On 8/11/2011 5:19 PM, Joel de Guzman wrote:
And, looking at the code, the implementation looks very sumptuous. I get indigestion with the old code :P
No kidding! That code was horrible. This code is still complex, but at least this time around I documented everything with pseudocode. It has helped immensely.
(Aside: we should talk about naming ext_ into something more obvious. Before it was a hidden curiosity. Now it seems its ready to take front stage).
That's what I referred to by "baking it in". I imagine that all the ext_ directories go away, and segmentation support just becomes a native part of Fusion. These things would need to change: - We find a new home for segmented_iterator.hpp et. al. - Sequence intrinsics (begin, end, size, empty) have meaningful default implementations for segmented data structures. - As many algorithms as possible should dispatch to segmented flavors when appropriate. E.g., ext_/fold_s.hpp just gets merged into fold.hpp. - Fusion containers and views that are segmented (joint_view) advertize themselves as such, and the rest of their implementation goes away. -- Eric Niebler BoostPro Computing http://www.boostpro.com