
Ion Gazta?aga wrote:
I recently committed to develop the initial implementation (not properly tested!) to use a different container than boost::container::vector as the underlying sequence. The idea is to pass a container instead of an allocator:
flat_set<int, std::less<int>, small_vector<int> >
and
flat_set<int, std::less<int>, static_vector<int> >
should work. Or better said, they'll work properly soon. Willing to try?
Thanks Ion, I'll have a look at that. I'm now using a crude implementation with linear insert and erase on top of a static_vector. This function has dropped from ~5% of total runtime to not measurable. It would be interesting to compare sorted vs. unsorted implementations. But it will have to be with a synthetic benchmark. Regards, Phil.