[Cuis-dev] [DEFECT] #copyFrom:count: for OrderedCollections

Nicolas Cellier nicolas.cellier.aka.nice at gmail.com
Tue Feb 6 13:24:36 PST 2024


I don't buy such a definition of polymorphism, because if it is that
restrictive, then it does not exist at all.
Beside, it seems that the polymorphism problem just moved:

    (Dictionary newWithRoomForMoreThan: 10) at: 10 put: self.
    (OrderedCollection newWithRoomForMoreThan: 10) at: 10 put: self.

Ah, but we can create two different messages as well...
Or maybe at:put: is false polymorphism?
Or isn't it rather Dictionary and OrderedCollection that are not polymorphic?

If I follow the same path:

    (Array new: 10) add: 5.
    (OrderedCollection new: 10) add: 5.

#new: is not polymorphic because in one case it answers a collection
that cannot grow, and in the other case it answers a collection that
can grow...

OrderedCollection and Array are not polymorphic, they don't behave the
same way, Interval neither, RunArray neither etc...
It's not just a different implementation of the same thing. Otherwise,
no need for so many subclasses.

All is in the contract that you expect from #new:
You are free to change the semantics of #new: but what makes such a
major change really necessary?
false polymorphism, really?

Nicolas

Le mar. 6 févr. 2024 à 20:54, Hernán Wilkinson via Cuis-dev
<cuis-dev at lists.cuis.st> a écrit :
>
> The definition of polymorphism, sadly, is not clear... Each language/book/culture defines it differently and uses that word differently and sometimes in contradictory ways.
> The definition I like to use after many years of teaching the subject is: "Objects of a set are polymorphic among themselves with respect to a set of messages, if the objects of the first set respond semantically the same to the messages of the second".
> "Semantically the same" means "they do the same thing", no matter how they do it (it implies that parameters must be polymorphic and that the results are polymorphic)
>
> If #new: n sent to Array creates a collection of n elements but sent to OrderedCollection does not, then Array and OrderedCollection are not polymorphic with respect to #new:, and therefore you cannot replace one receiver with another.
>
> A clear example of Array and OrderedCollection not being polymorphic regarding #new: is:
> (Array new: 10) at: 5 --> returns nil.
> (OrderedCollection new: 10) at: 5 --> gives error
>
> This behavior does not follow the "minimum attonishment" principle and therefore it is good to avoid it.
>
> It is true that it generates compatibility issues, but I also think that sometimes we should break that compatibility if we believe that what we are doing is better, I do not want Smalltalk neither Cuis to be what Alan Kay said a long time ago: "Once it got into production, it stops evolving due to the compatibility issues" (or something like that).
> It is also true that breaking compatibility should be done with care, and that is why this has been done on the rolling version (6.3), not the stable one (6.2). If we do not want compatibility issues we should work on the stable release and migrate to a different stable version if we want it/need it.
>
> When Juan told me about this problem the first thing we talked about was about the compatibility problems it could generate, but the idea of doing something better won over the compatibility idea... We are working on the rolling release and if the change generates more problems than solutions, we can go back to how it was.
>
> Let's give it a try and see what happens!
>
> Cheers!
> Hernan.
>
> On Tue, Feb 6, 2024 at 4:00 PM Luciano Notarfrancesco via Cuis-dev <cuis-dev at lists.cuis.st> wrote:
>>
>> I would say not the class but the receiver is the only one with the full context, as illustrated by Text>>grownTo: (and other examples in my code)
>>
>> On Wed, Feb 7, 2024 at 01:57 Andres Valloud via Cuis-dev <cuis-dev at lists.cuis.st> wrote:
>>>
>>> Be careful with that "false polymorphism" argument.  Going down that
>>> route means you can ignore the receiver's class when you see a selector
>>> because a selector can only have one meaning.  In a sense, it promotes
>>> selectors to operators.
>>>
>>> It should be pretty clear that collections like Array and Set should
>>> behave differently.  Why is that being ignored?
>>>
>>> Consider this alternate interpretation.  The root cause of this problem
>>> seems to be that some code has the expectation that
>>>
>>>         species new: n
>>>
>>> can have exactly one meaning.  The sender is ignoring the receiver
>>> class.  Only the receiver class has that context.  So the receiver class
>>> should be given the task of creating the new instance instead --- this
>>> is why arithmetic has double dispatching, for example.
>>>
>>> Andres.
>>>
>>> On 2/6/24 8:21 AM, Juan Vuletich via Cuis-dev wrote:
>>> > Let me also elaborate a bit on the rationale.
>>> >
>>> > In Smalltalk-80 (and every other Smalltalk system since then), the docs
>>> > will say that #new: will answer a collection of the requested size. But
>>> > it is not like that for Set, Dictionary, OrderedCollection and a few
>>> > others, that give a completely different semantics to this message. It
>>> > is no _that_ bad when the message is sent to an explicit class, although
>>> > you need to be aware of this.
>>> >
>>> > The real problem is when someone does `someCollection species new:
>>> > aNumber`. It gets really tricky to find out what is going to happen.
>>> >
>>> > This is a prime example of what I call "False Polymorphism". It looks
>>> > like a polymorphic message, but it is not. It is (at least) two sets of
>>> > senders/implementors, completely separated. This means obscure,
>>> > misleading code. It makes me sick. I fix every instance of this I see.
>>> > It rarely happens in the base Smalltalk-80 classes, but it is still wrong.
>>> >
>>> > Thanks,
>>> >
>>> --
>>> Cuis-dev mailing list
>>> Cuis-dev at lists.cuis.st
>>> https://lists.cuis.st/mailman/listinfo/cuis-dev
>>
>> --
>> Cuis-dev mailing list
>> Cuis-dev at lists.cuis.st
>> https://lists.cuis.st/mailman/listinfo/cuis-dev
>
>
>
> --
> Hernán Wilkinson
> Agile Software Development, Teaching & Coaching
> Phone: +54-011-4893-2057
> Twitter: @HernanWilkinson
> site: http://www.10Pines.com
> Address: Alem 896, Floor 6, Buenos Aires, Argentina
> --
> Cuis-dev mailing list
> Cuis-dev at lists.cuis.st
> https://lists.cuis.st/mailman/listinfo/cuis-dev


More information about the Cuis-dev mailing list