2009/11/12 Bert Freudenberg <bert at freudenbergs.de>:
>
> On 12.11.2009, at 11:35, Bert Freudenberg wrote:
>
>>
>> On 12.11.2009, at 09:42, Andreas Raab wrote:
>>
>>> Levente Uzonyi wrote:
>>>> But we now have a lot more proposals, let me summarize them all:
>>>> - (1)add a new instance variable: containsNil
>>>> - (2)use tally to indicate if nil is in the set
>>>> - (a)negative values mean nil is contained by the set
>>>> - (b)floats - (3)use a marker object
>>>> - (a)self
>>>> - (b)a unique object in each set
>>>> - (c)a class variable
>>>> - (4)use container objects in occupied slots, like associations in dictionaries
>>>> I prefer 3c, 2a, 3a, 1, 4 in the given order.
>>>> What about you?
>>>
>>> Nice summary. My current preferences would be (1) (clear and obvious) (3b) (ditto) and (3a)/(3c) (both with some hesitation). I'm not sure about (4) (I've only just heard of it; it sounds cool but I haven't seen no code yet). I would veto both (2a) and (2b) as obscure hacks compared to (1) unless the size overhead is significant.
>>
>> I'd prefer (3d): make a class called UnoccupiedSlot, use it as marker (either the class itself [I'd like that] or its singleton instance [for the purists]). This would make it blatantly obvious what's happening when inspecting the innards of a Set.
>>
>> I'd reluctantly accept (1) but fear it would complicate the logic - containsNil checks will have to be scattered all over the place.
>>
>> - Bert -
>
> Thinking about migrating (possibly) old data - how do the options fare with respect to reloading serialized Sets? I guess we could count nils and detect a mismatch with tally?
>
> - Bert -
>
>

Solution 2a) is most transparent w.r.t. migration, no supplementary
ivar, old ivar interpreted the same way
Solution 1) could be if using lazy initialization of ivar containsNil
(or using containsNil == true)
Solution 3b) would be transparent too with an ivar marker being nil by
default (you could not add nil to old Set but who cares)

I think every other solution would require a little migration hack.