On Tue, 12 Aug 2008 15:10:57 +0200, Jason Johnson wrote:

> On Tue, Aug 12, 2008 at 1:32 PM, Klaus D. Witzel  
> <klaus.witzel at cobss.com> wrote:
>> On Tue, 12 Aug 2008 09:28:22 +0200, Andreas Raab wrote:
>>
>>> Klaus D. Witzel wrote:
>>
>> [and I would object accessing inexisting elements for forcing an Error  
>> :( ]
>
> It's pretty consistent for the default case to signal an error

This looks perhaps good but the problem here is that #anyOne causes an  
indexing Error which n00bs and casual users will not immediately associate  
with the cause (and it is raised by a "remote" method which knows nothing  
about #sum). Users expect that Smalltalk's own collection protocols is  
resistant against indexing errors as usual, if you know what I mean ;)

Therefore I'm in favor of Andreas' #ifNil:, not only because of its  
elegance :)

And the essence of (x := y sum ifNil: [whatMyZeroIs]) is that the zero  
element is defined by the user, not the implementor of #sum ... (and  
#ifNil: can be omitted if you know your collections).

Regarding emptyCheck against #ifNil, your previous Q: during the loop, the  
former is a full send (with even more own sends) but the latter is  
inlined. For small collections one wouldn't notice a significant  
difference. And if collections are soo huuge (Smalltalk's huge ;) then  
people often implement/use the #fast* version of things.

> and
> provide an ifNone:/ifEmpty/etc. protocol for people who want other
> behavior, no?  Of course unary messages participating in this protocol
> do look funny. :)

Sure, alternatives are appreciated.