On 1 Sep 2014 at 15:41, Edward Diener wrote:
Do you intend this to be a separate library, or would it go into something like Boost.Container?
Not sure yet. My next goal is no-alloc promise-future, also based on spinlocks.
I don't intend to "finish" concurrent_unordered_map past what is already done, and I suspect no Boost library will accept it until I or someone else does.
What do you mean by the last part of your sentence ?
I'm not sure what is unclear in the sentence. concurrent_unordered_map is missing things like all const overloads and iterators (they are aliased to non-const, so code compiles), or C++03 support, or local iterators or container copy and move support. Noexcept is always on too instead of being conditional. But I have no personal need for all that, and I need to get on with other things, so I leave it where it is at - it is still extremely useful, the code is of a very high quality and is extremely well unit tested, it scales linearly to CPU cores and it solves the problem of concurrent unordered maps without breaking STL compatibility or semantics, so concurrent erase finally works, unlike most other implementations! I hope a GSoC student finishes it in 2015 or 2016, but for me it's enough for me to move on, so I am. You will hear what I will be doing next later today in fact. Niall -- ned Productions Limited Consulting http://www.nedproductions.biz/ http://ie.linkedin.com/in/nialldouglas/