On Thu, 26 Jun 2008 00:18:08 +0200, Nicolas wrote:

> Klaus D. Witzel a écrit :
>> On Wed, 25 Jun 2008 22:37:52 +0200, Damien Cassou wrote:
>>
>>> On Wed, Jun 25, 2008 at 4:59 PM, Klaus D. Witzel
>> ...
>>>>>> But what do *you* expect to happen with your original expression,  
>>>>>> mind to share?
>>>>>
>>>>> A collection of boleans.
>>>>
>>>> Well, let's look at #collect: it has a block as argument. Because of  
>>>> this
>>>> object orientedness, the implementor of #collect: cannot know what  
>>>> sort of
>>>> values the block will return in the future, no?
>>>>
>>>> How about this, let #collect: always collect into  
>>>> anOrderedCollection, then it can eventually
>>>>
>>>>  ^ anOrderedCollection coercedWith: self
>>>
>>> What does #coercedWith: mean? I would agree if it is something like:
>>> "try to make a collection of the same species as the parameter with
>>> the content of the receiver and if it is not possible, return the
>>> receiver"
>>  Right. It would perhaps ask (argument #couldStore: anElement) and,  
>> ifFalse, return a #copyFrom:to: of its array (Array should be the  
>> default, IHMO). And if all pass then
>>
>
> Another pattern: rather than testing #couldStore: as a precondition  
> inside the loop, do handle the Exception outside the loop.

Sure. BTW #couldStore: can be replaced by the VM's unwillingness to store,  
say of Damien's boolean into an instance of String--and so wouldn't cost  
anything besides the handler and its work.

> But write the handler so as to not iterate twice on already iterated  
> elements (the block might have side effects).
>
>>  ^ argument species withAll: self
>>  For sure this all could also be moved right into the #collect: method,  
>> might perform a bit faster. But anyways, things seem to have such price  
>> tag.
>>
>
> Simplicity of workaround: 'abc' asArray collect: [...] will be hard to  
> beat.

Except when you don't want/can make a copy/#asArray of the source  
collection.

> It's probably not worth trying to save the cost of additional asArray  
> iteration, given the complexification and de-optimization of code  
> implied by genericity.
>
> This makes my vote go to a statu quo.

And by code duplication, #collect:into: could do the job ;)

> Nicolas
>