Re: [boost] Proposal: Monotonic Containers

Subject: Re: [boost] Proposal: Monotonic Containers
From: David Bergman (David.Bergman_at_[hidden])
Date: 2009-06-10 17:46:36

On Jun 10, 2009, at 5:37 PM, Thorsten Ottosen wrote:

> Locality of reference seems like an important problem to improve. I  
> don't agree with those that say high-performance applications  
> usually involves a single container; mine certainly doesn't.

Actually, I have been thinking a bit about it after I wrote it - I am  
afraid I am that "they" here ;-) - and, you are right, most high- 
performance applications I write involve more than one container, just  
that one is pivotal and the typical bottleneck, but that fact is  
ignoring the "cache defocusing" of moving to that "secondary" container.

Anyway, my other comments still hold.

This locality across a few containers is something I get equally well,  
if not better, from a small heap than this "monotonic" container  
proposal.

> Intrusive containers seems to give similar benefits w.r.t. locality  
> of reference. Similar for flat_set/flat_map.

Yes

> It is not clear to me in which niche your library would fit. I hope  
> you
> can make that clear, and provide examples that show your library  
> performs better than otherwise.

I am curious as to how his ideas can affect the future of your  
auto_buffer, and perhaps creating an "auto_map" ;-)

/David