2011/10/1 Bert Freudenberg <bert at freudenbergs.de>:
>
> On 30.09.2011, at 13:05, Nicolas Cellier wrote:
>
>> 2011/9/30 Nicolas Cellier <nicolas.cellier.aka.nice at gmail.com>:
>>> 2011/9/30 Bert Freudenberg <bert at freudenbergs.de>:
>>>>
>>>> On 29.09.2011, at 19:22, Rodney Polkinghorne wrote:
>>>>
>>>>> Dear list
>>>>>
>>>>> I wanted an array that lazily initialized elements 1 to n when you
>>>>> asked for the nth one.  So I ran something like:
>>>>>
>>>>> SequenceableCollection variableSubclass #Foo ...
>>>>>
>>>>> When I started a workspace, and tried Foo new: 10, my image hung.
>>>>>
>>>>> Eventually, I found the problem.  The workspace tried to print the
>>>>> answered instance, which resulted in it being sent a do: message to
>>>>> print its elements.  This ran the method
>>>>>
>>>>> SequenceableCollection>>do: aBlock
>>>>>       1 to: self size do:
>>>>>               [:index | aBlock value: (self at: index)]
>>>>>
>>>>> which ran
>>>>>
>>>>> Collection>>size
>>>>>       | tally |
>>>>>       tally := 0.
>>>>>       self do: [:each | tally := tally + 1].
>>>>>       ^ tally
>>>>>
>>>>> which sent do:, which sent size, which ...
>>>>>
>>>>> Is this considered a bug, or a learning experience for new players?
>>>>
>>>> The latter. You should not instantiate an abstract class. I don't think there is much that could be done about this.
>>>>
>>>> - Bert -
>>>>
>>>
>>> But Foo is a concrete class from Rodney, not Abstract.
>
> No, it's still abstract since he did not override any methods.
>
>>> Generally, if the subclass must implement a method, it is expressed
>>> with an abstract implementation in the upper level
>>>    ^self subclassResponsibility
>>>
>>> Here, the subclass must implement a message, either #do: or #size but
>>> there is no explicit guidance.
>>> Personnally I consider this as very bad kernel code if not a bug.
>>> Does any subclass really use Collection>>size ?
>>
>> OK, I just opened an image to check and only one does: LinkedList.
>> I suggest moving  Collection>>size implementation to LinkedList, and
>> replace it with a subclassResponsibility.
>>
>> Nicolas
>
> That would be make the intention more clear, agreed.
>
> - Bert -
>

Eliot objected that #do: is already abstract at Collection level, so
#size does not have to be.
An alternative would be to re-implement
Sequenceable>>size
    ^subclassResponsibility

What do others think ?

Nicolas