On 5/5/2015 7:49 PM, Joaquin M Lopez Munoz wrote:

TONGARI J writes:

...

If the data sequence shows some affinity, I guess an ordered tuple_vector-based sequence can provide some performance benefit in traversal (with some crafted for_each method).

That is, a sequence of [AAABBB] may be traversed faster than [ABABAB] for such a container, and I believe in the later case such that the types are uniformly distributed, vector<variant> will perform better. I wrote something on this some time ago, maybe worth having a look at:

http://bannalia.blogspot.com/2014/05/fast-polymorphic-collections.html

http://bannalia.blogspot.com/2014/05/fast-polymorphic-collections-with.html

Nice analysis. If we can identify different use cases, we may be able to come up with several implementations of heterogeneous containers which balance efficiency and flexibility. I think it would be an interesting project to work on and surely useful if the containers all the a similar 'feel' -- James Armstrong