On Dec 4, 2003, at 12:49 AM, Stéphane Rollandin wrote:

> I tried this in a workspace:
>
> 	| uh |
>
> 	Smalltalk recreateSpecialObjectsArray.
> 	uh _ Morph new.
> 	uh becomeSingleton.
> 	uh delegate: #extent to: World.
> 	uh extent.
>
> ... and it worked fine (returned the World extent)

I was unclear about this in my earlier email, but note that you don't 
need to explicitly send #becomeSingleton. #delegate:to: will do that 
for you.

> I also try those two:
>
> 	| uh |
>
> 	Smalltalk recreateSpecialObjectsArray.
> 	uh _ #().
> 	uh becomeSingleton.
> 	uh delegate: #size to: #(1 2 3).
> 	uh size.

Well, I've clearly got a bug when doing this for variably-sized 
classes.  What's happening is that although the anonymous subclass is 
created, sending it #new: returns an instance of the original class (ie 
Array, not our new subclass of Array).  Then the delegation is 
installed into the methodDictionary of Array, which in this case leads 
to infinite recursion.  Not good.

I'm not sure exactly what the bug is, yet, but I bet it has something 
to do with class formats, and not using ClassBuilder.  Anyone know why 
this test fails?

TestSingletons>>testVariablySizedSingleton
	|subclass|
	subclass := Array singletonSubclass.
	self assert: (subclass new: 5) class == subclass.

Behavior>>singletonSubclass
	| meta subclass |
	meta _ Metaclass new.
	meta
		superclass: self class
		methodDictionary: MethodDictionary new
		format: self class format.
	subclass _ meta new.
	subclass
		superclass: self
		methodDictionary: MethodDictionary new
		format: self format.

	^ subclass