Isn't this mantis issue 7090?

You may also want to check how we fixed this over at Pharo (more than  
a year ago): http://code.google.com/p/pharo/issues/detail?id=13

HTH,
Adrian

On Oct 5, 2009, at 13:49 , Alexander Lazarević wrote:

> Hi!
>
> I think in a healthy squeak system
>
> (ProtoObject allSubclasses reject: [:each | each isMeta or: [each
> environment includes: each]]) size
>
> should be zero. This is true for a fresh trunk image. If you evaluate
> above expression after running ClassTraitTest the returned number is 9
> and increases by 9 every time you run the test. As an effect of this
> in a browser you can't switch to the class side of ProtoObject,
> because this produces an exception. The exception gets raised because
> some of ProtoObjects subclasses are not known in the environment
> Smalltalk.
>
> An example why this happens is here:
>
> setUpTrivialRequiresFixture
> 	self c3: (self
> 				createClassNamed: #C3
> 				superclass: ProtoObject
> 				uses: { }).
> 	self c3 superclass: nil.
> 	self c3 compile: 'foo ^self bla' classified: #accessing
>
> C3 is created as a subclass of ProtoObject, but then the superclass
> pointer of C3 is twisted to nil . I haven't found a place where this
> would be fixed by the system again. And when C3 is removed from the
> system it is still in ProtoObjects subclasses.
>
> The tests in ClassTraitTest return all green status and don't check if
> the classes are still broken.
>
> Should the test be removed or is there just something (I'm) missing?
> Anyone got more infos on the whys and whats?
>
> Alex
>