id,summary,reporter,owner,description,type,status,milestone,component,version,severity,resolution,keywords,cc
6324,Flyweight performance improvement,Koh Ohnishi <k_onishi@…>,Joaquín M López Muñoz,"Hi,
I am using flyweight for storing char_string, and when I do an erase one of the flyweight entry when there are a lot of entries, I am facing the performance issue, like ticket #4264 mentioned, in multi_index hashed_unique erase in boost/flyweight/hashed_factory.hpp hashed_factory_class::erase(handle_type) .

There are two reasons.

1. hashed_factory_class::erase should call erase with key, like
cont.erase(cont.key_extractor()(*(cont.iterator_to(*h))));
instead of calling
cont.erase(cont.iterator_to(*h));
as Change History of #4264 mentions.

2. boost::multi_index hashed_unique erase takes linear time against the bucketsize, even if you use erase method with key, as ticket #4264's Change History mentioned. It calculates begin() and do linear search the top of the node when erasing some node.

These two behavior cause significant performance penalty when a lot of entries are stored in flyweight.

I think most of users of multi_index hashed_unique are expecting not O(1) begin(), but O(1) erase(key). I would like the erase method of hashed_unique should perform O(1).

Thanks,
",Bugs,closed,To Be Determined,multi_index,Boost 1.48.0,Problem,fixed,"flyweight, multi_index, hashed_unique, performance",