Den 09-04-2011 19:06, Neil Groves skrev:
My earlier discussion of tree containers and n-dimensional containers was meant to invoke thoughts of containers not implemented by the STL that could be leverage, at least some, generic algorithms. From my own experience in high performance mathematical computing, and from discussion with colleagues in this field there are frequently special high-dimensional containers that would benefit from access to these algorithms.
Well, one thing is performance due to efficient iteration over segmented data structure. Another issue is just flattening. About segmentation. One approach is a new type of iterators with specialized algorithms for basically all algorithms. That's a lot of algorithms that has to be implemented again. So how the h*** do we avoid that? Can we avoid that? I'm thinking that some fat iterator with some internal book keeping might help. But then we have the problem of the end iterator taking up much space just because the types must match. Clearly ranges without iterators is superior in this context. -Thorsten