On 10/5/05, Arkadiy Vertleyb <vertleyb@hotmail.com> wrote:
Well, RTL addresses (or delays) the issue of persistense by separating table implementation from the rest of the library. We currently give the user a choise between an std::vector - based implementation and Boost.Multi-index- based implementation. A disk-based implementation can also be developed. (alternatively, someone may comeup with a persistent std::vector :-)
I need to learn more about RTL. I think the combination of RTL + RML is definetely needed as a boost library. At least myself can't wait for that library ! The question is who provides the persistent STL containers, maybe should be the Boost.Multi-Index as this is the higher level container. There is a bit of overlap: RTL --> uses Boost.Multi-Index - no persistent container RML --> own multi-index container and own persistent container library via mmap files shmem --> persistent versions of STL containers like std::map (via mmap files) Boost.Multiindex --> no persistent version I liked RML because is further along the idea of being a replacement for a traditional database. What I think the final library needs is: - combine RTL and RML in a single syntax (as friendly as possible) - use Boost.Multi-Index (assuming it provides a persistent container) or use persist or shmem persistent container