clear manner onto singleton classes then clearly the term singleton class should be dropped in favour of eigenclass.

The question then is "Is there an operation which leaves the singleton classes, and only the singleton classes, unaltered (except perhaps in some trivial fashion)?"

John Carter Phone : (64)(3) 358 6639
Tait Electronics Fax : (64)(3) 359 4632
PO Box 1645 Christchurch Email : john.carter@tait.co.nz
New Zealand

Carter's Clarification of Murphy's Law.

"Things only ever go right so that they may go more spectacularly wrong later."

From this principle, all of life and physics may be deduced.

···

On Tue, 6 Dec 2005, David A. Black wrote:

On Tue, 6 Dec 2005, Aredridel wrote:

Each Class (and indeed, every other object) has its own private class,
for just that object alone, the eigenclass. It can have methods added to

                                 ^^^^^^^^^^

Is there any chance of stemming the tide of nicknames and coined synonyms
for singleton classes?

10

Hi John,

I'm starting to really hate the Builder, Factory, AbstractFactory,
FactoryMethod names. They are not explicit enough to distinguish which is
what from the name

I agree with you. I once heard Eric Evans (author of the _best_
technical book I've ever read, "Domain Driven Design") remark that he
thought the GoF "Factory Method" should be called "Polymorphic
Factory" and "Abstract Factory" should be "Family Factory". Eric's
names make a lot of sense to me.

Wayne Vucenic
No Bugs Software
"Ruby and C++ Agile Contract Programming in Silicon Valley"

11

John Carter schrieb:

On to the next step. Let's rename Foo to AbstractFoo

Now I wanted
  class AbstractFoo < Class
  end

and you have given me that. But that was merely the first step in my agenda.

I actually also want....
  class ConcreteFooClass < AbstractFooClass
  end

  class OtherConcreteFooClass < AbstractFooClass
  end
that polymorphically inherits behaviour from AbstractFooClass.

John, I'm not sure what you really need, but I think that the singleton class approach that others have suggested gives you the desired behaviour:

   class Foo
     def self.new fu, stuff
       super "#{fu} and #{stuff}"
     end
     def initialize things
       puts "new Foo with #{things}"
     end
   end

   Foo.new "abstract fu", "abstract stuff"
   # => new Foo with abstract fu and abstract stuff

   class ConcreteFoo < Foo
     def self.new fu
       super fu, "concrete stuff"
     end
     def initialize things
       print "ConcreteFoo: "
       super
     end
   end

   ConcreteFoo.new "concrete fu"
   # => ConcreteFoo: new Foo with concrete fu and concrete stuff

The class ConcreteFoo inherits behaviour from the class Foo and ConcreteFoo's singleton class inherits from Foo's singleton class.

I don't know whether this still works in Ruby 1.9, though.

Regards,
Pit

12

Hi --

>
>> Each Class (and indeed, every other object) has its own private class,
>> for just that object alone, the eigenclass. It can have methods added to
> ^^^^^^^^^^
>
> Is there any chance of stemming the tide of nicknames and coined synonyms
> for singleton classes?

Hmm. It's perfectly standard mathematical terminology for an operator A,
if you can find a limited subset of items e_i such that A operating on
item e_i results in the same item e_i (upto a scalar factor), then
the e_i are called an eigenstate of the operator A and the scalar factor
is the eigenvalue.

This usage of the term "eigen" is far far wider than ruby and far far
older.

So are thousands of other usages of "eigen" :slight_smile: But I'm not arguing for
or against 'eigenclass'. I'm just suggesting that we not enter an era --
or, rather that we leave the era we appear to have entered -- of using a
variety of unofficial names for singleton classes as if they were the
official name.

> If this usage maps in some fairly clear manner onto singleton classes then
clearly the term singleton class should be dropped in favour of
eigenclass.

The question then is "Is there an operation which leaves the singleton
classes, and only the singleton classes, unaltered (except perhaps in some
trivial fashion)?"

"Eigen" is actually a much more common and adaptable term than that.
It's basically German for "own", and appears all over the place in
compounds and as a prefix (Eigentum is property (stuff you own),
Eigenschaft is a property (in the sense of a characteristic), etc.).

No particular antecedent coinage has a claim on Ruby's loyalty. If it's a
good replacement for "singleton class", it's because of what "eigen" +
"class" suggests in its own right.

David

···

On Tue, 6 Dec 2005, John Carter wrote:

On Tue, 6 Dec 2005, David A. Black wrote:
> On Tue, 6 Dec 2005, Aredridel wrote:

__
David A. Black
dblack@wobblini.net

"Ruby for Rails", forthcoming from Manning Publications, April 2006!

13

It seems to (as of a few days ago at least):

   [rosco@jukebox ruby]$ ruby oneninetest.rb
   new Foo with abstract fu and abstract stuff
   ConcreteFoo: new Foo with concrete fu and concrete stuff

   [rosco@jukebox ruby]$ ruby9 oneninetest.rb
   new Foo with abstract fu and abstract stuff
   ConcreteFoo: new Foo with concrete fu and concrete stuff

···

On Tue, 06 Dec 2005 08:30:43 -0000, Pit Capitain <pit@capitain.de> wrote:

The class ConcreteFoo inherits behaviour from the class Foo and ConcreteFoo's singleton class inherits from Foo's singleton class.

I don't know whether this still works in Ruby 1.9, though.

--
Ross Bamford - rosco@roscopeco.remove.co.uk
"\e[1;31mL"

14

The mathematical usage draws on the German word for "own", as does
"eigenclass". It's a lot less ambiguous a term than "singleton class",
and the fact that it's a pleasing and apt term is indicated by the
number of rubyists who implicitly vote by using it. I don't think any
other term has caught on as well since people started expressing
dissatisfaction with the ambiguous nature of "singleton class" and
"metaclass".

martin

···

John Carter <john.carter@tait.co.nz> wrote:

Hmm. It's perfectly standard mathematical terminology for an operator A,
if you can find a limited subset of items e_i such that A operating on
item e_i results in the same item e_i (upto a scalar factor), then
the e_i are called an eigenstate of the operator A and the scalar factor
is the eigenvalue.

This usage of the term "eigen" is far far wider than ruby and far far
older.

If this usage maps in some fairly clear manner onto singleton classes then
clearly the term singleton class should be dropped in favour of
eigenclass.

The question then is "Is there an operation which leaves the singleton
classes, and only the singleton classes, unaltered (except perhaps in some
trivial fashion)?"

15

Absolutely. Which I would suggest is 'very little', if only because there are so many and varied (and often technical) uses already for 'eigen'. 'Singleton class' has a kind-of-correctness about it I think which puts the newbie on the right track. I'm not advocating sticking with it, but it stripped away some mystique for me early on, and that kind of hint is important...

Also strongly agree with your observations about the the current situation - it was most confusing when I started out, and I know that a few others find the changeable terminology confusing. I believe that one official bad name has to be better than N unofficial ones regardless of quality or correctness.

···

On Tue, 06 Dec 2005 03:36:21 -0000, David A. Black <dblack@wobblini.net> wrote:

No particular antecedent coinage has a claim on Ruby's loyalty. If it's a
good replacement for "singleton class", it's because of what "eigen" +
"class" suggests in its own right.

--
Ross Bamford - rosco@roscopeco.remove.co.uk
"\e[1;31mL"

16

Ross Bamford schrieb:

···

On Tue, 06 Dec 2005 08:30:43 -0000, Pit Capitain <pit@capitain.de> wrote:

The class ConcreteFoo inherits behaviour from the class Foo and ConcreteFoo's singleton class inherits from Foo's singleton class.

I don't know whether this still works in Ruby 1.9, though.

It seems to (as of a few days ago at least):

  [rosco@jukebox ruby]$ ruby oneninetest.rb
  new Foo with abstract fu and abstract stuff
  ConcreteFoo: new Foo with concrete fu and concrete stuff

  [rosco@jukebox ruby]$ ruby9 oneninetest.rb
  new Foo with abstract fu and abstract stuff
  ConcreteFoo: new Foo with concrete fu and concrete stuff

Thanks for the info, Ross. I remember vaguely that something about the inheritance of classes' singleton classes should change in Ruby 2, but I'm not sure.

Regards,
Pit

17

Hi --

> Hmm. It's perfectly standard mathematical terminology for an operator A,
> if you can find a limited subset of items e_i such that A operating on
> item e_i results in the same item e_i (upto a scalar factor), then
> the e_i are called an eigenstate of the operator A and the scalar factor
> is the eigenvalue.
>
> This usage of the term "eigen" is far far wider than ruby and far far
> older.
>
> If this usage maps in some fairly clear manner onto singleton classes then
> clearly the term singleton class should be dropped in favour of
> eigenclass.
>
> The question then is "Is there an operation which leaves the singleton
> classes, and only the singleton classes, unaltered (except perhaps in some
> trivial fashion)?"

The mathematical usage draws on the German word for "own", as does
"eigenclass". It's a lot less ambiguous a term than "singleton class",
and the fact that it's a pleasing and apt term is indicated by the
number of rubyists who implicitly vote by using it.

A lot of the usage I've seen is from newcomers who see the term used by
non-newcomers and have no reason (because the situation is not explained
to them) to believe that it's anything but the standard, official term.

I don't think any other term has caught on as well since people started
expressing dissatisfaction with the ambiguous nature of "singleton
class" and "metaclass".

That may be due to restraint on the part of people who prefer other terms
than "eigenclass" but don't think there should be a
"vote"-by-pre-emptive-usage contest in this area. I suppose I could start
telling people that there's a thing called the "own class" in Ruby (a term
I've suggested) but I've chosen not to.

David

···

On Tue, 6 Dec 2005, Martin DeMello wrote:

John Carter <john.carter@tait.co.nz> wrote:

__
David A. Black
dblack@wobblini.net

"Ruby for Rails", forthcoming from Manning Publications, April 2006!

18

Isn't that pretty much how all new jargon comes into being, by being used?
Even in academic papers you'll see words and phrases coined without a vote
and if the semantics appeal to the community they will be referenced and
reused in another paper until they become part of the standard nomenclature.

I do understand that new terms can be confusing and, unlike a refereed paper,
most posters don't have footnotes to define all their terms, but new terms
will continue to emerge (e.g., Ajax) spontaneously.

The Ruby community is very fortunate to have Matz to sift through all the
suggestions, RCRs, code, and discussion and provide a certain amount of
guidance, but that doesn't preclude other sources of inspiration (I hope).

The big elephant in the room seems to be Matz comments that the semantics
of a meta/singleton/eigen class might be implemented via some mechanism
other than a class, which would certainly stir the pot a bit.

···

On Dec 6, 2005, at 10:11 AM, David A. Black wrote:

That may be due to restraint on the part of people who prefer other terms
than "eigenclass" but don't think there should be a
"vote"-by-pre-emptive-usage contest in this area.

19

Hi --

> That may be due to restraint on the part of people who prefer other
> terms
> than "eigenclass" but don't think there should be a
> "vote"-by-pre-emptive-usage contest in this area.

Isn't that pretty much how all new jargon comes into being, by being
used? Even in academic papers you'll see words and phrases coined
without a vote and if the semantics appeal to the community they will be
referenced and reused in another paper until they become part of the
standard nomenclature.

I know -- I've done it myself. I've never hesitated to use neologisms in
my own academic work, and if people find them helpful, they can use them.
If not, they're just something I came up with to help explain some
concept.

This is different. First of all, Ruby is a work by Matz, not any of us.
But, granting that single authorship is not really the whole story here,
there's no vacuum for "eigen/virtual/meta/single/adhoc"-class to fill
anyway. There's some fragility in the existing terms, but they do exist,
and the fragility is being examined carefully by Matz with community
input. What is the benefit of short-circuiting that process?

And let's say your model does apply. That means -- as in common speech
and academic discourse -- that different groups of people will use
different jargon for the same things. We'll have the eigenclass people,
the metaclass people, and so on. And newcomers will have to learn about
all of them, so that they can talk to everyone.

I think, among other things, that there's a limit to how tolerant people
are going to be of being expected to wallow in all this meta-meta-culture.
It's certainly not what attracted me to Ruby.

I do understand that new terms can be confusing and, unlike a refereed
paper, most posters don't have footnotes to define all their terms, but
new terms will continue to emerge (e.g., Ajax) spontaneously.

The Ruby community is very fortunate to have Matz to sift through all
the suggestions, RCRs, code, and discussion and provide a certain amount
of guidance, but that doesn't preclude other sources of inspiration (I
hope).

How about channeling these inspirations *through* the RCRs and
discussions? I just can't see how throwing arbitrarily many home-made
synonyms for language-level constructs at newcomers can be of the
slightest benefit. In the short term I suppose it's exhilerating and
makes people feel creative and participatory. But that's

The big elephant in the room seems to be Matz comments that the
semantics of a meta/singleton/eigen class might be implemented via some
mechanism other than a class, which would certainly stir the pot a bit.

No elephant, from my perspective. That's yet another reason (aside from
the general principle of respect for Matz's unique role) for not having
the terminology free-for-all.

David

···

On Wed, 7 Dec 2005 gwtmp01@mac.com wrote:

On Dec 6, 2005, at 10:11 AM, David A. Black wrote:

__
David A. Black
dblack@wobblini.net

"Ruby for Rails", forthcoming from Manning Publications, April 2006!

20

...all he wrote? :slight_smile:

Oh, you know. Short-term, or some synonym for it :slight_smile:

David

···

On Wed, 7 Dec 2005, David A. Black wrote:

How about channeling these inspirations *through* the RCRs and
discussions? I just can't see how throwing arbitrarily many home-made
synonyms for language-level constructs at newcomers can be of the
slightest benefit. In the short term I suppose it's exhilerating and
makes people feel creative and participatory. But that's

__
David A. Black
dblack@wobblini.net

"Ruby for Rails", forthcoming from Manning Publications, April 2006!