Remove algo logic duplication Round 3

[François Dumont]

On 09/29/2013 07:50 PM, Marc Glisse wrote:
> On Sun, 29 Sep 2013, Paolo Carlini wrote:
>
>>> Maybe we could use more references? The user-facing functions have 
>>> to take iterators by value, but I think the inline helpers could use 
>>> references.
>>
>> Marc it would be great if you could prepare a patch or provide 
>> (offline too, if you like) detailed guidance to Francois. We really 
>> want this change to be safe and nop (or very close to that) from the 
>> performance point of view. That's why I remained unconvinced and 
>> hesitant for such a long time. If we are not sure, let's just admit 
>> it, give the issue more thought and reconsider a more satisfactory 
>> patch for the next release series.
>
> Busy now, but please don't revert the patch because of my post. The 
> number of copies in the previous code was already rather large, 
> showing that we didn't care about it. The new code increases it a bit, 
> but only by a constant factor. We can improve that incrementally, I 
> believe.
>

     Considering iterator or functor types as trivial types to copy was 
one of the most important hypothesis for this patch to be valid. But I 
think it is quite safe, we do our best to have light iterator types in 
libstdc++ and any C++ user should know that such types needs to be light 
as even the Standard are taking those by copy.

     Regarding number of comparisons in sort algo I don't think that it 
is a regression of this patch right ? I only remove duplication of code, 
didn't change number of invocation of functors.

François