Proposal: Improving <algorithm> performance with containers of containers

[Paolo Carlini]

Hi Chris,

> I'd like to try to get this into 4.1 / so_7, and already have quite a 
> bit of the implementation of all 4 of these, so hopefully that should 
> be possible.

I would say: go ahead with a prototype of the approach that you favor. 
In my understanding, it should be possible to prepare a *small* 
prototype, say restricted to std::vector, and maybe not speeding up all 
the candidate operations, which would concretely explain your approach. 
As you may have noticed already ;) in this list we all like seeing 
actual code, more than "philosophical" debates, as a general rule. Agreed?

Thanks in advance,
Paolo.

P.S. I also look forward to preliminarily complete the clean-up wrt 
predicates... Do you want me to add testcases to your last (slighlty 
amended) patch and close the issue??